From 5a04cb254969af9ea17bb7ff2f764d6f78180506 Mon Sep 17 00:00:00 2001
From: sebffischer <sebf.fischer@gmail.com>
Date: Tue, 29 Oct 2024 11:39:39 +0000
Subject: [PATCH] =?UTF-8?q?Deploying=20to=20gh-pages=20from=20@=20mlr-org/?=
 =?UTF-8?q?mlr3torch@0edcf48311552395543c4e7140efb6c76f97d03b=20?=
 =?UTF-8?q?=F0=9F=9A=80?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 dev/articles/callbacks.html                   |   2 +-
 dev/articles/get_started.html                 |   2 +-
 dev/articles/internals_pipeop_torch.html      |  59 ++--
 .../figure-html/unnamed-chunk-37-1.png        | Bin 123258 -> 123484 bytes
 .../figure-html/unnamed-chunk-42-1.png        | Bin 123324 -> 123441 bytes
 .../figure-html/unnamed-chunk-46-1.png        | Bin 123585 -> 123298 bytes
 .../figure-html/unnamed-chunk-48-1.png        | Bin 122702 -> 122702 bytes
 .../figure-html/unnamed-chunk-50-1.png        | Bin 123536 -> 124094 bytes
 dev/articles/lazy_tensor.html                 |   8 +-
 dev/pkgdown.yml                               |   2 +-
 dev/reference/DataDescriptor.html             |   4 +-
 dev/reference/TorchCallback.html              |   2 +-
 dev/reference/as_data_descriptor.html         |   4 +-
 dev/reference/materialize.html                | 260 +++++++++---------
 .../mlr_pipeops_torch_ingress_ltnsr.html      |  46 ++--
 .../mlr_pipeops_torch_model_classif.html      |  26 +-
 .../mlr_pipeops_torch_model_regr.html         |  12 +-
 dev/reference/nn_graph.html                   |  32 +--
 dev/search.json                               |   2 +-
 19 files changed, 231 insertions(+), 230 deletions(-)

diff --git a/dev/articles/callbacks.html b/dev/articles/callbacks.html
index 76646ba0..c55d112d 100644
--- a/dev/articles/callbacks.html
+++ b/dev/articles/callbacks.html
@@ -225,7 +225,7 @@ <h3 id="writing-a-custom-logger">Writing a Custom Logger<a class="anchor" aria-l
 <span><span class="co">##     load_state_dict: function (state_dict) </span></span>
 <span><span class="co">##     on_before_valid: function () </span></span>
 <span><span class="co">##     on_batch_end: function () </span></span>
-<span><span class="co">##   Parent env: &lt;environment: 0x55c7a37847b0&gt;</span></span>
+<span><span class="co">##   Parent env: &lt;environment: 0x55f6ed50fb38&gt;</span></span>
 <span><span class="co">##   Locked objects: FALSE</span></span>
 <span><span class="co">##   Locked class: FALSE</span></span>
 <span><span class="co">##   Portable: TRUE</span></span></code></pre>
diff --git a/dev/articles/get_started.html b/dev/articles/get_started.html
index 1e40868c..deb9c3c5 100644
--- a/dev/articles/get_started.html
+++ b/dev/articles/get_started.html
@@ -236,7 +236,7 @@ <h3 id="loss">Loss<a class="anchor" aria-label="anchor" href="#loss"></a>
 <span><span class="co">#&gt;     clone: function (deep = FALSE, ..., replace_values = TRUE) </span></span>
 <span><span class="co">#&gt;   Private:</span></span>
 <span><span class="co">#&gt;     .__clone_r6__: function (deep = FALSE) </span></span>
-<span><span class="co">#&gt;   Parent env: &lt;environment: 0x556f0dd5ea18&gt;</span></span>
+<span><span class="co">#&gt;   Parent env: &lt;environment: 0x556c58224810&gt;</span></span>
 <span><span class="co">#&gt;   Locked objects: FALSE</span></span>
 <span><span class="co">#&gt;   Locked class: FALSE</span></span>
 <span><span class="co">#&gt;   Portable: TRUE</span></span></code></pre></div>
diff --git a/dev/articles/internals_pipeop_torch.html b/dev/articles/internals_pipeop_torch.html
index 1b41b29c..b7599b72 100644
--- a/dev/articles/internals_pipeop_torch.html
+++ b/dev/articles/internals_pipeop_torch.html
@@ -104,8 +104,8 @@ <h2 id="a-torch-primer">A <code>torch</code> Primer<a class="anchor" aria-label=
 <code class="sourceCode R"><span><span class="va">input</span> <span class="op">=</span> <span class="fu"><a href="https://rdrr.io/pkg/torch/man/torch_randn.html" class="external-link">torch_randn</a></span><span class="op">(</span><span class="fl">2</span>, <span class="fl">3</span><span class="op">)</span></span>
 <span><span class="va">input</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  1.0799 -2.3132 -0.0767</span></span>
-<span><span class="co">#&gt; -1.0029 -0.4515  0.3045</span></span>
+<span><span class="co">#&gt;  0.9197  0.6295 -0.9055</span></span>
+<span><span class="co">#&gt; -2.5884  0.7595  1.2294</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,3} ]</span></span></code></pre></div>
 <p>A <code>nn_module</code> is constructed from a
 <code>nn_module_generator</code>. <code>nn_linear</code> is one of the
@@ -117,8 +117,8 @@ <h2 id="a-torch-primer">A <code>torch</code> Primer<a class="anchor" aria-label=
 <code class="sourceCode R"><span><span class="va">output</span> <span class="op">=</span> <span class="fu">module_1</span><span class="op">(</span><span class="va">input</span><span class="op">)</span></span>
 <span><span class="va">output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.9228 -1.1292 -0.5401  0.8716</span></span>
-<span><span class="co">#&gt;  0.1488 -0.7452 -0.0094  0.2273</span></span>
+<span><span class="co">#&gt;  0.6356 -0.0022  0.3491 -1.0918</span></span>
+<span><span class="co">#&gt; -1.5718  1.8213 -1.6298  0.2341</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,4} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span></code></pre></div>
 <p>A neural network with one (4-unit) hidden layer and two outputs needs
 the following ingredients</p>
@@ -134,8 +134,8 @@ <h2 id="a-torch-primer">A <code>torch</code> Primer<a class="anchor" aria-label=
 <span><span class="va">output</span> <span class="op">=</span> <span class="fu">softmax</span><span class="op">(</span><span class="va">output</span><span class="op">)</span></span>
 <span><span class="va">output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.2143  0.2882  0.4975</span></span>
-<span><span class="co">#&gt;  0.2435  0.2812  0.4753</span></span>
+<span><span class="co">#&gt;  0.2569  0.4082  0.3350</span></span>
+<span><span class="co">#&gt;  0.3488  0.3890  0.2623</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,3} ][ grad_fn = &lt;SoftmaxBackward0&gt; ]</span></span></code></pre></div>
 <p>We will now continue with showing how such a neural network can be
 represented in <code>mlr3torch</code>.</p>
@@ -170,8 +170,8 @@ <h2 id="neural-networks-as-graphs">Neural Networks as Graphs<a class="anchor" ar
 <code class="sourceCode R"><span><span class="va">output</span> <span class="op">=</span> <span class="va">po_module_1</span><span class="op">$</span><span class="fu">train</span><span class="op">(</span><span class="fu"><a href="https://rdrr.io/r/base/list.html" class="external-link">list</a></span><span class="op">(</span><span class="va">input</span><span class="op">)</span><span class="op">)</span><span class="op">[[</span><span class="fl">1</span><span class="op">]</span><span class="op">]</span></span>
 <span><span class="va">output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.9228 -1.1292 -0.5401  0.8716</span></span>
-<span><span class="co">#&gt;  0.1488 -0.7452 -0.0094  0.2273</span></span>
+<span><span class="co">#&gt;  0.6356 -0.0022  0.3491 -1.0918</span></span>
+<span><span class="co">#&gt; -1.5718  1.8213 -1.6298  0.2341</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,4} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span></code></pre></div>
 <p>Note we only use the <code>$train()</code>, since torch modules do
 not have anything that maps to the <code>state</code> (it is filled by
@@ -196,8 +196,8 @@ <h2 id="neural-networks-as-graphs">Neural Networks as Graphs<a class="anchor" ar
 <code class="sourceCode R"><span><span class="va">output</span> <span class="op">=</span> <span class="va">module_graph</span><span class="op">$</span><span class="fu">train</span><span class="op">(</span><span class="va">input</span><span class="op">)</span><span class="op">[[</span><span class="fl">1</span><span class="op">]</span><span class="op">]</span></span>
 <span><span class="va">output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.2143  0.2882  0.4975</span></span>
-<span><span class="co">#&gt;  0.2435  0.2812  0.4753</span></span>
+<span><span class="co">#&gt;  0.2569  0.4082  0.3350</span></span>
+<span><span class="co">#&gt;  0.3488  0.3890  0.2623</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,3} ][ grad_fn = &lt;SoftmaxBackward0&gt; ]</span></span></code></pre></div>
 <p>While this object allows to easily perform a forward pass, it does
 not inherit from <code>nn_module</code>, which is useful for various
@@ -245,8 +245,8 @@ <h2 id="neural-networks-as-graphs">Neural Networks as Graphs<a class="anchor" ar
 <div class="sourceCode" id="cb12"><pre class="downlit sourceCode r">
 <code class="sourceCode R"><span><span class="fu">graph_module</span><span class="op">(</span><span class="va">input</span><span class="op">)</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.2143  0.2882  0.4975</span></span>
-<span><span class="co">#&gt;  0.2435  0.2812  0.4753</span></span>
+<span><span class="co">#&gt;  0.2569  0.4082  0.3350</span></span>
+<span><span class="co">#&gt;  0.3488  0.3890  0.2623</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,3} ][ grad_fn = &lt;SoftmaxBackward0&gt; ]</span></span></code></pre></div>
 </div>
 <div class="section level2">
@@ -363,8 +363,8 @@ <h3 id="modeldescriptor">
 <span></span>
 <span><span class="fu">small_module</span><span class="op">(</span><span class="va">input</span><span class="op">)</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt; -0.3948  0.0355 -1.1148  0.9649</span></span>
-<span><span class="co">#&gt; -0.3343 -1.2209  0.1999  0.1987</span></span>
+<span><span class="co">#&gt;  0.2935  0.0055  0.1483  0.6004</span></span>
+<span><span class="co">#&gt;  0.3216  2.2073 -0.5215 -0.5771</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,4} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span></code></pre></div>
 </div>
 <div class="section level3">
@@ -429,9 +429,9 @@ <h3 id="using-modeldescriptor-to-get-data">Using <code>ModelDescriptor</code> to
 <span></span>
 <span><span class="fu">small_module</span><span class="op">(</span><span class="va">batch</span><span class="op">$</span><span class="va">x</span><span class="op">[[</span><span class="fl">1</span><span class="op">]</span><span class="op">]</span><span class="op">)</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  4.6313 -0.4342  2.2361 -0.9563</span></span>
-<span><span class="co">#&gt;  4.2838 -0.3334  2.0962 -0.8742</span></span>
-<span><span class="co">#&gt;  4.2675 -0.4344  2.0631 -0.8572</span></span>
+<span><span class="co">#&gt;  1.7664  4.3770 -3.4375 -0.1343</span></span>
+<span><span class="co">#&gt;  1.7621  4.0245 -3.1681 -0.1304</span></span>
+<span><span class="co">#&gt;  1.6213  4.0500 -3.1483 -0.0796</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{3,4} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span></code></pre></div>
 </div>
 <div class="section level3">
@@ -457,8 +457,8 @@ <h3 id="building-sequential-nns">Building sequential NNs<a class="anchor" aria-l
 <span></span>
 <span><span class="fu">graph_module</span><span class="op">(</span><span class="va">input</span><span class="op">)</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.4210  0.2343  0.3447</span></span>
-<span><span class="co">#&gt;  0.4048  0.2254  0.3698</span></span>
+<span><span class="co">#&gt;  0.2308  0.4097  0.3595</span></span>
+<span><span class="co">#&gt;  0.1968  0.3787  0.4245</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,3} ][ grad_fn = &lt;SoftmaxBackward0&gt; ]</span></span></code></pre></div>
 </div>
 <div class="section level3">
@@ -673,14 +673,15 @@ <h3 id="building-more-interesting-nns">Building more interesting NNs<a class="an
 <span><span class="op">)</span></span>
 <span><span class="co">#&gt; $lin_out.output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt; -1.1582</span></span>
-<span><span class="co">#&gt; -1.0255</span></span>
+<span><span class="co">#&gt; 0.01 *</span></span>
+<span><span class="co">#&gt; -8.6465</span></span>
+<span><span class="co">#&gt;  -9.1034</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,1} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span>
 <span><span class="co">#&gt; </span></span>
 <span><span class="co">#&gt; $nn_softmax.output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.9275  0.0577  0.0148</span></span>
-<span><span class="co">#&gt;  0.9175  0.0639  0.0187</span></span>
+<span><span class="co">#&gt;  0.3961  0.4251  0.1788</span></span>
+<span><span class="co">#&gt;  0.3951  0.4239  0.1810</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,3} ][ grad_fn = &lt;SoftmaxBackward0&gt; ]</span></span></code></pre></div>
 <p>The first linear layer that takes “Sepal” input
 (<code>"linear1"</code>) creates a 2x4 tensor (batch size 2, 4 units),
@@ -689,14 +690,14 @@ <h3 id="building-more-interesting-nns">Building more interesting NNs<a class="an
 <code class="sourceCode R"><span><span class="va">iris_module</span><span class="op">$</span><span class="va">graph</span><span class="op">$</span><span class="va">pipeops</span><span class="op">$</span><span class="va">linear1</span><span class="op">$</span><span class="va">.result</span></span>
 <span><span class="co">#&gt; $output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  1.6322  4.9484  4.6457  0.6479</span></span>
-<span><span class="co">#&gt;  1.5350  4.6447  4.3141  0.4427</span></span>
+<span><span class="co">#&gt;  2.1381 -1.4562 -2.3068 -3.0385</span></span>
+<span><span class="co">#&gt;  2.0602 -1.6112 -2.1434 -2.9874</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,4} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span>
 <span><span class="va">iris_module</span><span class="op">$</span><span class="va">graph</span><span class="op">$</span><span class="va">pipeops</span><span class="op">$</span><span class="va">linear3</span><span class="op">$</span><span class="va">.result</span></span>
 <span><span class="co">#&gt; $output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  0.1811 -0.2108  0.3277  0.3537  0.0080</span></span>
-<span><span class="co">#&gt;  0.1811 -0.2108  0.3277  0.3537  0.0080</span></span>
+<span><span class="co">#&gt; -0.4947 -0.6350  0.1133 -0.4768  0.0452</span></span>
+<span><span class="co">#&gt; -0.4947 -0.6350  0.1133 -0.4768  0.0452</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,5} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span></span></code></pre></div>
 <p>We observe that the <code>po("nn_merge_cat")</code> concatenates
 these, as expected:</p>
@@ -704,8 +705,8 @@ <h3 id="building-more-interesting-nns">Building more interesting NNs<a class="an
 <code class="sourceCode R"><span><span class="va">iris_module</span><span class="op">$</span><span class="va">graph</span><span class="op">$</span><span class="va">pipeops</span><span class="op">$</span><span class="va">nn_merge_cat</span><span class="op">$</span><span class="va">.result</span></span>
 <span><span class="co">#&gt; $output</span></span>
 <span><span class="co">#&gt; torch_tensor</span></span>
-<span><span class="co">#&gt;  1.6322  4.9484  4.6457  0.6479  0.1811 -0.2108  0.3277  0.3537  0.0080</span></span>
-<span><span class="co">#&gt;  1.5350  4.6447  4.3141  0.4427  0.1811 -0.2108  0.3277  0.3537  0.0080</span></span>
+<span><span class="co">#&gt;  2.1381 -1.4562 -2.3068 -3.0385 -0.4947 -0.6350  0.1133 -0.4768  0.0452</span></span>
+<span><span class="co">#&gt;  2.0602 -1.6112 -2.1434 -2.9874 -0.4947 -0.6350  0.1133 -0.4768  0.0452</span></span>
 <span><span class="co">#&gt; [ CPUFloatType{2,9} ][ grad_fn = &lt;CatBackward0&gt; ]</span></span></code></pre></div>
 </li>
 </ol>
diff --git a/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-37-1.png b/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-37-1.png
index 25cd23b7aa5b38fd7f9e6ea3e8b14683df693d06..dc7f345e4249e7a176e18bacf922f3f4d67ab8b5 100644
GIT binary patch
literal 123484
zcmc$_Wmr^i_$@r@4^U7s=rkxvr5gkR2kDen>7lzr6j4%!l5U1Zx?4cHySuw%nD-gQ
z|9j5I_rrS*mt5=%huP2G`+n}V?zPr@larCSgH3`Bfk5s+U%yg-KrkF3kQ?uBVuIgj
zUGLWhKQKSMk$44ug0EB2N)PbkmgQ?T8wiBl2mO65XmM8`0(k_1zIv(T7_%|y;23gv
z*}Th_Y#hBcsc<y)Cfw80RPLT71O1wn=X{yGlG3mF=1yEOvG4ODa$njg3XZ2-La#HK
zt(x~!)e7AU#1@GzlJ2k38Me!9)UTbE-7b<1S?cPi#YFV4;{90$1X8z7BtU+344EOp
zMz0h7wWEMn_4f!elS=1!?eB5j8~L5<e-Aw)l4dZjUcn+(k8^dX?dwH&by$Um!E$xz
z`0W478?(CbdX`%8#zwu+{esX<AmWLY!N<e<t;gDGw7?PM+hJbXK5O|J-eFL98}pOT
z!}<_xY;3*fM0%`|{NxxEwp%Ym;Qr!ITSPIIExGkd$DRB=!3!2_d8wg-f`Sw>H)IP(
zP5C40PFrq5_4gca-oC9=YCg$Nena@y3+`^DftLt4H*B(^7@_-j`FN7Wut9daH6_^=
zh@1%br(rQ!{@u^HwU^GaEy6D>xydA@rIp{kOK^2_6ClT;2#t))sH)<JDp*=RH(XG%
zTwPs-cTwP^mzi>opX$~R537Fs_>n^9KiRhSc0~h&w1$R;wuwkJIxS;k<K^xFzpr0+
zEJ+;IB_(|?kjTaD?d?^%orH}I)5V2_*9ws?=BHbg+c4zWVO-#~g#YAsDt?=CMr81r
zmUYcq{{E2vdw4iv+Ze;JaI_0mo|>1(f?<K-uyyHq?(N%Y#NPHH|Mi2Qhu^6dmzUeY
zE4@m+3g0p=<7jUY*4f;8ZOC$bw0!PeV9Rg$?ODRbJeT6t1N^^P4*$Q^+5h1;ZrfOe
zOtr1rf`87un7;Mp+fSzzFNj-ibj@<7k6d0NmmSkbPGYYSHs(XVonu@a)_Bfe6E;qs
zv6EL(373f%+*~DMj;`!qTfM38uM8Je2t<6UOZMk=VSKGnPl*z3vfFwL2Y+|^B4qxW
z3e+j??zNzOWPfeBjwgx|)txUFcWC-uYo$0cGP13sgR!1nEF?Obp>lF-l?Wnv4<pa-
z;^G1i4{rgSrSPp(T%cUq0w@ww&UT}POlo|!moGn(-SwerZE2Cf{8ZogZ3c>gVTj{i
z80pjI+X1f1pb+~1ZaXD8xkr9YbD(mcM|av^Ztw?UjhJ#qO3~zbFuyNcT>V{RCJB|4
zbJq?OY0$;f$;oYa!r;WK9(fPXc6GVO$;qj$yE}sF^!z*`W1862)wSyK<UdPe{bh=l
z_8*-Fr)Ou{?(d0}zfhR)Z%wc)^IQ2e&T{MQkOYWFN>$xc2#)1z{|E|xv)iz8hx?-q
z%f$~V8twnGXFwn5LfvgeUN_~8RuvN8m@d~Al6HPW8jebf+3?2^)Gl;?W4FomN&st$
zh0^a>5X(LyZ#%WN*1i(yTCXv5cAcWsbFZ6^^5dUv;n{9o$aXjyjceS?5<&%~Da@ni
z*hoxBk7CuxZW9biPq~q0`)(MMe5Cj#8%<Dea-M6dI1`@>HxFcH`>Lg;3oF0_O$L(P
zY#p}=PBA31oN9wJ``y2G)Liq>TCt@+-QhJmaY}hm(mC`-w9;f{KR#0f<--4v3}Pv-
z`T0-(^~8Q6#3F%mN~*oPJ()D(9Lr%jV9!x@mWb1?1!X0|Q>!j^7Ced`n|>i91C6gb
zV=;TXtC6vc3AtdqYRdFqp_BqyNWNi)ilMNz(-X2&3~>UOPIVV<`uwC$>#FVA<o!~5
z0gjB%A=e?;et#RhC-%&*d}Ne>ezvRY4ZFi0N_|$s&44;saix|W-y~wAgYC?!!9$Xc
z8j_Iy?_wcKQslIi)xrVJf{#s%s3Q!D%C_($nJ%OFHrW=J7x8RFAf62W^Wq`K$j8vk
z$@gE|*|Hy7`C3e9IF*l%6&SpHL45;K_UURF*35hiS}j2v@^VT{7-B<~CMv0zlhjZb
zhis-mBM5}+3b5q5QfX)<;ny3o$^_OY{B78-LE^4ZOPFW*=QuSKGr1M{{!5{<SO2V~
z2LYtc><R!wZQci4!BaK{bj`clpnR7Rkq}7tRqIHI!MTa8tY@km%agA$>Va33D>+g9
z7OqrnUijfPCd88P3IP{m>fgz&%OI(sSSlX#dodRqO5~HQZ)ggGqDt4*G<P8o^Mt=>
zns9xEXUCoRHm2F>=AA`DL^-=|d!EJb@Q_?)H>`G&dzu}$u0eAD=f&yThz#Ci;_Rx9
z{xjr$y*2NK*jW5ex4@E4Cka6Up-ER%JN3f{5IMc8i&46FEO_{FE^}b0m#NmM^<-S>
zd%RY<n*>+)KvF)}?^Ro)5uEsE^d+!cv=|UPf~)1@zj|l+BhJGk{E2SX+RPXwyY2u!
zZ^5o1Pi&xHZnq>)=HkQc{)?cz5GEGIQe4s;X18s+#Q})tn-?y;e*uI}V0UnBH5O`e
zxV2vjTOadQ-)R+A6P?4GE-SsB4Et@GPrux8D|W0_0dbV~Jw`_DhA8{SY4<>4i*#Hd
zF64ytU#b51zGGC)YF{(N6q{=NaC%^iiJ{2h^fS(oxA?EMHIuD@gvX8eL}XhZQy`x0
ziT~~?E75J(WJbKY!y1C^8KTIf9XpkZTvgk=q~Xa2e4Xkz*kG*)=2KON|2zx)|CKj0
z6~DS!<?S8Y+YnEje?>V{lC74IYEz@!_pPz5yHnmsF9ybCnmWaGJB;JeNI!Ux5yF4D
zkT8al-1ofhDc4ti9+7k!?o$|)I^{DG(w8SZO{#f$9<{9U#iS)i!jh1>MY+GEO0Irm
zszQ?G&U$WD3?o#*Nb^P$*H+H|(oc|Cwyw*r<a}j{veSbqT4%S1Y9*xiEUh7R)Bno+
zWA+a&v2A`2znHr|PqIoC)=8)3O{=>IwW1@!l9+01ayR#<hUau@hw?=$50+H3^cDh!
zAqgEb6vXHU<%vD_`_hbLzroC!mn?I9RZwexW6y)pfx4CJk!FsCt44Kv4DGb1<>q+4
z%67G-#xojBi00jYRq`v{$EDqzMJ`~7E31(_zRjov52j;U&AfKBV@oV8sLpxpVy0G@
z@V|GFcfHx)cw<8!p=RFN5$%=hJ;yWlkcth2*>~>4Q-UNW6(8Kc24Vi6g8#v}D-BgB
zG!g&Z!qt;XjP4|U98yr)z!klQ`Z<3rh~@qN*>jDvJ-dxARG(gQj7@Cpkx|w@QZ}2C
zg|p8`zi7h8h0I`H6;)e#$gT55%MpAM4ITHFz2=?18U0W=Xx(4tzJg8s@ZTFw^xoj3
zf)4$3EHzXUWq`uFQfSHra3C{+g@_w}0l{O*mfMzw35rRKM@I7OQ~t^_OwAQZ4ZU00
zxygUmrG~22bkE8KG#9sj&E-WaZ4j*ffTh@eRf>G9P|e=uBUP&&YnFI50lH!<bR6;;
zlIihb+QnniRM5y)pNXz;yR)xU&!{G<-v~e;zF1eQ)BnsQq&n?SnxRv3E?08F4~1L2
z`z+Ex-@a-~&v*FaX!C8B8{ZIY?ep;9c1eqEJCdb+L$`ck;^%Mf)3>f+PG>*+WVhrk
zrL-d0XiVc7ps=A5pLR9~q{-)ML$E&L;E}iB=a6$L$q8?5kqvMWNF7t1%J_NG46nDe
z5AM$h)r~H;XrkIwXID|At;!`QjewnKD6+VSn{}Qs6}z-cZmOMmt55Evy&*1C2&8*|
zt_k|wVfE?z2m^2-gcb$pSqb9lv#4jdvgqRs>n<7e&FmE)t}T&3*WUZE=!?9GFz`HQ
zm7}scJ9XKO4w%K1Ofq7l!;JxihDJoYwRT%|Kir108nt-gh6TI!q{gfx0=_rK<y8cX
zw~^f2=(`OO-o0wxb&bt|H98VU7K)n4tik0-YV4bN)}sO+g9is4P87t3w-Rnta<24q
zebKUZn`|~yF16AaH7i#?X%!ybWAzKQ8lJNnt6i&QB?@rf6Pm_1W`QIuT<zqUmz5>i
zJq6{{tt9GDRUvLM8p>LcZ4Qm>wWA*PQw%W^Q1oN9Hc0B>T2}kG?2nHnc`DRus|Ur@
zGFq5X&=?sr<fe4a#rEj+(<$0%RVB@05ySr=p^R4r%Wni8BQ=yM*F~)6<9)HIZ9=@9
zC8Wfq{*X{^G3Hv%H<Ja6JAJE->xcb*0Yy$jMvDeiT0e(o>Es%Y%A*dQU2>w&vZ*pq
zGqiS{ok4g!Pr=G}XiWapu;XiU`!vp+C<aQ@yTO`0nTX|qP0OGmX(x43l?xgT_gq&-
zlq-RyYjMTdMX9dyWsMAPkzpa{#YGaQ6lKP5SPmYZvZ#Td@~EK<-RTU;qJHXYxGST(
z^NRQG+JwW;cH(Og>MQJq;EooW6=R#8W)3Z=6L%6sEdOA<^v#-9mMFbQ0>zr~az8La
z)8fv}A(v^>zNNwagC@S9XR|xz)RgL33!)~d0_Nx3n&d7MDK7Ds!UkM>*C3Mrnh}LH
z@3WV1JTk5EO|_RE9j~Fh+A9K`V@EwciY4vxatUUn;VUck<ZWI2$z_>#0|{ypVy{{%
zV{ssI1y{X(=ljHl6cj$n(zB^1=>8I(0vO_xZH}l;*jQS0RLSa+{auLg>YVh@pWC+c
z9RL$lVJk8O%@Tz8p0e{dZ>m8c+b{pZfv1mBrn)Mex$G#10gjs+Ox2{EGk?Sg@jSk2
zZFQoVgXT$JSK1_b%f!{5NodA}xWnQV0?k!S7uA>F=t6{>u8_skhv+t@-4@?lc#-iE
zUBt3!)UGd8hRt+6K)ELP9mov%zbi=i=pe{s^6u*q<GDgh$))-NQi4-@_)<9pAMZip
zWvP(-bwDQciLW{c1s)MuE>;{l15d2_-e?YQz6&(*m}2^cPHm8o3+B$Jr_HT^e<-{F
zfAVj#rx?mOHmg|GVWAhI)v{@}9bQsPR{kDw7_@HY6gkE{uixobTWabQS>ZNfRJ?63
zipGmjd;?jmKlcpMmxCLq;AmUbTzmQxT|xGYC$9(dh*HYVwDj*}F}7x8hw4gB6wCPx
z%PN~26m$p@2%I!H+O2z(BIa*E_&@*aOi9BDSOX`%%x4B0IO_Q`5}WG$GL-~DmG5E!
z?P`fDT`5V-qtBHz!==Iir~lV_AEPPBDi?wd^tJVHWYVWfE_!eXC^G>uLj6!9Q@O<i
z{{hA3YC6P|)$`xE=ondL1TZh!`(HpH`eawPj?STgt$XEF!?jo_54DF$!jmKh2G){H
zAWtBYuL<Y>8#EWas_W9c*=qaX(=;Y6Y5D+UJx(2!G&zblS%r}n9wPGj$^xMeKrg&{
z-yLT@PUgXuys4~TPhF^??l0wiv!^64>t7Ts8qgsJ1gXg<v(={E6L=I`jO)l{*Tb>2
z!+VX;jKC82-&?@;5gv(FC?`1pEpux+piyWm0Dxt^$f@d-6iCy;5$X-ZpfqYDOfI%q
z<`vj)#yeHJArMLinJ_%C{rj9CsyU}_UqH8%m4T|>rO29+WtKyeiqJ|p-aQpHRZUby
zu8VPq$cJ^~dEZfIj|Fl<&9-Zhs$2ioAc>HrUO^IK0x-(#uf5DV)A}Om?f9^G=0`=v
z(&!?#Z)V1zpU05D#T0{g5k!vls(?LJ&)hE<+FHBh8qI?;#n+>Ki-x(Li)i^sht$6t
z2xSpxU=t|*tZ-ep_1_)VrezZl!!eZl9^~;e>X>x3&wq!$Q>n7?p1yvH@odMfrkHyW
z$$yZ!!wUvK8ZeIxs^1O;XxVPFrfru;W!woTZo<C<k;MPEd|ou<Q8jU`tq#Sclb9{e
z6Q&4^tAN3XK!P1Yr+igVb?Yfq<PlUD;z~9qf%QfT!f*9Q_x3}>f>p?RAqh(Vl(RfV
zx$i7c)+i?PZeFhBy6|Pt1?bHF?TJvGmb+U6QPlm#W{nS0V<hFwWx_=olJRwN;?(V`
z#F@sAgX1}R`d`d?H$0iAf{q%uOvZ)9+G~m@L+jtmn|ELL>=Xanap>%C`PduxNp!<H
z;j4tM8+r}bHb*NY{S2g)<X(R^FFSTRtloVb={6zs@FBj$pn3o`l=hfWd2ngCoPW9F
z+hkkkYX$IIefB6{1`0^%l{6@vkEU^M2W^vW-R;2~MesHID%D--Gdk{y5-_xa)|avH
z$;8%Z5|4cBx%9(k{zPRQgfXB2e|FNz&V!uX3YGX;WcuCQ=p+VAWXs|k4l>RCSgH&=
zgnJV)E=2Oq-@BeL!uaFGOUj@M309(ks_EApt8AJEeupe3UQbNu=nB5TifRB@maH7J
zZ1r_Xw;xQ@@~-9NsXTkmJt4X}EB%3fP_B@3ve>E96|j!-H64NM-)z=wA;o4>M8p9!
zV8~B75xp+_Z04$eH+|?HQa-Ci@jv?1n_}LWmeFLFrzy8}sr7a-4V0*U=0r-+gP@H;
zzV0bTM5Daqx7kfhp^FMC<ISLMy2Fw1C!2QM*azn)^#zM5S?8WE${Dap_}7~d>cGE^
zV}=vam?%3<MC0;N+RU6HIg<$!&pB_DOIVBIL_5*yP7zorQ|!6cEgCT=h06B~U~+py
z|GT|!;`n^!VD`{^O!a$w&U6c|O2ZLPre^ZTE$4%r2Qt_Y>MMFSQ!@A_ZM|+Qz&CwJ
z;ZB91<^mFc7R+H>0-j>g5)j(I%DDBF4q<*d{2DEAfVw$sCnYa7Ze{?jMAJ3DKSNi;
zqd$9i49s`i_{550Pn$cwb39M66}=(+Vvgjm7F@~{e_TzKVH|I$|GdneSzhy#Ihvr+
z-p%d(s70!jRgd39e19>t-e{olX|0m$;TgIiqhY`-dOvWE^jJafaA*%FQ2wR)mp<9e
z{`{O;deFq63d!NI2Lq%+bO19OOCgX6(Z3aJ>DRwH&>Uu?$F6~%6EzAmIcsI0Si?s1
zZ9I5YbnKt%6J9#pQ9U>3TB+>T9%{o-v@=4EqL5mh&P=vQfv2|m!|VFbiTJ-0{7*-%
zW8zU;ZEJh2_}Wz9^aP`vgCX1cA+#^!m*^IjsP{vS@^A{$!{RT2CraZ(ObT<oZjx_!
zo_Uyj2OK;@zP-ygS81~U#kGXDSDSePZnacI1%{)BO|_bj!G?xu)Rh7__z`*W%Bld?
z07fJz{UcVT|NZ9;h3^$Af(;uH_5_N3`>q2R1t9p>{uX>4)3uk0d!?Aqo}MObobs4=
zO7hi+!twAmin)`nLFBgno$|wTqJeBP2S7z7M_zR2sDar%l#0V-Jo$fr*-*`S06Zn9
za2g&X6JZcDRx3h#%&h#%R)5fVB8__<yrcQQcSOGwT~jif8oAXBGz3ny0zwj5HBPVK
zw);qyRwH(N2Vh&YQU7J1p5sz9g<Bp=y2SqCIVWe=sJGOWR;mNNK>1ph%1}-vR7j6y
znAAR|Cd7kyT7UUl{&I{_MO&GVgQ}lOysTNO8CK*YaHaD}o`JithTs}^xd_-<ZXRO?
z`3@Ia$cD7)F{l25KpcfAIR5w!O=&2zp!N^D{mT|YnS;}p8kGbA%1CBW<#MPFhPX;Z
zuTeM?mnbNVm7a=cW0kI6WcJaSgT|T5Egf=i^<oefz_QTxzFJ7O^YnA2ViL{7yeSV!
zG`g;BQ}l!HklUD%t`~aU-8{8J!`+#eqF+pUz8fwsQ_j>6fVI&6y?EjLtp4->agUN$
zm8ID|YfVBX<7d>8(t^1$8&clQD)Ol?yTP?glfm3}VeiipF1*a(%vXY$+!H_t*0|Oa
z&N_J=d4W;1NN_dr;Ku-3a9yJYp^Ha~=+BslVFZqcti&$wT_3GeC!PqMn-DrS?oWSw
z75X1&WzFq~OMU<R_U3fH%&KuKJf&{Naxi;ZBqwhTA>(GQYH5_>6GAWr+G3IG6z*6N
zELVn_0?s5PD@dRt)~uquZ!a*Icpd;7Ndi|5C`$M<j1t!zO>SkAaH-+IfHz^zXP<$7
zkbSM*H%}XgvlS|_=M%X<PNL!hu^}f|fALQ4B{%uq&Wo)Z(x<LPto@2qQ1N87YqQE!
z#G!69@)|>oJ>eWU2c2cXB(}qBB#z5$tdVZ~q!$kY$wqQnrk$jshdYE&(~LvUF?>|y
z&6kX+DMryO=WrZ1J{_8sW!D|Do#h;xMj@qXpRLX<t>d;+Btt{?or?IjI1ozP|0+hC
zlQrTlrFp%$^AZ^?b?3I{KZgN}U77{rQ*H7e^~V$Y5fBKsYq{T=c*H*=z71IR&o$dB
zjTd&g@Lt=JI?%R5i#$?~axT<eBHS`ifo?C+Ywx#^kq0p8V^Qy|!W5HstW%-r#G{5L
zD$N4%N?}V#{Cg^=7f~BU^SlMdUrM^&R5gZ~`gR^~9&D@9WV#)b3jCNTB)eNl<HoJo
zPg~0X%^dhyx#mEeCK=!iq>G)6CkVxiTN-xnHjl^3XP$Om?xKpE4qXLMQ@4RUv|WDy
zE@1v|+m-7Y#W8iBZ@s2|elCR}HrQwPlq<czwnjx0nY-MswYo7{1ujOxhfk=#PJSGZ
zVEQe3DGWge%r?XNTD2R5)f^X?i;H?0PN)d>mBH9dkHep}fV%13$w@iWq8O-Cbx3dW
zlm~d;JKjB&0*Kg=oOSp4-ee1h+g@eCThT#eF!9>qFMeeco*j?HT<`?2vnG^OZ@Wr!
zRn_PyM`<VjN_IKCaElFd-Ai9?xB+3lnp74iQgD4t)+@xo>t7+Y-zEbx#!^X3RZ^4d
zWbU10qLqQ<8(IYuc9=+C1CNzy?G(9HR*VdScajgvbXLolJZjx4ARZDFrE{LdJB<pU
zTCFC;T0aRENWEVwrFPIxh1^3*?A`J2DSoPwDOu<qF*#3&muikHs~US()RyKjbjlHy
zuyN$2@EQj~cl+-SrfYd{cY7#%TQC$*#0nvP(VA<s-c##<(&aX$wF+^I;~2W9g5fj#
zU0*}=Me9u<Mx;Zx5z6are7;4c2GJUW2T}Li+FSdEDFrk{ZndX@?J9jXLk#U3#Jx=T
zKGL8AtAEFyfh3Qo&lD%!1P%kBqjVWRAf8YDl2lJ?Q!u>)xoNPdyAazyTYCX{`8Xdo
zh0=%i;l|ghSIugEw={|f?>>JCh(2s;8|dWhl}k=QaxxxFinrB6m3)wSYu|}FwMk8l
zQ94xNqR(|7ezMIK#RPL%$AYIk)A0}g1NRUJ;a@#Mm`9)GC!n>ftWmB?DA+JdzFg_N
zz}_zIP@1w$SahtK|1H0*iDH{|`wCfuX-I^@4<?nt^}Pft^a*{LG}KCAz3c#{DS|5b
zpJp2Hi4B?}QXWeDQOJCOLU!XXf`uw#9h<7Ch(~=`)1>zE_}XscZ{Rmadh{K1p32f(
z#^%prrTUv9P=&Cx;R+6fWQYkBR0H)xbqdhrBHeygU^~!f8fcUnC>z@!nbhiUdi}5Z
zTc{#JmOO;Y``3`2i;88(C^1mZ$_3iVo9v>!P$M4Z+D;KsF?4`sC9I6G5=~@X7rEpn
zka0PA3xUL3DgBE|_DjT2co3=KcAU-D?V0Qj_EN+0`3ti>9D9?%AZlUEWT+Z(T(v+d
zHRu`e3Eq(N7@L;<@@))AU13MuT`&!K?mcCADov`|)jR8^5gO$xVo(?an7C3Yl#in(
z$<LLsCNA)%65Zq3UU<sHfg+)6_9G#UZ+wJwr#-y-@zBs1nNy!%^hq!rW!QkCV9iGl
zwfGP#dvPO2saCAo&)<I<p@nI4arj$L@BV*TfD6$NT5%zOGawqbcE=UfMH-T*pykAb
z6gXmb=FR2vK!Q~J(~~X7TR-18GnjbI6n|a~>fSGRw-hs@JD`z$CwCU`4Ti<f!cBO9
z<WQ|11f!~nRm_{8nMHA^yM((lySR6W9Rm>tt@&_lqkvOF{|2rQXv)vWR%37rQByS1
zR%~n8O%p3v@G>m=z|N^_RmfacO22z5uJ6GE3ML_wguVw3nc!*-{00}*mTT;n16d%R
zhGFAg#{(IPXNSW&#rB)(vAlMY^78U)J&B*34^|RajI{OtgFxt*ZPEh!Pf97ns_SU{
zIFtsX4;4)(WeUDDx4$oawtn!QdgBM;^QH~E*<5s#5E4fKxrB$l-H7BHi`trDyuRP^
zGD0M{7@n?r_61rD!@IHC!$X)unsf%$?<xR42=AV}GW_LeMoVa!OMakPT&lnuq2bFx
zlyXa@pla$D0oxDV+t}FO<Tk}~1e~MQ?N^e+a1d)dU*ad|9dRvw&_FAaKCQD=ed{$+
z@UyQjvYu6<HjU1KF=3Unx2*nL|J@u9)VT5EXI)>xo(rq=8eX-9tqo*Eu^ad4T%PCa
zwj3ajr`ON7Yln-B`}H^@Dx*Oy$*t8~uBb^zpQQQI6rx)k$+5Gf;3c6pVd2YCiO?|X
zSL+CYJY*Kl`r><mgOe3hrf5(!|96sfZUEZ6hc(iQHAFenLbalZyV;<w2G=h<)isZO
zFFvtU@xKJ-9UTjUShR<e^!Rc$f_mOURcbP8!FFUYf4P&DM4F<E8r7e+Z+CXFDN$^(
zNX1ckgaIa3hCTo60bMkQ0-7CcTwnNB3S%_-TnvOVgJ^N5xCs(5Wx83nW)T`}D|{GT
z0>sJQ+9QbPoa)w;o>^PyGjP%9Xhxj#+A}v4@;FM+ts+e(HZw@@>~*#C;hrs^lAhM@
z>-||cz{5nN%lfxd*o$`qV(pVQz@7v8K#j^dUvul$&-G6cTYWX;>yj^C$i2|wI)R?5
zeX@n2u!#c<?8;FZ#kIN<=s$A3(J;7_=72^iV7&5{8)g8ZXwtHKKlR7OoR{cRwBEfy
z_j`En8w%cNwcFls@et9N3oNUU$?z-_3n!(-`?)$T!{bFjgyC2(9aw!zT5Q%;DK-@!
z<}77`;AetXAQOju`6wWDVvCtG(lRnb3#}p4P~Db5WrR%Z!uB@Hr}{^fl$6R@D$=+D
z2louO0pTor2DGI4Y3Em<r)c?HrUPRZr@B;o8Eatej7z`0d(ekvn%bvP+Y5T_<gKed
z=lN?m+M)$kg~CcTz|^d)ErIjlV49Gii;EkQRJ!taNcOTb2TarF`IbzYew{D=9LFy1
zeqU<bw2&w{!l{84v%a{9hg$X(G`(Yot$dYToS)CKQ}i)GWwv@xMFUQ@FSLP(IS5%B
zLQyam-hk-8E~j@yNJTz@lCT?n|DemzmGu*JxYSLbZXIRdjgi>9C|TX9z<2Tp(UyHy
zVUI;F@Ke;!<Fw+x2iLg`kAhH$jsrYp;~`l)-YLZe<X~QzcJcUhM%4r(l6`ga!RxoS
zmqU*Nh}cAdFoCY9rK)i)<Can#YgzR<nNH$<MHhF<yoB#J0~{^B00}%bv}_}N@@E)@
zlatqNxI60NZqC_}JkXCX_AwEP5~Uq|9lbeW*0v!eC1sMh{Cp(?d)Fb%c@r^hYqq?_
z@9M8Z6@n-P_+x;HkZ?i&bJ3*L0u|P4D&KFS$$#GYPOg1jyLZE(5-q*2jaj_TH~4Qy
z2w*)m6vp!|w4E*L*hB4HMXE1J?V;Uxk1CiOeV@gye_$By8-F;wA|$JgXmoKO?q=ao
z4l_V16*t*o`&lqfd)@>Kq}x2PCm&_klHiny=wFo!i@L+{q2qRwO77xn^vny@!wTi0
zx_i}pb1IvfJHbbJ7wR126G<n{)%bX7qBMTWJU}U5t)nN`JpTm8%?YEjTeGR8aDHC}
zu!N!p1L7&fz=g3L$z~WDAI}2y4G0*F6L4YIZ^JX2s#3dl?HX(-JM!#s9id)Kzx9b5
z0ui<vDL0O^F6aZ}5hZ}s3VzxWZ9;GWYiUkJkG{fkMlaI-djv(BW;-=G*nt_zJ8hks
zaSy?A<T=LZR%b}arz*t;*DD0GxW!a6b4zQ~Q*~-lZ-v|s&qkj(^dI;Em*LCG+2iiD
zC=H6Ry!hhKkZ=K2+m)=KG}B>+_?m1CHd-UJJyPsAAvUa%+bdj(>mm)zRJw<Yf{>ZK
zHtG84@NgQaufP9*7Y5G3ZVP2R>aDqXE&;P<x;G9XtS8|m39H_3YQxQue9>vw{Z8uP
zg{39>r%xL<C*t7o-3+<Q+_5rHg}<2;pZ>a#%L`=bsA<7?IgOr}{gxy5@!VtjG@)r0
zZnC?9K0;(uwbF1r<-%AEo7>_cNwAVZV1EPsUn^6sR^RFu!wZiZorIb6kuKyVT^O?E
zTjTlV!T#Ek6{+&NBsIBavgmn9hQc9c)BQY0BB?c(_S!+W8RS@rc`#>QTozzLp;k0v
zAM4-y`sQS&nK`FJ*K?QsCg(0{z#Nz7#(W=-B<K7~(4h2jb@m+w_rcuG_At1uCK&Co
zZ~3XX9y(VfHF9JQCWPG<Pc|)S&{{9!Z~Ohba-jtW`|{YpqrALaf32q8pwMY?_g2CV
z`sPJA9CdM0Q;Si0T;yZkCQ2Z+sFDYvG7u>1hf}{fXr+xe8Y?nJSB83#(QZ(oOe}Be
zXrZCb-aZJ@#h_xQ(i7?yK#VA`adze<?y?pCOpFk!Aka79+#3az7JLjX`0@*IE1`vr
zbD9_#-rUUM=n8>6_MPjGF0-?RWKyW8CF!T)heH?U_4n*ar3GPQC5Wh!&C?(gN&3=K
zA@TqI^I6ZF`ediPT~CNCo1Fb#b#jF7lf;i=F%c`9G)J}LC_>uet@mk6X!=JR`erTD
z&BjYAmg8JN{)h@RQz0Rt!!-bCk(D~`r>R&Vb^>waqSWyKPu#s^ybyb@t81n^wVf7B
z%Lhez<fEm^K1ZesFzcik{3<GXMj+_|uGkMWDcg;iyZUK}3ig({b=nL^ckRt(cWt**
zCk#je<1;S99ZW2&c$5R`63J4rw3d}Oq^jeV+USGZZcqqvPRmK_>LsQb&=4A}16>(}
zqQj(JM7Bznh-mdOua&#|V%oSvbpNW7wsssyD0S7!dHSw)N9g~2%g9P{%9HWC3#k^{
z^kLWXfn#4hOZz+4$je;^A72`p+$gGNfT)ZYvchpC;U5^3L*nD($M+@Sljaef1|cA_
z7hzQ9KQD1)bQGG89`mWGDK6Vvyc&}RGUqwgUhD`Cnr%$!3-VL5uxww9S~fXN@W#rR
zX$#2W7|Qf~!7mI83A>C4Qq7$b&fz;&osOnZa8z<lG-&>%8lhArl-H}hv#%btx3>oZ
zlYomvBZkYq_Vm0#$^3|p&YGCegu(}5Op3Ll>Wl<&>(>`G%botizIBTSwY{^m<ClBu
z=1p$YHi(LOxWWm@$g;oOA=N(6U3L#<BMHz(LS!rsg_dg&n4dn1<awR%hu8&T#qd#Z
z-=!}RLh(F3#zFP}6qA7V<(m}_4h8TyMIypnEd#Cu?%Vno>n>Z}c>&^2F}8(%w|5Tx
z!i^h>$%t>8<vvd`26+gK&#Ga)dMYX~FNKFCu-T5hBFiOJn8Ts4z|I8f8CpJU^=p3j
z{9@7RrPXWx3Rzu=X@uE2VSKDf30;9R1EAim=I!$0eJ{!RiWFP!ZVqHy2=&zQ#Scjz
z2ZXkQ=owY<T4^Xy?`Z8=!Y9KSLV`@1viskw0?`$4CRk6lz_E{qE<4F-RH%)?J_v?+
zTwX>LH^l2cq(ilJ2?R!i2q-=!W!~q{|D~p;J`=obP%SDfEJPJh0ALj(e5j*12;)Le
z4q;^`iF0cXcE(*>sq>4ryaQ>9C{~@VxjhgETQ6=oU#-i%1LiZ8LT%X$8J|kD*ub*B
zSD;`EWU?`SlOh$SGw8W}cK_F7gLLu9B?`$cCyim|f#~juglvB$jFtiAefOt;tl91D
z-nnS6I0$hmc4<Y7mG9^xcTTg$s_nw!oNRmn>Y|>j6`Y)9HHB4ox#y_LC{)WFSBCEH
zG{=WV2d6Zv+RDG%X_8F`GF+T|u6qatAPQ8*L)mJC$dkU*H`LH#lVN7XyfpCtj?qAi
zjiAbj-76x)5c5Fn0BL0GY@HO?;}zyRAx-4Csf=8Fig}lnd3QI)z>G$2i(Hw6xo`TR
zZKMU#d%3_AT028+C<kzs(w#2I2RN6N!euEAg{`~!$_zuaU8>>>SQQ4-Tjoi)VbLIG
zhV-&lg?s6)r5{Rf8`Uvr;FAmsN2_-C^L0-$RwDZ>$T)Yp45^^e231K_*`ZGiix%}^
z$fEo3%i}^^AZG?zEAQBEO{jsssb1q+P9qza9TP(aHO$K@H|l*7Kq;gRx9t`<c4Vai
zgmt(wb2ujxNPlR76F<a9Qq0j|%j?3F>BP}<&!#K=_>L&Z5xE2P-k1n=^b~E0e^|Fu
z*+vlj7NR;TaOoh(s+c$qK>JI53NmY(vs@}=T<_XJiId9SRRUxS&D_T(_{GKh7xNCu
z$rL4q`AbC}=N}eZEs{x}emI~W9yaTfA$OTd0{2y&eJkxrHQl`JNb!MRWvIJ#p1S7f
zo6fuq<&|o{er!XtyoNvIuf#e`1rNdoafL1tqAhCFkmtw4U<%6vDRB^jaol;wVLma^
z2nElsDjG=kY!=9LFn?}p9i<YuITr%gHexaY#6sNNl9K^1QJ$pPv_00{4i!-kP5I=)
z3EZ^21wx{XEXtTCA>}tCMA(RI7dLhBwNuNFOf_;A_dV#!6ZCsnrrpvNzxgNusg|Sm
zsOiaO@gF@2C=pC&#&a5yP+s@eL(zDYk}`}_VXg}!3Te)@7{g;V+8rDpo9RTUoz+O1
z_N+p|PT*)wFIVUeM3`797h4;esgwbyk@}Iqw_T+;n5i7nYozq*)o16UjSN}A(^p)R
zwjqe805w8!f?p^f?2S12`sZYduBRET*%TM+*82dT%Bg+~y)E3=Y|d7jFnQhdlF{NU
z(2TzaH5GHYij3~=4;#KYGZ?OURGX0gS6^|P*MO(Q)6^(ukG8P_ltC+dpX$?xxT>F)
z%f(=VZTBgQQURn|-J|zfsfX2Xcya6{Pb(#x+r&;(N-4(YWl6J1JYzfU!7wRQd{|_6
zq;a-eKK75kF9t2sbF&xx1`!G2R)z?|q{HLm;X7Ot38dy__5Jdog|ro~*+s3PV|1e$
zxF$^(&(x8dwZ&(ZEqEZhbeR~oP&6JIX}#e3P46h*s4iA^{U|Y6|0v&)lYhbhC_u;p
z9X<BbbkI<=;uJGpIsh?o;Nv7|nC;xxqz_pHx@`r<$Khd;vTxARB6fhu%w7yyAUtx~
zHtw8T1tz{pgeEEQhES&;B5B-C{FkJVN~DO|lz@I61$6xKvuZFf+(CB+QKDcV6JnyR
zrend=OX}(0cR-6!DQ84~l0eQGH2lK@QX!-1K)Wc^{<3<dZJNRZ2YM!}<x`F@6Lz_1
z8RX!3^LzZHsw}!diiwPDfA*%K^(P>bGK@uh+t*^?O(g*;-rhn@{&8TimWIk}_#A1a
zgY-D?lYyd^`xL8$fp5P<-|wvu3Nk5l@$j(n@WmgA6{M`FDJBvADahsFS53-_>se<R
z!-u%(FeYXTEQqk01loD>{2&h;xIyy~NnCD}Y~<=QNddPjod@J<_H6k>JqLtx&q?sq
z8aLHGdUTXc6zjG7g(}{H`!1Uh0u~KoTA^^;svg6!X62mZ%6Otb-$1yRl$Hr?oJE@w
z*S$o_htTf8-w;Dn+q>s<bW|cQAK#A)^m-5;4-#0cr)3^Y`h`uXjgBg^QbVUta~ETg
zt`p~DARJltA4r^O|II3%U;z(69q7DfT9tBu)YCES`-?Q{0-Yx<ccn*FdfawDlgrnJ
zmu#%sENtjfP*DQcR28;&3&g(gDU_jyQJNW?vF8jlH3o^x-d|c?!wb5pDq=S${CmE>
zMW6Pkkh!6i>!$VPkNX6~sw)D32}$5?^^8%o6@kD2h}D8XJ3ZqQ6^=Dd)%7l<GN|^h
z1)!MW{1qQPLgK=+fhIY$L5Y&ZD7o;<JvD243*3?QW)~f~yqCD4t6Tv#ys<#tzAb&J
z_c@Sa#HyF0eHod+(!!r>uZ3m214MnD>P!9QX4O2}Prd0WNCp6z;1eHOU5>D6YtXWX
zHSfn%#v+b#<k0xVPb;+SuMlLW3{tD0(r}6mrvr<l4rVRQv`SZ!D0Qk&4Bd`L+cU>g
zVz*ouq&Z-tt&6UpZ!Y{b#5@WN*XgS-2ZU2&1s@kSWmc$8CelmiDO;Kfs9lJnak1ZL
zcpZpCwzlnzZkI>N>YU4bcRUQoPTADkA85G83A*W~6LkD|1O;J@KeOHnFS-7mZy_Lt
z*0CTeIU*==t|~K>>D^!CzRE&!H0Nly>e51Oh=u8s<=%fZ2BtTety8W1!-}%d7kry#
zGFw_{Gd-^d%_fFYiKo&4R0B^QeQx{(LtukcurCAsY0(ldUEik$u#^bRbz9!U{*+So
z)A?oasTX`a<u}D+leuL;1|!l|W+e)Zb(Z>8Nw{8@iGl&s+j;h~t710pe#K@zn0hO;
zXKp||;eQ?2i}&YWsGS!RwF^e~l(Ht@71!*Vo%&oH!mkbWNMh&pNrA~fx!;G&=rrXV
z9jd)kn-x11Q<f<eR&{he8jaXsKB7f^e_LETuc&M`Bzra<KZzwar~r&(wBU`lg*iJN
z#VGx&6eA54Y&UaIH4DotK%dy0D3T&|cb9jIKSg?tdglY58f2R*&c-U4H59@L!K{U5
zeK~J;lI=TC!qVr0std%xf;5Zbk%SNAOixB=p$5-ojhj7MelN@;Kz~klj@^prUDyYe
zGKi;D`6=qiN#$I{J!f!QFq5+OA_&d9W`lLCTY-Gg_DI_c9j&K|<#m5;A?Q|ZTEbsk
zF)M9X;n0V%xHXN=9M&{OWWx3@dBYs-Ypd&2ct7I2ZvZFUUvt9F8IS{BM^<L%xlVy6
z$o#>kEK-~=&IuzsYR>kmX^V^pC*HURG7nEAo~#|(S#PtNMW34ovr$a)D^^AXMtBJS
z%&o0lp+<03nyqAk2Rg=X6AA#BY44I_S}4&!<pyjhtr7&#>|2X*45WJeQGvmE98zvR
zENJklNSHvffd~UH`InWk`wXD$`Wqs3jF-5!@j=WFLg(~HDE91%oBRQ+YY>Z&l6DqS
zTw#;%B*+35OU{$bk{&Rha(J8(QpXfog4iem9Vi7R$=c;_DU8p`%m-6Ix){9d>Rzzr
zX5TtYL^+awfU-t*IGYYoot~V)Kp+A=`+HyVgUDGLb^Ig^Pb@n92lQ3CgQ>koHO_96
zyF2Rz<JKIgMt5&!r&t*u&Wi&td*-uOp#71YH5f){#z)2_f$h%>z%|Q9Mqckw#k-40
zJtMe6@S;%4Ku7v=!6{bIF1yDCK;`P61f8(A!@FZ~w23+Nof{PDCO~X(?>&|g@RV&q
z7qOCys=2_odeecV&w%1HqYSW*aqn_$b6d$ZIgU5dQW(cTrtDzsOU~i8e25-vRFU<b
z9Mk27iarhHS<4|9REn2;Xu*n2^xNA9HXFsWfEeo=lh0qa6^IzH=|Q5CFFMm@l4!*h
z!$}QQCU`ia(&<tg#EIIA0umm_#ip@uhSEGdettP`S{@h?gXL@5O&nzhku`)5m6HQV
z>`|qrr{z(5rz=mkK|qe$&jcL{PiW8p?FFjP2+WrY1J&R`4Tc&k3$*+U1{*YQ_htJK
ztQW(R6Ogs^n}tr&F7GU&-6l$Vzv@9r_h8@XbS4PLF_)Ks<>Hat&noqqPXA<5(?4q`
zM4$UoRT(xS3OrlA=TTuH_L{UujJ;qo0|`EAzr0$s2E}_=KDp>W)%EWAO8s*4vs_7X
zH)L<&*iJvrnBcOOvbl3Wl1XU<2bB!Q>1dmjV2bXz!TLgtKD<8z*aZg@l|ostoaTs_
zZJ^NEc9p_Y4BRBjVhW+#hDB~$L`?ChPg|sy`BF4-lGh<}Dm&@70r+BDH2O|cL4k5_
zYa&w|ek1IsSqW!u%x15b1K3^YTzcQZ-?Zqjfh|DI=+M1_XOtL3CqPgaL`m3>5CUhX
z>kbF&xReYaWK3;ZyF*d~J`Hh5<QQ*WvrUaoddB2cJtw+NV%egIj#<~7))E#w_sWjd
z!U~S^qFn38dUlBJl<gZIR<AxjeS5*ocYZQE$!mA<9qbRecY1aaY-gC_bZOd#)LWVf
zZsWt5+Wib#aiD2jMfd2dLT+I#%#)_itc;m=-pcE{*r?LQ@_icKE<-jBRJ`=Dh8!NC
z41-KFz18w{>R9fiOb${l`yh_zuS)N@DRWJ8DWJm%Q=2BC@EAIuYGLppluQ*wMpW@C
zvyRYkS15D$T-IQJzr_U@_}I<VIaEPzOlfWP;ojk)>RTmh4$~=lJCfqy`nM+xo3XFC
zT^PSqI)M+0y$V~BoG2HjeRvm*&T?P=V03~=dw07$NSb*)VR0kxZ`i^04W~*?jk9(M
z=j#KC5;}MY(rOOTeTpHzMcsNw@*zW7gDG~ZrEE&<AhptDFFj{7T{_22aN*V7ybGw_
z<=>Bi;ztEF$r@1lff1=pfI)~Bh;T%BOGk_jZ*PcFtx{Sj_I*_5t<kLkN=WO*JYwd@
zSmxB4ftu(%^IMlKPK^J&!!_S=0e2AnF)Fa%rAF0OvF`-QYle+wD&@qXvkUW11LfP%
zIoyHU8wFOb0J>`bxU9n72G5D`09B<8+*AcF*D2DfIg_D_Y70&!_DhL5*OB>~f5sCi
zug!<yQ&f0-G_V7<fF@X|SGss9<l;a~K6uFS?jka>tn&T?s{gK2l~oe>Tu=2H@Wxh3
zz>_q7hEK=Dq#%l&>Q!0HT`%JGA5Oi9$i!qw`L+)PRVsbG^;P0|smd_@MB(x1ao1-K
z)2<jr#-)KVhv%$?jV?3nQ9_)lD_r=C9>)uEg~v+po^`^JqCFfL%5gCn74iJ_K<2~M
z7e-`_E!7Trt4JN2<Ld7Z$S*vDaqW+hO8wgwk3ZiNxd>r*^u8x@^YJf_IETuqU#}*p
zE*h9g?3}ch`DY33ug&!5%Ib%ug;b6|`5ZqDJ%52y=Dc8dF(f6T@dMYh+)b2>)UN4Q
zUt~iAKkEaLZ#CAQG5-Hu=agEU)6sdc@=2G@8zY5wA}P)ghcSBtwikZ+Sct_(f1ftr
zW8E50>}9gWtJtL*KP^o^*97zM7+X<&j-%n^|9;rM{L=+_Ty<8DYrBcklaA2doFF}Y
zXYgKh;!t-%KlR<aq;biO1~1Jd=55jGy06Vs1*lwb74p%qKdIjQb6v}JYXZJ*8sSz>
z(+-Zd?gU|(8$NF*!y;{?H4wwP?~J6@yKZDAv=!LcX2^b`$+>k)u)Rj?)3>q{!UQ?Q
zsBV$F>s|XTvsoPtLL;30TLRS|3pIqE2nc@8DU6!kP##F%8ZgnjGa=U6r}*&HbnjiR
z7O!_=Boh;6cZ$_<L@ygVlOX3M%}2f^%lQRZP2?`#{X|vAwJ_Pt+7*(N-j&|O<(qd+
zQQirk{ET@Pp2qHJ?rVv=n`4qY2iE;AJ1HapPvlkSGZd{V`eJTArg*``cZ85HJff1&
ze00)-yC|DL7o&cDcg&0bi{QR}&dA+&9dE<}s=oTU$Yl)Lx*EJdPEUQ(s<v;6s)WD-
z1n=<jH}1%G%}+T~jWt|<!h#IQ&%pQ;=bBZ~&LaQ*rQkk7puYGG;%DA;-2HgX1hv}I
zH>O)6y^@ROEZIrh8soiOtcH_)qZieu%dp&vr2yxx8|)+k%t(&-qB7FOz;4MvD$5o`
z9coGyEcSD{KuSHspLJjEE79Y#)EtwO1W;VAPd%nL(i#1P?_D@<aNpjZ&0fZwUvL(g
z$xuW_!fx!g?U~;-@6nKT98Jw>kCLA5@z;LHC0E)c2Vp0<V?a<mFP9_DP`fa<D*Cq4
zd4DAI#*eq<q0e%2yYHRV6q6T=2kGxmQLz@yo%ae##Y~dCFx{wJSWdBdxBJ|>Z@!%@
zIwH31<2EHNtw5Z6P^}br=vhA%(6@I*gro|+?0X`R05{&<y&`HO9GsC-%X=@e2ojBO
zwn;<4U-pJ|+qf?q5|Knc*RAK2>a-i5FDh{3vtjg2>PbNMJKDQsXy((0MNzOI1M|Hf
z5*xRYO&x0BE34UOK&?zq%>=<ysHQa+neNHf+?5G%>9byG!6P8pIme|`kq;{#9^n#1
z=(k*^ywZ8!Vz~LWb`e$gW8tgc#zobQTZz{kogEeyGyF0W|13B(;%d`=E}PE}&SSM|
z?oRhqG@&Ux3?V=ll8WR23iBCuYjDKP2_Bz-ep2iCUwPePSlj=y0B^^6_E%-GGPXwx
znqer^-k$YML67TU>Z0qx@#~?Y(<5F4qK4OhnvZU@G0M{V*k4M@s9?rT>NGdx2g~q8
z9eNPz;z(TFHK%#OTmWato&>d!);TSOm`eWr`nK-!GCB7_KT!kAM$&VZ^QkhhPoJ=C
zH#s!@SJGwv64s%tZs+VDM6e}3U6+ozEsVbNsllQ{e+uUFiN{%xtG9C;(}{jPpJufs
zd_GI(ns9dy-i?P0;4$_REL;=Q;5#5=kK48AY-+AC9mi1eL29Tb(5c6e9yd?q)^s_M
z*gJj!x8otHlb@Ft|G~j+`uw)N-Q&r-o)XK3c3p{v??kFgWGxqG9)Mois{2$YCwq?q
zTf8|<Dz-ANmJ8J}mKuN0+4)8yf(uWJt?)*eFKB9Vcn~S3Y^OPIT;j8i@$|u{C`K*x
z(vKhC(MH$K&BvEbg-%#CnpQD>*RXc^iXw5YVL|-^9x?Dduz90Fw$7el7xB?{^O(wx
zls#wF+OK@3mJ(u^H-+D(`|d+PtPAO!#i<PvE~L#=sUq&Qm_~=ql^64J`Zrmnvl&BZ
zds7y!>Rsm9r7*TIrO8IIR28;&2d|v(l63DP*Kh4i)(0f?r%*K~&|#MxF%RsULMyjK
zSI4$n)e9HDq8BGg`)`?&2?`lI#?x5wM{W>0I=V-Ed^l`u`z}5agDTM7UdRvi;T%im
zhbF3Ji?Y9R7&wMzI=iHvvT%{ULC~Zt;g9Qg{TjpfT==}wMz@)$ulGSb{yj^N;Rwh^
zsWZt9K+v7*SEJa!9TgfZzW>wJmHEuqO`kn!d#Q(h412jREJ^7SInnvd_q8s%T{}7o
zFviRMfguVWF0bp{PVXKPUcY;f(k2?1!vHvZYc8jqAhG9yAN`^Yr6be~ZspHjoX$)9
zcZQT!mwe)lfswA1(A(eAmCqdO9JG+3Dp+6lJ4IxFVR7-N`KqZ5#l$O-12b&wJvls#
zfH=XWaL%HQo_K_IY=AEL<Z+zah}t;Mjqj0U^0S%_Hf2M{nU>62kw4UuEZ)5`kksqF
z{E2ULFjY+%&DZ@J{mAFu)1svQG&uRr22rmcY{vH#X^RZ^^Y+W`+eQ_Iz8W}7qoRIg
zL%l=faq&EaOTs1J@o2yJGHU^(|6F&wo@)~;1c~x_Q1d+g(%|KK5YQ>VPAvaDUP9`Z
zbV+tx_eh+|MTKQ%mkqK+k5|{_ZGqFEW|D;tt&k95aoppxW&%8dg%CJV`|M=x3#-pq
zHZ*sh&@v=H*glx;tum8^Bz?kFrjymSr@xhGD3>w74jZa@K;IuAnVRH2AHHCS2zER-
z^|qdLqd6ok&Ghg4Q@gO3>=}G?6HFJU#Wnvhn|*6xN$<VEdZ222Y5F?be>hmUO0YX)
z4X*H~%+}`<D!F9p4fW(%+%9u0X&#06-Am-Kii-~)o>;Kjx$?30!R5huH@E{a$>ZvK
z3rWwu7$ndw=uzFnFA6e|P`yX)`aP%Ct+PDD<rFs9%9^*-+1~!X8Aq#(SaV>)1~ze{
z+nhxnq2}S=vGFoB=@D1R2x-erp5iajv>x*t-vp}I)0y{GVpC+QX_QZiV6iRY)y-`f
z+xKk_pGQJxe~hh^5N_vq+R=?&CtDwAsjEweXEDYfP5M}ECQ)+9WH%302gvYX@7vrI
zsJ`Y|GWkWxzzKcn_c+8=cCgWVSt%q0i)(lAJh`>AHEFR&&Hv(pzVp1+Csk1w+yykC
zmTcu1bId7<dk}kpo&tsX&nBz7+<%~%ALbXv#K9BTDtq=lINU$%4ciwF>$oiz<d>}D
zx}+s$Pf73Kasf0Pdp^Tj;c_DUlE(RYvaD_a;o1YEtjfz$$y%I5XjrqIFk&``A2`{N
z-awEgy>GE1isYL(uc(Pjd}bFVAYMwS@?x(4&D+Ye?~%a`e7YKgXUaWcopp6z#XHB6
znL3syO4N84>3(-oS)L;x{?6vk`r(L>Dbe`Gx}G01PlbZxEZ-o=0KSEUJrEl=7yapA
zowb1dX9vQ*T!Bkw{pR;}#u$!ko)mc((5%_Q-WB30uI9R5zUJL|DT_aYTH8C9{2Y8j
z;-cP?FNU|CKji%$5#s$xg7Ax!&_}!2C0u{2LyBKLk=KYccppBr3OVf;7kX|mn{D;#
zi&oR`iI0hFNGlVbjgKxr)p=ZAIG$O>kQ`G9artD-kB6PsP~dO79$}>`X$sZ&M!VNW
zu)LM$K5MtpOMUeLW&o{i(aD9Av^S1*MZVXa7Ss6Nl1Le!@Q5W#bJvPX*4lnuY5;a;
ze(EI2HxutqS-baL)i-8VO|(LhwZljYyAjHE^oBCM`W@ttv(4u3M1=LtIxq=c*l<PD
zFEO7FAC!FrElkCyN0K`7myYhN>@TS(!31Q;rzG`^EHiX<pIBHQY!|01!Op7kiE&yA
zD0vYUM&jGwNM96NJ$)evXUeLjx7kR3#L8H<BFs87(neV3z7fINgukReU-{vBq9NyX
zdby0&w%0iglI`XlZzy|zA=tjXHyk;h6h<=^JVN}Ucq9&1GSlyQZwJDety!(6u)p<5
zX?#_v`6dxER$ISxevA$_8h*d4W2c&7gjy(FAtowxK|bh%w+{_uN}70F5_xhy6=E5q
zNsQT8FGT=~Dv#&gR2a`<HvyN`c_oiWV&_JI6~w$4*V5AwUXr|g{pb6p<;6gcA2}Ml
zq@FYav`KX8fjQ2>pM?CWsOi&x?fNDy+Rt#7bQ?EH*G|94&lQ~{5dAGBYjAwFomF_C
zuu4kAy1CvtaBK?g;Ls^?>+{O0c>};Zj9ZX_2Od%@tGDfN<9DZ|!LQF5x1INLV7cn(
z9hd}beQfe)FDh>=d<=lFZ?c_!9A^z}{?+D<N#1O|xnhkTZ?+S=l|OhdJ|wme@GsKy
zXCgBmv5M((io*RNxid>y?v#2^?VYD}z(8_7q7=KFF}($qv4{~jP~c7+zMio>s<5ny
z>v!^FT(fo~s9ej9Hvk#uY$h>ZX-f9qu=Jk2J<d?>6t}Z&{YV>znL9J_x5QW4QxY=H
zWOBA&wSrWKzDHBYo}7}$@Rxn-W$8$tt)%~;KmVJMA#it|FmJAM8CNFz|KaMZ!>Zc4
zuTd^?l~U;zkuK>56#?mzZcw_BZfTGZq#LBWySux)I}cswTL-=G?|Htz^wD$9-h0KI
zV~jcG+O84qXL1&XGiV@go(9o%Nd}TJp0ZeyxF%EL#6?3!#spcftw~{#S4RFXR+ed}
zza)lBHVY)d?2!YYw}4Q6zyW&4I+Hb1`;|(%Ea_mrb@`jNnIbVwl-}=;>Ki+P{CG}w
zx1DE2_}E>8z3nZ{@M4)AT!U${+1b0y(MzrT#dZ)`jjJWZ1Z!T}gN98tRpdtz&o;^H
zanIIK3J8??%5`Zz1y_BtYHN3E1IrjtxmKs9&pX!`?J-ImAoxpuI#LGwhOx3N-+z{>
zqigDy-Nx;}5~F#E<vR91*ZX99dmB2ddHeC(d`q}OeB3BV#y5CbpC2Z2N8}F^OATo6
zoD<3)?rdq$_C8VKbb1tugRg=6b(?nM+y*XBEmnaeDxwsdN+#-QJ(f~lGyY5Gf{~5l
zI>?^RYXR!`UG324*7$HVn^0bEUU>96L`fL)mDMc>dx`nc)`-Q3S3o>Z4e%Bsfl@bC
z3wdhVgnEEMNg3HvCs^zcPficIejv<tclRvKq|3eO5DrU-@rei`m9*i_TFgSAF5eTS
z0JGpH9QR+)3r(f`@|!}}zzB(2-rEW{S_M8dG4?ZV`Zg_ZzJV(|dhWB$8Yd}#%b)E9
zWbMN8Gdw=zf{-u2$rL#zTVIjZZ}FjsIoL)@2KClaUoVu371j?9bhRM^tYuV#3`;G3
zcz^MtAN6n9?4qI)QaZ91f~ra>bSPFMfvAe5dCP^diwXP`q$w@c2MghzrB21h=XS0P
z6!n)p0jXq+d$h+St{@7g&~>#=_23#u>2a_g_zNv&wK;sdigh*l@-Ma&@oRdnAE#qq
zObFvGb4EiW32(IZwB-(*$D3PvP_uqdQj*>ibM}f38&$hbC+)s}^rl@ng^spqW&rm>
zh>yj1oRHVW^VuXhY<-gaw>;Mg)OW&TF@!}=G4pm=COM_7+7RFf)OO@$zt&iq4q>30
zl5pOBXn#r7Q}1LnR_9XO8x&g(e=t{>>Mkm%?TV_1>N2=qkJKgVPtBMQ(il+!`H;?*
zaI_K^eCRj*M6QFxue1`&V<~ibntZ;hlaIv+(iDsh=D)rg6pt6UQ?a#(OA~Wha=XZ<
z5gZO9uzV4r5|j|G8lz)X!aQAW`+|Ft_6-?TMH^!W`Nk>X@$0mtBoe39KhZMuzsXhV
zl?<`mAE7c!npH`8Vm?)!4_CIsl3XG>+U+ibK3%n9{hEG1<t=~al+kpFXmH&}N9TW^
z_;ge`U1HQ8s8zo3>B#s(E*xrqLrY!m<GC2}hIbh&>oW(-o|jnZee`rq+5}E0NT_o`
zx*wnHlU^Q$%yAskBv9On-&N@PRTUw*zt>%G)KWL3dg{!)n(c%bq0HeZTiO1cmXMe6
zZN(eFOxtGlpYk_`j|vCg<MSQxb~LhL2W8kg*O?8c5K~7496`8;jF6;rZE|V&+Cuiq
zl<HJ$+uO!gJXQ-e&-E9^WZrZL&T*IZryFXDo6boapLPLez7fkXtcY5)I(RtMXoC>W
zr!3gHKx&gC1>2_|{t70kJvM-u>_)kW4tWT&>~rI~kV33BKLyqcZEd6A#HkGBSogEp
zn-cD5nQn?(ePC(!bxQ>*flZIK>jgF(i|Kg&7$e=xS@WVDcjiQcgwB-mvy+mnQD?aQ
z#Ar>?QfF#T+^1?&XPobEu$OC{c`9!we}?~b_b%rRw%J*%%mBgI`BYR&C(TTIeNV8K
zQFC?1S<C;10vl`1>a1%pe)gPk))ngGH>8!}Ru3S&YjvW4d{LQYQ>Hrcz|G-i$FV$M
z;3Ank*tVGULh3jCS53wD4YOB{GdL#@Yk)H1K~9UP8Z5kMGW?R+@w{qP?mHT^tL{Sh
z__h`a=e&=bYn&3c+>bRWgnRz1n&KLi*b$c|=C$9Q=tZug)vM*+R|J}O)BzQtKL;B$
zGa`P;@26)6!u>mU|J2Zuy@!z4RT~>TSgLGp43Cpd_Ub}(`QvgGfxWX;jJ2=BMwcen
zQVTAN5ko?Hwq<T7s@IY;{y_BtD%<0u7H*^iRXIspa?J+r_^B+_s&c&M(3txwM>Og)
z3!b1v5{Vmeoh{#cB@RccjVb!b{-n=7wS`1djptNw`fP)*n5pXKjsabL4sHRcDyR6C
z&?lB#<q+EFW0tF{>N0!?tINWa)uO=M54wacmIFk`#s*wocwGg0du%A_Y>#>xY^!^P
zJ}HF#`LP@aI-y3Jw_m+e`wPiw18bj3^$EcKu)G$X>uY&&q}MfClOLIHl5q1QBSjJa
zq(pAI(SS0h`^DGbAhhGvu8~YQ9G(rjJQG4OLSr*Sr_Gx07w*|*1YEgeKD&+8j%dF(
zbB*7pe@IstP1S9@3D&GUMK@DNYU?iLuGdiaZAej$*(kO!Y12Kt4f+{F=w^QHZ%uJe
zEvF`Lyf^B9PpU}*_R3wL@6QB<6W1y-|7yqRv__sl(e#yZ!^Jdn>1#Fu6;qPMemp5N
zNvwP!XHm%gVZ;@ymH8dVyv8M_yI^S&1u0#1AM3+Z`@Q}jIJ6D6i8&u9KRrX?EH&h6
zpdk+pC$d|NE}PojyE|trAlCE;Yjzadq_u0RESB1}h(Up%wWTL`ELzL9&rs~rTm8aR
z%@_$%ZKcInyLtACp8~!ROWoeG$L=Db723rXwvtw4dzR7P#n7vsmV5Ck_S_XoF!$@-
z31@n+*pEOhfIJOOgSPHZh_lX+dssta)OoprpfB7PBgKWObm1Fw>UnG!`&H~HsCYsN
zlgde{4WKTauuXDYcy(K!OpM?#{ljS$X~q`cY7u9#@&3(Thu|Lmx8(JunDv5UjadqV
zv2_Qg(Xi=~>s+f&Bv=5ro%}CT$AWDfG#E+P+%w@15O?XZr%(Ltoy}|d0}>GD9F(bk
zw`)+G_6Y^k;`c@>OZcPmd8xK;IG^oO5vU(Dx|!H8zkR#XY07SCYy8nxh4uy;MSF6t
zK#jA?Yt0V{u)oj<JPGG6s>gN%+P%UO$EwUvO|^-D;<;4%wE-<&cX47Jt_+V2KSCu<
zE`iRc?_-*N*K$HGRo5VmW#3O;x6q_0vG7E6#&X+n!$ik?VQslevdRJkD@YG+g4Jg>
zUKe;il``LgyQJ65RPl}2r@Di;iT@Da`vVniP45jBp`;z;pUn!lmW>z*j{0Ml&rNN(
z{LClSXha;H_xUt%2E00wF^UKGs=3QaKq29*av=*)$dyBVNZL?p&My)VxCYOG?Xbzt
zc%anmDsylClo<rVHz3fH1+%g->chY+{*76_O7#siCR(NFi|*S-hAVeMW2~3p;((a^
z;#wG+i^X2NiPR8CbkRDfn==8Z@7>}MvSQMEu1SM0W+A1D=3vB^PdGJI$_`ac`nJ;q
zZ%!x7p}z(o$+9wy^EhEG@J~5v;6pM@Z3gB2k%PqTFChGei@Fb5Lv*gNU$x`!vEzI$
zG_Bh>P1(iLb$Dz2-(*0HP#R`78&TcFlQ5mLINB25JJVG|d)5Rrs|4$(e^$bO$sksF
zhU2K`M$8_lU7Rl8C^qjd!k^<y3aKzfPMmYpjAa5PtL%X1IXZK1Ed7PYm=P;Fxl6lT
zp-azVoa;(g!J=OvDk%jwdO2Xb{<vl+aieZyx!(>T{}|a!H+;lG$Nk`{fjS`Jm^RFu
zCmj?_r0!o;<?9!k{1U%znIsB`AzF%sIugpALh6tF{(YJwo!FBho4s{DHTvy<IaW+R
z<oQyqr(fu+F!a=6=C^Gf3*H?I2?k&U%uvMmn+_IJTMAOgu~vqPN(X524y{hNM8UIq
z`Km?!4R4(PJ*N*4LPrmeqqZ+Dds)>*)8D>*tFtv~&RiCT`dv)K?@?1*PjyT6k9;+_
zPojSGwsiaqQGjp2GHW(<g=b8+%SJW|i{o{f!D3|_c_n=qI?W+#&COoeVco$4iAgp2
zi=K{&Qz)NJ#~A)5Z`U@_^GJ{2&0#9Y$bag1Er^(aiS?__+|0oM$?DM%J7vR5rL;_6
z(R?9QrDKO@5V539nQs?-G^?!Cb%VyAu>ustj6`{+aSfGbxO7Z(f<-%dJ^m<X<bQ*h
zy%$x$hvY>K=Zo9sX>_MgVj@*`5>kuqdihR6*%zN2Q5>+~1y=F$$~f?EQZW;7pQd0A
z*s%KsC5ROp8<jty!50og*QM<0{V>?{@xtNdVV|z1bE5F@{HAhD;m>mQX+`+WjW3oK
zH=?(w<|JEl!)^qr`r4*aarB;BpxVT@fBmWSp#~E|`<JBhqymxb(&6AEebO1iyYvaI
z?J*ouYs3@lkxwCL%eUDs@%-f>C^USV)s>s0<!%V5xVc|I?W(xM@(tr2G5`)ZG1J@e
zG)x8zFnvD4zP?zB0wz9R_dMFEz{$Dci8}0CRX^bg!JE9+09fU@(<;tYHSQLE>~`cQ
z8Na$)b~p)Vt|+Rtu7gG8dV4&PV;Q(Dd{wR0HE)oArwuKCcP_3C4QO(fK@{s|&sWCL
zT$6#cR5K2v2LH+g{*8|-FEGHeLm=i8khFaQ5WXSfr2x2ofQSEhJtN4DI)%if2G9>?
zh-RyCxKh@4R&axo_s?t)(E4(asmo(+=Y31ecdZJDZx9Y29q#_30M7=<zNJOB;?5(T
zeJAA3!zH}}BIO)wPj-HG&*xB9xU$^>r|?Vd)J*etV01Yu?C98~@-jpUCvAGrt^1=u
z>7~laesxLqut66I2}!AXN!>PI71`Zs_b1@ZPkn{drE{SdlW}u>Z@^7$wj3~2*<U}G
zOWJSx#V|Qi19_KV^hX=XH#*Gt!@56EL=6z=EsXtK`=URj(<LtC?C9^4&;d=WMoOO8
zTAlli?XoZ@3az}*($Z-^=LhABFkIHp4weFm4j^v#xAR(eO9s-})ukF7Yw{zwzgW%q
zb<KKLaMz)=yCiz)o$wKncr1G&uy!*n97}leUP^mYNimZEEp2&q0pXR!Cc#NbSTA8A
zT_1&3>sYGVY9^1xZO&mvk@36^vV^Jq-n`4Hs|z})p4@^P(lRg-NTVM-zRZj|CgKR%
z>tkuKx8rEu42h~0dI;Jx;(t}>kkGN;91V_cEZ*J--|%jD^M(7UCN8N(K<N?Zw%HQ-
z2i;W3HO(!mX||6MabNy>=4({9pL2a*Q19OcwA{v9wI@yUcmm)9gLI&Dq_)H10#Dje
ze}w?&8^tFfq@z^50YJMaq_*)LC`Rg&4^S!YwGN~ia@X*lvrK0*5?pVKlzCr$MN`1I
zooo>JU;qb{*VN9Y;tN7olS3}s0%gAxe^$@Tq^5MB%S)zAY>nEEsm<_T(Rm|W^(8Qc
zosPAYOg`XKncn}c#uDvz{ok47BU!zW_4}H-2r;>u`b&JcM*8^c5cpi9k-$S2;k%A}
zzgo-q#)bm{TcbXBU2qqdcgN(|7k#Z$ZSiOg3*QqPnq71qkqLx>qDD?Z7p!hw8~WU_
zM*K{j{~QU?4F@5h<+qf=irQqFxMfJ#q~5KUu?s`$_w(Y5*+*-7JXIblN};{Q@(fn%
zLl@DiGfbxU88>DqVM$S~g@FMT+gryyNlIL-^ol|K*OD39oyEV&JBtn1qt*ldsLlQV
z1xUZHPhu;D68tIp2BL8?HLOK*DyXf1#^}5@v(kb3h!#k*`v{Tu-I`IL)XzKz*))#-
zWaR^uDjUGomc4Fts(P%AlN|BY7T5)6^+X&Y018H2L=EW<ag>CSPG1QM!@%oUbZ&bG
z^~?vI0;1(Y^J^=U@xWvQhp#gi<vXD@J708+n)3vmER3h(x0XCjVyC0UoEk|05ucj6
zokKQ@SL%<9(ExJfOBz)Q05$N1qpks}yC`IR(>B*J?)2Oy7LC&FCpD(2aKFG{Q}q09
zRA@wu!ac!5Dku*nD0v{%$9^16&S+p_S$5uHfihYG#Qx0DGb{Nno~tD20PV8J*`~PI
zZW;2mAy4gy-l>v)wJ&2kewqg!e-aBuF^AdT(2xaq4UJw@SQ<>!Aey_BoOlGaLZv{6
z?=DqbT^^6*cZ+S8Urbj!Qx5<YdfEgYKrUZ%^9HXSK`vsWBnwa+?po9BD|gk&F7Atp
z>}>T9deCW;`2Are39UN}NppK*{7QxBnK8lq2^z<s7RfX^N)^I|5?!~2`0~f9h?9sL
z7yNaYaD@xzJ8EhEHiUC3!LQofP&{0nw&(QA2C89klf($=zxe-}J_&-5ED7<z>{%Tt
zBL>_#xwIidMj@S%g=x7Qm!ENMyL(Tq)(>uR|BW|d7zUBO`44~__G^Pu7yY6Prnie_
z(d281BShOnGbu)0mXtu4`?1*EKiA;ke|i~r(UCQd7hAb4a`9+z$Z5;k;PjMcoru^L
zC;@e6$D*J>j#@iGzj8X9@Wv;BgmJGA+7YcXR5ST65&o4?b?i#i;8p4bb5pfQ;j{6r
z9o|q7de!)>KmG|G_HX`zg_RdSaj>7ddkeHaTL<<1_Nfv7U%x*+Lpm{5hm4drU>8b2
ze=71{dU#D)09?`y1i6j{1Sg70#Dj8G1uO+yp1U2BL{}R&!|vlYA>W^|@$R11>U#-u
zZ`l|Ne>o5dp5a{`bqM#$l&QSbX%z_wNi8Z`S}^{+Z~=_{Y`{JoJ<I+s&HQu9OyA+H
zGpH`fQ-(ZnHPip}B;k){w&c=d$@!En0;Q?(oXXMB$_gBs^3vI#r?{)N57!jdt(BE#
zb#=@nG&G1hIy$)Wk?$E~v&W~nYjqFzG*Wmt<&PEyx_h!L=>NpecmKuDe`#h@UF=Hb
zbc?<j$)>1*khX=7qitd&gPSg2(>0`3aevOL$PoXAm+pnBHc7^Y!dUjCmxt~`vJC|x
z>hjC-$ZBsCqjI0UvFK-;JvHI2GZWb@e-{b-{}zeb96tPWw_f7R4qlsqa6IdXy1mak
ztS8wvH1dU@IcPcW_@nP4r7ayDna9=<emq|dI=UPZx%hoRr%;0mynFz_h`BK8@x|cn
z_S4F@l{7ca4p1z&T?{z-gTx5=#D~!elAZka=npotD%<D6aUUARe<$;80NqhfSEj*C
zfI7XPr^IpiM%*ohd3c%q?c3sx(wpAX*eY1H0{7nX)gAd^hyD>VEdftgkN9dPgT|`o
zCbx`qsN!gPpy*M^MP&ho_t2=_!lJtI<c;cs#G?2gF#Bk4U)~Ou%F~q0r;SyrnFJ2g
z*=h^5a1o+{7msvlsz1!TTnyWT3SdYoCCed{NU9GdrPnKc%$OOk)^Wcmd-k#7p+}j+
ztu+yJ@0og{dN6=5iVWAs*M33<NY9L}u|6W45wH-si@CW;xJuSYa<&WrQEuh&L6EbB
z`I=nJ`UA`#!!cBs_GS6c3-G6q!1MaE5v#5yl)yc_v0lhl#pLQG<ueL}#t&wCFL@EA
zlFd9b<fXq}?SoxP>f|b%K9M%r76g+Oo{%e~2P@{D9qpaJ7U1LY&X!PL(KUl~$KB-?
z>&Rg4I8-DXLo>TXLN?b93!|JO92G<{T;1xPHeIqeWZBGyJdDtm=zl9M6qQ$as2nN>
zWVRyn<yR%LX}uVk<5)|TdM_&<Pv~y0b(DrT<gAGHk~2^QYBt6_ewSG+mNlvOIYu`>
zzJBXw0`(O6M>T*s`cn7u@?QgtPFiY_ia<QgSUkKmkN%!YM=Li6g*t4p+CR>(MFzk0
z0G@L~{z`;IBtT_j*UwM(m(qXPI#8l3scMFh*LY~u2{U7;4<*qcMrj$DwF6NZ$WK`v
zIbs(;^at}FVjaG~a&Ql1BGa9U<o2Z)!D>Dlx4JBKAdkn~_p|+wpsT90d*sxJ|4wvl
zUAB`ZAQHrxy~{j8k>yRpi5hNn{##{*Lzo`1@-tO&6yafD{Cd6`d}-MuzJRoD7Fv9{
zI3&2g5*aYP!MT_S(cN8mqKkxhQZfo;I?1+n^^fYDZ~@;zde(C;MbWPmVs$~fxV$k#
zOa`^O_njGPqYdHJMfyabb`FdJI$w8{V#orFHipaSAa1<v*oUYT)CT;j{O8Re>#Efw
zY9B0lcoZW{8y3S%S$lg}BRYkTK%pN3Tv;Jt`%fGZKHTDG`u}$DcVy%spnS3NA_i?%
zl8wSFRo?TOksK2(Gl?>TQu<3<{@{D^;J^_vKNJcYQcOSs2zYC@;)Q{|1FY=1XG)y~
z<w&?U9oP{{=~+*0c378ZSv~eNpuBO4{bVO68o)+EXPGBFYWu?tBIZ1Ek-b6vXZn2q
zn?4yZeL(IW1-a^VYk}3qr5OMaVX+v#{@Dl>!3Fk}`-0(E>O%%bi=8dT*tc*+Hpbqf
z`PtI+suygq>+i><yCrDysDVepWpsTekd!Dcy~IeLz$q@P_C`6zP-&V2U?ER7^(W&w
znA;D5L_X!*B@FxFuMq{oPP#i)>;>CQJ!D)fPvC^GFc;CLrcw=5eb(Bho^8G?4-}Fj
zKCGxeGMmlK5PJdpR@RHZ@!@%SSb8}x+!g0G=0HSJ_!D>k7>8<IOhFRz+b6{`DZkU-
z=p3#Egc}QhB}JpADA6r0MY40w3fS7vBy-cuf_KeX{m|C-<Tt}V0WSV8wjwb~2Ywnj
zuu3$R)eJ2Kb3OJG=6VSLqaT<UhXALFcUzU;%iUd2#CgJZ`~>)5Fbj%v49mR<?u=1x
zD)V1$&B>I_w799UGnW3Ah<cAqd_e?2*qC(0pX=29-*uiy%m5z-1=$W0uD^ElbYwl}
z4O+m5Hg9#fQ=;oRPhaD5n7S{K+wg{;0UHh&TZ%1C+nT1r;!o|tZhP*y&yzAaP<j_)
z9>cLpF>`@apz1RI>hv0+A&xb%Xcrp}=6r0>rVyu17{=Q8nfP3Et{<J?&Iv^rsv&UU
zq3d9J9fA}0%dNc>7VN#UH2!a7Rq`_$^u^5nL3sX6>&1SjbDCVDpP199z*-NF6h>gV
zS+Fw!c~(=(Eevjv5NL2mfKp_*Q0VGh74G9;c&~K@{&Ns5uoV;Je6{dzNBuEET4GXx
zT#!6flO5us^Ce65bj;MGxm4~8Ym4I@xxJ;D+xTy0@oN27gw{v%5X8!YCnv+H^B@tA
zORxVKHrg>(b@zw+e4Y*-7g#l_thAnjLRc@MW@wy9lTS7rG9IMErIR|KWo4QHXddvO
zn&{KdB{nN#9msC?=reVmy2}=;C44=(*~h^quf@is<_5an9(7&$8c^4{8clq~&K<7M
zs?Yqzhu9|ejA3F9VThfnGj|f7hmYc4{X#yViENa1dv*`xToQHtxvQPOaOcCA3QBBQ
zOQyPcnYY89CeM$(vMHucBP0mtb#LkKj|kcz2wqlh%-h{A(1~#ZYML=ofE;U!L8D*D
z)Gt@qPYL9057GSPO$YPmL@z&Y;N{8!c^i@u{F<LX%2ui`6WA(}WqIEY42}8Rn;$k;
zi!hii*0lu`QYZ?PKitZP|MKSZ@|n)S`<9sj0QR2SI!^?94Y?zBe93Gt0UX3DaoS5L
ztuu-yI87nyewdwt#<QH?1^M0J<b<xt_U8h;gt*g;vm+@g+OEwIVm?|$6M0y<$lvls
zRXKqWOcth5$REz!G~Q)oDPLX+mIt6R`>$!>IVRa()09kZ_SxyB*}mW&!ERU##z62O
z_foAh!Em$h5Y`6NJUKO#z`?&%Rj<Ud264C<ahUz>mvgk6?AhdPmrafuJ31<nVptaU
z^-1`aQ>xr(_qNz-d2BhqO~_~KS6T^hPJ!~($8_0I+tWIx{JA@JQf6-FClbOk7`6rK
zpAqE$sMI~LeDfYMxSz@((43R(yh^k>M&miEe#W`FS&SZ{v0QfmOJwpNh5))%xnEGv
z@j71`44at{?(n>Ib3e3&<EgFBvJS$7*Tws0?*ju0?uV#8JnGh}_@EjtB`QlB98{yd
zJ@#>vhbSS5Ol$S#pUZ=HF$ID)N%BK~R)}5~1q%!x{(CGfojWcbQycu1+k7<r5T}pM
z6jb;_iQK|}WxWDz#&20zGL++poX;U74XREJ;{EFGyHW7Kkpv2s$|-p2vFG*SgN+{?
zsCN!-?RJMFfH9`REdP_p4-yi#Iwn!{hs#BX|Mlef`3B?*5N7nL+Z|3)`4_LPeOX#k
zgErjkMP?KgxsUx;`au^zw1qF9#U_W(4hMvG>R*|u2Lo=Ei2-k6evCcZ?x({5>MC+1
z{Rs|g{{;s$RvJ~|CYeop!CXNZjL)$FZY&DMryi*pmJ%aINROoXKU}<dU^JrSf0o1N
zuKry8%k^EmYD+Tr+c`B1K~Q=4kUycdD((ATs;=AbsT?(r)O%N=P04$SreQ)a4Q?q}
zXAGmDC-_d-Jasc81)Xn!&;-8m*?>q;rpk604HRMiea0gMJ_QbRP@3jio+!dKtwl{S
zoovR5iI%@iBw_{s5zpD+J1T7V+;;VK_ftl|bBc<sn}L8kbA}^SsxWUn&sX4h{RGhP
zuoyhgRJxDy9v9LIR&U_g&glUFmdrCrfa?;?z7$=lz4&6fskW>pPXY=Al7urY>7~M-
zU|vOL6fhR?>bwS{HqT+2)Z9o#M(-b$i0SWmNrJJBK9*(IAvJ&0!(RZzMs4j!$HUsU
zH3rmfx0fV?jd!dtYw;X}GN5I+@qR6YSnnk^p~{}4^EVv8625>UM-?!$dx<~{M!$wi
zUVJ-i5dp{Ji5^|8f}%RbN9)}k&fvX!LL_6Sj=Iwq@Iq1_<(|^KA+`-r=xt8ORf3Fm
zf-)yYEfF@R=#e+N2^69V?8E&47`9}(B;jS1eLEmpP~)<WzI@J4zyt6F#m3AFxH&ww
z!n~1hvV}j<;20G{mUx$35+G}vdZEv4@pR>o_7bQY<^{e6AfxD=r5au?a=to*ih-qg
zI^zKBgI+(ozxaeWg}hkJ)cInf0{_>BLwAMm4_iz<{h;n?oyilNI-V+PY%pa>a5nR!
zr9gxGN8dhGSMO`OgC{Sr?X1tbWSPRvN=@#++`8BubzeC3@<@TCv*>ot)Zlav7l8W1
z2z*sCkr#xA_cCPw0|Ucc5<Z23<;?~aUmQ;~xcuAu{OYe7{kfFSxbJTx!Lhih2IWh7
z$nb>8`IsEI-xHsOL6$*#i^!XOC7#|*sn;Mq$u+3?#(0-Y4j#*dhDS8FNF}xdsSyrf
zyy4_lW)272)~ONP^dk2kzh^%p?5|!IYyJzMTW8_L*S_9T`<?X=<9?$BAYWQ-_i39Y
zUdwGhDNLnXQ)sFI?mgL2g?UCcMhc+CmW=8!Rt%XU1r{;KfjBS>2sY5hBh=Clf)7;(
zQ>7ANCQ><LMINWR59)`_N*wmdK<kkddJLmO+@%jCCzx52aG~dic||v0OX}Q!GJWIW
znTaRu>FL>Up&&nL>PgU2dMy2W;Ac&j#SnBjdTSY#N?B$+8GNzmmc*#rG4_p<Yt}I%
zgC>g?Ly&}<8y_AX{`7Qhk>|b_oK*qrt}~iLLqpGlGITT{SZI~tpo-+J`d9bcE8CvV
zJjDzFsY^pe%<~#b-rl-tvYpr)n-<%Lg`BmfSd`Czh74I}XjkmQ3Z$W30%`N<>1-en
zKWZAQ@PMgC3{4v2!=Ty9WlmIAuVnhg(5Aih1Vh&69K-?eS_X<d;G;Sne>+V0gwy?+
z&yt-<`v9dXA;^i8c%y*hK54E5G)6>)Xdd=PH0c5}S6Zi7<6<Re4yDQm$-8d<MU6Dz
zf&}h?we<^ZMcwvvF_D1va&45Z!9o2s{bryIP}&89g<^dgrGIR);89H@6}ern7!m2l
z8>4QFJ^Xue7YLs&;FLEctLYONxsg?Y%hOJH`^wtEK}}Y!x*K+^fmpbAy(y(fxR;Oc
z;8$fkGdq1POx7W7YLIqdvb)NKFS5y!OTn|UvKlffhl`Gm{!AH_bw@O{)F1ie(5=}d
z<c8h8w4|h~r{{6g++3Z4i%&eeInn+7y~#{jHVo>1{y9_p<sqh{dGGBZ^EuS>2c$|D
z_8UAA=e1u1)<~g6KP&;gk<Fm|tTDXr4Dm~VL=baK?g)<l3q~54gaq`T%rycfT2=;p
z%V*%U6EvQ9aw`-&&3nfAy82dNu-XN&yj?`{;aC74The|O7%O0H-BO^bb-cHE#c&^Y
zby7%S30%2sHmXdg0#L*PXZb%qRgzID9Rs7*gEKLXqPr-gob&rU6M}6{`^P}nN4;b5
zX{p<@YR~NWXZ2z;4vyz;CX;nx4gppjgROngk1@wrS?=5%(9*p~9!5m`p*Q%>Si<^3
zXvokt8o~s+0zf0fh|%TV4V&GgzJ2_osoEQZp?gE16UA_^TRuC1<YenSNG>^lUE05<
z_a<*r#<6sECjt%_3WU-W&&R3E=zNFoyAj|_l<zRT9n#sA%zV}KI%5ytzODE@s5;A7
zOnb8|i4~l9>Z*$bEUJwz{GPvlPsw&}tm~CJ-~Ue46gU1?qQ931jU4Z?ijf1yYNbi&
z@m){DSzFqYkTjeVI|FcKCN`Lxc6QA*^c4dW9MC;pM2PnH9A<>ZB+7b6M5>?e8^Zw=
zO?z=AWlfJff{Gp#qCu^df}euWJv4VSC$HlK@ZzO@Wg0k+5))c&efB6S$y~Z7^Zqxm
z;(0{SYCQp~gbYSxzZ-Z{nA%KmqHsN1S`R_)1atB(q)|(=krx=irz*cHV=7QTlwa!q
zunRw??%v*wopd?4Z+I82@QVErZJ<$Ps;2zIu#xi7FnG*-&2y>%g}}ijv|@`R$_t)z
z6t1DF&fGOU{SuqATiMS2#fsD|>7w9?q4tb9h*mZt12!Q~`9`MldWf4}{?)cFoC*pQ
zLoTjnJ=o4{1Sw<yy-Qq7Z!^|ALS3U;{V7j}x4S10V0D2(<un4`1X{+*es}*)ri`ou
zWD%I7#MWinyYIJ!n#&Vl!JDo#qOayu9(P70+nu|3nwY?o^u3N2R#8}OZ1|{Wb?kX`
z5K!;>Yu>un3N#}0(V9db@7tOXs^u21j36Z5#u;PZ6%~xe>)kC-+0}uqA-U=brv;1x
z!27Le>;JHiSU<24z-07~F?m<T259OOg9lUWTUm0b5u!mC?iZYOr-lQGx4CVM;FVV}
z7zb!JkZiZqU@jO~Ssopc?PPE5ip=%v{bj?MAQbW|Z~!tikL6n3f7W~7);ZU)&@FMH
zhUO27o{#U&bqw?Z^((P$23C2G;K6LR``_W)pG1Vv6uu)QSZsXcz5$=HGHl5stjH%d
zHQj$nPb}PYePMA%<mDYs@uWim7-@du*hXe0y8z8TMPYaN`gqWQ(clT98-WD)eEh|t
z)XhPw(P`oV9-!VprdB@LMpEOh`M%q!xIC3bgG!_hON?<zVvahLGEo;{c-ENi*NJp7
zY;NemHa(5lmhzkpja!@LSGsaC5yx=tZlyNno?@e}Aghs59^ovQI?3z(H*$XjC(7{J
ze(B1{R+?X#!58m<yVt6ONwm%tga|NXRh6y;yzujA3{2&=?GyE^$J+Y-H7-G12SHFv
zUcfVGHbXqq`RpTfjeXqYWXc}fiy^}laQ2m9{;xM9F|7vW-oU`~A%FNznglpAVcs3?
zH6~yk#N@lod3|73ygZ?U1Kj+unie1pPv;nYr;HmwfuTg{N21456z<?Opxhkhv)CDP
z{wOMqd&+ZVmv4CV_}ktYNB``p79_`kV5&ImZ^?g(qtd4zrdjYl2@=THMK8~Zhs*+8
zno%&g^lTkipniaX|0nTQ_5ahgzlQycM0S@?*nst+nE1!!<8L3607Ia*E_86b)N)uG
zPO0{@2@%{bSHB1{x0v%crQ#>%I?itb6p+sP0%FTBWeT%i5V~ca3#+Srqm({%rs=7X
zv<70)3u1HyT@(#L8Na}1?M+iB#Iq6{?At*=CxdkYSfBl?-2DR9%lBp3ui6DZ9?yoR
zLcV)m|D__*B9a~!n|L%!h^4t}^O@Ma*W$C+oFsTIE&?Wd>IQqG*$VkncTga`iv$F=
zjm=;RFwZ_)&*?Ax`2g(c{}(VCt#|5V5aMXyd@?G2gLJmLQ>}xt@S`l%GoN~<t(qI{
zouiRXu`zD89qLd?mRE>?oz*Xv!L+S|(%Ktmnw5jy_(WOjL3+@!cw0x%w=cGk)7sTn
z(1++af5@w-3W^SRH{Vi=r7~mb)B6sH+7!p#=pQ68o*#d1fu{wWHB{|95p@~;7W6~_
zA8mVcI#IuhO=nkIo?1098Bul(l2N3sNPuc%Dglsb13<eGCD$Hp<;zXRO4<bsrqVj=
zEESda-Oa_-&wru%Uxj8O;Ja8b`;hA;-%o}ua&wBt!u$c#^DYh0q*Mdg3|XxSmdNQ!
zXP<%1(%kUg_a$0i?F|<=a(J$E6ukEWG+p&9T9<pLj$83kx*vu1zM*JK&NM%2#H^z~
zGp?z78567eNFKWR9X6P4t)Cv2OB?AHNM;lCcV6!r>&X%^3eB}X#T9ruT#rvzvC_^?
z?)B5=bq|PAke$2)Lc?Tr#qAyo(@S%BtxcuMDH}w6JGihk`fZ+0b<mUcNR;QYZ=wN}
zx*U&;79S&^!cf{>;As)xvkxVrl1ABRcl%4-Tu{|J{77|{!GsxN8&pX<jkGRkz0lRP
zq1wz1%A|AD&j>tqop=BV`mgIfg(o*<asw09MaSkRvo;bFMAKWErxUu0ashzWGTP6v
zE~q`^^i{dk03}Af-km~_;<07JS40U`50?9nj^(x(Ell!fJQdS2LxSp&x}VB`%%*K%
zlxAJ9mNf!ZdstNVIDhJ{_Fik7>m|3lB75_77;pmU#&fD@O*xXx$P#(o-QsUJS6=^&
zx^DsS8b+#y^cS>SemfYjTtJnEvIh++pvCU9o0X$^&|0oZ97{vAfC@NL-O-nY?1sY#
z@B+F(-b{bXP&-AG`86Vj;JwZi?q2+y#nt3}Eu0X!o_+b&hoR~;3o=3zl%>*Li}I}_
zcIz9TDmG$wU=w4x7W$Gdgl@OWQ9K!vV&dlmjz=(q-!g)y2&|wrm`+KOEax+ct=#uk
z$tPJ^F$0BQP>XM~yRrwJ2Jh2bQnd96cx$zd<PlA^`G9r@3z=ls&pH(wTh{N8w(qXV
z;ehE{w(|Wb9<LTkGU!wUO&nxA7Tgk!^##_6N08aW^<0VD5EI?uz0GcMu_qDF>gbYV
z9EN*NAEz6%UvFcyn5lgk?i?(!?*Fg=HPX#j8J^7<)g_o}2m_|p<1L=|pyAj9{K^CE
zx5xM+L)m})qMr`gR)v4c=SDzk?WBT|6r?|i4IDxk_^Q!K0!JytuolOCH`-S`YfPh!
z(*X@0GTB$14ac~%(91|zXhb`1ZFt{lHJL@JLVpm0_zv{_iV^4<eE9&{UO0-~;DIk$
z3euVJ%ZbQ6(9H$mT~vcqvTTHNfXZfr6b7rkqK|+9+wTn^w#VA0HMUfX9Ht6QHGUxz
zwleE9DY&Bd<eW^dB>cEJlBIkD?R%!djAX3xZLhN$-OX|@YV+x_re~^w;m|U!N=US4
zwM#dmP5V3#OxtUl=NK3OD$>$0E~O)(5$tL(n*;s6y1Ra5l;~}>Lnts6whWjxxAnaQ
z_VL_g(RmZ7?jxE;m%=UVkm**h8ZFJ9JM~YVUn^ls``EE*O!*}PPi#eQgU(@^^&$mF
zkqcSWeWFxru&&H@Q40s~oBWKr&@o(nH>v>bTyXJaS9V1qfa-latT#;gddT!O?cdj`
zzf#<hEPIbOJ9wL~a{L~rCMTC5d_Ck96~p!ZfNQ$F;PggffuAVrb(sWLa?_PsR8&fR
z5+q*#pRjQ5%>!jb<p-ro&20xi;X&Fo8ZdZ!G=6FsA28>sszoWV_$*U7DU*g)){r*O
zaM}$|AcMGNI+OFo4cpN?V)$FUAk2f-kr5yAd@lGR^WoB85m+&xe<Q!N%P;U9=y=?}
z6T}S0ziU&?`b_;AdhK3UR;qVMv48tMwbCslB8ae+qVDr@Rjsd7Tl+lXYsf2MVQ}c|
z+sks7OP0pn*miX|Rc!tmspwYq!Tb-#UlCPDAK~1r)#`0JFxNWQ>7&tpNFN@76eFCv
z>d1J5PY*6}S6!?VmNjgWi3VC#*_-fIoJ%4TF<&{RtEXV?JB?&O)YsAz*0*Lt`WF_Y
zGjtaV2enx6aIU&?F$bg@7tT%93oz{R$By*S?HQ|9R_Nd13tyIyA)mcwpa@6HL?t6V
zTR9V}dbxjC<x8Gl8x~;^j~>^^<aER2^7ICa7pW=B-#m3#Wz06<0q(JTVZeIz{ZyO4
z>r6PRtXWZsnvtVxhS!KvgA+-Rqbr`OQV>@Z-g)!JMSkYGPneFA$z}c0=2L2Y-->gZ
zPQHobcFKTy%ZMq>XfhXQKubf<#QLvUduXv7C0&m;gk6BDM}*CRznV&a>U!S-`70NZ
z<28J|dObCPvEf(GxJ-VnH&wd=--w}}-OTLxW%xo#6mEzS&&q|I`Vm&mzPfry^y)ib
z*slqPH!>B<8gJP{#bXz*2(;R|+tZvM<>T)kV>h<+%tZQ)A6nQ#3(3v(S$-8CPU?T+
zDi3lOqQhm`@aF0p`NkQO?F<=6>mk-a@kk^;&mYfOn)Vr5^xXULMX#Q}_#lkM6PP>v
zqZ)*awMC{zqWn=b4Noe3m<Vg#9DkHqBmQeP{PUiad|W%@Vbmt0{)}3jnx)L6#z7f9
zx%0#_92zHxf={K*aayM?KYWq>`Kd8|k8Mm)5U;sgEzxIcdM~=Kcbj%U<M8ARZ^3N*
zbaU&9H22YF`88Z7p-AMatry`S5zP}`p?*Nl%^1K+{OvWTqtiM*PEeaqp;o_d0#%<#
z>S6LzWFpn8z9G4vPJaGeUI~Hx^W=Tm787QIk7tlG+SM7fC`U`bEZpdsI+dzcAIW`7
zZ~teJF>$G2wQr#T63t+FZu(x<8QIr1Kid<zfAUooX85A2pJ=+yX5FzDog1c$$ouqJ
zj4vIf+q=#23#@2}6zmI^R;&sW6_Zq}J#;tQqyf($T7KBy*sW!lFyB^_*sH!MHiz$Q
zsTdat-gss>?603O*lxg5WOM#0IL|eTN!>Dib}c<gzZLtA(@u+#{(!IUd|I>&Yol`M
zAq!>|{%(Txz0R$**yfXPlg2#R_gx(>vM$%TEU&^p%%}&xh2^c3{D@%F(Q4vCmxgB%
zDiS?0^Lfz=@$>w=qPN41lmlYmipezH6wG{KX~z6|?f4Q;H8}~>hCj^?++BXT*9f;7
zF(|9MZ%;a{*?j-3gudipw87-o3XahUTQ(HeH$_}HrojvB{Evs19*ukl9TSH_vMSO2
zMV1UFbF&bdy?UqV_4ch9R0s6@uV(>PrifC)#)iEWjabquI<B(;aX*`-JHQ(YK!>Uq
zs#Bo~<Cqa$N^1l-N_m}sOwW<ksxyJq;S3_LG&^N&GLIbDxDZX6yGVBT0jDNa?JXYF
zfE2}Yo6z~)HA}Cixp>R**(CzodU(?L%DSRuV8E_OA<fs`BuMiI6toz&G$=Ke5l*2=
zt)<{ywU7&<#8P{RYv;&FEZ71?w?X6|@9#)(EnjQ}K;Gh^U~Hya9Z#v8(?)j%u*ZfG
z)F(7`n1TUW5Gy@@sB)SSmUP8}d9Vxz=Z?nQ;mn<->>$cy|9f!qynC3Nqs-bSBDdG>
zfb|!rG5y;+|ApUgG?w}jlQ>H^H^oxMaDsejI*jRP9qdmsk=Y=*yO9cP;to(nJBBfE
z-ux>#k&o{`f2c+IGQ79_CVuKui+|8MCnPDc`B9L{$q^Fs){w4rL@|mFQ;RJ$NF_gA
zyHiGQ{mgvg{NS=;PfD$gI+$?AXr#g1mXQMWXLwA2UAV?nEuuVkuv^@FB_(c)(;=*Y
zz-K!Wi!%fDJ}2#|7Oq$4tRO&2w(Z#YDczx3oZXpf8}=9ZYHF|P7U~*r+IuoJ%=r5y
z^z6-pb9z-gHA;EamU|%t3#GwK^+1})-ye2k@gT{+TAx+qEVD*grpZKau;Q{fV=XBu
z)}5-F#9>@sceD=q<9Q&bdhj{0S0m{MKVF*utr)ao&%TfH&ZLhqyrb3$;plG1y6q$3
z3NUoM+4zD}GotfD2%Z`QHx<v^Ok(&0r?8(P4GQ~p;5du=Lqg$7SG4EF=DfbP1<Nfl
zpGro;rE^)Us&1aorIIE%p`=S})`EW^Z|@fud&58hU)hAJG&7Nhm}eKh&?cDonr&ms
zRA#B(dC+0}C~?#v{|LFk(+eG6Bshd5OJsKDQ=!5oi$Py6bJPnr`T+P@MXS>xg%P<G
zpd@#+_v14(3VGQQe~RL3rP+r#r>SueX7d=Rx2Bc`J&B!?!H5u3u_9QwZ)TM$i2R4e
zsCid2Z=+@V6RL@3`*G`R_b0JrN`7p8c7{6m4_iZb8eA7dK53;lL6Q%i-PD0k9+j~H
z0RdeD0}aE5pH(f3&w?aV7KXRKZJk}?C-_u({D8(j%^KYv+NM4bpf+=fI`myx>2czo
zK#bY;&tk<=nspkZ3=S$Se91sJoVR*zJ{HT~IIpxFJ;N1ZT7#S{5fz}19Tm9)H=0;W
z<!=c$Red?m!7wYfNzcl8pHZc6L`}y@1gE+=!)@2?k9}k|RMv596z6MB`h@@TwI{Lj
zi?m_CVS|tP4JEzGE_EB*RM-)~VKrntArVs^7lsp9>eOH8U<xrC`lXGop*`iLNKGI&
zJGr%3h`Y0ThFzXJRdZr&!tz}sGc7M~kL&9jGq0QlC_XaO**dwXC^R+odbA?@_wTiU
zTd(JpL{vu#;h?gU#%Ssz?Q*)vPF`8s%P<r%t!JfSrvJoqIe^{7ihpi_zELE1_vW3V
zinxd+i7hI<{CQx$pHN!60c9<Vb;UK54Y&7n{n*U+Ml)Al;|Xq$(fxAba56)P^1QWT
z#-xl=Xkdb8X?JO;`LABh(MIPvR?Dkf<MPO}`-a-vLod<voO8&T<FEmJeIEm}0^blu
zl#zl8QlO(-(6^eG0Z5QyJB#=7zSZ}&p^A%x{#n&IiHFpfSz|kn(2MU`G2ro_7cL)N
zu4_~H)?a&N;ma=NlN%cw6A%!XeZw2qqe4QA0;euxU}C0>8}%;k1<uMD8X8_v%wl0-
zv7h;Jq;80ciozA;tv9~4NpIg%gW@Ar7QWu;GCKjUKTj9V*VK~ZI>q^z<kX2il;z#c
zs6Oukm{9s#ZqSnFZnUWon%&p)?d&N3Y_%YS#{~~gIuuuNRVivPp;pnodlTS83xi3#
zb5rGkVFA@HQK=dh?$4gSQm<0x4^o7ojKEr7+Hp0`hGzctXO0S|opMh#<kVF9@DTI~
zNf{baMMg%RHH5A%yKEnR9}9~`6qD1jcbsfA{;ZIpf4Gspw^$K|zRca#EspT(ab>03
zTf&+}u|f~D7jtz2`3IF%RaJG-a*G50SuP5Z3K91WSy|76ggga#w`J&dgyavQz*bE~
zOZ(3+M~?{k^^5+qFP7-e?<l;oaI^Jbab>c5;4D!2@{H!oZ)(j11!LMGJG2kRhGN>b
zFN+yF{en2bu>{Op%m$A`#P2bLqjF~HS<J>`AekYd_EyY6wfrjkw=dJ!9z7pPl`Ges
z1BcesSzbJQ#k<Vm%_%a)b@iEBp(WjVvMSwhlb%`!8^Y(}0lo{@lPb+N(4aKKAs3b;
zDcZt0L3bByBi5K^8ddACl3sX>WTdcvYoxjwnegb-j~TTzj+zx7*};R>iG`<&?j7Yn
zW2@FqSYPuXc3dg)O9!LsJnKzjIO|@XcVmVI8oudBP;gC_@$e;^GNSCA`1u@P;)4lG
zF1DM<B=_MeKyW#Bov$8*S^Pg(V63==i5J43)H6f2`1tvd-rN3bK2NFA1_RprZU4kL
zt5}G(7-#XZb;Vt|JD3XlyTI{(e98(~TsJiYrmJ*N^=+XaSIt5w7cp7E`MXPf;Ds6X
zE<Q2!cm1EzWrwpbq$lWN1VI`zv^GD(`ra$XE!?wBIK^Evl4p8B-C3y06zGs7(V<1z
z%9}+>cO%u>INn5q<w(;f@|u1{D!4F7WX(gOjwOY8luu2(Gto0QLe~_@VC*HsqR;O(
z!BAYce<s~+#{;YnmdpW0_OiwInE0Ugej=lTj(c@)NATSLbCuFR_iW=xWmJSegt*QK
z)H3PN`n196-MR<0w~U+W>xiZV+FtBlTvUNA>XEXKYninTU*1#}EUo=iue;tB4NuJc
zgWl2qq#&Cv0~RK`4r>#oV|aygv(uPU{06#%h7Bd*Lh>@g(<l8me7%VX=+zeXoTC(H
z%cYvA%j=U_EiPrXx^Mtui*S?%R&|ZD80fotu<0FsD4{1xv+NyUF?I2so?Xm)666?s
z)~g3YF0rT11jdL`kjFSdFt8{PllWdY5t?~uh`>e*P2^-w638!3p1o9N*=OC3MQ{Mz
zA@<B6fSi|zROgJFK>dc1xRE2?PS5nWO(u7LZ!s$1Ts}M;-e9yhhDM^c_-#ObbZF;S
zm!TI{ph2i1(ajtajICq^Z2-xP3`JYD>{=QGb@Lc~cY73yXpI!03;5`x+w1s^agl<J
znR~a~X-<6<CF3QxpybTi@{s=CXF}IauNW@xA9EVZIqXhzJACWTwP!D!7GHRs`})U|
zbV1MjeA6JEKl%Lg-6f6&5ZkH*P1Paj;HA>;<(`NF3mnqgmKTe(rU~1Sv6ryNxt$Z|
zbG77fR5$cqWyPaI*NA@9bZ`|C7Kf*+5FB&-b5HY>77k4+QTfm*aDu!sS?`qi{-1O~
zPLsO>SLEw}K}dQ099tlgwwR^una-BLC^eze_2r|D0cWCeAu%Y`uEESrH?_WMZh7Rl
zszcHP9Cm!qlpisLJnw{zzAJN1M7Xvlz002n#VL>JVm*v2&mWfV2(CtxjZW}OE9I6@
zvCB`RH8=O{l<XKEMLXAP4tN7$o4{Z891YFQ-1p-?GTBjN7m3|Itw`IuwGCpDy9mS@
z%Kg4_(5RfNZAFT67*bI6i$C!?GOX^pw5A@0(LYLvc)(J57O9}O1MTgGJlr*3Ju+g5
zR(K_-b$!FVVbut7dX8`XLn4<tiF300;_Z`=1aT2<tX;j+*MLcfpND@l!r$euMRzeI
z(M@uWicG1O7&ulq?iw33occ{jrEgoj-1`V)GgGnt=WtiC4f7Oq7RjIttB%JSx573!
z0-60}6nm7CW^~lyGz#$U-aLZ1Gl!hIC1_{7UEfSaLEhyRI<gPbXKL0Lxr@t3CzEgS
zdUEC@6@P)ILQ+5eYj#jCo7+Q94cTu*#a(pGGu^ZQx?xoosc<AfODhSdJ3jj}(|0>_
z=@WgT#4XO{#1z-yIK~qfTzvj(fvZ;NcapY&(;#`udcWH0u<+>a24eOJk>EdD@cSPs
z2wt|gu4Yv_Vld`FyCC5hQ(v#5`#y?M^HGt>y)O=mk7y@9(>iDZagA<?e>34al5iK8
znD&bKAXe8}PIu}KvTAZ!M@)tggN%2_gT})*zeiw>Fu`V9cRY^wSU@ElAomFKilwzk
z^C8Slb=C<3mQXOHZi8DQQzweWWn~+H1f)6C5=9)WUM}y)bJbw$FGjNLLcW!sAQK*w
zM6yYmXz(^_^sT7ZZy~|Wpg5o-;R^>B+<37P+it8Bm*XH`=XUIhA#v943M({>hh28D
zaya@7{oX`(|9Qyoy2GXKc4IJ-gT&x+ix@e<c4W;X&RIZPr&lGEUYUhF@w*BA=kr8Q
zpp5Bm@`R)(oQ><6yJy*Gf6@5lT@gIgI(`dBy_EfKC})b@4ZS8cOl438rfRPUg%ZJz
zN&84R-DpY%M|WK>y^mw&L3m^|7ygU6mcx2QD_!&Jq(K=E^jwO?XeLpxG}GE@mR8a%
z&fEB%YP^ez*4Ir@OII%Cr=4ArKOQ#?G{*>MI14puhbVG)s(Ju@A*pY%Hy(GeWdG26
z54w+lN`FV3*gS7xf%ZXHTLR?NZ2b$F_^4;^-lbuQyV{X7_j7;7b%Rb|!m%_$I6oUn
zV5SxJ_Y6cVnW1x3uj2!Xs@>FuuZY2GNBw;z)p<`1f5@zB_8%L({sD*8Za?0>3@E-=
zL;_7B$$YWRdAoE5nB!pZC{&}E@Xv7^iRXzE#bOGgw-WfD+nsYf@qA>%tK7D$FbejJ
zUf*J;qInR-`5L4B*&7b!cSuearT-*63KLuU$hQ@N7N&x--I9Dk)Y8)s8(2E-A%{=;
z3z4+0|B%8trV?{*`Ma4S`Q~~2Z^OnNy61lydy+o=*|+dDBQGTh=L5^@xerjdCn+P{
z;+#Y{O9OSaXLI)Cb7Zq|cx;(?5r3=?YVyM%&hIwiFkE7Xc8<oTjBkxj7y?A`0rhEO
zK_|6<!p=Z|{TcQJxBF{O?OG@~rz3awF!YB^_BcRR;k7<|LP_seW#W|bcwOze;U1Yj
z8K>-YpSZBH>SZTDnza=`s#BflQ#>)W22KW!8y5YpQ%|*jRYgw+s(*hxe8p$gk2$NP
zG>tx_{RWnMnX6Q<2Iuk0C`jqQgIsSgUwml88$2r_9bVyF%<(;87rWG9JS3i{1i!T*
zkEUa4igRRnzylfog8$@z!^)E>f9mQk^Ih(D>CIS#oxR7kOw~ouUk`isUrVvj$FSyf
z8N?FiFh)FN4tr2#9XC-(P1DzfV;T{Msx*Y>Qh@m0e7VU_Lbgm`*lh{wvMXDA!M=8{
z5*yyL_G~LmWGPe1@hg3d(GcF7Mh7nFR2sUD{?VwWrN}AnSm^lCkFh|5G6ld%CMns4
z3&k#4P|Suy!z_p{Hy8d%H~-cunPxa;$O7`i+8bfHU*TRVx*kE{Xafy&Lk9T;0qa}i
z(LbvGT^S&XdTdT8m&<xuZO7wsKq%#EhUIUW*)1y7(lYW)(N6OgpH0|U`{}!X&T^Y_
z875?Ysl#a~^rRZHE@zg-cS#K7=p@5pd0YScXMeUFC2bc%Q%lW?ccu7O!?ZTao<UY8
z)d*2<fM^%)GZ>D@!OMXV02w|;HF)f<klQ{oGP1`Zl=c%Gy_dH)_fCEu*c|B>r34W+
z7Lm%c-|{2EA;lRlE`m^FAI)KXcfz=?7gJP~ki~i)OwVFCzAluhT{7|^y&lna+}P3z
zL6uC%(Bt&s0HwD4|L)1c_Uv&oC3|{Dae!4sV&Q<925YcQX>V`$E9XHj`gs30%ychc
z^x2o8nT~7ZKKRkp6xUdftEgB2hphpPjPQAA;>(bNvVXAc%)ft78BOVYrN-+l5u7fY
z4SxAGHz|VW`w$iNyC}`58>mJ@W8Y64yVxV%-A@4_Bos85s@%eJJa>lU#SHy3wgPF4
zybRcO#Q%?`uMCT_?YgFs2I&?-x&-M)r9-;AJEc>)Te@2qhVJg}?(Xj9ySShCI}Z4P
z4i__L?7i07mF0ALHq^I$vR2ev@#95aUA1#zg-VRq&9`~l64?zJHABXQ?S?8(*1}Tj
zSUS>Z%Mhru?M%uqan)atK-Q>Re0ciLPi%_B{oh$5N?)m{aG;=iz-o;3pf8nTGJIAM
zvBYb);zQz~;ScPoS~DEm{Q|fQi^vIeRr|SUQGC9GlS`<)2B7_gT7@p0_uB=^m!GiM
zkrrzwB)xo37!fJ895iLJf*;fl#0`6iJNSwWYV4uK+H_Z20O>e{w>(6r&E2m`PH&Fn
z+a7V6CSSu@j0(QX8>_0S_KL_x+aH2DJ>b?Gc9XDk-64LYc##jM1f3PB(aYIj<?U5E
z%<A|vE&Nm~eoX;dm0^#-J7%3tZ`8AuK0+fUh(t*Gz$aG}M{6UV?v52OIjg%`H^I3n
z3k3d$9QZ!NzIQUrK`oj#7j9bV%8-_<ce%D9a3atGaf*Bggeq?+^hGOOs{)s)`D%Mc
zfnk@wEG?)+)#D??i<~||*m&X5%1|2(`*$W4dOc{zs&8}(YdSynR=b85$BO&;hF#1k
z7LsX-YW{vR<v?yzhPJh}3GVGo0i3M-*fANq{ud1QIm`-YJ_E{@u6OBMUS$>YTlJD)
z&SxhKcNf<P#O8_(x_B49Q;>aWS#{C^DgQui8!puP_DX#FEx9eyApioY8aFlV<76?t
zuWe#;>Mj5%GT&@Y0``2R;)f(l=q0&tXQj`0!9k#6&!HRb>TGmlC7T*0&10s6v><eU
zhDueW3oOzZwG<c(Eq0(8V1lzixpwmVMRi^c6(K=0N;1?_&k+YW%d|zjQJ({Y{p(+u
zv>HkN*VZE)0UEl_l*o5w+J8iJWho&d-0RA;|JU-zP2~Ts@d2MLkvlsB*(Dh5!=bGJ
zZOuaqxxwv=u=b-XQ0yNPB&$nosMf~tXw5ye>s^Eu*!$G=!msNPamCq=X@6b5|GB3m
z_Tj!~gZF095-!|hFz<zExgjN)X-yWeXt|>X2QF_^^9}6v=9=$6e0y}raei*!zy@DE
zr8$eWy8Ojm)$;ilyQvuUn)`6_RL9w1g~I^f99+}VG_bc`89r~j8Clvp(c@Q}{-nfR
zTb&nNm)VdCsMqDcFS7wX8sYH1eFo{Z#`o__VGKU$fbSX=CTWXbBBkT5A89Sdbe>hf
z#R;+2#U;|`FB`m6>|R1m**2-3ybz+Z%}CU&IXxq1{go@FUY=xl5zO6^uA|%<ntbJ%
zJiNp(bg|6OZZW^l7%8d9{Jg-+I%dzaU5(VRt6xHx$1zL)Rm`iavj5oInjJ^VRIXcF
zD}KMI^=vb?2<pI*6;{kM80+|EwvtflzxAW+rT_Tkid7#$<CX<tD}&AZDRt$xrj5Es
zi-*BuQj-QptEApNISwZS;SWcukMalrIqa_2|I*%J0<f5A)g_vxC6dwLEoxPkzR}q3
zjwNC?6vw%xeSPj$Z*Ox9cvAUu;UG9FQHT5yB{R5;GpYqCv)i=UDK#W?UlM7dO7}BF
zE@f*-LXA~`0WgJXt@Vzx@bld|4&GoiN0;fm+4W4%inkq#>5vaD$JF!a9#Ct)*=6>9
zU__$iTcOF?Gloy@5kDC@`*P_PWbi^$bXebP_JpbHDTs7rRePGL0^oFW*E;dexZr^H
zD-^KSa?7wf3fg1GFpe(R=&Tv#2mOlH0sr&jHvnEZ4brc3RqRqFmjv02LXSHyeo#D2
zVQ~8&<k0y)$UzPi%@pdqLDE@hd{%DhIY@VqUakGanlYk-R=~*=0eQzK4h;J~WS)AY
z;-HJ*%H%EaFnWqWH$5_a?B&zavqy*fj=$FYMBO;Ki5Nr>Ap(SHVq$Zk7EwG?frb0E
zGuO@JaA70ehC7>8Q*Cr{X9MV(|0XZc_+R(|#|%VF*4xh%02Uy%7*$4*d9oHo3E+p_
z6F27dRw_DTZ`;1i<Hd7AC&t(1m(I0$?$p{(X@CNKG!F(2B*ejh9sT{|jznMG>3pJ-
z){{a3oiPy08j#`DoAiR%B%doc>A-{t_dMjYJQ=QU%~_?HI#7E#9K#z#KG#IQW-|zM
zql0%TK`2rx^?n1mi#;tdzwRTRW!j0ZG$OssFLlPG&njp(qTlTiMjq=5dwhOPIs7jn
zdhh=PknF5ptE9_UFpv;(OP|)A0W13}Em0jPq#?&eWil9nPw$uDlsS@dX1T}*gLOJ@
zx-T;YvWu9};M|2pO|GDk&;1!k^9sAGycX`TQF-dOWJl9DgTuTD0pve)yiyNWox2iU
z$4r1s|N1&2(Z|i*boGc^QsH2`3owlN`x!y|*j&05%9~^D*VlTU3(>dEC4nRrV@6R6
zxr<GC5kFrYlNQu=;w=+{Ekol>`Gn9@%w1ihqfWqdDyvm{T^;>)d0@+7=J5sS1)19(
z5ptjf(?8fOy1^ZQo6SF;Fe4k~FWb$%gnL}GUD;#Z$aGJx81@v>v^c>lwdgJ?Epl}-
zGdcduKYlztlL+E|YWQkC%W4XuU*B`NU;mc(SFza_BcGw`s&C^5$tnW@^Yc`n^y??l
z|2T#5C%*qen%uKBXCGFRYhnQE+<u^S#p!+S48;@P-8REF@BZywQj?QmDc-xaBeTR;
z!;ONnog43SxtuwpW!PT{1;`{2^@N?Kb?I+=^reJp_qD8jYl$X9lWEwEw(@3x$a)#C
zqqEZcmAvcBH&$`O%2jQ^26W1q=Tg*+MnyIo8HR<wH$D9IF^9e467(0Q8ed9qi3~uy
zU6Zo`o6X-?m_$Ajw#IJnpFmV2pNs&y{KmfK!*G96oSr7H_s6@eNSJC8MbK77_3r|s
zH#2{*TO5FNFUDC(l{u9`%2z9wrUe2siE*M%Rn8l6L>e4%f$1U}D|xuJqpf&@j@u^P
zF?cmGS3!T}<)a~KvHkj_b}!yMl|V#4Up_YIgD-W3Q8K{5{Z7G(1aY+kUgwtutrtI>
zlXWL~Tgx!1?AT`;70TnGKO|-23$Pxb_*zCaZq9TteZLbesgJR9oyomPl!2Lr(hKi@
zVWdN{{|QlZLcp249I2%K`*Kfql;|xxyb9ZWtYte=fs$Hn@a#Z3bs7noz_|RoBs2t3
z;jyjO4rWfm%|-c3H)?Y!CoZv&h&hyXhL_5-I+$*!|24vt!Ro?>t$eX*LVa;_%^2O}
z?xi#4h48@Zu1KQqiQTkwFRb(7=?<*+dPUVu{18`9?yc=m3eY}-Ot(aSKgz5L#g|G3
z%Q}{)uSH!iY$@4z(-&TyDlOvkr)_i$=TJp8>ij8nVIw8G7gs7wF^Nupf$-rj5PJu6
ztqy^KvV$vDV0!XDC$F4W>fvGXjeS_#I{#hN8MJLfNGl>5)G;!C-oSPlX<FjbCGOzD
zxz?doS3=J7D`UqcR$cdAw!TKT>l^?qRo55#4KKgvRdzFo|L3^7e{UWi_H-w<8V|>v
zeQb3mmcwZ3mypcMt4B9+wgEuHWW}B?0L)q#keU*|bZeVyj1RXO`zi`LxQo${(Yy{T
z!2(%lFcP6i7s##95&6;)PCE-lbie`$>o1$d^=has*{_j<SzL|{fhd~9cE?B_RbO0Y
zVnAFNd`2q@!U=n{u~k#D;(QDJSLKB^(OkwE?#Q9E<`f{A<A?sSS-429_?<!92Ic{W
z$LEcw*(&P+dd3U^aUpGXpytFK9+#l*u=D(a=_B4`fk;Eh<CiUATD*wCtC|9DE4Hej
zb0U_pLPfQ8*m9@pol!IL{flkE(bLGyPG8eY`AX^mrryyJj5$GBf;+Pao1;%^zy9ll
zRQK)wj!|+;E>g>>Yqa(quS>2%aiTqod^|i`a*KO%$>b`4;~P3(l2!_D2^jkTd7s6C
zN0N^4Jzx+-GGkGAu9Fy5)!UaRueY@=Eo1-fCv3y~B2RL&**@_dx?3L~+frM-8GvhP
z(KOA^9#Xuy3PR{lO3nB`#qU<yzQ;nztQF+{<0U^@^o<M(d8yf_c~#}`A!gZC#^73;
zNr4b~gmW(4DA-n@Kum^nsG-P~g*Ep48X*r5O@WAA5^<q&R51n6TOc5z2DGsB_uzNl
zm%f<vm~i#no|!IO4%xRQH)A)?+_{9mI^>{$FJ&aH^=|jco)PqxOgA6EJ$n|OXE7SF
zIzJu*C=gk%Dx9i&O0t<|Wp)K*-&Jxl^RWN%S2-~7YqV$|!z@VXi|+@M=H_Ls4;Ki|
zD4AX~1D=Y-)aB``_>=mJ$=#`GXc`aifJ_s|eg5b|?Ken(>cr`5TYry9b;AKdiIjVn
zI*+kXq4u$Mju(QtY@Wuu6Mh==_4cS6?sE!?wmbAvDrh=@Qr!vYu;}AOKWiK?XH32k
zAgnb-j(ER{b+@9#V`)7I`826Oq`RRR^(PICFIFd^W+dWZOZ{uvvv3`D`=OIVu0Oyi
zQ_A<l<6bQsZVS8Z&J0^D%$<AuBAZ`zVXky1?ETaEH~h~B2D|0EE}a4jak$=sZr-CO
zAJWje3kmQUU;;I@K=!uw<XLh^n`1v?Qtb&u!T;R7lqojmxgpg3Q)+GJ-&)vwiI2Q`
zOq&%cE-U{)5SC8~mXVBN%O}Nkm7%m%IbX4e$hxJ%lVoZeE#sWHRoe9seT)Pb>J<*+
zx|c|84vgTT7*R6$|NTqUS9(YF(N#C@5*mrE+7^NXK%K%8G|77(7s@x`*~?NS0k!FB
zzM83`l~+S!x?wxa<l(lZj}xO+;lVe7K%JW!^y^v#xU(;syhneB^}8z14mis=UbI_D
zK}tKYUQxQF0CS`X9}M!~n^(Cy3BcftM-lXsi73I%jZ9m0im$O@`piKCJgf@#TQ#gG
z$bQ)FCk1iTmjf4i_dTnuH;Y_Tcbz)I)aw6^F5Uk>x|{o(LIA=F4xNW>V$CkJxPb!6
z!nVXj^1&%_BGpJ{ZggjcZEn||?X&Ebn4eI-bdpVul6H4BK?0V##IaOcXRsQ63DcWV
zf<|-g1oG}lF#04Rz|s#R=W?SCW)Rh=%H=jSjms-~u1!<R>=~8s^k;6(y2t!S?$tMh
zS=5Xf?t(ZC6UgxfRey(&|DOwRS5DiP!C#vhrA;IvV@!Lm+HnuDe+tE(LI-brr2Czq
zSK#9UKmvhq|3$Bgh3(ZfSUIxOu@B8=FD_B!S9B`MD?4)s>8Y(;X!F}&ImDwWW_s!E
z^-d7Qj|`LHyD_~z?F(f41QF>to+W$32a<apY>`y{uop!5%1<DyHg=SXJLZ->2d$;6
zF78^7R(UnCvM{9Zn}A(Z6@HzF_6K3r6tRg}$5)RXs*h29_G8yyE9+C=SG5f3e|9NH
zJJ>cw<+c8V^(4eiA9py)FH%MY7pWw+*slj6OMc>cp;x8*wGIwbt3uUsp!lT9Z1G7v
z%`+vnhR{!FtV5sxa{S;RgUIf+V)EkZ^eMXA0k?esYSkJFL{!H23ncG#r^h;<f|{AR
zHD1Qn{AA}G=yhq@!(;nryn>JGdPo@+?2>+prt?%!he_HH*j*%D596Zi?9GpXgWcEP
z96i`fLn@!9@Bx}{qv4a$?P(3C26ayxB+2=z*(j+%C*joT`h<Ws=#Ec?X)5u6hJmK1
zcNR{mrtq(hPH`b-lE96D1d^n#KZg_|j5KURlUnC@uR%7qyKe|Fdk-(gHm)uck_n^R
zC%4@YpH)tK?nOT@@1UOk-Tk7b@8fC*;plCi<+?PRUEZ^_aJhh(kN(<*zJKC$;!B$q
z!_fQ@>dvL}sqp<%lpcc`92SF*k}Ll^u?pDDQ2&*>*V_@?S^u<+`WyWVnRSd-=ZZrq
z1ag%JrNM1cU>q83j6H^4b7z~=H+82!o>;k)nhgXDQd>!LWBk0O$ajXVc~kE(xc`~t
zchz~d#&#Rv#<L^q)RqDUS;x7t3tn0XwAp{zlD3E?cW|yfhT7Y^kTKQdwLN{iKuHq*
zndg9lXD<RiYpvuix1z{}(I<PjJ$A2EA++i4p*>y%8`t@5WC6;!)`vYZFRih^(HS>2
zA(vT8tvKV$m`5R&GwG#Yp&u>d>1%4!7mrz}l*Kr7*9>Xaze@bP^FgQv?sO-Jt0C+*
z9c+FW{5(i@KVg2GNmO|yjQwfTj+O1LdTm*Yzq-+@3Jq&|o0-Mnw?P1=+>dDwC7<*i
zvdg{G(GUM;6(7a&^}BH~?QWfe_>#|iY4SO&l;kCK7ZHL7yXt;_f?gxi*rDXhAQFpf
zwd$9S*Z`w7MgOii{1e`q=aViV^BzI~e6qpyv5A}oU4~I9(@fP_9}18ehYIO^^e<{>
zZIP#gTq67NjKdq`Sd^X$jpp$Ll$i#4w()=O{#VD0_mAHhP&z7a<4)socE+tXAyV$p
zK#m^Umwb65RMag_nWnnhU_%h%swI>R&X2NX(ZOODR(6~MM2+V3;d#XmzyBaW?pj&q
z2Rf(tfJ8zZ$Z94xUUUcpyqiR98zBmag0^G--Nk@|ICAww;Uf9V!k^oygp>q~MQoBA
zh`uL7QnLH77ckhXGLFL)utDhp(=_phhN)ftYdopw_kW&|X`ykR)vM#%vY+QCUU(9&
zoNBFmn~nYt-Pklb7hd!qN3qb+35Mqz;wIEpS_QGF`F@-|nSYdisv&1OA@XjW%K-Q{
z1v$IsLDK`%Pp*Nq<x3PGHEVd`rEPP4r}G|G!Om=5?ttIvc`Y;kU@(5cQ%h@@4?Fb9
zqrEHkPq^mjDYy2)cow6ytM2TS>b$DmHu&!0>V!RQ<&6$ddR4+%@vbRE&vb`MG2YYA
zzQC$Ey5^BOznrzwPA%MRiR^Ub!4-@kzA#aNCdZ~t4cDHfcWszxO}@Pz5x^?jR0wlO
zy0qgOB>O@mv8!+^S{QXCagAZl7bB$6`S+3Q$<TM>(2PxZOVaqRVd`}P9wkB}8g~m<
zs%vzlTkYgGJsY=gRg<hw9X9e+nRaxB^?0XmHET$>;)TaMM)fZ$=057sP*(1}Jz07<
z0=IoYL!+XhImk@Np|jf^PHvPCe`i`8_=#{NysgJ<?4FhJ!TC*JUy0BJ`uq|T<ULsM
z$Uur}M2O?6<_DKQydP!o1A8a06S@A@Du<`sAV2Ne7|xZpg{*P~?Wa4GJDs<$GN4BU
zumvoL1G%pZYXuKJzu;P5vc!>u)0ff9c=7htT7(IG@(CVY+iehtH0u-AGeES&tJh_<
z?|*-`wI#ToIsfJlfUi2pVF*QKfz;XoNr9j;YFr4QB<s}6hyoPoSj>_Tel?JnDXDBs
zTU#U~SJnnU9I-WAn+wqPA~u%0tXhhnJ#ndNup`F_ujMmr<Cr+zm@QvoeQhE@j)5Rh
z7~d9-lu;jt#k_zw=A)5#IX!|VLwlFR!>3QViZtg?v^Al%<V{m=H*%^xv!D9ZaawkQ
zFSFW$SeBZ(6{&%Ai>Q5um~W?31+$m`C8XbjxoOUUnVc#7o{!P-@rt+`?_A1`u%fa~
z?;IX;|54Su^KE|M02DJg*Q4;7JaDuy9aZ35liQV~&G}fbYypUUq1Kr2<FBu$M6aGy
zg>^sdzh3h8xSH)45VdHT+M8%BMjS`lq<)Gs*rPC}M&ssGp{uhePG7HsiOt=VA~;>>
zjuORb`!QUGNu=TFlXTM8FV6o{=8^^iXw(qV$X2NK_6vA|^<=n-PZ-7?fWY}k+!?-A
z74H8aNye3rUs%jh)<lkyBXKNxgbtQyP!}ZEupcvuO-)nUxso)<{tx(H|I$HytM!}G
zAQ$Jy2h)4Fgh)c8r7SP}9iQO4_;3MI^ZLE6-<^+UZ@%?_sEI6~(IfaS7`tt5Zhk;O
zfc!*HAI@&I4c)V(1j2Q3u($t!f|8$=mGx{w%<Y)<JqT4*ZH0)A;{EpLyxQ4pOTW#D
zOuW^|?T7u|&Ns?cE~&V)WwEHWRszJAgNx~wYnMZ9$1nMYlm!aPIeim-u<nJ2Bgza@
ziM}=JW|F@!GX^b+W1sb>>VPE3gPG81ZBV;u9w#{XPdG2&w<;Wd3o1pZRk;=~3__o&
zyIAcR!BSoJOOC`q*)R-BwoFIn)|O2)2wzUxOT9w$bPOf)D+ErM3^ETox?L(bFvsJU
z-`RF{_Ezq4&15&nOWxI~pp2!Ld@)se<JAF-{Pg@l9F<zFC4MTq)#m-h7LY-;aeORg
zYs;LSo$YwGnqH>WG?^s?t$33McGqGdh}d^*{0ASZp$)8!$w-Ql_OjJ<5r+HiamP|q
zN=wOqY@tlZ!pc`taL3ce!cV=VqeC*er?94IcZkwY=s?I^4(SA*J^?TAkZ}=UOedxq
zTi~w51a0-|ksO>=PAzVx$B(WK@~PMPiK8YAtIj)fD^dhV3^HEUY-3y86FU4dLj<d9
zwC%zXt$A*H0Ez1s0BWM~*vzqx=D=--&I1Dj6jW4ICSxS654(gAI855Z^FU|6TDopf
z&MZ>1_w$A5!<Ny|Uvaztas}%+Y6xmXBFJRXxVsWy>*$G($NmHpi7^B~@o}pAL}5GH
zE5BCFJhSpE9I#oMIUBTLE3G<nR^fzJCPof)R0V=y4#l+1+23!1*Bg!fM^>6qhVIoa
zghU85VK+L~Tm~x(FFpyR(2xb!v?Subfl%38W>!)QQ9>e5O|$tLeCbnj3;DLap}{`m
z=}xxB>8-`p7BB7GMdP!LACC(ZJIf1m>nHMZ_*e-W=q^`Sbphx({Im=V`3`%dY23dL
zMGEE5VBp|Vs#g>0Z)V(l@2(G~cA_-uthJW@yW3>e+_RCjSjIcw75S@w`x6mAr$BfY
z`n<L=4Ti1bm4B6{IgPvwh~d}G58QHv;Cxt}{}bchpRbHab!m<xX5%0CP*g*p3C(Ab
zp~^5QJEZqCSCY{4^2%z!y`{-g0iVNFX~{7wQ~Q3#xue=JK?!Dj;cwD9r9jzT>O89n
z-xt@PK#kB}Om9-*m(kC{qIWrWSKrGWoJ(1(U2T>yUC#wk)OF(V)}<hE*p)dSl)-H`
zBfaF^K}b)sG<%HW1&_Xrd^&u>c{jlLr|E8>JYGk8xd!MdHzg}zbbr2a|M-|QWBwIL
z{KLVCjg9S`$Q6C}9_6*QtsQDecAmhB%&o0uCgB;$!AFQF^q9WmnqUo`apI9+bwaqM
z$pqmOo{7O@6$`fIEd(viEMJs)i%+6xCP=zAILG}ynOg~OZBvR50^@F~bA)cZklAFa
zJPYzf$t6TVB@Y48g2@eOV5%Z_B>I!XF~W*@ze2UgWy^eOn-brqmh3Rr`z4=r5t{K#
z8XrD2tg}JGt8k+qAinPXg|URXYv!La`jp<hH=xQ7$r7+FD_Ko+{_}t0V6ozbJTQ*O
zI?K7z0c6Y7u)&<$>i*Yfd|kqT<9HDniBQrps0ui+s52f5Z30LV$o|(hY9%dYM^%*0
zwL9bBC0%#!Weq<x5@6|4mltaPvS11aFHPa<c*wm1Yu2SAO{oZU(_30Hu(6Q^E;_kL
zc>O93qFvXFutFK#iuc>gPSn+<IcLfDcb~h7Grkv)Eg>*^W5#^qFOStSMT}D&giAo%
z^PQ9~|HGaQ2<uTB$=B5?0l#ezX=A~_mKQA|ATSc`wjsMOcecpPS;v{0wG)h6g4-P|
z-we07ww8MvTaxYIo^x;13FJ~ZXHn5xqK&}`iJBO7u)u@l_fL*crwF^!*vFSMx(zCx
zUEV|~nTirp;|HSdyz{V}HMrWF+0R7`_O}YUoyTdPj<%7<IQ#=n)~(SHSuRyMPZf*;
z&AwF6;#3n%{!I3Lau>dl5lP1Ik}I=mf^Sq|>XoK0vQS8SSq74Ce(WkUiFsgH%JdDc
zrWivDTW|;g5t)en8#jFe1IYdP3WJDN8WIc)-~9Y9K$a#21@#+y2KVn!v#A0MEG+-@
zbYiV0rybzPG3saxVg&>Q(5qJm$B>I|Zf*HRM4*hMux%xqWNeIQf2ZuDp{3P7n9L6+
z=J}hI^{KS9^!(<=XuaLn<LRVshwDv4Q&aQdD>0v2wanG@fg?~_8Czk{H(4Y{4!qRX
z-ag0k=~gP`TL_!QY|`)5A1D8Apw3vwcokk<x&l#Zh~M+b&^HUVNRIuyn{=7+5^7<h
z;M(cP*5!MKRlF_k>$$G)d_0V@3s7_2vqA8zH152fOv<;j5~(4cYP;~NcxWbw=Vl8&
zICoB^FEAx!bc_*3Odeo*{W~diDJz&rrL-eK!uwL#YkP7IuCOn%-lEB6mp^N+zi_|0
zxZymQaY<x*8zX%v%E9)DgMDVg%lMEdrP4dO*Gk1hmOBYdRu=eSMp(hCmds#P)?kOO
z^20BZ?Axj5Q?>S#s_6mG$>cFuc1*Z7*)v=c`P>c}FWfQn8#*CX)1s_AppRji-v}7d
z(Uqj!G;_TBAZ|x)VmLqF28X@LW><&S=SxXozknpy?RH;isSI9x9UUE+G)_6A;kMUD
zBs4U%-U!0JP;5Fe8JU9m`lS7RQ{%A=>Alf(d~QeTpFii!yRM^VN>sK^Pe+U7a_?X6
z)<Pm8@~f-k^YSQW!8&7CV**iaZEs4j)83>CTD!^d@jv|FQ0mWGpWa`~<^9T=pP#?I
zyKBGQkFh(NJ_PJHKC`Z1!k*N0ggVnX91Kjk`3x-trF6;<b#-;Cvv<h*zFGSxIlT3A
zn7#KjtZvfTN*(bc<01!5=A*(IzXxV}GRR2zONYB(IN?x}RJYlnyeT?!VCbztc<fhS
z70nf7MOGH?CrGjOa5_BsnU})@{xD3R+cSTwq1xDY1rSix<VKL!g2ZG9r|Cu(mQ@WU
zNgHB*=IJk{P<zCAUIVrKF5(ug`n#2&cr=_<9xJz)xexTsALwGsohps?`=V$WD6Gjr
zL5>AI?UB*yP~glz*YZ>B!*g9r6X*(p!0s)7t?*u<1o|7)an`6yH#wa^%vYJ9+1c6E
ze7tJ{x;angiDSR=b8sXAm_80bn;S_FS9`36{pg5DNOUW6DQp(?M`caG=;68Nmr}JV
zJa$Xf1iODF1+GBw?eXSF6ll_szD;E=T7%RqE)G0uuED{C|M~oTWmT2gcoyWE*AoMk
zYzD>GuN9LgwC<blODeovI;sCLMBN)Nt(uu;{Ce}$NF7_s?1MSB`)*kwMcbv?r#Z<m
zcOLt>w^P4O78g2#)}~(%mVNb~^Y3pAh}7wjL|)e@09vnA=Rp_eat$@1@wuZ*K3UI-
zQIbu{`a}Duns+u+!>#K$c0%RT)9*n(=KSmfAP}00_kj`T2MCFLN6{ny`%=WQ%$Nxr
z&5MT)53GYhLPj$-m&yllroRr8zMjDV^cru3#jh82h_l9D#dz#n7zB>UzDZCnyLhFt
zN{h;xVHooX#nti%V;y|+LGZ1fa1lKbnN~~y|7p5gbq7D&iHUsxw?5KtY;3U6276r&
zP?>%s@B)AhurgoqkmKXy(Q$B;SK)*Ur!Mw$qqH5K{QY4_yr1nL($dq_%5@<_LPB<`
z{K7&*IX&)~0kZ<DyVmMSuhn?{k8^J<bFX19gYNAr7=t`GJUp+c=!0K}k2aW!oZJtj
zQO9CDj8_#^#eOLzBST3`i|8{8(!j*Q5doU9!ue$b?In<K?{^Y)Os5JW7Pvwe<xW(H
ze0_bDWFX<G85wsU_A)uR$|h7h!YJQ&D>zkgdXbr#K(9Wk(p2U5{gR-piiDG)t<xNW
zk&_<Z_{)GV=R-rq?hWj|@|UhfMbopqVNR#2Ux@m}`L>*dGTXiw0F1(CJ2LaofM40a
z9!ghrFTPtABQ+vxKKxMO$2f{b(!bmJYAWnj*J$3~z$b^bgGC)y47;FgBQR@@w*h?x
zR2s7*gN+k#^PXggBPD%!EY{S*z05wA%J`y=N}TC%AdTab<->GVZW!`IFrx6{y{uHZ
zu!Bfp?+H6PdqXdfv<**q!h%&B30pY}uOm(mp(2js(JU{o2jOv_sy;Qq{K8Qpyja*$
z<x-q*dUvSefynrs@iYqehb7?g1IOW{uYUNf#t`dUTU{&8Ya{R9>CUSmU@&;Vj{NiI
zvi$=KGc)lZRO0u&BIa@0-k->|oG<sizjuLxDtoSTdb~b>07?LCPnWsU(!rXBGmDEU
z<Kw^BZB~Z?W4KUnM{d*h!k#M{)vCW-0&Z1*y$@q*)jgapR(aOWomf9SlmIq=D6D+9
zD1!HG1xJ78pVBw;>HU9rz`g;{lEr$Nv&sRhCU;ij+kaOqn6BrTw~TjxSH4PcMT@DY
zq>*a*X4>4=BxCIoBfa;gN7x{{+9HV%NGG7?lfv!py5K!tCTcj!kKt|FcyUu&4)E3?
zp950koE`{>!TJJh&A4vVOMS=nw_dL$D=&O5)>lFPk*6Wb`Mxd5%v8!|;EM${qpvk_
z51cOCtm_dIye+pYON~JCL$CZ$bQH-hpfKid5RZJ41bYNuoVebfBYSv~D1QM$-K|!C
zTx?ejdp+Jtv77<q{MYkgCy|KXBkA=8Z0QNur0&iDBzyrcUN0}N>%*BapdS$J7T@E3
zE;i%5v7TN>l)$U%>s|yQhxUbgx=bJNV#xdJDOoUW%4@1RXctts3iXMb`!Q7COk5m6
z@-IDP7Jvq%l3DTxn@8j{TmiG6H9?Jug-XJ!S^``Kg%kW{fGHJ^@cu`Wtko89r-VTN
zUc*O*2o_#3!5IK^)L&mhv%f1rv)vH<W>$#VFF7QXJEMDc5U#Tncwb|tYV&xa-C?RY
zS{wn_Y*FhYsWeygyw|~|=`u71M?Y=6f)u&Ri2SMxbtgqS^zcF@9wz=-fDJw(9|vC6
z<avpsv(QNQrCEeSRnU}q3%6BF$c=ID_D75pS#syFM`^mHoz1#<mo<a!C1@iE8t_uI
z*aQTFfhdGPi>5sQIudaM!{eHqDggd;`DYuThNjE)Uyo)%TN&=B#muw*2$=tt=ZpZm
zv}1)TNJFc0s_+D?*HtGHljcO-?d8_c+Ak(LI=VM4H#cr4r$Gy!?rtEDKsKElTZ!SW
z2@K)|m6pDn50~=}czhlw`b_UvuC3mPT#LCfe}Dh=-CgAp*RaE4C*L}SLYbQ6pm#a9
z8#G`FWImA+Rqx)d0{Rsa1Q5Y8&rl(rP*+2rEWsvgDYKUSrE?&8y&o!Jp}h=iTlP$V
zZA`;}Zd<Bnql~?tm(SUP4#<2pz1K{;W*`RB<wBH*C$UUkhlK{7NjFuM(9keuf_RSZ
z+FG{fbrIkas2!i29IP~Pw7B2xdcQr90y&CFNqFxl1b`JN?du(W;*yeGJv)nQNr0yk
z%6|e{3<B>T9TT%_p~?h&ztT1rHF@AjS;N#)T78}QlSe^bevp*IVs>~Nuizhu7qT~!
zSQ2okNn8$;NVtp?^z^0A&eqn}Qt8|YA(&K~J3C5O-9Kx@=Df*%v=}kRV_;&7m~Xn;
zS4{y`R2?f#kP#YAy`7aftNvlBffJ?_ugeG@^>YVC#X1HTaX4fJfJo{eQ2JJc3|Ji)
zAj;|a43TUW5|4$udAu9o{mTJJ>;%uqp4~Lc134UGKq1~kLoz34!^X~1^g0glgz2`C
zpi$-46j!6M0REoBICkRZ@tKTm-7bu?4Wv11qh_(9D$l&U;;$n5ZpFBvR4{#{(TS;-
z(10m(tOrX*X}ux<)5|OCdMB1(+iD}q%PUHS<T+&Yl=(wMnW7U@B7}4T)0@bFym(ub
zZ#Htwpah(V{J)jYmH*&Xb|#^WXJzJEMRw-OvXGZ8>cJj-f1PHHv4QIi5WDyBuy$L$
z9(OCw_YV(WJs%wh2M1Lv4H0Dd9}EMQl0M$4zw%AhP`<XnDOruJ1}qL;Rql?YUIBFS
ztIaAefZ9VsL-k)CZ-z4kG7qK-S%D{by*`+bc-?QPuYj4QL2Xs;d+E-DHaP$nZ5yq{
z5MFKsiW=3}-^g#RwI#G=dX?JhYHJD|z}dX#_m03ywn5^D9_2F&1PS1Jb^!Ie(SNbr
zXU^Evkro{vy>Cx>XhW;q*YBN?=v(|0F@%2p=wAK#rQ#<hn;d3tNP)S2=m~CI^FZpV
zK`6Kv4>N-DmOxtHrovec1O2DaK}*%mT_3f1-YMgsM~~5sG~aaB-O*u2dgrs{LZ+_1
zkx$j0XFQ}?7&4>_9Vt2jd)G0<`;K*EhG!cq{;hs}U$+MaD8GQ?FahuMMvi)-Eh{ef
zY8AD_%{c`+%MEwA6{vTAAZ*EWkJKK^;PSS!?NFAuvdK(-Sbt`qrpNkrKr~zq&g|*!
zX%*?>c5C`&U-m&>qGM_B{{EvRgjsB53hgMP?pc(#gK-xbBCEAGqb9ii<g48;zoRe>
zGC$uWD=M!0A?4V}zC;zd)nQOEm6hmi-c}MnwGz%u6iwX5{*LhNj${9<k0eTMu@u`}
z|Hb1Nreyn?KR*cHoO)&pk75z(!=FzdI)4K;f-Y0l@@%ywfzkyN6BFUXhayL!7VR+r
zaY9<BDh_vT^}-J{IiD4%fVD+>+#G5wEG<hH&f+;#(%uSwQJYTYAxNiu(*rK+I~yBL
z<BEIVeZcj(-5e_IG+-YFE?_oidR&UTE;via)%04Df9P!K(`7FAx=^7{enKVqjnC2m
z1zzXpd<xVv+6!#tOk9pPmDhKwG@1JNx=<NTCfhu!0QO>}#W)j!cP=v;6&+g>yX2?}
z(B-kM@Rujfi_hlC=nH{P<+9UdO?OAzNqA#ypKZMyfId{v^I~6X<A@}8XBQJFR}gkr
zIhXg9fLt)=R1tG~{_=OkJxW03IZFKORjN=MZO&Z4ttE2Jq-~_JM<;vkNRTE7RYv~v
zfHuN|x)^l<HEkm3Xt(NW>}R;gDKQD*8&2h-V*aWh%cg?refM7cwsxvwrx3ykQO6V_
zy^;LpnI&YSzdqwJ-_V>^;iU5+b7LEG`C<ihV^FjkBLeDfyDig}A8!)0pl(z&0mw^#
zN1NTd8u_<fvLY9eKH7*?K0DM;?t1k<w=Vo&HDuANZeTW|oJ_{uWpk=o$U4-55GU=`
zULJWrMoGNcZWxkj<%-b$8vG=QVuIA)oHRH)rfk1IcY>VxKor6At;Ffr$C@D@<}8~C
zPqhhgT|=`;M7l}EBO=AgW?*3EDeuNDB|5F8Hq*y?(@#bS9Y&~LrldtqP)^h1Dd=YI
z<@M?8$B!S1k1mReiVyZnwN^H+!*g?W9<Bf@7on2z^stsPegpv6#de%*N=@vXrk%1V
zV<M~3AXb5FCTe@*@1rDMSNrs1$)dFW;17SsKkeUew|d(UHN8Ee1jzq9^0gUnd~;NE
zs^hSni^u+0Wxy`y?Kw!=Hnu(apwRqq<5pTKomMVl$~^o0$Q5QX65Sf&=vVlh_hY}!
z45ge735U|obdEcgQRHg%m;t5iGaCb!@14%6sDylTd@#WoJVseo(K?EzJq>-Ma7UyR
ze=2u;q0UpRXs;A#!fV>-iwC<WC)KSD6d`Q*wVz(bWVhU#a#4U1jj2ju`F4<Vy(51p
z4e3jy@B%ZY`|p$I(<yrI3-A5kZs!%TWpZe^X8P6MlJveham>$gBezU*(KrqP8?2dN
z`{!uL6E0?N)>6rUAijJ<Lo1_VJyA+TB-jx*72YEUE<fLq%+mGoPOHhU$YwCPp6xY%
zl#$K(EJuBMR5%;lNDu5(TrLymDx4}yJi~ueGf;-CWj;i9hVvdkhnWz3$K3<uWKH@&
zSn+MrLF0)eu*?0Foj=Nv`p!KAeD&V^<Vrau>?5$xCna(6aPCzV<;qGa{%D6Y1o;WD
zi&|>u)l4pU6ZT>j(VH;psHQ3-gMdx=WBYc0zZFHy(yIIaa{-1Dm+EYE4%`G@t}zt~
zq>9V$K){^p|Hw|OPMcR9ueFg;Pk#)#T7^C=Ag>x-?@wGGFI3G|{8OYA3CHUgOrYa(
zyQUU+y(h<{lq$O6O!%sj))z$zOf=674i1*U9Os~#QczHch*V?N_7gkT2CvqX*9*KA
zW5Gb;jHPs(wt02UE+qDgAKBi3pQ9=Q1|w&o*RsTeFf3DVnB{`%SdUulq@FB)f&?yj
zH#plo)If5%Gh&(dz~9FTMU2%KPUm^)BbyyfL8Ik**7-GqZ7qS%yAG$7wm#(&8`XU=
zdQB3p!w)S<IQ94~F(CnR(xC(EFh|}i<B<kmd|-S5e-$}47MyTV`v+fWd<MTu1y}X{
z<Qrm2_rzhgx%C1Io+{6YiOiV*OzZ5q5YjmakaE&L_C5$PsodMp11;2l9Hsa8IA<`Q
z3ZCh6iR05}GHXbj-r5J4`RXeDWOFdEa4F=n9PzPXv^Vb3(|jhJc&ZsJ0eLRMpwCxf
z<Gyks4V87`tx3#*q9K#q@dvLRiQaT;KxC-Tl}Zw)i*)&_T_vV>dNhe2+gG&KUsQpa
z&c-`*0xZTINBs#XTX(v0#{80*nP0v*lR5^?LFLucO-;k`TS5{}j*m5TKXTyepB+dE
z&J9vAn=P{8wcQF)5QhcT6rY<i(1B3e(<HOl#(xV7C^})+>6F`7XJ0AA9H-VHc-14H
zv4cHzc9m#sYvL)j{#<~Cepdnq=+p8DxAW;RK!f*}8(4*eggAM4fR>2S1BQ6WaDapX
zkmwCS{Mj25c%xsl-XG6aid-s?Cd6UX?C(Mo2nT@}xYHv6r76R87eA7i2Tst?kmB{V
z4mZl<q6fEBwF1gVyT$F|@d(VS*9nvI%}oCF%}=H<H!I8MYHzH{pbs@(;)_Dzs(Ss!
zL1AX4{f-0#rA%7Cz}s_|_v;nupFe-f%gft!nScHI)eiV7fY|8V0|u_52IQ(2$F(^;
zHdZK`$zN@=mH}u6QH)xR0MiSVq68WSgwM~bYP)O%XpENF?#(>4Byn+Z?F@~igm4u7
zu@&I{ZAy29;~kJ+Lk?YF1!ie*XLY|TKu&AL<WcdywbALBtr3{9%M;cbk{mVe(rEi5
zMfru#-0i6{{o|?)q#kNskEY(zUdbW*xcdC_03ypMm1JE^+g$VE_UXuH&5aY|PksWc
znqorg)(It|AfEmM0aV}AFCN(PzwPggrLES(&~1O&yG)>NI;H(vaJq~BY0xSRo)J<p
zIY?7USZ-PHJDm(dZ_ijD9FPzzal64M&zpr0D{DVpdJ0n+gm|X3GFa;m$zVeHXMZ#8
zAa41jkFDCj5x4#?u8G5;!e>lf*asyMwYU7Pug!wQfOzSA7Q9`!D@e*g1F53~yo2uq
z9VL)>j&90r{85#iv4e971N(2c)L1x(_cU_mGrttzbYG1)<Nc~Kd9*S;%Liwg+Bnfg
zNmb?(QIO*DPAtrI*y+W7=oRO!X^`iyCH`lL(0BmqK2<1N2!Ib4cXtCTs|Za-QOY>7
z#34Id+bn>})KNHR0>?yHnAXOO&Gpg{m_-N>_TJG^)MTkZ`C@<ogabcA$%7YQk#H32
zTiO8KAD5B-?RCkf?F%O-9>A##S=1g}Mn^{-52r%_K_NOV&0~oZpzt0qH*-5{Olp<>
z(b4Gi3=Bh2V-piGBz$h?LriVcx6{cCGCl@<kzWB_Zh2*8Xsj(E0o!V^`dzYy0Ad6x
zsj9e#f#L|OfZi~gDd0UcH&S)o)7!gxdw&B44$|%Eiu=&9+Z&@hF3iTN$@tO3%@YU2
zKKCs=5dj-`=2rRk^ha|!mXFcZYHn%uCBLL_SKU$5jp@)g=1wH;*~-K+dAK(-ICjKH
zhPCgv`PYpIl{hR+dA9a5+BA?a##NyjEQ<MItjKr53OUAir2@2C(uK2?S4+PqIc#8N
zy#7!#VN1nOM$Ny@5Fm4Vp47I5rE7afH*Z_o)<%zNSveTBXz4}6;~|$OYu*G&k&Jla
zJ@&9pa*$>Q^{5Ccj5IJ#b_^qx-y=%UA(3SA6fe#qEVXGeNb}y7l2#WPrNUYSC|;CA
zS9q^dhE^e&*Lf1xdr(Byr_0H^$I(bzct^mw0cQvJFCYbXmR9A)dd0tkln>Sox|tB3
zE;irkgVMlUC)J*w5h9R?!27zDCsE=IuCg!RxM*N9(6d621b5<I$S=kJGewzfB<=Qv
zBzbit%3yf;w<-d_7xdHbtj5E-tIe)TWvsxlXnna|k|3?CtJ?zt6Za2FuXVS8Z7>Fq
z3}BlGiRsz_T}idk(JXys#@alM&;8c$bh%-EeqOO5BPa--N+zwyi7(qaX*!3Df+DZE
zMffM}p^<z-qK?->ye$Bhka{w>#DtF#G@m;W_BjWOzVbpfP1c|M`Y2lpq=)^!CSYD1
zmP2$FRue)foFj$NfaN{4TK*)1bDo47_)~?m)&mW#^&-jDuQcC4(pfU37DId8bUSl-
zg8$r^yKt5$Yb}+$qPnGpg~Qq;K0fbwAI8dD74%@ErS)k(GCawUkvOvCQu)?>0Jz-w
z0aPC5+FCC*2A+Bra5NJkMPZT^)y2(j+=jaFB&lBvR<1KA7+aw<SagdD<A?8RV`CG!
zp*3j(J-N#Hu@S518+e%QZzWv<lvi}PslI+Uhu+AJb$cP`nnQ$h%UVqmnQruRbS*eE
zpKf&8u1lZn#M3GM9rm^NspFJBl5yc@cV0(7RzlR2u?vSG3qIlMEB4Csk9#_;Mv?UN
z4uvmT<RGELt0oS@SFr$nVpMHLUj93y<<yzuJC$5rfs9ux?jfqt0$q1XqQQV5Lh+VN
zXv;h_e`G<4;&0iDkBVY6--|2PH^zD~xa;HW_Jsf2Y8D`Rl*Z){NG=*_yWS2tn$9CC
zEDZJV@Sv)${t*vPOjQ*Rh&kw=uXk{ADd|vujvH_}shI<S$NTvx)u@ZSzIN>F>T=x-
zp#qFMZh0NJ@e7U^1ChLf!a}Pj&(L26gSXcg3JMBfF08GyGu5|m@xj3#-0n`LU0hu9
zw|HOgw|K9|zn7}lg#JF7t?+*H0t9$lCnri?-ZWomcovWQbM+b%W|cI=fImPiF{Uun
zE9K^BuFB)V-sxoVCm;X=3@TePkujWrEfn~L(AkokBBASb6jW5>*XIWqSlAh7YQU@N
ziQZa5d@4FpHDuYcq5W{k{kI$Qh_iAJEoQw%vgeG8{6_F(lTxXffNqi)M~lq*X~0eP
zy!WJ@{mD8n^}=tSE;|>qaYapYo=|$|wNAE3aaFk^{-xgsewnNwy8eXfEDQ#$T|)I_
z-{^SkDaGkr_IG?119k7+r?^QKH3kUx``p&UyH(s&C3K&0K?`XAZN*|afR2PZiO{$;
zVp&8!6Im<nD!(gw%}lu56ZS8$!)(yPJ%A&e3SRN5)W&S)j`)k@J}TBXSrW*gLAzP}
zNBI^jb98U880L8qR=L4#zP!w(`dshGgif3jo&07t-J!VCf<mF+?8$}cv6&xBWA5+`
zqdD1G6*EKUj?`&bv#*%#iJ86uQL;>8Ox!(Ru>;g3C$wDc{sC<iPbha?{ni!-um%22
zNdTru=W#9oz$p+294Ivi3JPL186^S|qzta{fbb+T5oa{8@+nwhEiHUoeNiUQ_ZMud
zy=!Zve%Zz?8B!5bgM*=RxguZ9<Lv^}gjZN-I@~X}2bMC#^z`)hZzfLy3uV&(ImUhl
zybch(EGsKxdsO^^?(aU;le7il0(bY-j0n8mI@imiXD=oEA{HSNkJCvkP1r998=+|U
zB`=#sPO?sL&ablbqkSw(-ajd{DN6HP5c%=wXEyGNP8jgQ=t_+?sCX|{ig1rH1}Ba}
zNfmy9Ku6D0t3G~Rn%ADhocW=`oz+I<?U_I&_hayaM_SlZ&VcMlHU$_@!<j10Lkl6)
zRJy<YwjOS}`6EaEbAztom%HB#CxTzvE|&jMHK?%A(9s9{axeDur|BeKcjPag2$`XD
zf#m&8z<N7JLy=xDHvD&s;#fe`%N#Wy_RG8*=f?4#Ow}p~0e>EHulH0W$Y}4hzblhp
zItmy5<IA1g=3N>k)z`S3W2jQ16Wepg0<*fmFn2aVHZBUecjWK-18}IDq^u{-Z4d);
zcsmGjaEu4ffq#RyXHft~GqHmR1}1wE@&{`@csEWJqBE!%>Ql+cj*Jn^jIfNiljB{!
z1}sGF?S6f00at4lSB@`qYvh#ms3*x?c%8E0U(8WP+p>1Cv0N@c_`|Ma;s|bEi%Lg9
zQaAJCXBLm~G$3ozQ%ll-=FlVtsemSV7vDcJpArlh-q#?}%YLx<tv+pJUTReuGt-D~
zJENK6GH$T>$xW)X5lZvj>HuG_BgYH`j5@8Br)01~^+5x}Y)^F^pF9zq!AG*N4w}7}
zOH0yt!Zyri_HJs#M6&r8NhiMzGxSbzcG|p$AJbTECc(3Zn@Ljmn~z8R^H=g-yt^NI
zugOj+{8oLWwC~2K;=nQQYxLpIue0Hd^_$B}mb++RdAfkI;0+I~p9|uu3xqmm#f$?#
zOBj2*sv7WfF;6Mn?<`JWS%#!Y=@mS!h7aBUYO5%#3SY1F1uuEU##e>NCuB+m0=c5v
z*{{{Xmf!o^QL$z0l`)zwx8&xypO-XwsCEoq!&-vKdlq1$b0Kjt|GBVEE1m9XZ{fI{
z@JR;ZXE@-`WjfrrpC0pCyRE_|G5b}lLM#v{-p<ziTV)uf**?w`4PLSftxlrDGQ$0&
z(qG}=+|AKU!|-_teHlzFp)`_FcuTVD0VDv3u<aXZlCo^ZaXszT;GiV3QV-B|PR=5V
zGT}ls!eRoimgZ*T;KI;|PaaDah_dzSqnR9|QixB8>J+BY%B}T<5kQv-gYS}4{T5uz
z{l8uDCqO@2h7l@K5h5ezbhSmI`O0H^)7DaFD*YvM2~|*c=yn%Ty1uJf_$5C8^I?3g
z>)tn3lB@-}t{{UN^iTFP{xw|WbFClW+me`_%tiz}QJz4|LopYCb<vwnD3)3^zXLAL
z(WO$hCgC?NkA>zlU!HjL<X~0|LS0hkFnoJSf_DI#d(9;k!!BaluwN^s6pqtp@Zig`
z1p(|6S=<ma1=jZrI3j7W$jtart*udpVXSn!=I<{d>h6M}g}DeJ!~4R&jxNkqs`mcK
z<Ql!nrY>;UU;=FPAc&=X0xEr0aCL_Mvdq}-c8r|Ut!dnOkXc<|xna>TA3)Z$?iPdk
zh}l0()Jaz@#%#uvBO*w1QL*xNDo?_c(pGKnMe-}<&113|<>N3if$_d9LAWPbc0tD1
z-tN;8Yl$^k3^V6rXx{dKVEOIK+{fLyvKVEHPj?cOfO+7m+CU%mI{opwmg|p<?yUg(
z-z7MO``khILAd2`-r+-u*Mt8(beQRk4I)eOa*^HU-3x8RoJBBK7v^bJdj=?)4tpPM
zSB!OfiSU8S+yV|_<f*pUX$CNQvobHq^~8mFKSW76T-s4Oi3ES2G^iE?DLj<3`E9PK
zJ->9we?;wOz8*692imn-twVvwwG6OIKF^D?>x(N4_})NGpYZL(&i(A^=!pwZPAmtQ
z`8zBiYPEQrV`eg&+<i4^`f9|<S(pksZ>8|8x3$EQ&1qbz<0dW!K3u`6NdoUgZz^a=
z*93?61paUDi%iP}_dhTGfSgh&Wto)kn-ea@&S!#X2d?vj+7gOTSVus8i=_nBnJv3T
z;mdIT_X=dUGjXiH32n1G60kxT*=Z{-Fiu9|2884dSfjuk2lFUdRUbQpY2UmpT-A$9
zqEMErd^Eg3H(_U!^ST%>D*WgS9e4*_E7}mzv8$neS^Vb-g=+j&Q126s*pSgf_gjp|
zBMbEob6(8#6XX18FB|Q&#N0mE(LU0o{4k-x(d{48%PK*Drr!3I*t=~ahz6jA9=g?w
zzk56&zydtwXUWX0IU(Sma&yT6c2jM=k^&&5t?g-Mn@miYj^>mACrau(=!*%_1{a|E
zvMCaDgJ-t)Mn+>!4ir5(p7K?r=3t$|B&4%h1v+nahkEM{D7j1o(@t8Vl#2f>Qylkl
zoC#r17a6L=eeK&Jw_CS@1?Y)TEjA2gkc`aa+y|R`>c}Jzl`6X)y8%vfq^6qdPrWCq
z&z#k!Jb?)V=h5(-_;O^APuSEm6Hbmc^j5TGP}Z}BY%=_1+azPLAHSW1n_`~J;&fwA
zP|xGOMc&Dl#I1q_I{&G7$7eXaEO(3s*NL=I`p){uszUr_XS3LlQ%Un!F`eZJziP_S
z^&|*q{%n@63YE(GYe8+oSapt8MF}^d^1c`>9@RwK;<aiH*NUA6O$d8!8`CPY+eNj<
zRbkI`xXD8Dvim?w=gJci5&yBdO)DlgR#}9C&|4hJ&o|QGX*H&`mMzA9mnAIomB6Z4
zq|rhzr_C|Ln(})Jk+m2(-ZrF#{KSZ?;R_)Wgn9fn!wB8(993KVDW$0yybU|_7@}9=
zJWY0D$@OneT>hxN^Bm%$uo>Rx0fih<@PEc%7~1$2=InB3Xu|XxB~ZO&5Oo2FCt(?_
ze{7@T1`K&`2KSnIzQ#hgwS|{>yQekNmC2WiiEGm~?eamkV`B2R%NuE_WFtBYo~e-T
z4g>A1XY<6As<qNBS}R1*NV|nXB2i_3ZXT=rmfZtog5qPF04ZT8Iz<VYqb}gBCeUm`
zlN+zAmPvW)3>1@4c1kcF>_qNvZ+T#Szt8J2dnr<*DSbQ=_`6tjv1p&o%r?J-hCco}
zojaUOVYdrlXyMjC<Ez0L`FnOqpMJ9__9U43baa+-Xc@1P0qH@1*{zK(9C>VWX{!;R
z5&V8bmO$?(j}=&Tq?oBd%(pRoWTe=v1Y-aBIgc&<^lR=(!u_v$=(93TN@cHo;Q{jY
zbrH}Wc$!t)7WL4>)+pf24Z!*QKbo!rD(mLyB1j{sNViCLcL>srba!`mhe}Jgba%Ia
zbc1wvclWpW{=akZ=t1Cl*x&B#%$<AhoFH80&D%dxX)cy&61Jn%k{Ip7-!S4oO*A{9
zl3R)T^QeL%XVmb)zh<WVC}sn#!eh;XqTxP<j(b=Ak%7WMzN57Q?SU;k>{&HH#6G=n
zp5>(qK8uc7I6QcL%nBQVt&t@yO;7cKG<|cUu*IeBDz9^AsfOdrzkv996`Aq=QE9Qs
zInVw2$YE!6H7;FUO>Ml^Qhl!0G9-4gBo^;*zK)EW8y7JICswLRsifnDX_<g&A50D>
z05#F5uZ{1%MZ=r}ZE?r>Gc*L&yjs&s4cFUyHhC?j2-K2Et%BW?mcIm2tE@Nv5T{4P
zTS+|Ka32g7&jvxO-N;aX!B|1tli8?VRV+=zcaH50aTZ522SvFi7hJIU{0t@g#Xjrq
z)f44~yZ!4hKTje`f~BxnEC15(htu!Fe~;3ompH9ejkc%4c~>_nba8_mC;qPmpREu*
z4>m%LgJ&%Ph;J<ONBCzKYo_A?@*;Kve4p~8+QxOagdpxDvD6NO5uAUBAQIfQu^?2t
zxp1L}Z%tWrO4eLxh*A@ki%>NxML9r=vcDC)$kkuT@N-U@F0Gtely}3UFg$pALNS%^
zq(pi$p#}&4M26MFP0K7*9uEp2*-y^!4f5^!%QD5)2v5@U_nq3{QWl-8WDz@BDz0MU
zp2kcS?%9fnU*4Hl_nkS+`?hQk<58U5znJvNx$L~w?^!&$@$kltCdaj_22TFJB(6Rh
zWLAp}_GAFz?irHWdo|PI?wpyGML|Qe2}s?zd|oMQYkGhp_tLe7kd~IlWw-IEtYjQc
z<55q;1>RnOHVj)chr~pN=r{T+EbW8{h5L+}#I@kz(v=RG#y$0C^AA+xy)_MKs~aCa
zUK1HDGwQ>}HyfMITg0e!5lg57FiyXo=d<|rBj2YGq{drem3<RN@y_c+l(YGWa&5E}
zdbeswU0hVE^|&3(my_LVLO{WxsGdfNx#|DedL^C1TdAQiu;xdPw+A%HVB|9o1IcGn
zXuGtc21rhHaXs)OB#m`enq)xcWqHW}!o@{9rrMX<sJk#}PMR8I^88O3DMh~Nw4EDO
zkp;Pz#F{QTcUhCi+xQzh>-{^oW#WgdSKGfs<mCM{)y38h1l~2HA}(vaPW3$zX&Te-
z_{D|ez`_}JD0qzlD-fk5em2}pE_J0GCC^hV9Tk4>@vhYp%JpI%qwuh)aoZawJJ9%{
zpNyn9mR}un0Z!_Bl7E{;iJc-2hI`u0t|BgXYK~SjYNn11`VM8iFbrYs+a+#Td%7&y
zeypp_D(0iN3ip`%{9;<~f-#OG?}%ki$~5h+8+FrGjHTAb&NVo3m3PfHr*{Go=J@`8
z`Nnf$pU2#m(AjoK|55gTM^icyYj8nVEpK8hM*<@ppGSPUR1;9Eyk*HLy#x!U_rHgR
z8h}p)1P0n3%rF2P89OZKg{OssgZLZJ?f}y+fuoZ-sa@w6xKSULpC1W`MU=eMtpw|i
zR6IpSZ2-Ux>{83CtG^U{I7gl$^KbvHJz!jM<83Xkq=v+Zg)h~#wuTWDo62R_#j%l_
zLQLbt=GKB9D9|^Wm&@d~7TmkpJyh(~Ve55Usg@;N6!lY=GPZDhS+gVcT^-YpH`b4o
zQoOH6{|G0&?d=i}7tM(gZ{aReRIhc*C-&#V4(Tu5W5t_ee537@5-MEg8;cjZzYRa`
zMZoi;f9{8Au#q_0FGX@Ha$NvFi=#zAOd*);L!>FmBh6y7E6ZxRnr|Qd?o1Caa}+B1
zY^`%${N^C_9>c9fy0L`oFaASGj`*er&8H8YvjX&(vW3(;k9Wil?zPLF%lb_BcG!r*
zo}cBWYPM5KAm=y*h+|_y2b4NzA?xK4d^Cu*eX)pl7>|72LL1HG_L0kXJ9A>TwC^Lc
z6H6{O4Tx&=>)j3=v&Z};h3|XZ8CLirlsQhZL{W=87ab|tsdP=<*MG(klBD*e{cfn>
zDzmt+pWA-vYMAE@Z?8SskE0yKjPK`1i5E%_$#*6~?g}FRo*hZm)#eMap64fqYK~RP
z6VlA>p6kX=jDq!ad;B~leUEk5)Tw~FHkSaiOg3t3p$*e||Hl$#TWe_Teu*htd;xm3
zTRRJ!q6Z3Y^}Ev!f@5rgW1igX>ev2ScWnLL-MML5@AlNi#2Ca85vgN@FXk|gK&z2Q
zd2_w1tLp=XgfkaFfwSaULYsskc%L8akqLMQ_a=+>xjuY;yZ3FjO0`J2YzuJGARa)w
zO_5Y|!;S;s69AM>$j>K_qfzfpV9*Dh3Z3B;jxWN(kmU}$<1o=HXs<Jlz%_~Q_Mq6v
z$NqB?kxzel3N2fEM&>T?v98#7dqr}4Nzm2dK^@q7g#Msze@ME9UynUZx4D@<Oi^n}
zFAiX?ci5-t5RZlSjmmKFw$%wkW$(+Q%b=i$zLUdFUVx_fJm2kGsqdPd(x7qpq<A&O
z&1w!Q*0c6fhhSU_Wi6)L&%0Sk@t86-<D{njD6W0q@R6}0bg7I-@+J$+xV@$uw&a!W
zEAU{C^&W7mwL%U+I^YMGfZnpGqy@st+VZyo*tV?|I`!H6_2m!77^0q$&q4*2gyqIm
z__rS1O2}`xDo|OgsfO;Wb)O%GK-u!G`CQeebtle|+ERgepFkM=FtNigW-DhIDg_9w
zRaK|o1tZ^Ay<*+?WDuF0LQ|ug?^luxrH?L6`jlRU`{0TgkKC2{ynSW9F*_bnOYA;w
zr7&W~=g~0HA3c$3KBnLNQ8Q)r{`LDSA>|vJj>|^}hEr^oS{NBL0JxA3Q~c>}*lQj9
zhgZOLq#9p2H@7`Mx!>B^*x1zU!$L2L4nKefQYaP^jE^)R8a;aeI)zcF?Aw5#;l9ae
z<2<hcrf6e33f><V8xp{jPH%f9RGpYF38qdk!$jq|9MQ3JTv#aLyZ;V@?c3S-@xI0w
zab02FxGBm5Y>qX&x$G&i(T5oGK0<9K$=+?_H$7+BT1hO6*`9ABp4s80#IX8|X}>1K
z1?N^Hsk6aqKylliUeoi?{_dL&1CV;xuj`&jH`XZNfhKMr0w~IzebceR?OxZD24*#5
zn~AWyXM_G9a<^^W)`=MDax!VgRyMYUb*A59J6k9%6HWby&3faRhn#gBlVR{DFETpj
zvz*c^N-nAc9FBFoJ#EXbqqfb&bH@#HX!x#;ybXoq6=;4^;(TTD&n*HJROC@wB3Q0F
z%xVV<c(y2KfVtFlQNJg8dyL43I+B`9qS8n8_-McGoT?ZYF`a-!G4+=bh_Al*DgV2k
zGhQ*e2fYMjHkh*Z!y`83o2BLwU;}$=r>2&qs)zb|_2^^t0K5slH2?X@9kP%K1i)!l
z#{XPw7~hucIT_ucx>zFVlM+;xE3Gb9H0&}c2$n{!<4wIjE+=(tEprb`4JJ-a3(qc=
z+|w+omen(ryD!{>5hU$wNn|@3YL>z;WPYBrUm8X#u_%?YQ%TZR82VB7bygj9wC2vw
zFDUzAVK!-XVU|M9I4)y)*LpuuLroU8ZU3z6i}`=M<xLbQzJQVT9rEV(_WF*>W%06*
z5Zlt=a`WL3aNr(9O#wN+&2p7(gOX=l9)oAOhA9$`vin5O`v6X*7{%X#JjT7!gTKhB
z2teJ1oHoq7I}RCPv{bX#ZsGpOVck6)Rbqa)bd2TT_ETD8qrVRkl=_*|d)Wv}MoY|x
z_<lK3ox7ecOzQE;!>P&e1elmU8gG=sS<$4wDQlS~n|GUUeDs=z|My;*zZl7!0K$2E
zER`~3xz!?Pn%CVf<@9pB?RsAaJ>LIw0chs6l;!R8yDvb5O5zX)_v)(G=D{{C&7bL3
zwG&p1AEjZtMt3THvLayQQ^k3*qs9yd&E-&uj|3`;o3zKba*;;DB0|+0Yw2li5e&)V
z9p(khEa5{e2uEA2V05c>K!*6R{XwZdyhC1WXGQwL>rxljoc$<u&EfI7D^L%vy2Nd>
z@KdbMluwm)SDSYVPVz4H?i-s5%tV2uU+UWbLls9rR+mbzE8yyP4X*b*pgpWMn=Ax$
zr?oXqHY<(WYr2*T`P;J`iTM6_z!e_E%YAtc>1P(Z!6ijJLdJ+Wh0J8c#?8a>f|5j6
z_|h)SWwU=2{A|c}9o=fZ8vO*@tm3^^^oXSDCl`=YVq~-Zi*jtp;|~ybi@o(<iNqi1
zZ5+fGrn;1gZKIP@$Lxe}2qfYw2urKDRKSh*pO{Tzs@LsNxs(Z$l&|G*fQRO%0keuc
zx+IoVEJfsR!IDD3@BAtLtNE3^(Yh|oqcbxgcrnwqa=!gb6_eF=?B_W<I<k1<dnLx>
z<o{6(1g(i9qTUTgq+mOKbBLTMp1LeVA{oJT>7j~hmWuTM3Q0d%GO#L#6L~05Fi1KR
zxl(}pE+8oQ!WPx<4$T1UKX9)AE!Yd|4-uW3e3_P!@h2?4%ddxfz}(34aGfdF-9r8f
z4raYf-v>K2p1)C@Ax^(54H8zY7x)K$wnLQQ?~#*Ip<?3#6yOZL>ie?2@LoFS5(<o)
zLdpitHmuw&{WYVq49Uqvd4$W|^=>^3bD@KW1P<B@G}y9>`4ELb9D1<dv_T0`0^?k0
zB*fmf&rGG2<}Y0mXf=G2;@yS%jb{x$P)#K&)Jg~eXG!bJaIEB=v?=B>S+Wyid-rVk
znm-5F3rTZ-y&~V^6^lXtn*0@xR;ssdA)*tZG;Y6g?0_kMds#eQWBx(S-~7|dcG=3F
zKQs0WmNoz3g)-?3*WuZJ0I!uO**#Zdj>TvI2YQ51I;U%0)LPB;<Cbn$2YrCm8#!RD
z;9YGwABRSz1OXbpu=c^O%UpnZ(kf^|kAz8HLESEI_+4K2Dn5V^?u~2&ygxet6O#Dk
zGvW~g{#KT4Btp(41m%Q*>7K~N$%N4DuakK-NJ*|dN>;PbS2U$I4HKX|<`G&OL|3mQ
z<63U`(ljqgm5^h}2#B@Z6YSN)^=}>v6|fGBHJb>p4LkW}K}9=$+Q+L{nq<qHI9c{c
zPU4(yjTGT3sF+?fL$R9JJvt1}`%V8C#d~$__%pbs%`x=xWYtth&vSxj9!~9Uwmy2`
zLB`eO`>;arX$5h=0T19$R-jBy(UsTXyIt3EdU#GwUt^m4_tql;Sq<#BjqjdR9@Dx0
z%d0c=gqLKLSwJ(r)?&5;B+RCO8K1Ks7#w^K*q&AH*H*<P){l9=yq}*!F&X^c5eqN;
zyg_@xYrg`f@OzF!x!u~$>vh5mbRJ5&4`TB>NZjoDf5Q55UuXE%mtRx&#2vwiAl@+7
z-)$T@#@Qv{!{Gr}HBrN!Y4akDMV{W|BRZr2Jv>mG6d1CgFj5lmwn6;<b)EL7qw%b-
z8(4qtFRS=8)+41|L=^^_t1%NCK^!wfHH5Y;-KY9pEF|@)1apQSg#qB`4|>P;>9472
z3qh#z3(AMs@URdbl&58V#nL0ZsQ_u2%Roq)W>l!ey(CyyW(*^(FQoCOqp<q2D&J7r
z{{uv^3poc;`ay=~vhwPs!tnCqWnf?cbf{p^R({bM0MRYG?b>T_2>F#lRFhx6s~z{H
zhH8@IArgTqOAVCT9XAk&9m85Fb8va2-Z0%cF|eNFBXawL_$IXdHz1ga4n!x3wwRg=
z()vFmzQ%#>i9aSHW60{Cyl;Mjqrry@-jGjlz<%nB`xzNKwxcV=*ldd|?s$>N9kb$P
zS+vm8K+_|}>k02?u}aZ4hlj$iE`quBy;h+!O%<m|`x#kQg%azwQv#?4B(^Qgs{fW|
zP^y~UTyRrM3KUf8fXHTZQs^~a<z*e74O%)j4PP3CVeJcBg$><q4iTF15V3h4mvF^&
zMt*Eh{8wea#SY>+lWPXOH5#?*UqT`1&U(%-myRq?VrVcgCg?VxHW^#`Bw~Pe|CdHF
zMQ^J(`S+i`z9Bo;b%)WQfc`bIk#Uj2p6|7e5rB~0twF@X`3}<JSA3DJKfBs*S(yuo
z()Tq_gzw6b3LF#@1%|(=qRw25F`BNGqpW9tu^><`I0OM&fg)WC1|(1FPZI-?R2xVx
z-pqKy*QKl}(M4C{h%&cR&34zzX(zsKOjDlL389u;aRUt>m9ysD*-sDl>Qs1bS^MPT
zb@V<sI0hZjzBQqef>6y4%$BF1XktR`znF1nXWR=Ga|vcC81vc?hn9J`pu|~L9Rw}P
z)kPH(T{6uqrK@q4aZJJz$-=qCG<a1rz>Nv=cwpA;Y@FJ4{ohJSDnB1vz3hpMG!W1~
zYI!F3SP@w~+o6f|(22HCQjOIMzKxn7=q%G*!3&jtYj*T7Om0flF?nOKb9orAL1NWY
zK#AH+Nb(l8<@(c027RwwBo@*Dc`6iJx?JpRekYA5ou#vE);SuJFE&E}#fq@AqOjjd
zzWpnpHas6?(`MI(+Ls4gSf~&@l=b-5_A!#C876mPl6!j{Z|T5&O6m`QeMZPbirmpX
zTdH&~r$XtvQDs~0LYRW5T}HyhDrANt_Z@7rT`oi553W@+7x53q^poZLFRQCRW;uHM
z=f==F65BoyVE&>!$)a9KF$T6K8==bid_(s^WBGw}-)aQ}pyF(KRn_tj|M)&n{edY)
z0}kb{{4<0{?tgugg><j|czZ#jSC{^Y=ZZ2g(Um*~0i>9x@YdoXy{z>xE>rBGDcML4
z*fas0e&c_pwW!?LV+6MX|AG<<n&NiObpfgxXEG!JYWwAmX%I^Utty>O9e@8kCaI>H
zPx;RKYGh8<MA7dabkx-@EIVVPF}{PI7BHpwV6>H0uAp~(^v$}7U+qiK`ohej2Q2x2
ziN{}#C{Uq+rw5UUG!p0KQXAvu5f>|lDqNsYbD&EJcbQMmq^)LCL2pTz$Q-m~Mgj0;
zYD;rgxnxP&&O9jyUQ0;H`DYd*uC(NC^1)sv&tgs;T}!d^R!ftmI;qqnk=)4jX<j%$
z(Q|Ah051bRR)-6!UYAMd>4P-cVxc|?bg!d-{`6~?oeV|(2?%l1ilQ$^ZS<Fk1j`45
zjZ6wT{p~HEnrJ-6hFhwDnjACGQHyqGK{VE8TS&hDWdXq>8fHW<t0pU;D~!+Bfa{LZ
zgH1+w>?l52vtKxs%sW%VuvevBGCsZ*ry^-*iImfi)Tj5~Tls9lzqf(suAzYo^c|!m
zBssJ}7~m}iG%1N$5rL$L)M;J}a1$Fh8E}lb{M_nDJzUt59N<|^`S#Z2pS-N$x1CdS
z*|)E>l50zmU-Mx~KRc(@oBwRttx)Cw92r|K=_qHG{kR@b;+Vdh4Z6xAwz+EZwx+AQ
zuAj<dt&vPR7*sLiF)%N;trgUj^BzID7Y-j6FiYI^&k@s)Nhki_yQ~*Uc*JuDw2Q;J
znymbt?hR~!$L0Yw-x{m(%YwEVj^eQ244_{PE{Uq@7F-BoAf8l0#^P48d>})AS91j+
zXO1%ZLNf4V@6I>aEIoeGIbLWRHFq3GvG}lJ{Kg5Pce*21{zv5*N38`4neuVo@n2nl
zo2Fw$Of`$2|Cj}#DRYV-=B^#`Mh#H-c6>t_E8ok{^6|y`Bi4bA)Y2De_YXOfkIDtR
z)4bnMsjfT5G<H_?9{9=5KO5&?&)=kaEC4HjY<gZ4JC+rL4i=uDh(6lx3}Tu89W@(k
z26sQ_tAp9izQtUr<Vx#hUTz+9-o&7n|6VX>CNZ0mx=#dIgue!iKlh5R3|x7XTdJ0F
zexN7E$vqnFQHaq~c?1@N`$s9xy8PMj3pezL=eU5!?d;=ut`1RalS7;R*NWIYDZRml
z%44Y%I)Jg!V0ZSB*hq?go*`L6Nq|y9J3;6w$H7LngWP&&AEq>}t~5a&CV674OzcHy
zFUD!>+p{g*F}zse1zAQm<QDwW83n1;Hnx9il=bD4B-Ae<gT*WVl5k1dEFlB-p18xH
zwux1Ef^`^9WjKrwTdRX`_h0d@{vTe>0N7)0q-@td1Mzfp+FCVcv7$!^VAEf-_%Ho#
z+@1c}6sAC*$I`s?A?b;#0a2c@w2kZGJ+dGVMIV*kuKjW}dbC28Cyn<lEJZ`n&(8Uw
zl?gz7Ii1>|^$En~<Y=ER(J}if+s|INi;1&zH^~iaNh*|hOYL=sL(wol3y&Vu33ogn
z-2|aUiMb8iR_&|9^-}w6%pQ-%B&<$7HGKm$1!$VZ^HQsLbk%>B-t6?n1z;yU-okA2
zE9{H`CKST2y}GW91_kDPKEn{_v{0jux3379ywlD9{c_l)2Y!&~7EArTD10Oy9iN&-
z!ZL2_w|96h-z;g?C#-L;?{j_mOrHp9O&}uP8~|!l)@wkd040qw5Q?A=Teik6q@q!h
zI2C2Zv`a|o6w@WivkZ_83V)S%)Z3r=srko>3m%>ErkUhRy;2(i*61=6ugNPrEHb4-
z5~K6eBwVFMCQKEkO|rAEl{xyZZ9+qZa*NcfMQ`D1$SyVBd$-hOAL(1_xV2rCGw&|U
zDa<6QHUY}a3bAR&eCyX>mT#;WrLWQ)Ts}hna6Hk%jhWQzV1YXM`Z0y(cUA++DzV#N
z)s&aFOUO)3Vra&DV+k+S6TW?$xLjd$%sc!^<0BJYL<mK}0grv0OdM_AMD{e+$rjfa
zP09S@Na%JGP`KXR-Nn&SeL%2Bp#wh#&N*x?MF5;%@)|>zHlf4&;PBJHVd)Kk?VV&E
z#+KzxS6zlEj-N?L5&mg)y5C*-q}YbNGV=VS)MU9Yx-d)*dd;FO{&~>i+vT~`n<vX#
zD<<vWA{PsjE^gw-Aq0}ZAn)^Lg@U_TXqD)XB%DR|rEj&k5M8n7`0zi^5NqynPbIn8
zbi;Yfz2>(dlH@^5WXa!9+y@;Jm1sWzs1R>A2WaSC)wTZjTl0%yC0-;krlO$JWmHQJ
z&5Xtsdpf`=lk$kHR>$|HaYtpXi=4}SVM}7?EI+(~Z)|2v6jCp88SzBfTWa9~{H5%i
zoEMJR_rWLH>xv$HDs5VFa{B*mfob+yhdRKXgSbvo1fFL|E@K26j91Uqx82#M&vrN0
zn)xC9zrMdgx}c|>5#ebU6p~Z=?4Q53$Yvc~eNvqM*DkXbmu5w*KzqM6oi&l6vPtlq
zR~PBggBu|ScS|d1diZyS#^@8(ggFE)dgnC8%4Jn#38{sOGEDQmi@%xSF9p?;n~1U%
zoTU$IKUNaImhT{BE06kRiNYo`t;~Kc=fDG?mGFr78w_t^ZJdAWv7k6^XSd7ZicPjg
zH#*r9@^vJ9HSaSDJ$*U673);5A)^I#Oh983V+pXrNW<wAdAB&a=fHhF$@Xxwy?p5j
z=Nzh3B$PGw*BR3fE5Y+;O}v)d_YvNEB%h&VHYUA4aQ4Ul5eE?}=B33EQIba^mZq^g
z@W?sRMAj;95XdyRYTVkYPa;JQDK)K)h~nRmg_PPD^wIj=*D>)Lp`sb0MLInA{WY&E
z3OR+kbPjfn|Gw;qohlsC(Q#nyj$K`@t7!{w@~8Wm`XkR4B`6nI5qA9{$Pme=Iws|N
zBMWs)a4mM2T*>j$4;Fv@Y?_{JoIC$(J@G=t1=JJNM{DJkIx6{CXX4nuX`du$NQ%oy
zD*fF88E>*G*j*A&xZnS~_e(P-cU=zrBgFc94OGtg(UEr}3X_@!XwVRH9j#;%f^j;P
ztv70B?M*kA6Yw)ij%5G(>2OvlpWNNvRseVh0x;R385tP|TM%D@jLAuL4mP-)Au$l)
z=ri4}jZ#w}!03e5j$kHK`JBLT>D<HZxhvB#mgK)Br9dA5>;KsdPggu`Obt-iW6<J$
z;=4~;oGwavbS9sz-8p0No^vE!A`2Eo3p%_y7zv8Ljm+q#`kT%h9elPIF0;#?E8{PD
zqy>CuN1qSnnwvKT8$-R{8uK}eQu<~!*bu_5{V1=j`PY!zt1}IKx@C!4cuA}R?t3#!
z<~iPahMx!@Juud5*YqtWUm~7t&%=HMo1mG~oz_iI%1r}api}2cJ}xHa1MvF!(?j4r
z2ngR&K<##WHb4iuJb%W<-l?)YGrPFDCUSjxS%CU=1V+0YXc?-BpD_?=l%|8#3kU>L
zac0?`%<_q~1awpoe@W05uDoCY*v=(QhZ+HEmC|&f(R{+@+1wN<Nkrx#YAg`!Ae=WP
zADJfcKa7+oMRrzw+WA|}7NE+t)d`Ou54{7KrhGzFh;&KkaOXRBXAy(U>C%4}v$KO6
z55W_uRQdg)e@lhZ6z}Gxg$sFGTKHCP*Xyan`H>U+#UVVS8sOjon#OwVe$X$9Y|xB^
zgn|O584w`z%LDr2SS$CfK`cOR>;!kesF)ZLJ$)poxKH}2%2w9w)&-9NPj(a}XlGYQ
zyu7@<n`}#bC7Mlf!T4_Z5AL901*F4$3k~+OZQgu9hXM2tClBW%On}6yUdytn!i@a}
z=xblJ?gtAE(V&H~wzVY;1R7;6SJ;64&ajuVgz`dGyS{GA@fHCi`pTNm#KY>?zR8Hb
zJU(}y+>N<UvPwpx%mvx~P*SEOs1jI6zhZEf(=t5x<cr->m#S?g6}(c6IGo2ZH>cNm
z^U4+0Hui&}%hWxC_3#<7YD=$X8mH{^KL5KKk2pm1)<ye!vy;o%Qd`2zqe!809jr!B
z_n-78(^{Ol+lwZ4UzolCkEnZi0L|JIVB$vC_V5!}!>IqPA>14Xi_|@s$Ygzc3lzp2
z?pM@6tO$R9b(jS}!_A1rl%@$V{L-OOy}L=eL}{YU+ZzyT90z99Yb|sEGZ^&ueL<^V
zU*CRTGo%}L;BUd#MBp<+vrk4!TB*|xiB7HZ8}NPmL36;0%US@h+T>r-{tDD8If2wn
zpU|sv7=Xy1%0s18!ABTHXilAge4_IS_eLAC83!?Xq{CZZ7xaXFBJO#U*x6^Zxr)aM
zm2@!eTztY0Grm_gSK-)&Mdv+*mPoq~2i?K`-Ah=I)D%v-IF%xg3&dfOdi2J1mpj_i
zds6>fuh$vvb9RzR<SsU{K4R<KGuaGr2tv0m&f@^InXR)XAR;3AA|e93*v9rRW8&i8
z<KYc~Ux})j$!n@u^+nwYv=GR+?EFBs@79vm+{_I`z`<%Iq#r-}b#w?oa5^1)Yu6!{
zxgjMZ)2GMmZ-YU>*WkSZf{8((c|W<h*Z>l5fIC*J{c|88B=iC53`VyLlPX*Zetrm$
zzmc)9paH`vFC-w2@0^;7V_Wu!uCHgePdjrDnx%4iPEGt*Qrt&{MboejJAEKS`<#J(
zAbAt*F6~CVHB~Lwb#GUjL&zheR+beVGou*|N%b>BRC$%PSOQvvgnJcA5L+y|SO^0`
zN4e7=AGFX9v_)oahdAxtQyn#U0~IDH%IEmTlGEsT6tcu6>VBIaCJm#3nWp_v8@cwz
z9bf(d7ld#E{_zQM$oy1n%F*2wRn@;L{CQ3}8a~9t=vaGh&#B^Z*uC}vCOb>xt^8n)
zVWpu&>)l?_d@XR=F#yEj`o>1PiygUQ>mSqY?Y^J^jSpNB{(MJO1=uTH^NGOi`m{Qg
z!jv8n4NVwO;>P!E?eD8UsQ{}vb|9p~$Hy<$Zc78g5+G)h<ctGGbjw#vu&}WBt{d;3
zp6;!I_O`3p^_&ufE90xz);-~PLp<64VvMH&<4fUXcKB*09_n#13`Zm@p7al-9{`c-
z#Q}(%qUH{9W#C&EtB{kd^0CHZXBAR2wS5`p;DhY1zyLjp+)fUF;aUs?T&s)DL$yE7
zDdEI5=Eu~0-9~MCjJfbvx*DB;R+|9^$T2&F4M@nFzvffBR)yv{s!r*NjoZGgBY##5
zKSbM$0R1@tvlD)OV-;7!uFLkFD&T?<a5AkG6cGRA0C3mY2avn|o*3p-(LHt*wSKt0
z{UdEz@JsvW@<KvFW5A~l+p@}jTl&I{$GZ8nn+DtiF9sBnDeO@Iu3nFOZS7cVy^Q>O
zJ%Z0m!tR<A0|P_L^HTA}+wSRZpYG+~(cw9ishp0GFJSn`cJ)_dng7nnp6Jbq)p*CT
zGQYV&#4X<hk-DJl7n2;ZeyuX3SBXfqMjhD`fO(WkJc#1&oZRVuG`gT9{#0A;5A1t&
z5pRf)1P+K=;YfwspKuNAuolw@WDP|@osS;=BrksCrry3yvzn|#zaSxg;imriS4qrC
zDmq+SZJ^f?=!KS>94G^F{nyPK&{1U-Y!Yt)W~+J&u246MbF~kgFS#xFxTeN%=Ypt|
z5SPN=zr!p=cCH<o1||~~pj!;QgD#Gj(-n)9qGDnKw6t$=v8JgEjRzBe5tm`0_H=Z~
z*w|Rj53&~r8o+p0Y8-|_;fDAP46`J8<MtV(lUc)9E$4KMjg`HSYHMpDfX~F=J<KDF
z7nUb@Wyfr>cpZ<9WC1g*!6uFSSG4iJ>Aph%zE|`3Brpa$J%0Pz&HwLDT~Xp2c1(|=
z0#*|<DDB#SR;x(5%VWz^9d+RCNV(jtqw{#7{lbF(*-#~rVaYOCL5nV*d7VSLQ^Tp$
zc15fTU!Z6;{@C>@|Da^`*nvIf>*WpFYyymkz3rK8&SJh+&9c_NK>E49=*C_AaB$eb
zAdy+xS8Zss*(2rf;K<~XRuruQ9v=S1G>k^QmL8;i*r4;X<ju;%^H^&s|25}&(yav_
zgqqdqOZdN5X`&^hm6h4{yIV@PuUn?c#0CZjnQv8=yq`VQYRtN(wY{Yj6h45RM@rE7
zxEB7&MM7L$R9t**tt&*M+2vIF;2)uAFTLHT_6v<vPZKQ~x^~oE$JC!3Jr^z*@f?|Y
ziSrNuR<|TZR0BaNDYVbpRUZbW0a5(VGF^E`KR`62ROYL@e7#+Z8!SxeQU*OUojE@t
zD}%WzrK(He-PTs!uUL|Lfcem_vg_fAjnZ>w)|!r&NPqv{E-*!;)deOnXO>I;zg`~P
z8_6~&`Q(bQ)7;I}+E1S{$M$!pGg3J^3RwsU2+(QNzJT}*R82YnBX|u9dp>KNzIu4*
zsjf9NJB#H5dKwFia%w`_@pRLrI~#6iS8kbg9~&APAnF^JmhS6Wd){MU^td0l1Ovqz
zFc=}>;c4;6WC3PJYL-ELMaR01aVsyBIiQCDb~+Xo7B5~W|9RppRW3O%d-j0a5Re;S
z)Jkf9j{VDJk5H>FKbcl2{~kb4u%l?dw?g_9tI!%HjAH(EwWcak3x+1gY%Y~)TB%X&
zvl=a#(i8EazTH=f3=JG{EJ;%Y1T0wZS#eV>l0E}}U+`sO1JRJ#VsmEZ%>*t5_4RZ~
z1HnaUuj(dO!Y{lY-;qzXt1~>0EO`ctC>o*QJ`YRwXS4l$3mFhO`W`MrTrLv(UF=8l
z{T~RdukU4EYgQCDBEE`36M9Yc?c+x{NJ>cg*SkFY*dzgw;%=!|_v=BTpB#-k_}SC9
z`#OgP-A8zwjYJuZm)x}sS?xc+@ut!kZ*QKUXL=QCGtIRQzc*X^#DVb7!wE+^;_A?3
zcY7cmgSL{2qpKKGIQskb6?q9v_eIlW((YukQV|L0!GZ8@zcWg-HIg3DuCoF>jtZDd
zfl(y{kS-1HaX9Q{7r&|mwt8kg7PFPmKD3&J+C5pc+m(*{N)RvGS2!FOm?F_RIXUGk
z^d~0+osp^=XP6qWUj|!vXSt0JP?<<bNprg^!0sMQVhK^LGO`7Bhe4t*Lt4teuk#(s
zvEp08Kkw%cx-CXSMd11FlId;=&A7q?YM3?FNgEUW(_0S@CDrT$Uw)T1V&)0EC!LQR
z7==$84;y8$^gBrOgEFSUI^jlYm@9&+y6~s!5lS4NE;n2mQRxKo5ic&7kJ+6l>$MX}
zJPyv*zxsQ9J<>C{_`-aJt&#?bf-jFyk29`ojeIa1ls>*v!tQeO)<U#rZ0I{II-kS_
z=KPZ!?LS+fQoI`RuKo3A_}}(zmRYiSMb4o@!Qr;_EUCSbQ=vN(^K%qx2ETn?`7lmg
zii7M9TBq+P$tP0MDXQGOyfZdWpj3{%M!{yz|LCNqHm7*Ky|u*%yiFj2PzbcP*OeS;
z<L!Yq?Cx<8Y<=bJ8@4PMij#00HXu=Ug6NN~wVY@DyH|+fc-sZ(t+@Z<s)Cc+XYx^3
zBy=1yBRFsT_-bO9z9MIQYDK4XJpQ_b-Oy?8;EPH+%q*wjLr>~T6nek)-crAzGW>kF
zqHA%0<z`6S;-01I_~w0&bxEh~9+$+8xMwpzH2fs)_VGvI%dOq8<55=!g1?63G?-hd
zLDSOh`xW%hNC(XUsp6e9C@5=*QpfM-+OnYG|F_}BZrY(5YV_HA7#S1Gt%ynP8n4Y4
zBVFR+YY%=)XX!po$Ydy~p23#6E_q&Sg0LwJglt`kN5G{A*!iWotieMBy&A=0`hP9}
z5|+FvfsBld(QpbpKkx*6(f<S49olt#OpMLhh8XyKA4vNs4{@-t;D0M?dj^*N@N)rq
z9t!UbAp~fLy;yRg(V)K2D<_LEK(zswPjj^;9-A7TBRR$+X%H{IJ)OZQ%GBL)v9Vns
zeNW_ZcS<GbNsgLK9>)VzOpt3sjFxHhfjcWetwcC$Orb%3DVEL$KwxL%5_GCuBrkq}
zNmPw7GOD`Y<!2`2ynA=^gKUR%EFk`Dbe?+jQhvQVaGMoGBr`NBo?hmyvC1mOOiqoF
z+pM4^$x+5Ot{qhix2J6xXRmUd^DiR?cEXtb7mj;T`|ox-L7%B!znlHN^Pfm3%EX5Y
z4KQ=hsuiC(#6)PHS9ivAD5ahHKgTK;?3yGrCasK-D!NM6CuYG`%`W3k*omq6qB&3T
zH!xdAJiTqs2i>o!J)W@Kf*tFf#I%W6S>vX)+)L)}1qB6*ij!L&&ISw&3{IBZk8PhH
zua%pgfjybh?tL(U*NZRXlGnZE%O6mnnbjl#Pe$dzUd?6~hSSs2m*ykj`fH;x0YxI9
zcc^*L0<=2^ST|TKx3=^<p%8dpyn8WrU|kv--eYnBQu*^Ahah`Z;<5(~`{_2NJzlSe
z`kIM%z%|MulfLNTtNjQiLO6D8W&SLb?};S6<Qv}}HG!C8Fxret?}NdXqBk`x3%e<`
zMB@W{psFH4-j4j6ZG~ff3ZcagiQ5{a;=!@W@8!CSSSKB#N4LkF3y(*F3r`PlbI1qO
zU79b^5zaR~k8FN9UGTcra>kh<mNuP6zvC{fIE-(o%TM4!n7ARw5LbLaXDj`Fr{0Ca
z`fw)9zlBF4x8;w1XQoBmeWp-|N}UxR@S7k6GH)=|Jui>U9~Q+39C?t^fI|4v05Mj)
z3+P6_g@ot>Q#T<0h6i7koQwnNIpeu%(}TsP_?HZ6jt+>(fp?ef2+#G(=4N2K4u9T+
zF5_bjW3^I|A?UiiTyz<N!N3LE57=T{&08DKRvO}R*k#LGjx5;A6)0FNH77>!-hDec
zu>tyYa1(N$^%CRqxW$mAdyM~xa9G#B#A=RDrt+8XYLiqP`s>8i)F$8CoA1^g`QkEO
zvZ2}h<>6SQkKOZ~0;Pw4^0*QI>!Vkf=H4qiTiF&LrXDNH&0>2-8@ZHS>!yajk3Z-c
z`m(ws@;1@%g=eWx_So9Act}rDJH*MsV2Bfnz)G8)*!t)g=ls$H{E&Epy#2^lg`Tcx
zF$3-1FWR|1i_6hIwKO)SYgUu%38AEn@8x@&5~Reu-VJM&y0BXNIZJ4lx99kRy#S0A
z8;y9EEIh5z&y`AauYMfgvXaJG>+2Qac>^EmhwCvBIBPVA!=J@qWnOHDFZYbybHG`B
zi-#xW;NXyY4QSQQU9mTlaz&Fb>Q^AKoa89w*w!;P?%C*zP6VzovDezFF6gnUEx&;S
z-0Tla5s@CZLqMoE+q3}N<QpeF_AK|c2r})mw5`?Z<aJvG5}6D}BaaOiRC*JvSpz5Z
z@5yOAF#D1s=XEHlth#J|lPr>t`$QofjlK1rkX1iPPW@cr-kQZ?_*eOTTL@8FUt0Wr
zr-FA=xgEKZ>|Y%&rAwW}QC9DhK1WVy-y6iN?i6;l3z>VhzWH=}V*2!Oq5QtK=NDD$
zopto@Z#o})u5tzv)aB+A`OG=51<<)Lo;wP{U@3a*-KMIQCA2x0CZm^7gEKG7S6u)3
z_t(Rl;^8LnT2V(IHAXwSUzO{Q@UGEiN5-XwB}X&8;$6B3+3D@&ZTW73XR+@PSpIDi
zu{kX*Eju?C5_r{88to`e7iIu3LywERw>CWnJd^s{PlqAf$vu3scbk5d5Nc%3CyuhM
zBjo;Bu7Fu}*5h}1pkA5i?RAAa)D$#@sjP1;GEkb4GLHe9hd?+u^zi9=17fsLxry@F
zYO;uD8-<i|oton0s1Js)-ua&L82M;1O*6#Xw`JC*j_+W|x8b`@fyd_{BQ=xb!fG3G
z26E{21dkgUC|1MYnosl;aLC)A6Qfq1YIiM_Qo+vja<c>@DGrpaY!n2YWlZ++tAF=W
z`X_kM;_`F`W*Og{PlSc0oRbd`t55H<lV*KCBq&*rh?v$N#fnjF)U^^vF;afBu}f4D
zc`$)>pi-XZsco&hQ*@2$YP}4@ZvHQ?0waAV17Ts`Y_xxdGAE$^^QM0fdTFV}y~^#<
zjNi23?c2Aq>o-)!4_Algk8HHFv$NcwZcLCQ)zH*bZFDG@jCi>+x*c?GomiVsSxB?Q
zC_<%8GN|pYuHztKqU9eR*0^)Ff84GSjDfJ#`6ayI_FCyazf|1US+VdIdT9poTVGF0
z0mcw#ZQ&ieU{j@jyPHZpOjdHzbZ$m3M4N$+xKLAjlp?aJqXbv=(QwScP~gZ<9U+7S
zAN7{i_APO)pf(u=TISzD%r~siUgG|LK9+xlsUfhtT!XLu=}r`481WP__$Nf%{QK#U
zt}JR!uR95<%}0V(&XJtXU5`tCX5vh;@zc^YN6v6?u2DwlJ~~aTo(Ww;#%>oEwsupH
zlye9E)MirQdXyN~r=!1!tr)NU{BoGVc;)Yt7^*ltZVp{n0N63dd+sAkSFnh{VH+-N
z04oH95daqgBzGhjs*BHH4B{Ah$|BxC(DXn^)AdcKpR_oYZ{b~+nZjrf4sW>6>Jtbb
zUU$>VpYTNUPcb!*n-q7a8+jxWYt*#eP^sAwe_MVH$|}qq66AN05V&^^sFuH>&3Wp1
zrScQ0fs=D_5w$F?RS|iNeP(R?l$<^>W__k;b&~bOP2EIO$QI+b5R`s$dnTO8dgcd_
zh^^Jhj`(or=@Rm}(S8Jpo2&1);<k0ku+-r`T#u!+M}BtUt5)<>NTkrZqd2#xIV0>%
zd3R}6Xp*E;Ewu_1N30Va!KpfWA$u4l=DC3w>YamED`q`MbYA+~YuYrN%Kz4f=nII|
zkjKF4c+@P`%MCOYs6k!S?^|dxR&G8`3ml9Qed6OWfZaf)87-I70U;@A5HPh?V<~+}
z2b1NjH|EKBWs@!<ED&U1l?N$HqEZpk-UKzpemrJF-S!<*Ika&n6A6;U$?`at#{Gmb
z%_D|-(vAftj8BxuIbhkFhjj)^;)JA5;m47762$L$;reJ#=uQgZE~a3f=nWx~RTGR0
zsj@3V51Js>-Tytrl)}}KRuLjt0^`=vFUx9jwU-~mFX|CTXYrPleiYQmzsiDIY0wFU
zA*M=y{Gz&33L~QQ+&4z>tY8G!c0H{xS44|XUj)UTKg(^5))y=@ZOQlFI{cAp--Ffy
zm`WU-HZu9)J3u77{m+*bO&<WHT`&IH1tgt7wzl6wt}caufRMmujsJ21_eK&ag;2em
zu)5PWD&T?mIUkW*b>!(7Or2L6olC6DlxG!wmR&#j7S*zQJIbF^wpYb~=4oXfQxwdm
zw;Jz|(z00pyiQSrF6cSWtp8^aUR?L)(8VyQBqL7I=#up2jJw$xJ65f$<vi^-Q~8$P
zApxSd62dxY2CH+M^Tktfn63;t>EwNMI@d*D8cLY7FvL2gz<tuqlmF{kV!4R$sN6V#
z0>hfIBhTB!TjJ$Z0uIsf@pHb%xmmvaMo9!Kt@yKE^!K4LKJu#ke<KwG-ul6ipFjd?
zTW02^fRPbt1oxR}4x}{zoo`P%QK(wAK^_JQjv6!M^R1!IrKL2oG#7zH(@)^Y#c(}z
z67+Xx(G*$r$>MOKNm{`F&N!}}=H%X+yx)D?ktHAP3{^?+dhev=w??G~o9HtkB;J0x
zDp<~ib4<vGe``dEKh(S*qjYnEAL@3@0E(s@kt0+QUBNr;kYAc{sxamgI%~Q&Tad`Z
z;S1Me1_Jkv#SC6@POo~Vj+B06tV8a{OejLu*yIs3t}!B)Jajf=Dpfo0qUT8U)Elkn
zZjHZ2Rr3+<YAy5b$uDcf`wMQfK8c|q_2w9%he3uxlq7SGI=b+`U;DPYKmd8!9n73c
z+6IS};CJ-X?F>2v+v4nUT3eaJ2>5~xrtu_CmF(BuIamU7*Ip2nU4hLG)tu1FcS2y<
zVvti&E2_abT+;K&qR5<7<}o(~6f>FLyuMfFQpk~GGPqMoa)#HPD2_EC81(UP=(o_J
zbI+OHC&7!srs{Uan@;zo3+w3X%+WG+Y}`io5QMtfJly<r7!mBc8Gqh1OQG%RmNTkB
zyHRVC`Q*qaC19rY0gX_1pCJd+t*CBXeOg1#tDH5gen_O=aXfXary{2B%R(G+|2cj3
zUTdf0+A&<}A_Yf+#Q2ko+NGKIWaP|{NK6<xVtAHZJE8W?9=B~W%0#sw;<-Nje+LCS
z42J3ivf5<K50H(|0MG*TqreUq1GsnJ+}}?CU^pdZZ^pe8ALxb>Sj^Ei8yzx$ZqTSp
zR7z?S*bpl&8B?}>k}sS73Q``YwAx&@9diPQ==z5gVD^2{yKcwCq%|fnX~L{0K<jH2
zs^>-Xe7n?@wI|fRCl39^?e}qF%8esm3qeGoV?3)rvB@G|=+`#<k5=dJhL2@_(Nm#B
zj|KQw5%c0Dzjt#d>o`90b&e#^SsK|VJxn@X8OLC6A`1uKurW9sy^*WYk)Z9@-Hsfg
z^c1jwE*z}8XtKnTJq*Z%@><lj+4RWb=19t1^5LmMc@v6FE*gSVy}cxSG$N{m!$nDf
z!+V8Au(*RHw>wx#*KjR+M=|=wv0=36yiS@6t5o$i7sU^>b-n{_WEXBN8bnT2O_@?Z
z9<)EmK4}SPA%+3-_vcGD&)E5CDF1fZ(^h4;F?JRc-)lyIdjo_{2(-iso0q$~axw!2
zRaFy(derJ|h)76Cns3%ZjC$A|_dWoaSU5}{tTXWBWxL0$RZ=pWU0=Z5b)252)74d1
zu8n;cCNVY3Y}i?s#v@xKk59aB;nw0+T8lNF$STMQrNfNLuyZkEPPMHcmOe4m_<EvP
zqWmnZFQBv@$(;<gK9gQCeSISbU<kDr6<;YBLXCg)N{w{S4op_M81=VC=2)nl)<h<|
z?e6vbHIPT>8^#_m;)S}Q*hkyQ+wMKK!x@k5JBB;OxNRec0G!nzv$>0J-5I>c&mkfE
z73D`Pzb1AmrHtMQF!N4l|5nFIl{Svm>o544H&bbd2&%XC&dv@jMsNhNNuo0zBGk<X
zv1uaY-c6*u#bz-JLqiCykt-;7uC-v8_`wFF<Z(bhXT|!x;MH)C!cd13X>?y<4};oC
z(!X&>;0)7F3;9lCgH|K3>Zb+B2s+&iIuI;?q`atk#DMYiw1;4QWyPnes!FJ*g2de&
zkxt0+i)6y+{sn07{N005b(UQqL=qMh8vCN3_qn9;l{N|~YG3fk+qPRcvHSHtnSsyB
z3O^)JotxEf-;K8sAHq1d&(5<lk(R+O(Zo){8mQ=YE5^O=r0Oj@;sb*>h6zMJS)J>?
z{Kdxu>uHNc+lmjO0_t2Fv1Ra=siy0rnw6vN918=ro0-HIOv*59XKfh;?D+@nbEFeF
z*5VsAo9pZ1vznE?{m(~%)Hx%SS<u_+Pcu!oik`S*<)OmFF#fN&+_@CcFJ0cTakgkF
zkL7%RqC-fP4hBNJtlAUGH@2EZhmX&(sMEN3z{|HZecC@YHMMyfEi=lT8G6aO^)_f>
zK?9^nG)LdjKYaKD+BrqdG6>CKGGHa8{Zn(t9^&5`(eFy4L2Ge^EjnI`sA^{8at&BB
zwQVmf)%QEk6j5+Q3r{67uX*!zZ>jQ)ns>alN@WInYrx21>DwIg&L9DSwU555c*fX*
zHWulMC_cm?j8o~v$mBUJ>aFG0^(O-Hhkao*oxI!)2TIkvuixiA0>C{rx_!#E1Fn(-
z+_Loa=~t^Xv0+)RHpz9jzCV8uDF=H8qqCb>USnS%oV3pl#AZ$eMwTg~XD}(>nrEoG
z6j5p;N^w=oA~cJf$3Wjvb4&B9NZe{#LAm+k1IV|frR9T*wVj)PYbHt=IIV6i$e@aD
zNo_{Qu9{nV#y~?;{NRFzi)-W1?DaHUTU8akbSl7X7nYmk2rx|_O5-K1*6eEF-3Lm)
zS0_M5r96+S0_IjQ!^T@1cP0<SKswsfENr6i@MILAGgFc?@>{&J7!jyr@W(@(Lap0`
zTSw;&6ZsWUGSQZs<ZK$2n#FS$^PXK10&g&72savKH)n>6<L%q2%B>1g)TXwDKVZ3g
z446Mhq9R=@BX~ZA_uDJv2S-Df(}avmB|bboMf|bk?eHX!=^h&K^ZSKXgEyLZ>G6iE
za-||}YvjvfO2HqTnb{w+lj-7FxkEm1;l*>%>QyUN-sg~;J1hLOSJL6tdy}R8M3%5l
zq)rpMn+roUCrLibbbL<l{&z#xe>Mq1u5&~XSfIz20s+iFW2^hBX7OcAK~jqZ>-oU^
zK<vlWY(mU2h1u4hAux?@<l4`00XtvD!%5C>ADLV|F3s!jmNYyE=1~qiPSdYCw#DA!
zDLtz;lg4=Z0#wBu@2RIFeU{OgWu{;?QHg%=c+aaCRI#gN!DOL`0jlsC>~2AHJwoR*
zh03;P)t>5a*t})t-t0*pY!R1}q8*tlJrl+|l;3?c<9ZXNZ(ZZ{^YZT@!DB4sF&l=?
zH#(Ccay7ap2t(nI1P5OaL-)Ze&U|S3xwgy{+f!15rTofr#4kV~$;sri&i2&hvi8f>
zz6Nt9+>uu+ow2lp;5j%&0R|XknRdLut5c<#>YLLd1fH*bLP8|EEE(NFWXDWd^!x#G
zRcVlz<?)0f*EiPRJ3obkRP^GSh>s>1EgECN0g-j_JUw`9#yn<VF<v|0+ovpg)rA)u
z7SvBXt=$;_Wv1-X6k!?M{WZ8yVp&nLrRGA1<mMXX$?G3G;*~l0(1N`1v;dbw__XP2
z*lp{%BXuH;8-s1i#{Pa}U+2o@ymzXaU`}W^<6bLMzPT(SHC{)G5@b;q$#%*$ow%B*
z^_FKgBu`wYvcJlfZRN#Ryp16+nJ={zrc}!R%}9Q!kC>Ant)*zMlcLg?1ip3uF`z#?
zZ);OT@E)o0qrFTBG|%&k7a;r_xY3!%(QIezzv;?TIl)Q)(JfOxAJm!DG6m|oBLR#P
z_NI*aI)B|Gbhl+6yXaw3)6$pkly0|j-on`zx&K5qb0PWQm%@I&O(AVDqkZj(GkWX4
zDJ0thuf?S}hFYr>2;)MG9=;6~>BxE6NTG)EeC@)W<l0$Foe&LY==ktk-yMBW5<=(8
z#TSjF`f2uuxL}xfb7jPzFgmI7_F7dIcRwaLy+^pXNTLsjobTALP5&>|Tg$ci`ZC2a
zlY#Se;1zH}j>%k=ZM%cDQRti~ftYz(l}WIwRBgsP-s>xod87QJIi0MXii{|u{0?so
zL3a+`aH|mEj?a@#*V&=y6jMa`Sxnk4cN{yb4Xa8qmH9-03ZmAlJXh~CfjR?rsKv8M
z>e<k5(*6#E3F!_$z%nZ`B15(43lJ00)edLmYN%n_wZ&ks*&SsWXd$KLwR+GWs}e8z
zxy-n-yI#(gcbDm{MlmR@V6+y8jdj=W>F{k&ZQoV_n<Z=6J7p`XBsOhE2G2izJVwt|
z@dr}*C#*CsWYe60#o5&MiS7jN%>R<xZ^8v*mtf@dx8$zx=>x_NOBk}x_?7X24WC8W
z&ccoDS3z>)3}2z<#pdTr7y75p@#~c2^ZWI7jRX%#RS24Ks3dlB=+;4b_7qsLE(ntY
zme7<(^3_iFzu3=z=wM^s*t<<?O-(`xXaao;F%K<to0IQWPbGMq@PNE6VC<ZLU-;0p
zLj90{U_S9H8mHPG8sxw1t`)DZHU+F)PwVP102b@ta3al3=FQ6=oKb@zT=w!qxLfa7
zlz6Y<w^2TkS4AhY^S^*Ea93va3c1v!80rwDHfJ`gX^9b|RyGEy5m}DTTN!2AYun!5
z0~u&sI!{<P)M6ach&1)HpTqH`F*-81N3rD0|6aIzy{_Y7(NJUhm-&P7yLxiI*T7M=
zwm;L5Q<J62t2o>Be5+WqsrK^=tTG+b_h&biEGF|So6LaT<yFx7ZOW4CS}sVD9k9nj
zfS677ygwd9V82xjh7_t+vy|?oa3{f!4cl!f>np`2ETv;We{!T^GsXX9DE?@=unyzX
z2RpZI_+)e!nP=OzG7o-L>LYZzoQTsn$sEnjWp&nkMRB4V-`{O>-!4vGbD@3r{-e>f
zvHyE$IU0Vsb+9iic&6qCgBF(qsx4caXLuyKm%(%QT;)T!mGmWVIJixm+wR{kc3<k>
z2Lop4JAYmGDaCbMjcSMbCaTA@R>U&&R-@GD|BgfEX;uDjF6*Uc7`kE5aj91`{R|JB
z(ndSPOlx$$6Xx*8CfoHPBrYc4YSe_L^1=K3!DIMj!6ocuM+&h-<6+7~=(oa)w})0m
z&xGPHjnYmlKB6yC-TZ*7srkA&j#7)bB;~nrY#yJq&QI*KO(6tr41c(^Dc#tgR~?Hn
zss+k+<38!cT@-;kKL*O3?dpxTWvOC+Aj@g)zMYH(lTLpke~L`~`8`82BDsz1)uGxb
zs8h-Xw4aSq)fkSV^ML&u(JnuvD@GMy5dE2*e>1WrWirO{I;=&*g5(7p0>efVo6TQ<
z3|U(%*Rq*5jNw5!{4Dmopr<OfC=eIm8)#C-$K@eFel+ys>#0kl+A#-Z<*B{GI!A$`
z_&`J>?I-73$9Q!YPc)aZmqe_IC<?6%8f0C#2PNr6{vk@L9{&rRM1wkGO~av~NamYh
z+hpRWaBIedaZ_ms$>DeQ4NRV?u(3yved#}~%1#sgNJfIFK*<f&3=fC+@t3}A%UwFB
ztEN}@MI4zDe&=lW$e8xFLEyBetG1T0jO*h63bOzbYRVHg7Z(*O6z~d{D?6D&`>T>-
zW(*&2VkDM^9CP6yM)(P+44v00s3;AuG1@A;XOF=dROINMoc2#!hqW6Z_AhrX^i(Pf
ziQsYyzuJkG;5SydBT3%pbp1c}-oh)&Eo>VnR1^dZLXoiOQd-hD(w))*(%l_`7^DJ2
zhe+pubeD*LfW*+<-JRb(c+R`N^;_S6@O#fXtTn>SJp0+t-uJ%ay00s!GMZbCtDT6^
z-#aSnDdBDX(*07Mw4xGeo4;1%vUTY9JnJNw)EL=}%Vo^u+cf<VdaBu)172Ox;|6vX
z+T>IO9*CT+UfIu4$n604yVv6Jl&dK(zk9@u%Vvu{+TrhkTR1mN0<%CWZX^vGv;eJ8
zFQ<4PV^)NXmfYj24yc*r+)VR5iQb#_s(Z7wg)ZK}WBmB@DQ+cH6?|}O3O;<#=YL6$
z6+I<N8Ou<c<;#jW@l3gOMcCSHL@PYVQzoj&HHO=*^Y$p=2NAVeE6<swFC>px$p!u2
zsaG}2bWCL7%vtF+oT^QI7#YwU*x&U&=;(|$<53CFu6P=K;V9<z4_>SX-Wx`<_)S{~
zR@+-r>zcS2(C(!Em=8S^7*FYw=KH4T(U@@4g?O!pYXpdSb7nmHZhV48bZ5MQ9Rp5i
zft^Lxp<A&+pKF5^gqn=l>e%b79GM%tXVoG*NqSRtHLQYA3PKGJ{(OKe7izU@V^zIB
z|24L1lY)*O?>`49)@CdLCVre+uQ?I#mg0^t;JBR??BzKeMzok3YKi>m-71(=%CJiP
z2>JWL(b0@&n~>Rr@lCS1e+v2$M=RRz5|CYB42?e%<I5k_6=b|!3R6bLMfq7v?%$le
zM9mH-;N+zg`K(*JvqEx1#$orzb6qsJv`yZf{b(rD^>!^uo6;1x!_v9*IZ?7B^X!Ny
zX%r*W+k|yhD-}za!<SZ#;jz}MfM+Tv_f1R!qxr@TX(4<$btEB}?iFa(#XjaEkQX=N
zyX)=X!`tURcB#ktjU^1hxBKNIb}eUS6e>s}kLOCD4@JYMQ|;`<^CbA?PN&=W!n>?P
zEoYY`=PJagY+fI5wXX`k6n>6N8Wca#^Yg{=;i{wCqj6&8s(tz=%~8eDs+S7}Dr>X)
zmKfb(kEB<?nOL?Ix6z`%X4zQRH9d%DVdyvBc&k6u2MQTK-PV)ky(&K~pBXTgk$(UB
zAx%Cn2ib}F2j#Q8b({RyLL~5ok824nWU-!EzVnQeS2LM6aSpnHJ25b2#X~s8)pm<>
zFMlcZR%TcF?W2i@`r!?^W=;ziVi|Lt)8rqxIa7`XVTXRI7<2z<a6KnitorXGnN^5Y
zsO<}`?0C-|Ar~2iQ{q(Db8Wwn{rC?>^(t0D1C({0`fZQX1enNAX{^<Q`gwPRsR8Q*
zNdgwFN+*RmE9U;nmp%mr?7VTgldhr7!!*9I{A~RyOIUZ+aNw-QI_Y|||Ab<7-6uZ&
z+SiM;oNt!K6)}&kuM``;y`pA^H&acRP~Y5*>P@q1QzFu*EtEBw>5a&A8}liHyT%cq
zB!}E(JR$y%y2dX%?BTlm{8xI@=sV;b31^<zFnI<^PdDsJ5Etb89NE48jUzKTzxYNt
z1ixuuDDV?wYJu)1RtRCjd)lDqLY67P)_f}K-&G9mriImyn(aTqW<ls6?O#vFihxCj
z0o*;`Lu45tmG`SlJxQX{6cQ2al|)$Pnhb<C1=vC&^UDN;Tpy63^hc-nTWQ^YRM}F|
z9_LaPi&cN=T(Co!cyOB%^m(`8fjzj#L@HQ`bGi~!;M~rj%T4hf8HzwM@Hm%REh#v5
zZ6&@nTDGj##vYhs|0OE=mPGp6qWsaom1@QRumA)o<0ag-!sIw!8K|l0Z{(BxGddLE
zwSJeY<KXQ(^QMWctritx4Y%QW?im(zd2VsZ&Y>s^i9^U5(-Kb|bD%eINxWFUt~lGI
z$??`80-eUA?GL!!eg52^_q=U3(naC<BPK0W2%VCcd5ld)o5iq$ah`p|`z`*uZJTwS
zV3+o8<xG*7+GWPP1HL0bIf<HH0sxCQ%{|QFvkfAsV}2a5i?nr}`X$qmnEmyFiZ&r6
zAG#f$UPwIQn;idEWM9q|N*VFZ%+98O=`-&9{3Uv7puFf&8^5@2ITni}=pIz4(EP(f
zP_nDqX|vlFa@P}!OK)ZJgwU~Gp*9c1e&I8Y_(tT*^WZ^%)9^rN$$GMwa;9s>mr<jr
z^vLykkAiCHH!)GKobL1L-?c!E-|yWYnh6oK#f4oAF=brUGJbtil=a1#^9`YzvH(~T
z+T7+qR|0V3e0yAO{1{TSN{e}%H>7A7suSDR@y&ncgsv9aGxVi+2axMGG}Udso_Wlm
z+xx}yXm-EE(&AtPc{3<i`rR1n1&jXQ>`t*BTfY4$nf}p&^0<63Nd4_ytPPihbw}ed
z^muTXQU^t3D*x)$N2p||<%9!ZNXJmczS#m$lEcn(aRF39@lg0!&O8dH5^DFph=<ml
zDs|ep57=Qe+RiZE`I6;YCT$8D8~x+QKhR_FtGQVRXl&4+)#>RsXkDO7_+;<uUca7c
zL7++f(WQ299G*XaUOE-T7x_pBi-Om#l)(ZJcfhp52hfrScgGj6P(dTL8;JHnn`6H2
z?>wCbC-}#wkEMtC+)<L~FNeJ9cPPVLy@XxmvNH)ODnm#)r=xai4m)Eb?VT%gIbr_L
z$Q;v*Z_K{c@8j2K?uw-S?%K&iR?ApA2gKx~{!Qrjh5A1h)8LKB&A!r!YN)(xLL6Xv
ziT&nB)ta+t5+_X!ja!d(FEFa)qRC0NK|i_08Z!GYKsNBlO&P`V5lD~$`sA_W>aci8
z(~lo${VKHffWzvr*xu?0(hsPnU;{F#7YEQQq$-woynY15Tr?Jo&v7-|ZVb?vJPymZ
z0OLx-$hee=Wn+J087TY!1QYn7qZ4U&6vb}%3gifZrr|E&r9C~*{Ek4c=_5!m0c;wv
zi`K5Y!)=iaiDa{(VthlxxvU@+3R}J2U8jNioq>KCINprzT6n~_2&);-@UgXqRW~XG
z)y27vpJH2$VZy@D%Ui~JgHRf!)%-NEXgW*v`R_KooF&@tp)oK&rLX!mKK`rR3Go`4
z%i%&A=03<JW>yNZvIYaQ^%S%O1XCiA@i9W4f@n@708jv?Bovs`AkPK`Gqlz#^er6j
zt;s-Yv8XA5O&=F(CYMQBwSGV*n=R@VsEgK|?B0RtKvVeNq?MLcRt7<?<mBXc@Q-!R
zvpvuB#m@Hoyu8tXkA_TSZ3wl*ueP>EBMy0GWnpMR1=)Zali6h;cJD8@XOxQJ(ubJ`
z{ZqgTe;}ly0h*<gqrI(KY{H%=IHo_cS2w<2jIx1>%w+0~KaVdX)O;6YbJ-MB4Bw?R
z=kXG^1hm}c=38gBocJtv$=bK2Zxink1IAArZ{dp{Y#<Q2@D68S+3l}SbM13zY=&83
z@#}l~7%F`0j>&Y?RGQWcoSJu%pd@5eq^6wxstz#ULH-gmwyFS9!2o25MhceEvuBt9
z{>Fl?jODQ!Dq7k!zavl<a6aB>0&z<IogM|Ztu}6zw$myz;4*hRJ(yD=Hr-np!l{LV
z0%MV`m7QHORGFZe0a&pjV`5qvk(F4rP-0ntc8*5(-QC?n<L>uor+YPsU65Iu0`0Dx
zUn$O8HG?DetHV(+9aD32O+a%dDk^%HgQKvQKl^$?c~dLNn@dR8GckSz5sb{`@1GYF
za2eM2++5c@s5jNaeywvMdpY;mb!K-S<Iy)F=d}2DQc<7N()5FijRY1wZlfVA;7rvY
z`h9OTOm%ss!=mRLP%nY%a4^3Nm_)SZte3XO!A#~60KV{RK><W#VRy{)j3TM?7mSyI
z*b!;BR<(zf2W3T*;Ir8nOJ&vU%3+~Bz>eAT_(wz2$rZ1Oh7T$XtG#bv6JiA1;m(!6
zx65W}X&LXntBS#8)^F9OtGwql;5~pmgw^w!4fUtyk0glgDF@TTBh$#}`*H8QL}VDI
zTCt*`BYbB)YqHy!_|G8VbnDVn*Okek)%D$T3>P8Z=|LJu)S+*w)UUV7t?T96(aqNf
z;#5ksd!|BPraDEa^g%iR%3KxS=<q&=gvfXDj;|Gnv3BIe3A*#19ro2^2AFa%s4&L!
zSbrPTD*@nU%aBMgWgBGHona@bRyZ(&dLmEBunis{g?6<1m!V|>1JRzt6@k|Zg2h=5
zzu_0ofB<^}Z1;fH9GTL`Gaz+(@Zdp~S(sVIdpa^am~RymA)XtgmfX0kGcVrpRSr>-
zFRg$0RP&&ciU1k=Zt9@)(B)66XWDxiNVcNyNA5we*XG6Q5WmEVw1c<efuEKAeEdm7
zRC4S?vOq(V7f=(m@dB=*GBQma9fV#`6(I0}v1=B|n3ynpC9-BEwaHTGzW<tlfIw+<
zb!DXuOqA0MLIqS6=)n|Sc$$UUggzp)qM_onY79hnUgCJ^D6nsMUZC!)$Tz!*Wf>I}
z)yHID9~{9fN~i_H;*OhsU>G);TO#n=NWBCOsvr}mDrSD?Fo<U5gA$Td)0FYUvtaZ(
z637ICJbT=}s&A9JpfQ+V5?az1sJ6l^;2U^xZ>4rOxVLWAETt;Cj9}KgUG$gNqrAGA
zJqCmylutG%{aMi}F*Gvq0nKkRjXGpezg1XS^>xYs2>r-m`IUc=W?XIU({Lu$dO($-
zD@o`MD4>Kv_rpNb&`nZO5iq-^0`G%^3D;E;OITP))mTaOvqC~c73fckJ~~$w8e-tN
ztpEJ^GbA$7XonSU?OWo>hP5+c$_KT`zNFS}N=n5;j+6X63xW85jR=0)%RX5R_mK%J
z${(cBq(3gWA+tL=CxUW&C353Il^N%$ogD`T==e?p1bx_H$XuzNXQemi&U)W{Gkmcu
z)Q#J>n_*2ov75gJX9g&)t+(^7^%$S!Y9gBKL0;Ql&>&5^MJY7mXipis#+WgHU#`*M
zv3c`y@GT!7pPX}zNNcHiKMAqp2bW?wANBMdKb5P_<N5S%rKvMwKCgRFomK37?SPhI
zI;V++s|k<L*^&DGJk#$h*cC4AdBY)BnzoF)%IS3+ca1*tFWrZJAB&N4YY=Gw!9a8;
z9;VYmN<=2Pklnm226}!$4_sPQ^jDEM0u%cZ#x+t>m9&i)@99h<v$d+CLuqBY0B5JR
zcbmiL+SgcZJG*VA6$%?$+bDn*d}SI{CdI$r9sWDM4$uY9hd)A2$0t#eP?LZ03}*~M
z-g9jssQMofNEs-%qR&6{s)^5?I_%`gPWeu^S=s$MyNOq(X=~!7`>VWl)6&_~oP6_U
zGe8DnfJOZP_`S>fSv-f10KS6fe$xv8l?A$uxSLyBjH|5l^b;`G==A_VK@?vEfZQWs
zQoqE;255m2xFH9b81xE!qK!{~T#{*XvR8A4|1_9VV1MER6&(WuR@iAR_&8H5C1bfg
z_4i<waH!m_dBJu=HwnR!NfY`B6l0*P$5%S-0Rw{{V2yj}gWrENmgssY3PPSz)U2H9
z14ahE7jL%c0V(P$N{N5(@Ooes#hxd>uZ&2|i;16CfA!X#`O=wMDmEE;<1gfmlg|fU
zjpusb>)w#7*@Y_QH?`~h_7Vd;JW4oyOqy!`Kaf+t)siufEJv37{QSz@K~vNy6&ab@
z%JO99@sjr0eR}#&GD0U|kM({kKAE+d*#F+>PkNJzssru=grK&RrWzU=RLgAi<7|!6
zBa82&!{+D1iY<QP2z*+`V_c?SIki{1&P;!o)VoGVP_yeC*hLpw;2znG1K?pjmhFX#
zq2i?|uWfmNO+&Jlu>C7rG#Y_>$j{r8dZ0bKxTuz~|HOH7nye{HkpX%jUw|h7GC?Wu
z19i#a;o*5`m<C5y4=DOT^dWEb+!q1aj>n)S<?Z7GB+E`NE^JQguVFbpmz2E9!op%7
zl%|}e3J{K?y*0!7cUL$pN67<)U%;MmUe0goFR>g`V!jC-ZU6Q~UZvnA0$>dM0wxQs
z9x?^eXwdN0yHrVbKQ%|A;C3a%p1Nd?QYMES=z}CL%s;r|`<I*o{#VXdBWPB42Bj9T
z-Mk^W10uG<+8XcukvLLn+$n7<*P-IsNBPgc7Mpra*@M4)Q>nP^>=VG?>hc!vXP2iB
zFPPv^U;1-~s=oK=eq<tYRK(vRk*lTD1kP0Vsy(pT_|O4Fj8}+y>E`AZxMy0Vh6cE=
zX%$A0(q~DRSd8q0on9iB$x&uE>7f%2B-1Y<|NTjI2#y(i<#Em^Ni^9Zuwg}lr>C-b
zA~)m#H}TJl4T4PtuF2mTs(a%3;#U)tJ?zP}LezgyB^>`AO^NA@P}F=qx}QMIv38^W
zYlQI4o~^^R9W($2mha9l0P~|qiv09~XJgA_vV?M~Yk&j9POYr$e?q4AYy654kX??C
zt~}GPJ0ySJ{|dSnwzjrz^E*8PAJ_cjI(i-lj)yKzj`sLh_M3?^Ui1{%!qQ~&{qdYX
zhFgy1EVJ>z=Tmm%cdz>2{CO%bgsA2xFr|vVD!m9N06vBw2`HUTXL+`#HtoOt=VZwv
zhIRdKbp0c+1+aWoG*=V*-jy?B(fkPfwYYXmisq)AoE&lE$m&Q%!JFnAJ&QF2J0O7g
z*nUwF0N5C4PaS$=4R#h$z0&A5igd}8`?_x|^<4)pZcVwxU48oFqgIi}AXUxIEeXol
z93n>?uD8hh6GdFtW^m$065|-RH@rVFPZYF2;=C}t++SjcHzsv>IjPRbBOZOL=xHP4
z%p!Gy#n%oEyH}n9)@Y#0l^c41k#w_I!(f1*k25qf@*@CmC+dCCu^1YN#zz4ko@8*R
z^nbpzdbK9CVKnv-XP^BJm;U(l!t*sX5}jX?mA?QpSpSv1Hu-KO39b}*=Ge0w?|aqj
zI<!W))E-N69x<wCqw4<+!-7DWb)7}^d$O)8gm?(t3QtlfeuDu+_lSP^@@0BvM)lvT
zsu3Elx-BKNES8;p`tV?9|HPJ#)PTn8%KNjUXXGO$ugH7e=#H35KqC6B_1j0h7>k30
zjJ4NCjR|udmJe)hk_N?QxCZ^tb;;nmtNr;&x_tcnl`W4|Sdtx_5#=EvH>&q1@l4Ym
zz$7VVK;ms-Z+k@cjn@J#!U3O}v+y|`S8dlDrx_xmJEbQF*Fi2(YwGb0haaB{g(%f{
z>KfXX65g;g^Y`BHVqVM1JUjjGdSn7wtlu3#kH^nkMWxTr97TL`MT$nDaSz;e;*vAq
zaOK|{w(XS`LqF}Vm5vdgiW*v=3%N9<GN&WOFuaKWW!1At?iZa;gY&uw$@C&*3QLzp
za6XaSn&xZ-pO&f*>>J)S`mKTBZXpe)ww>)zm`*X68trMV<oTa5P{SjK;FFG?zHD*V
z%L@YzA)KeM5?^lYwE%W!q#`FS2-b#N5*%#T?IhGa5{@3wdOX2c>Iy^FC}do&DB#1^
zzUAx4Sfe%-9YAxVm;#%Rc>DWTx-7Y8+Lrr|5s;}_a5*!dZMyOAH9O~DRn^`hBNJCv
zCc#KaNrCVM<Np2o47imM&}l@sP6B82a+b1sl^%Lg$}uL4rv^j`eF6s!y^D2!KiQNo
zz+At)PII@C+T0?zHszCRoF2&>k!*;{g}U}}4R@hKUgNam_DZ{Azxo37x`yxh7!wUJ
zXZOQx9XLEN-stOJzjf<!ErsK79UL$hYT@|11=`~nIE0kh{Us1K{JY24A)`|sP=#R8
zvrH*0B*G985h>6meiP0*JgaZ%tcy;IoHxRtuQ8J?sP6!(wZubU)sg*_>;{sJyUAjq
zufYF?V-F9S<g-gpAcaukzzg?9tN$`3xk+n$IU!9{FjCY({t``6#x8|#`g<RLn~Qjj
zms)*@Z@1`0tD3d{nyLT74k;b!md*kqgBkVsf2%{!KQ4Kr4AB&uCAvpwCHWs8uVbLa
zmi2yKMK^Bv`S~$e)G$Gd#?e~M*x&*f)gnpZn*f%cTPNguQgB2h>^}26HVesTj`On1
zb@G~<cg!H2gmse5)(!>&7X*YaM3(&|!yrYB?yjENug$23sQidWa@lT1U?(?emM)&A
z>M&oaDc}?MN%*5~pt3&y-`T!ARN(Tque-aT<z+9%sLK+wTCrL4R)>%bXd5-IK*00{
zR=p7i35fXt-d>y!@jKdB`?Je^v9dZ8!Jvy*Fm@Q(y`p#C`}}~yZ%<r1`+6jFh(_uy
zljm5KRh801Ld=MAAbR@6r)%1m%{6}$7IWixeT_F?;yC`}M{!J26%sOhe+CU9pGmH4
z-`a*$e9uiWsO;!M*~Pla^z1GqEobLw&&?i$k_OpY<GzbW#31iU>AL(F)Ez}Z7F}Oo
z9}smhBV)3c@BO}Bad0_H>lHw=B<afqda%k}yhg@SU(GJQe!-hcOuu||=m;=1{bAaU
z@yRKJm>-S!X(yjWQ)(+{h*RH3l1|%QCvR&=`NW6Dq4V_ot=j9KajPQ+8sw@U72htj
zI~;mCE_t1)=IKN)lZTVe^^CN3_W3hm`lG+~B^^CWUqOQ<qIab_9HMVaxS%H|Rcgk@
z@ZOK22sgSfhz_@aXNF!Mq8o%_R6G|dKL{K@xc2_AKE_&u$L2k8sL*6$3TvWuUURUF
zwa)@qt`MKlf$yZ$ZL0_&JF6URjHD|h@gR_MIQ>X5I$C|s()mmzrF2P$5Lxn8dR2)~
z5i=(akkXHSkd6i#&d~}@fpJf0`nDK>E;+&K#~*LPE^hyND(p~Te`(S{f9GZIsN0s5
z!SBdA=*=XvqXHX-1YH(oNqW{*X<Tm{Q@T~H<|fS9cYxM~mVoaERZ95rSw`gNz~g#R
z{h!O~)}OW1_Tr>!zdhuM{UBqdP=1o+8Q5^Mm;yZjp8{d$DrJ@7lM2fCzkp(*6KwdR
zUI7vGhDVQy%W#L@=<J}w^Us1EMkNUL;TH#Ad*0KRkYJvC_W`NJ)_Xb3ouL6ae9I!A
zPag&<-rxx|5&kG}4t+1C$iXl|H7~RIrJG#ygqGix+@R2RU2V`U?AC-&cOLys5pLR_
z?rbSPPdjLQzKj{TBi**1n_jlbbTC5<mengfgBC_JKdt6>mfF?E>YSAx<(nOseCnPW
zXUIpE6qGd7jn<JEe_DDs`ub~jArY-1ts+TIES-g2BQ<qofG|dET%37qAo?aO9F!B#
z(MR8dq|sVPYClWz7E$jtpIse0;H|B~ytQ*_*ZrgdWlIzoCf+!*Ciu^EmHR{pf!4>j
zj#Cpxl(!@iO$On0nf~kR_xi&=#U$WN)wo<w7>^zzqh95IChX?dh4n2V*x7k>0*5({
z$Mf!6yr%YO(ii786Z|MnM@^0|mHe)Kqgu*oq|Z45lCiIZO_R|3XK;2$Q_T|y0SY)j
zhJgLeU}TI#cv3nQIygAq(^0d-*J&L_5m~f)o2!l|V+!Huzu%{hlOJa$<;v!hLyiQ;
zBpBvwOmR`kajmYiuQlMw3Lf8D#7m&LUa|1#?e#i?OaC1g*33{siiwNU%m?PNJ9iB3
zzcc#HLN1tNW|vj<N;xSt=vw+rSDBrkEuQse_S`qs$ys1Wn5iDRT$4pB;?Xa*oZ0}*
z0#7}2OLaf0;AszD{rGh&ee{Ce)lA!YJdqadF--xc*8B=ux9IJ@Gt&Q-PM?53ZY4B(
zM@k}CS)uiqcnBhtbd-|g+ec)9&@%_^Mw-?`z|+bs2F_R}A|XD>Y+pUmGd1CX2Hia&
z0TtZJZ=>X~6bp|%?t*o^5c+Ad^1fXox@w!4&%M@X<KdFaX4}_S-&c~*puO?$f_*-G
z4NA!GeSMV*Rno<QGGvL^vL%c@JQ-XM6x)1@klXiy=S$XnN;8qqBPCJ*m3)zwkkr2(
z`Ul1RFG{Gt4?8=1G`}6*_-|$QvUQsVseg76D9-|efe-XftEa*IB?HilRkMR3%lWlt
zSd-OO!Mcspz3d;T4(x53Uj=!O7?I}x&O_<MD-NGMAOcV+G6K?*S|biDtLb0YImo0C
zM_?;qokBBUU0YjQ2nQq}ls#6k5P?$%RwXqDZ*jM+if<XRsz8AzCp7jN$)@QYQDHY+
z@(!Qs(ODeccpNWFqpt|If6<|rJO40&Y?~ulVCuKpJJ+x4h&e$R6aBeb$7p#4P2=_)
zyHxRHLLsME8!PD(7xsPG4CYF5t;>wYP{oYgIU3(%YQ?RrF{IzPWM3?A@b}OMk2V^m
z&JoM!7zDIE^78W86l!3}XUD%$gZT#!ZmsS3K+)u#elv#MZPys$f}8bhE0;tjmuGN?
z3hpNUsk|$)IsTd@{;go)>QG(z-=0_48b*&u+QxIlpFZ^}dI8735*qWR+q=>TwP?Ey
zul60I{MT5gj$~{t=n`xANQz-;sRvJ~5x9+-$D6x0`<7U4fAT#FdTi|;@m=>CN}ziC
zb9B*@^6#e8xk;1srF6=bC(jm}Ud(#>O-iD@Lt_)`=XI8o`9a59xvL}^7ioXkk4IPf
zFtV`Fu(RKclg4=p=^|vK4gjZa0LI{b)hWi30h4dZ<ZIzODSy=O_<(Wuk&1#8-J|2U
z4g7@cYdbwB3D_UUTmk?wvW`?(7gk^Yj&yI@mUj{$>5{l|r#J57^(rz!)xQgYEVB8{
zWJepLYVkOye8^-bd4V<R{_--#$8yITdjf}Kt6J54JQDY^8x}#^g&j&K8f78S#XPyZ
zED7N%>?5?wX;<ml$(pBRf?ynoBq#N$mmZdYHU73#>ap!^6Z!qsAY<BisO@K!hK~41
zH>by!dXp!oC+bUmWu|85oL6p;d<*I!<J_x0yMRgIGyadDq~N6*eXmZ%9Rin|*%ChV
zO%rz4p9u`I>GnSO8Zz2-z62(JD$8s9S~k{;5;zfrfv=mp;|ct1H|TF2x^>b{zZ#XF
zFqCi{^hc1;rzL2oRcJiwsQ!ubKB0P=MKN>^1ED&Up>&gYikUm+gnRC?qZ{1Ol>}(S
zNw+tXle=DE(1!K;cjRL`i2m=%-+2BJ*GG#iP=-;6t8xebkcy>V#`sSoMg4H?5T&@(
zXJxD-n+db)7*G_o9rJE>D*7yA*SI6GH1YevSmp2XzHlqE70Q&AdP6ZQf6@z2-o{W&
zFHT5aUv5(_erqvxXxp?jmCWgLez=uRAc#oWrkXJy#`b?O4OQO0UGup%FHuG6&H9x^
zVXNSyd5<;8-Lpq=?=-a^>vcWVa$XZD->K>r9*B_vD80s$v0Fjrty^W+Ie+n$0aLdU
z1dLa2)y7y>H+x}XgOO7Sq~I`;ZQKPKpOjk0ccnD<n})?jYl6(qtSFj#ms8DT2$q}6
zOk$x}>7G#P^HI)S)|71d8_vD#pQFuh<<<m*`>ap;ck5(gM{eQ?rD=0Yr^wyxLOv)F
z;`T<@-`)Sh-zLunD%RF%T6Y7d45MjAZqi*;Z=rJyr+)QR_5#lLf2{rSw4YfK_NZ7-
z)_z())foRLieVD#O*6DCtJ2y1iQ{Ts3m(KRemaz#EOX!y)X?ymzr$Tyth?}u>s<MS
zfpnSzRh3(?gSl#ki9M^*o{mr}_x!NG-SPALsDpS`Mb6Z~aCdHvHGWtB(Nmi;mW+{k
zEG7KX_7%)E5Z2g%dWKj?x!uCIC$_V{T3aOy3d;JuN}l;#@{O45NI*hAO8p2m=F@!d
z`_Ju1T1%D|=PY|q5qx%LbNAxJMZYxmefvzNq*&pOt(0+dVS()}+b83`r;Z*crEJov
z#$#RP;sP%6s`R%v3Goy~6_rp6{h}WCFymQnVqjoc#UMOSvKtF%8ShqxmlMWVk6P{7
z1tQ5)e}_F7Tpzk~IGyuhyQ}oc%&Ux}wm7~L`L>njBfWntw1c=jj`!<ZA)TIES=lb|
z^7ZT26Eje(DytyJ;9)I5soPJ*q_>%CqMV&r?Qab$<Xrl3bM>42<@6iLsa1mu-;a>J
z2$|8ZL3=xjp;<%MR<STJAbZIh7tOQkEGGX!T`0^__MeQVk4s12y&vcn=`<z`mYT#@
zNbsjY23|Es`63QZf^Bdndd5tO7U4Mwz{1_hHirspwsICI;1+?LE&d%YNl1|)P$nqI
zHK;npwi$Fo*Lk`@TQvRMyHCxBiczkZ7o=7G!vd^TrNzdQyYCJdepkyu3$%aP=-}F0
z*rI%5^AEb4Us5sv3k$0!N1giwQb~qY3DgU970z1*_*8-#fn=N^2?+^{y*XKs>otLb
z$IBo%u~5*%z5~U&ba<S&moLl0SFD!JQRZmZoMzkJ@*PrHkB`sehFZ}2_XVbyVkg%#
zh3C4u-zr)$U%%cR=BH4r5TXAdP4|Fa3Z5Bm#_ins-kmoaTh?R$NT=N>rc|uo_2ISW
ze>o~h;)}}8RLjvLmgVes--1Vrfzi_Vb|j|IvrpLfWsz)7cvUOM`-p*WBz-}DzOk;N
z>c^^;wu(oFHD(4bUbv7B((%!}_L-19{07u%M6SaS%psvK0H2b7KVYc?Rlvo{Dg!0Z
zL+BIdgscp^MZwTYNHRze7nu(!g9;UEr8#7c)T&&H(4sLr%L5AV?nV#ZGc#M#-+bnS
z9Pmx0RxI>q3dd-6U&xr}0=rw7lcbmp54p?q7ZDEd;@NY@DjbneB}fBfL4WVwGvEVY
ztz3a{meuAMn%X%a+bI?K=vk6*U>CYoZ`n<FeAmknl1RQ0?r!3wwbRqnrDbFo!eMTA
z<-p18=@shz75+tY+QE{_UzgP~m4#3f`Riw=emw1wVrsaDpu|!tv+`j6w-ZNwHhSLN
zT$YM66VA%-`d?Cy(NpN<r?WGuO)d5~=csIZl(}5VGD=d*cS=)~P+<FWDoE!-#hc{L
zdOWYaI0WDRKx@?l#1Znr=)wR#8HXs0Qyyle<5hq&)Jl!O*D~oz{{qx)sX)fxvS4S@
z=tsObQc(iD6(ek8NHjqQa7=s!>XQu6eW0hM6@lVGucjlM?g{S7=s!Wjq6Gh_x@A=g
z#NX((U(g|_kRa$zL`0Mfv}O>b%RnxTwM*_RNV>op4q>S|wo%HEe`!#lYHVyA&21TP
zQL(}mc<b?VD4Yr<A+o=e3XJ39<10OX?wx#GgGH46M?gL6QcRn&(-Rt+pCdCEkFKRP
zWYqj@>!}YmXwh*LC1K5QPgm^I#w`|IaPE4`G?OEZDpoZuGVfI@sw&EI5fqX&kXX=a
zFYZuHyWQcInB8-=`C*3j*B&0xOx-Uw<nqsh3VD@q`j&f>ahfuBQ}9t|RY%0Hn3LUB
z`O%a9_4f^emGFm3fyPr^Jk@WNh+k0)H`DI6gojdR1t;*eFn$3Bv4g`y6Cflb6Y{9U
z!fH3m;<BBU6zhU2tlMaKdgH=qRmIO>D(z{7U6CO2PAI0o$$wR7Ho)fl?s%(u?|Umt
zH&mZc>2d?yaTPP+>KqosrS+aihlh*kViT3;apI?&v?fq0D_<>N66a7hF)^9&8-u!9
zI#fq|OH`RiP)KcUc)AQWqe;PnYSZQ8<F9uXx<#P)AA>gKGJ*&H07V?j%1G9=;ZdIw
z;HvSBkm$0R*e0$4XLgkZ$Lkr-=5qAzdQ<dS-PI?R)FaoarHmg`4T_ehg>!s;&q1f=
zlKAX&=Y@$x-KZL;b#-<1)`+gkVSHOx$LSxKxY-XCDoUZ%exZU!G$*ve<@u?0u3_8%
z2pH>{E~LDdXBJvfuhvZ1@s#>h5fs+<0zI)9bV)tQQz43}qZMbfcgU@OMw0r0w*336
zx9^KJ?gEqaN(r@?<A}XV)K97>HgEEo73bPo*SPu7+r54S3(P71gEht+Sc=8{OGn!a
z55z!gLgiCaG&`;5R>e|o#VT+G5yKLUP(0SNAhIBIf*h5$)4{fIo8cDhVNxw%F^d_|
zB*5^A<TUwSd|a)Tx%$AppGjzZhnv(h4<)g9hd$7hrSZ~yZD!oz#74)o9Zv^e$E>!p
z$!?4SH5KbZ4Sq81(oBacd1!KwPFMe|d`TuR;&b*Ij%RXq*!K4^IrMFM4ISW2uO)=U
zxVOyY+Z1dzt>cxhyM8A!i&wrjOTuF_{Sn-aG$7cd<Kilo)ayfP*Yq!OTMW~Ql|$jE
z!(|nvqOF<wggOHbibgFe5)vA*mfydbHeQ+bW`B7&V%313lB)+ErwdfZdY}{X;-yP1
zCPnZ+Np9FE=p|+}Wz!8)uJ11O$-rX^NroyvBPS=P=QrK(AXdmY=Y0@)^(rzK2qqmP
zuxl|K+Q-s~wc8Nhl_@3BqV~rU4!ZAxw}tgG-Xn|8%Cxd-)UDbZx^F6Q>ir-+s9Uk!
zNok*;j%jL=aG_ByXOe8jxw*3>aZ|#95`FLE-sxmfH9``L`4lXHdL6BDhs`IO8#^jb
zWt)%c%O!wGR4GXqpIN(Fqbc~Tz}pFU*Futk>9PE`o_bENE>cEZJY-vDTrKwF$M47h
zt{8vqN~K|w$u0VM605^YCe&zrPzDrZT;}2&3w0VU>A!hHWHGIrcQCXNQ+qpb$aVeF
zg$wg=c&LC;M#RQuLjf?5Qs6KkZVi35fWsgStSxdA_UN)s2Dd);YqyTIAjPFV?bfrV
z2lgpsDQ*;(cf;cY5%0~rTJ{W0bhA8gx3`#g|0GgLt5}3fEt1mP^BfW8>P9^jAZEl@
zOKW3SF{wwDWZ0j4KS2@wl+D|q6^cQx0zahckVQ1_sXfKux<8rY9kp5O8^x@$@yrT2
zyn74^7z+%c&B;JzA|EWU8;LgN_g=jJwO~h*JI~zOg6eK!k$#pJ6C(e6cFL1c>`pIU
zqBw#fL|ml<LJkg_X;H91ixq3t3T|$d_4V~{xMK9#qQYWgTzl7t=!UDkRa({(<^H%#
zKsy~7htZ$gn3isC$G0a|VXiYNfkQ=v9)qDYC8orrd(>Ut2O~q_ib~I^dc)HU+tM2f
z@;g%@)QyxOQK3n!bH`CNWP82o-`l1v!!4@*3E6IDtwESyML%t?A<n?1Z?)rA6r<Z8
z<b%xUw|HU<YiPw5|Kuc8h%**>lcZW!?SwxxhQN*2Wjmc0U|N|<S>^84K$9R5Lc*%$
z)~jdG5x=tQ-1-e%rI6$VkHd@~bH5XWv_rPRt;zu8Po{ECG-zPpVXt`;{}xte!9+fU
zv-}kf0r?-O{Cf220Uinod?-uRN2`oTHVj28Ay)Nx&xsb3gRD2Li!CN0$x*G!M6O%I
zMN5b{z0koA;`dnGoDxlmn;*7mRLe`x9k`x?0|*{VLP=w$hob;>T6f~im=IH1(l%9z
zP@ctm0?R|q>zpQ8t*8>j3k0<SEa4*KZecJleNal+X(ZJa4FR5~Pc~6VAa)B$hLiH?
z!othFSb#YM18pMGCaR(^w|tR7tcy_S_$#n1h0n)&AHkvES9Jo+WY#j)l!_K`V=qNS
zmcTVpmX;3MT^-2+$q!@Ts|5kn7&dlxPViz-@L7Hd4B5Jzp(@(5=MtnbeP`~chu$0Q
zTrzm1b61K)AD;rjUe`vLV&Bj6n<{Oy$NrFEkaLOmyMB<ew6i=z<G0x97k3KR3yxp<
zISL&HF)4Qxo45idkXj7RjNblyJ?fL27Uk-SieZOfWn<b>!6h?-l7bCy95_pA1qOj&
zycAis!e1I0rE7T{=!?li5tjqp$a<;4WrvF0wK1fMmh(&i2?5#~hn+<N@tQQjhZiAs
zKyM|m^_%@AS0VfTlD%=}Ae4y`=m{vn=kXp>U|@@hO1Ia@Ha2>2V;VGfWG5LO7&Yd5
zJ7<<BBBeRA3RxAdh>2xVI!wf@Vxym!AJ~?@g0;3k3XYnYZ3SZry%~CoPDebdC(qkG
zLb@~DO;b%f_n)=2iK(1D>qu`PK4I19G_)SE?-L6VJea;I)&<p(dqP4rHZg1Hd53Q3
zLWdPKaQoH*hRC?g(g9Cl1d;FNCPQD*H|ot{+l~e3E<ao@=-_x=>!lkErBthl8E^>f
z7rI{Jk+LXiXlQ(a)0P6<@Yd0ivinv`Dl?E;iWc&$1_v_{P<Km!Ls})Ld`hj@kElaA
z;7E%Rq*Y2b)?^cNg*ujU1wL$!*EW5*Uv~+8gZoX?kMVa}&f1g^AL&+91?3b(db$s_
zwG>S1ciDql+oZdzQQg}^^Nbv8p4eeAp}BX~+_g#wo!ed@66ha2S`S#a0p^GqUTs;~
z&mc1Ntusj^3)ltNV%khbE6cPzPwZhqNpqU?eBSP$F^q?*V+z!Y6k#5^pe~yO>4B&=
z17Izd*c)+7P|nrJfQXTl(vmG|8<2P=%l-K#gM}HDm6dkQ)DcEd7aGjd@j*-Y5RTs2
zvued1g-yKx*t1?J-}nzJmsn_mr|Ib1nus}W$C>%gbP9~mH|Cm?XR0YPVowNA9~nH7
zihr#iYd%9nD(pVNeK^=ojt&$tWOaJXd(@I@R8v^TJmv8`j)YA;oV3tCBsH%OE{ZU)
z@=2+t1vM}H*2YvUP{zi0c~dxOUpCr+NHB)vBDEl3NXrbX%mFV#P1exVBm!WuQcBO`
zjr%3UX5Z1z{YHm-wR45znn!PydAq$QE`aES!Ic$SvksYFS;+z89Lq_^eMrg2!YULa
z;93m4Ret!6DaXRUfhRq~<9I_vB@IjVN|EWGyJz)kq(=j%?jA$$cRJWR==)}$=!6m&
z(9_(1uH>V+JcozABRbN=3DnMdtM<Z)*d6<Zkq3$u9(YmYWxV!gAHGXd(>>&W9_Mnk
zXA0?=Mwn)nYc*l<@;mUh0vEfCa#oS((Iv&JEIH$CEiLjj9w)eGD{E^=;G!0VK(Gsa
zLI3~@%8p_omk7wy!HH*h^yp8KAEqKhpJSyWQtDVMI*txM{ho7FyQjRh!3?+T8Y*#L
z2G7p8Q}V5chq!iJ2vt{j5$W2p=PbS15qi8B>7Pwd=9xnus&nB7)C4lJf$4Mqu+^;k
z?vVU#VF(+42HZ}9Kr=%*M{RR=SJ}a#7|0iq5G*K#Z~!I(RuZ9#-(+c_w<El><=3yz
zVKq~uP+s4L6|%TmvDOO&F?RdwIw)j9O?10K0m8)iiI@M2n8W<VKc!y^g|Z$cGZ)#X
zB_rrFiLWP)Qt%PEehdbM%I$LH7F%)cuzVpW@m!9B^h!w@f&L1WfuHB&c6!(RCdBjU
zqX~=0<OCnUED^)@mLc(fNfRpO;o(s}8*Lf}j)Ez&N*RoEURvu9)l#B$V06t{-+XuF
zCeaO>+1z)4w>8|ZkAx7Fu^Th}og_`@XzAMOajV{*ge#i_ckPv!ZD9}X1eHr8_o!M2
zj`jsPt-R{Jv9^4syWTfDOD~|}<b~r`t5Va*UWy=A-OdNbkPe&4^yF{{3{B)>ACeXR
zLx)Q6o9?xmvs{1zes?LsvZWu_ta7oRn1RPi;wlN25|<UJR@8!6z)+Nri{?c6JC4#A
z25F|Im5ASw=%;nmPTlS|pr==H+?mf+N6x#nH#GQob6EfA-eikPcH9#cq6!u75)ojQ
zaq);LoCs6v$w}%Y99H>$cs{B}AE5<c9wM0@xZazRA^XW6SELdIHGy|9`K8XqIr&=G
z`9F#Et*57_jg(VUQx8~?$HqZaLaQxy7!3Jy7FDV*T=qiPIaOS2vbO8A7G6nj;H1rp
zPZ9p&FbfU)oFo`w|EF}~g~T@YTqc+G;qfOCU>czg9AFfvP#TQaa{CiA_1Qy!Wu)er
zT4?6aqHjA>60CH!eSJlI=i3HjoQ8#y!kYI??4Nex?8r8QC*%YwsaUej&QX$wR`Z@c
zmD1({I}1)JCpMCYGm%7h?xHp#o>!%CRM_2>7>s6l68HTTlgn<q`p$9=_Lj;__#cW9
zr0mEq(UF~9zQK~iNQey~D&V{lQqL69_JxYYk_SPxDNfd3<bS%i{sLF9m8#%;o-p{|
z=5}E)gdHKP)J`T@yDJpZOuM{&2Wyr}3mK;Z^zPR`M+u#>32?3pqx{m9Q=KPcI6K2T
zgnH9zPIFle+d8A;D*Uo<?e6U;b?o=LY^L7nsg`SC8>2n{cCXFnwRm6aGstda_pMPR
z_pO#m@DWh;p1zwBmy}X+lKuQGOGzno@NtS-zAKlqr6vB#h{Oxmy{<2zVuO{I2|*9z
zKPLnSf6Ee~V=0bz%TexIyLSG)PoG$;*B$}toMd<Kqy&InHz;Vu)+)DxD#Wo~TfvUN
zPx{vLh#_-$E>o3SL&*ml5fMX6U*pNc_wL-A@bYUn$6N#ddM$kFvJ>XtnfUsqy*iBK
z?`sU%#Iizf?#GWiHSPcX4TgAgiWm38hd=)3HzCQ@=KPwNugMFI{(Fzl{qc+czW%>o
z+GOCc1}ymsaPn(D_nin+P>oXkez^Z@$ZXf{$=|m>{gm%b`tK*N`g#37Us6_LSYbYi
zlcr8hWw=ML3ZBRprF!D)WDKmK&D|+~?*bbJ;Dz_P6;Ah>1W!~zN~;~+#*f=2N`!a{
zNP?A}TKm3q`;%9MN|(<s7Vr@+)gq^+98P&wBeT;;RWM`4yM+wc2)*yN$(oS2&|m$=
z{d9=x@VG?Cu6rhqb2jl_<_+@cvTx^?^U}Gtj`j4(fah_k%h?{a`DlqY?sm9ECjCx2
zjjwt;|07P7Xh->Ir&pn*$0-Ot2a}(zteMt1Mjb>1=a;|gBh`!Xl|0$t&HSLD_Af$5
z^-B^r)x@~_n8;Se<v+fJNY=em9Kti#V4C@8!?P^CW_r?UtMkvkrn^vyAxl8%<QY&*
z#s9DfA0f;c{#<<h?G?|0ZQU0c>Pi&i{E0)tRn&&h)N#%~;-w42^@?F0`L3iu$`X>P
zmSU(!ZK=>K!zJdDolCTmcCf@fA2Gw&nj`)95^ivbaf75Z$<`Vx5k-%+k*>{+kkz)e
znxoyY42$gZ=mA5K3;ix()hAqSYE?8s$yMJ?25=T{vG}SN+HEPr!zxEUJ^wn{4D{vi
z<ct-Fs6Ht^4n$?MRQ`y5Sab!^u+X!aGc=-1b^g9|jh6KR9np|B0|f(v)U;MQcI2>z
zOoXfs52EH1LV#69`a?qlsy38-*R%T6*`b15-${_1-+}+^Ilguup5OMDhZN$HiY#LV
zq_I<L_^VFKtQJAtT)cMu6fMLZAN*D8I(#7~8KvlVx|nV!J$3<+aP#xS4<or6TunJW
zHtk!dXWM;PF>N#zBU+=>h*zMM6~m2<Ajh=4Ju@>y&&M~^tsfRe&DzmTt;pl>;e`X1
z)^VFfNfuL0h(1vz%d3&~pJ!{twF|1MEi<mYdQQu7wG(zNbz7Jr;D5HqI5<%X(cVo9
z-Mc~hc6h7uOi18q(SR9S6hk`|)`|lkCaY}o_4V*CH8nN$4GjXC!chY!$N$&1QOnun
ziV@Hx;QtkFs-R_Y8pA=Ar8LXNPX$(DTAOF#J&Iq9OH4jz5jE4Xw|p@oH1vVkCM1ij
zsMjCD4ee{J?{c-8<hzvKcao1EVNsV3IzDz&t#kd-+S1N)Rbu~k;O}fENKL6$I=gB{
zMRjbQ3bC^ntSeLl9-ny8Q4}b&ot--qZ>Im=C&0nrHn~NC@YIoro?fJCE=iUMAtMgB
za{dt4S<hF6oKjY%l9>SgJ(0mLTDxz9W<`g(i8XUFqx&*eayQT4`IP!aEgpsiSGvgF
zF`hOB!_i%Bim-<xs3ce4#j(HEgn<N03&W}aZKNNg9)Rc2)zKs_RO=nul|V~GNuDfX
zIX`!bb-pE^raVJ`+pJ#xd}|tI$5^4|vY@epZnKP5$iwdW5w<8vva|gp+$)#qN_hY7
z;f?F)Jv`r&ogwIKS6UexGCIl}ILaKC7kBcmN5gnn(0H2Qi7mN{>w489D)GdZ;jeMJ
z6;_glhvC(-*q)WE7`SdV!n-I%-60+}8~<7D!}OU-8|~TGP%h&WdSfNOu&K{A$Y2`&
zjYx}m^0k4<zZbrDZam5}rDBU^_7yE<;xo@qj<dQ{l0_Ni{mAF$^iU&JuJdK;mdZPM
zvV@Ab=d<<KVyVK}Aw1$*B0TC`L1NLLjN)@#m~RfX;fc$Xqt;+v>g^=XQETpKT3nF-
zd#yUSR_(<}m=!p#(U(k<IR66%;zCUJ-&Yu3&tN9bFY>=H|92iBFZ93b;e0pz|7ISX
zUCS{X#l^%r4|kS|fG^g3q+F@Fxf%D$6*=fFLlqdxa9RxacT@kjIffLYz4kyMr+llO
z&tD@UGBO=#vrQmGHvwjC;ecEB3I>+n;{I!0YO5HeYE9Xqh^VNL&<B&DRpo+<kB^9m
zh%n=dfmkyIk{YH^$vT9psQr+6iF3KeUt5G>rJAx(cLz#Sko8o9RFn#WLFsR%bb4wk
z39>c<K-nFV3`zm57A3BwYX5yqGv$g?M;eHMggk^ex3G|dP~|+`1b@%1Fv8OM&>-hD
z(A3oAgZTFoUapGKl$CP@KxQu{4|%iW<KuScY;Z$dTwL-?p2W4+pCwR|-3e5XKKDk*
z&)lS(HCQa#yQ`0-H{5OV8QrSpoUP>ePO{eQHr-15i7Q!n$fcb%Ia(iD30AT^4f)~v
zNqo(1?JO>>Dk|N`xo)8dGd-5Y&#TX%z!;i!&y)NKd;8Q9;q%Zg!!!5z_U+q1*=Hl+
z9zPx$HM4MoAyy+ss*FoU<*i?Rc<9u^4cTXJHe5c{3~X(NAtL4$*km!M-x#`<Ur@Yz
z9P9OPW_Ff=o4c=D|10@S*yPFC07-zexU@8Z>vc?oMvAuUxE0Sf<TV+eJW&OWk1C+b
zn_dX{cQR_38Fp~BVB*lv1zM*3fQT5<J}Td~pi{Z0+<vKGwA!GJ4AKGMEjS`v>YO*q
z>e+e738E@1-)!35yY$(x#B`u{MXl60G$@uQLqS#5?=rqp71@`?fz+bK-jCaMpY&&~
zc=3)n!kV`3F|$k9_U3j+YgQmW$1;6rNc^?_s3=`QL}2g-J=3E<ZS3{LK{W|@S~Oa>
zSsZ++@;RcM1U05(Lt+VVYD6Xi#%?<adVx#zVXNw_WN5{=hCQ-v7I$Y5ILoPT)-+%s
z)Kc)B;-u5i-2NcLEDFo0FH}ZQb&|9Y1eV+(ZE8qS0_{uC(WD?M`jJnf`s;Fabv#RA
zy;TdONYEaUb4*)2OIm$P@xV4Zv``OCrF+a-)UOwRY*#tRZIa%%TCFgmRwUvu1R0CF
z7-IiCQ>?5dE^q(qr!^sSNc?|WxCoL9TFWCnj^e5#In~RY8$CH%mS<AlYE~Fc$CcdS
zeTbB_Iko7n`x#5h{r`5?+}zyX+;1sM*Vi{B-7Van*<1A2GEfeO+`9foZlNXeUb9>L
z-WfC#LbWOE;la_7O8VHn9A#5q+*r&~?Udsyiq}?IUgezw7d~Ep;Pe<<BX8=D&JNXq
zUoi#$!_Z`l$DQS=TIP-KeHr>qI`e-s<w)`5QaLX7`k4>eqq;j5R^wF`l5}27>y0B;
zS0A}|>W-Z3y0-jkSN@F3yc6A{P!ehM5(lUG;ex2kdlGV<<OIjn&L3r4oc0s<11SU*
z9rr3*x>Nq{MC^1s%(aCH3k{9fc-Pe@B{}ujEW)OaE4@lG0$E`JSM38ob<H!Dr>hn*
zybP*H)XCOb?UGX{Ug=S=J?nZWWMI12nRK{Q6ft93#iLRg81q=6q3JZ?q4BFBWwOrv
zxcU>n+|9Ba;|Fo&U8#HF$WHZE<%Ls6p+AW?X}8CQEXw{+{~d=<hGMH4dU-g{Q82^g
zfM`w+9&6>(QoPdfS2li!Rr`MOY-}9a=8Pvc3S+trBUoBAp_jNUzEV39DEfcQR4dg9
zr92)_st<moC25eOmL&4ior$VIHS|z3s{}c(VOFYsx8DTiAIE1jE9Fk~rf4w*S&d3Q
zz+-*gc)s6iKZ12shbN;^DhtWMKIodd;)qEqu(4stYd850eW{Y%c_!N{N3XPR5T#S$
zXdr*pVz>(!=6a}?WyUh*ykfJpu^X#cYn`xL^i<JK6x>-t><`1UJls5tXA>jaoDdNg
zmtd@Uer}GvCX}K{@SzpFx8C<4YXnH?lOabQyVDQ3Gg;`6BF*He{miv8lD0P6ra~+v
z6u2;u)1%(pbM<P^(Fki_o6u?I+)vL8)4`%DBL|C<v5uPgU(FA9a<x@TxXcw7w~#q1
z>ON{5sCW1=JG;{vWcF(#GqTj@cba~cp$&Jm8{+tZmED3<$HyKT<a9Zfr?IqRE`{!~
z-mU#FN@RRoOs>l*-Pv31elOrUqN?S)i&_?<7CT$9GmdI<VyUi@45^M{mm$#lo5iVB
zRg5OCgjSH~$9nJ6S}_;`wIh&W4|gnYsn0%^3_0WsK)s)$<*+@#Ey`BulR_zJ0DVK>
ztoxCY5WiwumJ;(8kKD7jCfe-st-%kCPF4l{{F$9&{szrf8|h(;PMD=XC+tj`CK_<}
zwv<NxT#NWwUTBc~(0DHK!75!&{gCr~-op&dK+%XM2B&3*V*#4KPqhX-YP$PRm^C3x
zy^npYNttyxruXn6Z3Tb$b#e*mFAuo|956$Ylf=YWg8vP5Xd%>jxqq}T8p39?r9sKq
z5Y8xuyS`)X%=BP=kJ)M2Mw629K_876)H7|Jna(dG`viTN`i`c|yxOM}QXR6&@uv6I
z;=>d|lB1O}OOpz5*FtL6=HkN?Bbo-)_xpV-`2Xg4(qkd}LzHE^W$=AqMa^pbv&7VK
zzG1?{3<-Y&i}cJ%Q&nUx)`{CrU)=+ioXH}-xaguiuisi)i~Y&@cNj06UmS4-lhsRi
zr!3p&;lf$38PnFSD+{*|t(z<kUNJ3prWRF*XZ>+wBGK&5jr+c}xsCdnqfco#SRBV2
zg-B^{i$NRrK;pkysg`$Mq)+oNIjnqByN=pEig%ALa`cS)>EtmXuRk-hP`89s_%(Lt
zz2?<;GM=X)+lL{<ykDY}7Pn*u%v?5*`XR}N7S)|OMSoX#i!b`iAZb=*Jth#6T;Xy0
zf-h8F-2M~>`G-5EI8#mZV|>{`>Ly)&y`O(h8XvM;{!^4H+5Qx=rM)0xe^}$gi@o2m
zDOL0t<;CuN_LdYXL5RPePtt?&i&o0qnxm=>M?1<k4nZtu4t}#MLQbQ$o8$ylrB)e~
zaRpCFEDKVr*JCr^U(oh4VGMsvD3Bh$vuLuB_@I2T!k9NU++);M`i;TDA-$0CY;Kib
zj_M0-{+<4d=|t!Adh0gKtF6n@sVirpBP+GO{!+A<(BdqWC+eHe2gICJthI)?bWrt3
zrIhtl4JxKpnf_OYOt+Ss)<#>jQ@TwDtQiz5vMPBukj!MJQ_o~~q3>|5o_g?0n8tKJ
z+?0!ZvpDC$He=hEr>K-^(|m~>l7XM@7dGOrMSQTaYln%!z-3vWN_F)lc#zZN>-7Yw
zKS1`i)I?E1JgKL&^7-1ugI9>isEm>34Bte{x!E=ji+x$nT;i39MFB=OCP~{d$`sP-
zeKo3+?SfIQVl_)v^_=LpZ~bry!o<41Y+t3|N6?6|xjGxK+xN}49j(cIPDn8H|GVwp
zKZVTcKNdBcxI6welG?wclM=$2P}+C6jZZdy)DLA{z*!jFU9npt9_gvFXN;1hAyihk
z-dD?UF<q}B8{Bho-(6aZ&bKS$S{p(>SU4**GqP%?hWNc|wW>IcO!%d9-d{iQH3#G=
z#F=XQRM3gnKciJ|9C^Ybqcat5M`coyWWM(Fd_H3P@zcjYj6SP9@S`YgksCxq-8^BH
zDeP2w?3NwRLiqQ25yA6fd+h2=4|o1WM<<@8(F)c*DpzZyTmSqWUgBPmZDhd!pQ@<B
zVQDU<wa8>nnnFrEykSBXG6+SPDk@dHg?%Z3cbff`{=AHB`Q3xl?b%gWL$a{m9NO}`
z{Y%L}_Be(oRHvRc9PTKWgfsJN8u;4X-wE?jt$uFXo85gn)r1HM&s`tC3m4WD7%cAP
zTp413J=U#%zIHWWS}|tKHd2{g4huHaitW|E3~?<FWC<zf^u~myrt4Rgli7FN-7y*(
z#8y(~FRtVlSRefm&Er_T>}ff!a+j7{<@Ue;Ol*_^lV<9?32OV`X7Ie9GP8i9d1FXy
zd@gr&Om?uxq3V;evX>Uxg~kQd2e!FbZtgqt`0hLVf!uld=SjX>Fk2X5H(@GNXg!b-
z)rZO<J?Y}57UqMpBf8&|7<oi##6EPh7p~Xdb1%`a=v^62GEH!O7QDTd#l2dgK59^4
zxaNhkBIKYELE!oit-Op_h)d6>PW(V!RIYF`m|Db_n5-)LwC$!*_d}N{E(%FW#SX*x
zK0WU3?wGx?DOjPVVCNUPH;E~D;_0*@_iplkn7Rt6D%N(5A|N8DARw*ME!`#3-LNSE
z>Fx%Ll9KLjHk<C0?uJc=ba&_dIRCxpzb<s4E@sbs@xD*Z%<!iM{q1~-aT}kjD=KPQ
z%>+z2vyK72_)<qd9e{9TG>XLq>Q8jp{A=4rSO3{1k^v9$ad+!7KfD@`W!3r+UZ<v0
z$P>>IeW^?T=KbRvY_*1X->SgTF-smc3pj6sz^xdV?-7ncURMPxEoWA037P%t&{9iw
zwx=&fb0(Z7ve-xmIVolb)rOmewiLH9gEuC1tm!qgs-j^dJW4sVTzuLoyxg<G<L+?a
zJdEuI7)ORB!5Y!GHs|)Vm5;%vws?AC6X4`O{F9GtjM-#?Ykb4`*Ci4!E$;>aNN{=v
zuQK1ma!jSG=ha3)P!>(OyQ=HxkIkd;@{0GK3C15wmk0v^UJ3~g_WU9d=h@8MR6&{e
zCV6WQfz=jlEg=B$D@d;V!AZA^jgPhPSkYnb-D#$dr^nPejJ5GkbXhQ)w(OU&-c<=O
z$hTv69!9<f;4+`NwCq&bukxyJze1#C&KpXESwMnv0$;3Nla!THYr+NbEYS|6{5KEh
zwuteM@C>GPGf$2`X6nmGX5&Pz$ft446asK<rW_~XDOFWEmqzJlL_De?QL4O*UZ*gy
zE%Thji?dJaWYhTh$@#0iTh$v}MU1(~`EHuzBF~E&8=~gA0PX_l&p~M!*)2~>EXrZG
z3LM^jj!|u}mTB7c!hb%Qw?b?CpWarD3l?eh6!3&->1n0T%PsO{hqLxgr<v2^iLG!c
zHFc78uFtORa*OpJ(x=wbFTyj2)tnTnQ>^nk-;C7yK*U6;Xeq@=`4blRFlh@sJ3}^)
z7w+VXLsj-2_Z63Hs=k7c7UQe!He=uZu_=Ik#{W-RK=~ipT@(buBi0xr$@s?8EaN&|
zGd&XvjLNbZA?dUAyv|W-X}6{NSr01*1`(NPhf|Nf(V<ZKyUg#}{LwjDElozZ{1XtT
z&nqml!NeNklX7#418kN{mmtFuQ)FZB^Dxf;_)`d0<-3dTb_Kk&Y{#M0+?=_6%RjF!
za-~klCHuREep&L&s_V70-BoFXXU?B)N4p_yZ5HM$N%7Ac5pl&i440c;7_6E}T!tpm
zUX2w|N?dua(!`kmr`WvtM+3gbJaS*F^v_*8Qbz)>UM7~Z0a%7=gJZV-HD6#L*5?u`
zc2;F@weBH=j$)?RI?cxe<DG>ETroDbr(XYz2q|c|2prtDG)>N?D)%Gjp|@4}cb6m;
za?&-NS83TsU&O=nqf6TVY(68dOY#e?2Iwl0(^0Z<-UvyZjv-j-pgJiNh4JBv-}pYL
zwu}6oN{Z!1>xDO=*)zNOuZRnc9L=DzUmU9jB4Weyt=9)eLxlYz6Qc`r&Bn(H^$<_@
z2^p;!mv85~tg6g!NXkxUenoM|x_rD-h3aRRZH3+40>^S!F&G=?4V8pscO|yF`$sol
zII)!SM^Rrw*_`)?kH_Apo0aSmDwf(Gbv}Par-Ckuec=qEVG+w<SVFbwISMk_zYl#y
z*d6SWcS_a<cOEf^f)n@AAshV1Bg=gbu-XrMkZwPl(0?c-^b3{RvN>g4Ofvdr4;|D5
z{O;P)KC`}l@FnfB8pld9zhEbcfCftq?88R4D^yUrGW(9TsN6&TV<+*6h2Yc(1Xfv$
z$CTYgg61+)RR57g;OKYX+A)gp@c5CyQseFy#lMaiU@cbe3{|sjR~(rfl6YDJbpAt=
zZ9<zqzU#<Yyz%zS5uSmnQ;ae!pZ@xV$xJs{j9WJd5$(!KgpsiYh1jXNVnZB0u|2G^
znTYVQ5g+-wMOXI{-t~t`No9V!i1ggz{&2Z=wSuEUf5u5d@yqA0h5iAmu*J8o?r{!e
ze8S#DyanLc8)3F2+pQ#CBUPlyigmu|l5B(BN;zY;7Y5VztWCHUU)$4Z&BbWh>6C@|
zRfz_>#^Qf2Lu|=}piVo=qt|yC?uC14p(J=~XV*In&Q0JUwsFVCdc`yE)AJc%czJ+R
z8LqFI>h@6I*4(FtmzJKk7fgZfm%P?+oXuc4c96Kt4pq_||M;=|<Uj#{GFoaL&8oo*
z?d=e9$#4eLXUc&5jz8cV3L;6ViIyLJ6@M>K`B=@9`L!x!7fu=>=q;@$>vxDp6_D`~
ziRB~D4>bM>bxy-O<K`#Zd!BCj!ZXaFWgA1UvS#d5*vKZc1Wev^W82bKn2Z|T*1D%i
zJ=$liJCh7A3^HE@#`VLm)tx1mbu{IV;0ni12Sb(h&GI3il*gely4iab=H0ztXzX^t
zAh_xMRqUT~7Sty3hUoI2RwFxXDyO>w%Ab2COrl^U8v@$Z>B5DSqw#Ua5w#K9_TO@c
zh7yM?e%&UGl>0M3_J<_fesHuJlKIS#(?%<#%DjrqR50VpEp0T*>?N`43TNGD-sX(+
zX1kA61tqyk60z&Y6uvY6y)D~SjZSuV*F4>v-iO%XW?qf=1LBqJ?pBz=A3v3h3E9HO
zf_Z!UB!;UJN(>#NMQ;@S@Ebp!p{~u6AO#~a&UnXsd3{lRIIpHv<=FJ`MZJJ%F9ykY
zX{*@y-*9|G{PhNGb25pEa$0wB3sw$?UlYB$a$}w59A+*vs)k$b36V)KMh>f}nC?uF
zH;EMfNAUZ4BwwP4i$HGo;qd2G{XsaUl9Hn?)Cl6xDYK<OnLTNv77hyd#(KilM!!{3
zWI{|1&jKOLP_rt@Ir(JALYw`@)^KelcNsiZ;c8FQ(Hm;i?pvpLw>8GZSG&1Ra81?L
zDMnklJM?E_-e$A0cIIo0?r5!cogrHS+uvGZAE;rV0ehT-LB<t?X#?jpPgf3<!1IRd
zMljIM16&Uj2cVqBxl(OvE-w7}%BZykcEPz{sLtd-Ss95Bi+7arZkVk3!EArb>eXr_
zF>+irw=+$|{?eL}an4OdO=2X&9*2r&P-lzDYqCVM?40;_y9GQ}Pk2`IGZMVqDd@`i
zml`NiI@<N=W9vNRd_3MIia<j+ALnC;pKpIyS@L+mH9}r?;C4Kp+1~;>2y%r<I=H%S
z$oKhT3W*m0*l4ObMAI|MaWq}|mfHPVcv>L1zb>|=Y?KWVRUYk2xXO)h(hV@HFLf$6
znw`|dDU0z^FaIP!uK`Iv?wLuKNjM!{x`8y-H|AKGy6rikAw7oS$n3ZWyTOCQMmLq$
z#Zq|J51C%Q0?Hxy`&e-$>;#UTBvpQyC^6z6vU{SxjsfC~z|E77h=q{nmz2KY@W=&>
zwc<=Y9r>PJtOi3YU{KrjBzEuCFq&*LAOV?bXrijVr{K3vlk@NIx3{PBzu2zqmZKGu
zuK)H_T<N#ssGy+^2RhHH@+1Az!+MY&F@FBwT>yarb3E6}C6~WC-@JtO<Dj8QLUb)d
zWe7$RXIcr17yq{HgIuATz}~yGnzetWqqG(3p$lPAWOUr4dXQfx6;(C(k?U{b6T^x9
z%?rKIx1zSR%DOWd-A<LOQ;J5)N>2tNH;wt6rgZ}ySH9dCsb$o5#MBNLb61Y8R-{M9
zhoudArvSDa+&5u1UO#O8pc{5;I_FNTq&RTW-zKheCdzs?m_lpi60`eed9~+DYjp4&
zI+60Lm;3>@TnKVe1fmu3#jYG*)8~NFqoV;=*;;zF7lX+~IqUS{?K1_`jJ(izl@><%
ze`X)S`JpNs%?9a`NiaHX*}zp%*Y(3swpL?q`QXw(4c8BKUpq#<q(_CYDa{T2W*+Cx
zbd(x@`h2KlB@(mZCzV`hxY7eR@n~o`oi*I;WfhGrlKZ=+Yb_n`0blO8<y8tlQH;SL
zhtynK-{0YDTP*2sn+|AZte97rG=CE|JL*<*vl#eD&+qVw>+GfyT~<PDWRy*>)yVP6
zt#t3+F~n}i>C|FEv~0npzs&j1k3!1@S%Joz@8553Sluqo*sf2^8_GO_D^sMSZ};~d
zL3NS|pky=Klw`YFk@Bo}%E$*K@?4=PU#TpcxbYv_`E^7;g6)qW(9&3)=iHDW^tFv%
zbIhTm)soeq7MRGi*jmok$<c4YsqOm0`!&fW9^|eySC58%{Gr^&jsi%|dCq<Aeu{f`
zpFz!5H^8oBr@2Z}uO8hQ60b*7pr7{Di%3dH#mrg7T<X4aSA@jlSWl<t-+9&M%Sy+s
zn}Rvt^7Aw6y+Z`uOvusvQ#61|TeJ0fPTp)nx1x;&Q8P{^v+<|+RYnb&H>rFq97P{?
zqR6~SNE&!IF!)BAigF|iKRgUXuYjBGpa;7UOEkl%JsQq-4N{;Xd)(`#?TpABLN3|s
z)`36Z;hk3vKVD<K`uo3e7DLfLXm7gCe)vUNmgR{h#WKU@`Lrh>3XljJEg;G=U><N2
zVq_T_(F~a>B#2LZOew5LXEhq<wwx@R$g<U{urT-fJjX`SlVzKaasdTI%91NAGfT4y
zTjQ|Wbl+==2t%9LGoa6S>ZCBU7XE8wAoDc{Kwqb$xsd0%spm#}92CO$rePBlvLh|%
z1dcQ;!rfC`z_C9nQGcyj=4-VFWMK?K`RwtLD3u?%QmmdU$B;VnDu9U#G@uh(i-*fV
z;#B)&lB@q73N=oE*QVLm$HN(}N!~wHGlXQZhsz6#014@sCrGuts?Bf2BErGr<1Y4H
z&A{Ckz3snCKPRu8*q<xVTT^f1h0mT(K3u`7W)&`~ODF|-jbri+Q=nHc1BGqLDeI|^
zF`u^YyY_atl9mNu1M>!Oh5jiSRFrhij~U+jjk>~<tH|}Pcq|<AcMM)ry-BOSf%mSg
zp1EoP{id4z&%*hrk6+VJO~=K$`HHXWvknKn=YFk!uME^iYGw}?qXm;&-*qC@%0>12
zT;cztYB3P19)MS^fKp7ssGbV~@g`uYa7QQ7LEF>u|LTE&Wr>@vQ{I&pTtRg|+>;jm
zbZ}wr&L(~vs)yD=_2#(GwdG5=0~x(AM$TxKN<Gr?Po!uok?j+az|*@Ek_xK{V#6s5
zP*HtIdabo+A?MnoN-s`oE8CWRke^U(D#v5g9G29T+qS;167lts2**g0w4dQ+%YS`~
zpCpsqnl{mNs}q4kH#VPv+SA_~p%e}puAqQ`qsfT);5@f{QhaS=+yifAHT<g^U*QJB
zAmk}rIdSkoM$T3J^s~^6DCdzjcb;}DGQ(h*#Ws7x+3q={)DSvS%DigBY4}fqcN&B(
z1!H8d_IIlyFRnNAOFg#PWv<OjjEXd%h@DtbpiCD3`O|`f&f&C~sakT%tNo7@`v<9>
z0J%q~vOk-nF{Qvc=PtD`kx=3i`T=@%O+B$UH2>3JI>APIp+BLhmhJuN72glmRH>Zf
zkeLtMl8vg+XhQC*`<*#Qmxzg(KZn-gwRJ<$+>lhkz`DBYueTI`wBkY^*&FsDv3n4F
zn4L>A@xg)B)n$jBhnCk1j+V;=TJWT7PYd*Wu9WO74Y=+J*L(~J_;%+A78Ti}Aqgov
zF`6iSQzKD5Q)=8%mZw&qZYM}pmY-a~t(B@r)e4!Hmj}fMdKynTNaJ4fD6u^2p{0uG
zMGX%5JGJlirMZXg=4j+NXn#Coz^ksxRK^0rEru2DQfQBIi-9c8$ne)pmH{)>eLJ3D
zsN(qWBSn-Q9W_!G1m;Yn!*=hz#|>wBbbgvmUteX1ZfJK2Ytq2S(=##=bTY;B$;#Sk
zvI)4wWq7r?+a4`~cHm@dbugQ*H0p9Yu`uaHu>xNR`ogS5*Vj!;OZlEX#D9{r@yea0
zweHDb`z)pD>hQsIr-X(^knY8!zf2`nsw`6En2G)UR4`?EvKV_7av!-nt-*-|_Oit=
zym}@}yC0pG9JS%GUFJ<U9gS$X*0=p_g7(37y**C@L*CJG`wqgbzi18adGN0X+<!?j
z7JB*QpRb?Zh+&J^KPjT)BYNA~N$l>f(4S{1t~ptBMylR9ING-tn_=aCf5wHfjCIM!
z8!mh(A$`4}lpILdA9I}CCfK=VYShv3eCu`Bz;AntI=t2~_!TT9sAu3ED=8YIaSnt-
zeG7ZYcW#}bnBpjY9(A7C=ex{A`Xd+9mJ9CXrn%DK%O*OpEY;;NRvpJlh<MBU!f5XO
zzx3IDtIKy4#&R|`y_B`^dZfgE6N}psx(ao*cR0JfOr%tI-gt_#geAeq(A?glRb&9c
zbtFR2ltTLgDU5>oAT`V&*|73{Z(uqMIyo+Ur9`r>I2zDRn+l7e)ETIaC<agP9Nwol
zT5y$NX6zb|%i*px$PL8}G0H}u8aHcUE}gXFi%($?d5MQBmq598c<#Rcq14=_VZCC%
zMZ^f)MubG3#adRpc$tb3bavi60bhf&{pw}>Z0ta|p#9Wfum&x$ny@>bL&w13yzrEG
zeKG4pM`Km&uO3Nh^ye=5O70ZpCm9ieqp(a@Sq;2veXI+v6J~IVuEy;&7(abW?;~D>
z-T857&A4qn5B0jkCSP53foL9Ttz!fo9e4D0T!D`p=r_LZS8r^3(*aj9=ETQo*4=OH
zPL(AhIXOZ4cXE-h_V3WtU+#UYU(!czJfm1IT6jA-?$c{;bhx|db#i1=7d4Ai%+9W_
zFD2k4gV$LpVNlG%^!M75$Hi>Z*%7|s>C|U}FfV}jvF78{B3J%5=a6Lh37b<NEWX6~
z`lP!vOq}e7n7F-v_69i=@!;0SW<*NTrlKrEigvDhL)U3<Z9F9=O%dt}uDNg^so#8#
zd$g3M1}$D?C>gHGJDP{6C&$~;GCwDXzF;xm0d1liooXtBq06$vm8l{<Y_<}0IwLey
z2XKuU5sE*r1wodRiFlrU^`RV)&l7{GM?v~|I$0++YDzM;gGjF6kI#~t{7a-ej$ZxX
zWHNuof(TiV+<V+bhRI#Wd8oHQtd($1Me5Ozt4#H#>A+J~RAk&-$M0k7o3hVLU`Fd_
zv#=EGjSuZ&WG__6orUa!JuC`b-%<VUl^a+(wGw(A6UCx;=(oe~+^Dmh*kt=7Y`<Pi
z#E|?)LxR2Ke1^wJ5oL4XfOCO=B?p_K=!`010#{sAhOfTa=Rd>6Pp|sDGjx`32rLD#
zT2@~0XJ^(YR=GdHI$V)Xe(|iW?qN{OSHP>wpxQjI*o@f+QINuQ6<czhacx<7mxN?>
z|98JW5gPg-e(>xg7I^b@-_?nfRdy!xSZ4&{DCMG|rHXM~8h>(PA|Dc^J&Wp|K-VCT
zGZ>gL>fhgp5W)IiobvP|sE?Z%v5%P~APCKgz>*6sZ1A-t7+gEq9m4RDY(KHAqmEW%
zH%W|;pr+%`bM?9J#7!B;bLV16UiZGhV>^&&N~R+k!?hO{PO=2`YkrHkFP^Ct;lk;d
z-6@d^TGpT4j`ka&$w#d8ZY$W01kJ>3W?;54=H}>`ba0T1*K91rpi^aBC8V$CZ$&F5
zJ}br_v77;0W@@B3<sJiJl<go-?|u?ORJ32vW<WLnlLhD;`}6e&3&jVPRL7)*Qm**8
zQri|PV)F$YF#p9Ixbo_d{4yDRBb-tGo7Y*A?asni$KW($;R?s_Ho=Z-&)Vbd1EJh;
zGb67fBqmc+?wQEm)#q-zmXS2rEYp?xy=}vs2??7Pq>r*IOq>6TQRwM?>DtVN*4KFW
z46xsea)c|z-cH4U1E>ZOwzxOc>J=;pBOtH{BR0MnN+ADnl`2mYchpQtW^BA|LHhLj
z+y8v=YJK{y_A>sVFz)bqKho7SR}CeGr(RX_v%@V21u*kIwS32hcGDJml70yI)beXW
zr+IuJ#8F`)wrc$?c2BL^QU13<VTOaV5~Vi;_@!RWnz3*2HZ96L4?kMa(fRlF=v`dV
zXX3_;ZM&Rxbg+1h%D$$-R=PgbbKsaOS`#?xUe8sgF+wU|%VJ2{ViT>9fZe!O8LN%`
zNi1p_#v@3H_M&}v*H(yFQ0tMUWAM6XReGv4uvLZ4wm&GCih;I>?JktG^Ed5laWYJ=
z#@EeKD52x#<VBJ&qr-rg{M?=xd7r*3W-T2zvBU9oAv(>J-j)cTr~HM`X|mo$h+jB~
z3%FHtWN0z02NXqb1?d(m+UxSq<?-1+M%?1V__N~#bFGo0d?qF3x_++0ZLm&%8xnqG
z_gy>b(sUBxIljDET1FWutr$Ae=gl>kBL(c<a$c+P`%<Z$&Ne)5)6(1uA-9($iTeZP
zhp)5g?{Q!d_fR9Mf5(0n?R_)tQ<;??j^2+COq+Dn3ASYu%|<KY(8^{iD4H^V?pIn(
z;LY^ShY*1meYeA-$76~{ELU}Rs__C9rCd?iY}nb7(!4F#8kUVeFtIL$RxP~E_3=)!
zXJ%odSeYqe;$Zsg0YOeuyJ<ehZKZ}On*@h4QhnJ={t<NVjv`{m)$n+_bPqP+81f3A
zs5R>6C%T~Lm*w7r+-fOi$B?Q4P|d@%<-4{J{mMJQy5iS@)`-G(EERH<`)2I@yEUC&
zfyz-4td+MCnrr^#_70`aU@1(Tb%Hu4H`cK39L}AB+jx)qRcxlBV_@uO3Zux7{i)ko
zJ4I^#Dt?l<WZHdg)8$H^EBI7vas`1x$wjl{9>!i!_cr!7n=@4%bqDM4B(v!7CQ-gx
z<_yYv&8%5e1<J}Shqj89fZ}DFvWHHLvIt(-Zo|&8yuVCY6CRt(CYq*rI1y%mK!-P$
zGccYr4SJDLY$)!ttnhVcVXJrZYPCyc3Tj%M50dryM!Fiw{`uy&-~cA*3&QvOq!|Im
z|8II|0ap|<d5jttr&Ws=92(4)apQRcUc<Q<QV>)P;#oPaa$>=1Uj+qpKQ`XS$EA2Q
zcXv@ySWo#h<Wb-9%^en(w#n0wXXVCXNl=sF;5-^?=?`|aZju!fSa2Kc9_)|$=?L98
z9f4{|NgCayE!W%cyuM9vUo<xljtYAfmu09WIrtU;6a=sbSu(_q`pe|Tr|!dLG!bI4
zhw$B%M!C}~SfaaKYL(pQdF{e=op<%l7QeCS7gCVh@^Sf`$zmj4FE}L6IL9uuWK7z!
zL=$lsn3`HkRT&jl$0jIb%m0>_Z~Oa`>{otxatAXWrIAe9$Gpt*cytRc$M_h^5Ga#u
zoR+MR1S)gD&(Q1BmHxfvzUju8V^h4KTX!DWHKP;jy^y*nBuui**SPk)Og@qC`}21z
zG0qC`Se}DI0vFZE_{2c@jucDII0xTgw}3|CP~Rq1XC3|R&{Wu6GBKfq_)-<$wTr4$
z{{oocR8^>Jy)Z$+minnS0Vv&TPb!yNm5iqVys(VtD0)@4_U9cahI4aTr}L99SO<sa
zCtc1|ExFHii&rzVjgDtS9yvx2<Oah6nC{Ide6!Ytr@XG{7=wx$o(KGIwR`&g;NT^y
z4wzK27FGf*^z+jZb0o8h=rbpm8FhQmJ_^1+CWOHpJL-5iZB}~arts`Sz}?q&%*?-<
zislMlO$-ml)R6ieH@*AuVT|<Bz(7xS<tSy!X6H!}<8?3~-k`=!__O@(-k8oOhc3hx
zvAxUdKlJZK2s${mI;gke7@0`m2u&n=U$|C}fT8`Y&LDB$4`VSYS^G_+7wv@ok=3m;
z-~d6CJWq<h{5HyhO2RsZ`yyG?3KbP~LTUoKi4GU6=8*-R&y6PIVm}+}w;XCHO?vRf
zlQ8&Vg@<QX+*ubxS?qZ#4M1nFV#g%dnM;Im=wF!3>E2xRpgFgJ=SDLh9erT?kg97R
zikz4iFM8o-OZG0HNiK0Levyw5Uk7`l+MKmzBJhP=(&|vwhxugwijMW|P-5$FTQ@h>
z){!p%#FmdM&yYmjD0uHBFns1*y|>LXQ#hms{p_e%rO90W#dUY@?r!h5uzy~^rIsqQ
ztH?^S%(x0epx?3Z?`3>TPFpwdgwe$8C}eLwYIv?LTvPdzx&vKa4Hs1;P5ji<FuWIw
z1Laz@A37b`*bp~T#c|nAULfEWL_kfwf`c<W+Z~V^eHXNY8C3p=P3h4yNmfTs&1sIk
z*~>`?X_PosjGD<UFPBuA?bz-_*LchKVoIaO#dkf9XC34(Q*`}n!rpnYv)8{Q^bhS%
znh>N3TeLQoQnjPW<Epu+W2;*}9*T{{Ul#uwKz|!MITD&!%G|<ICPB*}1W16Hp6u5j
zd<-~iY-_XW`Jv#1*!ZQo`aYtAddZeMU&cPZoVgB{wh<nkTL9mVl?JGpg=8y>8tfwp
z=e;H&Fb9XKPL7o3u`UulJ(v5Pn8%evAHRVg5Dfnh+W~*xasAVl1%<}f4B!+U!F#=x
z`7g8yVB;MFBjJz0jiYWDjwK}@jOTM-$&AEkrRA-ggQptVQ$~mbocz8>%l&&5cbL`;
zTl{o3xzfy-wL$)tx$@lluH7@VN5xFTIz4~p$08pkpK98gD|s@$#vZ>Y;MRl5HLTaE
zkjPmQbgGc#sE~Ftb1=6F(!fsV85n30{{XuC_~yPws;}=#j?Cp)O7a8KZ#M2JUC;Sv
zTQzCv|J`DX0Z)n!<N@^4l9608QGe2{G4(n!a?!*fI21}N{GI!+{!I0_>G?0=amksT
zky88=!^p^^T_1UdEaNa0$geczUr-;eR?i04a!-PwJ;-6aM*3!JcgNbioYo7#i7a!3
zFOD~u!plkl0O%Z^j#nin<`wW_h4JF6GW=;9dg_|>{o%0PJ(2197#)<DDZ1jS8hB0=
z6fWu-BMYNM0y8{=p;~!l-~}QY^tAp&1u%r&iQLqW18EnZ3RhRc6coSbeS1`hvvy+R
zoK+UcqS&3}!;xR4;FwZn4rB=Wq%6Zk-~7E+jb6|F^|fb$P8!Rd-39oakX{=7q1grM
zDXx{ssAUVQ>~qe`RctEde8|2f0U8=6OIerMcY_`pdZg5>I`<#bD$1&MbT`atOiX0P
zM(@JJZR{OVBT`5_r%^szm`rZe%%O$ARX^oK%hMILwTRzysV>-0?^*sIZF=(eFBxWi
zi!fM{Wp^7){E&0piSTgqV1l6%L=s-^z(SbjT<6qye~y`O?m=IHrZ-PgVoyZf3(N#v
zc~X8>(!e>U!+MuRD#UtTbv84qvuujoF*62ZL$G@eZ0zc-kBeT^d~7aTxcf8x!!J*n
z#@xe;Zi%^RXFB*t1-7mLU+CuTH5{soVonS25$a%H%Y<qKl|XWAOM1k+L|Iu_g*J~E
z%KYxQ57S+v_}5;I_=#(Jv#H$vnbkA%G5oKw^^b8<p6m6Vn$W1wz;=Pe;L24^ddywf
z*(T@<AO%F0p~Kg=dU|Tye}9hutD|4Y%dn{nwb@oEJ8XMH&3>6*jfW=_6BQLnWJOJx
z)ok4S)N`gj+I^t2-RB@y9qh3;OS^gJoUc@tU1~VZZ)a!PeSmuLp1BlkXt{mcU#n4{
z*0JA)i67XG^TX##(VktN6rs~ClvTl_`l7S5`|y~8II?A$=X(_QFXzza3mHD;WWNfF
z6M1oN<j9dLa>}Q6TmLVVd-_5E2o%KCLG>Ql9v4hG?~4##E}9HJ)wPC9PJK4CkOpE)
z8eZH^P{^{438$ZMtE=0Icr-7YSP^>bo)6$XVUlD!Apv5A%FND#@8G&MA^61~n7>qq
z;_}TV{MFYB0n0Tl_4kj%fJAA@)z4b*lLc-b2;^aTB^k6lNavAd3G=9T))yrn?`7m4
zI_?SKb}~z}pw0f}*SVOQ`fU2;laT~<On#w)k<hy`{@d@??AEIn6S`yIDaZZ{%^ppb
zC^Oa_A6h07jL(UTw8z*G;x#pxdj|cSWX~oO;uYl;n#XsvoI2a=xo8veG@V!`>$sp|
zFAT2HWp-x@VHGhNZ}4aFz5;}CX(4FsRsUvSMocbEKChUG-Ke?<S399*QD09s?xT>0
z&^weROs<H;9s)7hBlAf^_7D;gJB3#OjRH~0RJ;Fja#4u)_SE^O;g>LZdH<Yb+IPo2
zw47m)-=Da+cR0Y9bOr|6vUAEPxsx|7x7Bo(tjt)~pDZ4x372;~N8js<R?3-jBCS+F
zo!?bievXDQa&;L&>RY;df>*f^WT=p`=Vq;5Ex_=oMlrxI!=S<J<36{0$W)4_gpdR^
zKtX_bZ_LaRaY&&FmWAJ#=rj8*MW~mWqBZT$4Nh0-Ku<Y|f2Ii~aoI3MT5=c=L@mxa
zDIF7{L)~=#t=&KU7FxrIu<pv1Mp)i_xJL&rGpbGbdf=mm&7ul^Q_V)=u^52lkM<5n
zZ^{jDCn6%I4x7{cmZD7EX{C^z_u7CkW?JPzkBa2;Mm(!;e>x5XZ$==Lr-0pS4R$ZX
zGAO5rum%ly&gW=&K2+$YTzfvPn07ql#lrFYs+g5t7Ko+$d_&v(vFYYp@+LpA!}$Of
zt!i7W{kZ$(wVs6!qY4QZ<K0*nde^e-It|@XJ6i9V%WO+NtVcBR#f@YmF-!EbE{NT_
zm2yxdPlP5e3TnOQQ!S#b+lsomjG#0l>YQu-mi)+)6Wi5?iJ;P!^Bt-|Tu_y}<FZex
zAgrjOX`C;%tAKEx)9yRa;Z8K_9?8+lv{Y7_lSOo!tbLY~bizpS(OT?GLp0-a(vD87
z6VH<L39jLSS44`$@_l(5s9LlfScYrHr{*|?OCYB%iX-ZaEWI%<Y~B~Ur|r7gnW919
z8dc@?WLYQb!ovS#2~+jcC%rG9|5OvtRdhVxokdk5B`sWADZMp%ZgA^Kz#Ae7TPtJ1
zMTk`mRapV`tJQwt)*k{}JH)VbEizVv5zIHdnGG9T(#J{R?9K00dk|U#8ZHRso|8;3
za?PV*xdH6EN_#=i8-A(Hvuu*>NnXbQCg%pvk%+~>K-6*D?S&EbzolRn4Si-$=ViRS
zf(G6)PNSTnDm#|um958GBqY2JA6bz))Y+srZc52?`E%#rA)DmOGn6s2Q_n=l-Fa^Z
zid2d4$yo4iv@K4Y)4gnatb*vXN{e{T_;%0Q4CM6VBiqW7zlVC$YKrsLwwrj;Zug`Q
z`T!!t{vu3*TMY%j`wQ=Q9$$`;O#2|*7gQs+#*_R3CSF3KP25T1EQrh}BlaZi#HVq5
zQJa>U@c80L1kh+X+1zPbJ?a52OBL4%&}C6Wegbdoo<#{AKwh-@hzS%&!bGNTKnU6{
z(7*>}Eu`DPwaooP`;lbfm-V69sQ}YkGdW@_jd}#sV&<i1VpwL3T#5;Sn{giY{LfJj
zW^Q84gWcV)l<A~pYQmFxao^$M^U7MeaupHsV*EKa=p3t^9`Ek(#q9{r`k1!@<R1n0
zgE*zUabE95s;sDI$;5n^94PKETgAM{iEQb|pYPJJAlu{<s|m+xKaGv_jPzSjLi^!(
z?gxrggWl()zKYbKHm!mcR!-pzWdjmu%wI$eCcCGA$QU=?ek0ZUi=MpPFf`j8#zpF@
zKv%Tlb1P=DQI-BNYbnkX^twD3*QE=;!Y4leVj6j#hw>iV4xM;~ogye8=d2n+{3}lH
z7AMjjqFG{aqz*)H^@{tqJi-HW%W-hIEX9B)za_FP-f)D^6I-2{Q9hdAz-hZFOj1Au
zn){ocFXwjor|esEKFTz&0XJBh_>0I?xL$v=Yca{>`n-~&Ln|b3eOpG;p`&hw2O-|P
z&hzL^l0@Z<1|HtwdiCEYD&razFatwV*DS&zvUCnPLXl$g+pQo;a!jynI_g7q<q<DM
z+TUz&;?mE~3B3mxfyHvh3IHs)IpJ4xt{Un2daN2nR7_MM`Wy3<7-dO!YiqJU=~(Xy
z0_q|Awnwxyv9*ThF#O?VcQaiT&D~(r2-ClN9gUMubhvF^i_o(L?q@4QU3U*qoo0jD
zXWHYOiQ}?a?&id@|L$!FD(bz$;@Bsv(sJk1vV`k4H1l-Il0Itm-ZOpf5NS+B&A48|
z7@evJy|gv)&Mvo7*tgBDvh1!BzDff_zY|zQ4gEb)x7$!|11Dv5w)@f>{AQlp4pW5>
z>+o6Buk~}1Re29gSc3Zp?_Xys2mhyJJ#A@~Aan6SdC$ZWNK&=m);IJuiq!A3dMI~X
zm3>+H;`O`RM=r+g`EHcVds*P+Klp41iZ*XO<0lwv-0Q&h8ih9}GuI@ZtHLa=P}zs>
z0%HJ34~i;z)hqT~bo$T4eS>9z2D5xwkK^!`W3HLRQ=FKX<Mpr%1Udr#r`Hg631~1t
zjZjhac0roemS7VrvYat|p8f581Q)27O+!Ro)wRB$b-iT)H68sbG&sRLIN^9)am<WO
zlo=1e#1q|fj;5bFg^r{gLj`uXq+9q66Bc9@)lU%fUROZ+cLo~^Dr8)w>2iv+nnxy|
z0!<)s4#jnK@nNnUh&$sVS+5GFx&ks^$E%Q!;*Qx4Y9u6-86G~ipr=#E!RelCs=|s;
zvqx;6X~%udtT&3`RsANUq6SYBx5+OlYcDDd?5*Pk?;mj`OvXy1n-y+|l;LJg*$6&%
zT-L-~-5ES*Kp>Cfa3d>2c`-J%2DEqa-hKluG+iU)ayTBe-(YtJz=edB9Sg77lCtLA
zeIx)3E9oF#v+}3`HDq?}WxzxWbf~O}OBl=zo>71G%BLZUJOMh5_V4`N69PGm_oX*6
z*DtNK&pMrFr#dj#Dup|K2PfMl$&c@G3l~h5@YtoT6qq<7lmLhQ?4@?gt`janMz@ib
z=U!vC@GUzrS?2)=BQQR3Y#SK+RK%ZMX(}Y;t|{5mJLR}n50Bliv=B>E1}s3hsEzY$
z5sVGg&NNl%<>$E+hyf-uvqVo%OGjJmIuFVgTS%C62M^!(M;E5+g^*>Zq;bOSJ3K5?
zJ!2^!#qL0q5X_4i2tTi5^k{b;kaxnA)!U}V9Qa?<2p;?raP^pY^lQa#mX4nDKGw*f
zLP<fPf}&GxPvGev9=_UQ#}Si);NXixMNXKvMmsG^Xqf=Rw~cy_kwB@Zy<PgQ>SUxG
z7Fj?bs|ic!lqZoJ5MckYrWy?-XUca<=aS3OdwYp@PN*>6S+3V9pZpf!M;jrY7WGaM
zxjTf+l9CTqHhX)Rx6)i-HyWm=Yw`Hbs-N?;A9LDm-1-N*7=d|V5qR?wjK1U&>G_>K
zi}3w)OHo=3YJb;zSoKeE9vE%p%=bgKklWH<Kt}=oNj-}J{(``H`M}+{85`}{9?ECE
zeXG;>2=;MHwOyWEXA&9MkD?bjjhD|z)Nq|dI$srApoi{r|5@7}*6Qyj&&Z4fXZBDE
z3U=L5Z!GD7foZxtmB_N-72K3?M^KLX?=pDdJNZS%<!wIJ=LpF7-Ic30*`GGmJ1yzK
zFUDP5wdH10pUTXuNh)^*(VrilwW)Uc>O11CE&If!y~A^LUBQj{oMC)aN3B(lhyG_p
zRErdBl+5Tm4~Hv4f!)b=6M6y;m_T;kBc7~^yJeH<s?UaP*g%LGy}BIf*L?J5a~oE+
zW~D89eXUQ$C;d^B!p`S0=?|VJQO5b8?DF)-jOMhpwcHPW2u=#`LnHGCenqR&`o%%y
zS?PPlfS+979(k}-lOTX5TOFI+4Ied*9fBGMUuYY%-AbFSh9+bo;gWd&JeTt=W<SCg
z(0R7rDK)*_TB`v$&?zeHdu|de{az1jtHy?x9fiN`d0C7i2bRyB%|*MMr8H_?X?3k`
z5>C~ui|_K!NYwjeq+JY_ICbJ!AJMYx610zZ0B!z4K=rWYT;S+>9+6N=dQDw__cX{;
zOY7qHG-h+W(lW_&L0W>Gn=KeY%pEc_!)BsWWxl>42=+)J83krSbVw*Mv}NCiJ-MIj
zzLmqH!j?sf!+I46yf?2LWy}kr;_!}hq*vMx4VXBF_x%X}%SM|qQFZ=)tB(KeA0{m-
zX51-yIy%>{GODYus3a|Ajca9ssWmr?AG65B^n19wJpfoYKeg&Qic<$<m>Q6k@|1zy
zF&Y16v}L|h!P601D>_V_2N941t~jaezHPke^g52IH#r|K1<u92yQnU_L$utnoRW?-
zu-Ep`;p`kpHd``PwRJU$+GF~9MTPY<P+9@4r_1X@PZa}yh0(G>?a;FnL6>#`qQd;n
zu+I$bwUfR5s=gZqC4<<w_(62<YFl>JvYugprpm6KvlN;a5x>7rLv0BUMKaKDGjU?%
z>0$t>_f3JJ1Cp@0JN!Ij#_XvxIK~W`o|aVpg0BruHbwn)7pu-20bfsC2D?RlDpb6<
zFSt1R)`k-HVqy>2ekgSy_$sUgw1`gJ<mKKP`wP=fhC6PzAW%7f_4Y8a{?@ZJl09dK
z0q@zRwmjb(eQs+%uw&jnwPwF?eZ`iZdNWPp;Cpsv$Mt<eJ8c9-@%arBhSoE3(Di`D
zKv+<4U}tM2zApP5b&E`r`Cn9e(+Wfxn*|^_gsW)<`9&H2*sh)LZ`UfL`M}3mM4g>o
zbu#OL(9kP0J5aF|YekX6Tl~aYOB5M_VQr0d@7cFNpEZ|fl9J_KI;tv+@Tqz|THbvl
zBa^Q{srIqkCD1X%ey!Y}b2!n}K1<lO<fg9}{KVQrG#hUF<qw+BxB8dL_eU?|ZQth6
zpnBB>-Q4i#Tdf<JHYff;SbKszW==SfUz5GgzQa|@%wk}dGB3S1S|_xN-_zF?t41le
z2-r$|X4BR_n&o<7+6kZYd0?;)*XY%`A>_XCg&WPkrK5%KJy1kEjZHp(m8QwkO3P)-
zR@aeGcET`iC%>4Q9%`A#A*ClMiK2}{cyOuMaI+*ujD+~RH{&=brx({2<(H$buRi3t
ztu}p38#5=ag7Q<K@u`ht$3#U$Cj@@XYa1WHFt=wRLzm8+Q6_K9Umtn1s<Pfz9DPg!
ztt|}Zb2ei=hV7~cqoYIDl&(G1bXO&1NTMO(BWoQUGVHP<<DA(YZ2Q(&!WlgJ_!(}6
z6v0&5wgSLfv3nNS)z#bD*Y-q>N3$W>Ga}zzXrhKTvR8tZM(F)8**44!Z~cBS$!kIA
z-;4b|qz?M!tTRBGxICJQW%g9jYV-Mi@k*)#_SY}$$==1?9R1>%79T2zOn_&T1D_*A
zNJa5OTI$QMM(DX2d5X{}yCYU_BCZYqbpR%7I*kK^rY5LKakAG9OilwMttPixD~vOQ
zHnsj7ru@)6nDH%p=40`$zLglFx~W%om7+DzUWlQHI_<VZwEAIgtZNqC?Z66ti<7Bj
zJ&AO#A6Sqd?=&zr%{#S6Q&N(or4gs1s;|ITB*AMZ?~3Zt44AE}nVx!ugEjMMLR<;?
z{hwZFhw<HBeDG4de&R*Gt%sIb<JCTp4Zlf?>I=Q1si;+e{)L9-#5FX`d(KJ$6-C2y
z@nzj$Qjx-U(rQDr4u}1YdWBZ=hR!`z&&0z^XPt~~E?2r2p<h7c!S)i#*<;PXpr1;>
z8aNBB3j=HLh0{w6%>VNCY-M|i%(t^BeADGZ=!BBuPZqNfJ5BCx?JBiG-!;`g*2KaB
z)*s%+st^jLSkUt?pe7Wr%#tTIUivty5Fqb{j+gHbSIRVW@Oj@NYbSe#<&k)`06jv1
zhGuhdrAv_U1^Dq2X66<3Hbb48zs3+gcE3}x&<W_m)>ST0MetoH_99iaJ_3%$oEMpj
z<K@_)rmT9jc7$E-Wx<eQeUShSkYKGU&CGFUxng(D@kyeW(3rZZx91QVZ0stKYa+{$
zcMqPf<5JgDNK>|ZFX6sBdo``=DC>q|!if8FV;%9&x9T?0kPK5@O`<V{xmwn4&SBTf
zwpCo9H~Lov+6NDIp6`S_;f#*8Ge0fmS2jC^U7KEgjW3!2v7XurD-kadB-|0^6JnIR
zdwg&US*^k?YsJPGt8e`!hVZ<HUxgorlDgtYO%!)z7q=ncV*-;XvB~QP60O?u{RS4@
zTs?LLMBBJYi2j!8cQ<F`@1KgWY;lHlK)N<BuDKiQUW#gYJp#`2$(c<?l^+54u5AtX
zVAtovV{gsg<LgkMgk}Oe7jRNhWMl`c;cA)3|Jn^tgmm?hiVg&WHix5vJLL|{)yYy4
z`5*KHgk{Z(qmMRXwXF6+PBXvSMkJB0moYC+f+)_qwNW^u5_>8MN`QeZ>V#fs-v^Y&
z59iy~dAS9d5rJ6N4lneJV<?S$@9*DGjhlT`9Jx62cr#<wanMDXVt?2n!=H3qJ%AmK
z<+fw%RPujWfYU#M6w9Es#Eh_Pt}JCvxL)$TfA$rg+#hBJ-lJ02cjH09aOV$forl=P
z_T2Z_xk5ItN_OYAEZmslztb`RME*$;3&FHIrJ3PZ7f)F0cRUlww5Nm<akm^5_dY*4
z8rpWT>*)d35B^tR)lJ|ZpKOFZM7dOAiVh%EZ>68(H$v3W>Cgh7e3o3`hZqnOu>V|d
zx(@)YOsls$D=L|m!5eKtD<h<$z_+rhv}I}_`ec^(uGTK;zM^O$^o8>b5tf{&1a-8L
zQTfr3*InnyMh9>=qkTk1vf{i5JhE_b%Xlb=fz*QRRN(Eo=)Nni&Z)$I@e^ct9s((K
zVjqB8T3llZ9tiPtuxlx1s25oHJ|IW6zB9gk!+&oGGo;U-f~XJWcp`wEe93mg+}rv-
zpl6_YdeUF5*2zdTA=*WKt_Gj8q}abAY(m{_53{%BJ$SI}y7%skqWn8tf~fxo8Y9&o
zxC4-C+Ipwrglz+We5hxj-#0*QaA`mL6ySpSuvw&jWXLk)!p`kXnSTqaL4WP;@|w%g
zP<np>*BJ!`yrGuUKKkCX<fo+1ce(5HrIiBepwsgP(i*h45#k%0^FgP}V#R=m1K_R!
z6v04M*!WbFGL0d~t0C0eu8{S(6}+$4lvH4n8qj(Cxm(HJp(9$Q3lFGC#)Vp0U@g6p
z8(6vJpED<O7E%&80cu%cUsAt`H_3xec(zuTxwf)J2q1yn`v}P?J5pbz?km`m=lzNX
zkgChA_P(VgK;f~&bG1t*1~PUEOba?s$ITOi)vRs!>4~sNh38}0HOzy;zCHTE?ffZA
z^%y>^#Cr2gNLuHSfYx)d#~#N!N);XSE?>A2ND{fFgnZ6q796ORs9eq`O%albSn}c3
zoqib2@Rsi&-0>846_%NXL&yVyjLoANV)|D4z$%xaXXW-~Lc@tr)9O1G`u3CW@aaQ%
z2@I^vX0)KOm=`|#SEaI?B`eeBj%7-hC1}yMRyU9L$R`9H0lV)2|AV4MX*OU1q!48=
zszb9q;Sb>=HfQz*zS-);sQv^oIfR4hl>iz0#s9XsRFntl<*9+NFt>6;ytV0<wlFaY
zrY@=TUy4KZ(RXGOvN!y8Pg5lhq&Mc2j1zwLNgH{d0aG^~JHr<+{7$!VTB4=h?le5C
z5?1D=Wr2yZ?H#V9!(LraRB%jLvaI68$xUSPmr^vearK6{<WW&iuC6}3E&efZq2{g8
z;X}Z&Inw9zW0cZNMtrWnr_29zS@c7Lr4+cABJONBhdbhP2s@mYxmW;!rE>ROO>Xrg
zu_xZ2w3oq(7bnM5ZDP_j-U%HP2z;XM@%|@JM7a(2apIy{N8Mi_9-N(+a&vTRo*e~(
z@^jYKcHF&rjE<iJtf<U>o-o=dpnjy6dGC_5t_9<NOZdS@Ez%Awcw8Szm3>=fGd8(>
zzP5@)t%Cl3*ZbkPijHaZ*s{-S&%ysFRqnhm%Pbe_JTc!ego|jmHuwAjZicD+jC)pf
zK_Y4(f2uTtF`#fBOsUQdW`fQRPAg!&S8GCw|4yPd&eRGU-z*Lg*_M`ME9U8g`~*)<
zP<wO1o=P<k19#$?qnqdZ^OW;#O3Jdli3Xwn>SS+crT!uDS*zGP{7{MFTYkF_FoKKP
zo0Cd<R}feo`=i;bwh^RVZAag3xLW^uHY^XO$!d}a$cPsa)6=gL-nZyJrDyT^I205&
z4F^PIn5a?%+Dn(WV!yZji=qL={v*I{OU;KR#iE4aeVI3Kk$dqZdUrk$nEnJE!MP>R
zG;jgqlM_Z685sUHs-QOMaa2BVv07B`uUlsHB{FivxF4Az+ZfK}i?8m+A4%y<@?^0=
zgT;*ASl-g3h`5iK`DqrU|3Zf6A|t=UP8-;V%m<+|k;<wgb=sQq$xJH7H(+hcL&rg0
z4<rs-w*BPBXfN>-bOrAn32Nda0zbWf2TlD7l-DH75XQ^Lq=<-To#Y?fmYqPz)v*Kd
zb@!_DLakp@;p`v40XEc9#x?SnFKsj8#&U2ad_+!k>cV<8rW`{ufI4<@;;O@G^BOTO
zD0sgi*pttTteT{8zTPLZzyc7|1Ecd=siN6|_Ljt8EYWvS;x69(+(*C6?-iB+zF<B^
zJ1DQzKaEX$1Gi1KHYSTGG3xs;OHVI8*uSj25{2Ba^lw)0WqddY>%D0Et^&;i;hXi%
z0V;2jhAWpUH8l(ApqjXFNf(O(cr0Kja3cQiNqMN&ryJ=LoS>AB^tJ|XcT-ykV5PG-
zx!Ak}3IVHP-p^INFoz3EB6HFrMXI99crTy_2Fm>$`dd};P?kX!H_kfdv!PM8eGqv<
z<wNWPruh>1>FE5#ScrHe=c5H~cfgC|Hpo(>UiVV}l~cx&*G_(-QT8q8oS1@!C#U1t
zD#-Hy|MEnq>>J2ae1fSSpTvgMYpN8cGBh<@VH(sr3l}OMT@)e)sW(L#1az}u7SmM3
zqnYCImRi2ZV7JIfP5(Y{=S6sNgOzj(9W!g3Ek;E>m>asAd_!;y@CG+XGi?{mr>g5{
zWXJ$aVoky)lCv;jOU@Gmf-AhkY4lmx*f-UlIRl5hNlBCUmtr9OW(R^4mA$HE029KX
z7WTb8mzRo99`(YDca5I=GUE{}xcfQH>o_K{gH_!$=pA>+=MEw`;^!*cuA7TM{#_Zb
z7Bzf^1i;0R=eQeKUEhHk=iP8j(J*7WfUxfsLHn9+t{e$5FH-LMG=0zUjLp%(>5HnM
z`H_^QdZoKH%oboog0CVV`)8=^-Y`E6E~c#5nFE1evq>Q|?%BUJ_2cNvgaXtTZ6K42
z5KAyATSSIof=Q>OS+xj?mPdmrEv>G5mS`Wy1Ojf7U1b_j5h4@g5Qbonh{zci*>W~&
z95*E8EaC5(Av!eBnP7{yK*`Yb73`%pn|-}$9<D_Yk?>KJ(ARtQ5NWB{+}R&YV-#pk
z55dosjQW-I&Y0=9q-%kvW<v<d=ikou^#!n}!6EOcX$f9*PMg2cjEl3iUuDkrI-%B-
zOVShh^qTg4`T+=2O~CPKNlBTddaD!TWF)dEB+Sgdw7p>D?cHfcBw#J-51d~uYho`o
z>F36vBzy3s0APpkO#U@7lhhI60;kfID;eICskjzIIB-75r)|8jUZhRQ#~7Nt*Y8sA
zsd?p6eRwk8nkX*$cbMq<Lv**v{>s;~8ayy3z-9iV1{+}maVuba?4Rv?1br3Yl_@8R
zpY7Z|ln5tgGtnF4O-!Hf5@w|(R(CUUT-G<Jr@5z>txN+xJ<by$*Ba;%RZ(GRog4mk
zo8t8e&AG}gg#-9Ua7LFvaPK1MOm|D`Qm<}2s2@}&fT2o0Fq1W8Ms^igNYY{>I9UB1
zERE4kZ3uO4=&YFt5!1>}&5YCo_U<%fdv0(xWLMqs^4W}>-@#R+1RFmi5SalY15Sdp
z*>2G}=fmVi{pjR+jV~(UdMT{bs{Zfxzp#GwpxlTTx&?du1}}DTcEkhyk})lYOS_kL
zWV}fB#o?L7EGK{b86|1P7WP7hj5#yMUAGzm$MJHAHkpwToXYw$RmAYXOkh!Rywtwa
zsT5pCPLd+0*q?Z>znG1wuuh1c%p_!UVyHLqSCBp+?cB>?YZKts%Sc1;kE6Xhr=SxO
zG<5ItiZ?>|lPoM^LEJ}|pAF>)a{ZB49q3zYQ??nddvzpJj%S3fBJGKd<`EoHN=jEw
zCY_9o4PVQmgICMB6dZT+i?V<`bDbjxOv}emMqNXoe6Ei&@0pCY3`8q<&GXLGAX-J^
z#5YaNi~#h}RsEo86thn>{F(;J8MNuXr*hO-z1R+YA|gDXujd^z?Ye4^^84lm7@oCE
z%JgF9#ja;Pzai#50kc2w7`|+8zI?yD6u?MTk8NON(Kksf^Z*2w-OmISLm^}5qp<by
z9NPsnq9h^^SX|wy*zUsqh3^<QsM{oz^|2d4cON;*s5$}YW!`emz@7%SLBfgULZQMj
z&cCemiKqXAa6_$31B^ZlqinYZd`}*0f_|E&dd%Glb|t$~=u!bbTS#yT6PRp(AwfOe
zPI}|+c=SXkiT9gbO``v#OCbdXfN&qbR#DQ=&TbVHxm874$S4lY)Y^91yOZSA)SOx^
zZ*TQ95Aq|weMcJMU(E?_A3St+&iPT~oU}^5)!Wjx0{^u$o37uYukSA-)d?0c2LF8Z
zG+TiYiy$OqjnmWO&f75eiOi9=e-&jE4fbX`d1g%HXS2`#P#?UrY&MR$aG+oT3qTw;
z{S?OEakeoqNj#^Jr~HYW3==Qqt72Xzxb0!rqyALZf1w|kfg)wbr6y<6N529k0AI;1
ze{Ia889z0%J^OP{NUA%pvFTK|0WJPZRaG!Hm#N{GmDLILG_P&0Qqx(^s`Ie;2qm%1
z|A$HlUhFrW_;|xc>>eIEEc4qWfBv&{+E8PhhN-KFN}^syCo|iJhd_{zVt^jvx|RNy
zPcf?W8zcF83H5WZ^7Jd0_@!CX0f|#CGb%Y_N>KCh`9f<aH6Q>eA|gKKVa-p`TrZ3^
zH>QeO0urJaA3FgQa2#xO0cT*WfU19*Uo@j$iGwo<QJ+1wjh9rKU&vOOs3}ihTXPy;
zj%9DRSLO>z9t`{MIkQHCrhJ|R1Xs*}pF!b@W2w&QI*Y3Gs%T~LwDga(b}Sx=(=s&A
zDc*mIk%Uo{mA^)zg(6crCxlC>aD8{6Cvap_7L^Fd9?Z8tv0nGso{eIyE!w&s$V*;*
z_w&lO1m>lauyn`JkRfNeoQcbEr<3c2b5A>{@DMnEr&d_9e&Y%(tnj#$P@g<%AOoG4
z;{G{c5SrA0-!A_*PdQ}s58hj&pa3@>yh;ueNiEBmzx|n+`Dk%7)05nk_QeY)-^67t
zGj4F#fm^6OfNnmLB4{t18?H^Bd0ANj+&hq!4bV;)d%CZgxFk0MJwb8Ygn$P4L4<+q
z?Bh><2o2ZN;k{rxHuv;Nn;*#!>GtNYZ-XO_PH!|#^W_u-0mCBE(y}}n{qp{Q+I!2e
zD!1-kd|`{kHUI&Et)Mh0siXpeba#nJcf+DVK|w(lC5Ut{x;q3xN~A$jnnicR8H@eC
z`~Um+zs{F)J{&H&;F4!O&wS<_;~p``9JdXdsawCCks{~#Gv*PhC8q}8L`Za5Z#}-I
zntlHPv$R>emFo_+TA)gv+S%;jR*ZQgA@#q8Q98pUKmb?hymBN1gw`6X0=;#qnLCA-
z@PCo9CrUmXAx_K1xCAYcjih^!wT-o+EVDI3<P}<@ZY{Tj9sFQmo)S^Og+;cv>Z!PD
zw6{ifqxUMD7|5A-ZoP$DeK<r!0cB^e)vS&1UKc!wNqAT#`4<ALy<CRE!f0nd`6DR6
zf@E4^v5q4np;MK13{+L!<6+fZwSlTMTW~#yEiIusv8OW5Gc&P(?E&-2yQnyiXMmwl
z6T9#B)QX_<^NZ;i^*4MxA071l*%{+GFWuO)Ty*4Vci-tZME?7lVE?g*?`+h>fq;EL
z85NT#9rvYR=?R!rNUZR2nZN7*OlhHs#e{`vC4%p?3!E~L3=s4eeQEr+J^(m!yDN;6
zg6)decUob&#i3@}v`)ngh4mkGC#O*~DPL3WCsjJ#hn0cjN%MXFu&+MVGl%%h@B9*4
zYa&J{?%W=H^a-l-RNja!^ohZ4TAC3HqF9&uy~wwv%(;2rP^0ZHOluy>!|JE?t_cvZ
zoaENj@37y6D&0a?%;O+Fw)Z4H=NA5q;R$MzpZiJ?1vQ!m)R$}=R+3-~MOx(<j_;Ur
zAmrc$=7s-1J2!Cl*7L(P6gXo#;L@`h?LG^`_@ywgOG>ynj_W}f%uAM}fte)x@!D-|
z&>Rsv=K5~w(t}VU1|r!Nfu|-ZK53#Kvx;vu9&gQ0c*9amIlsac)WNf|K7`7wqdR2Y
z<?pY8p@4djy*1>iXCb6Q-K;m5{zR$cTh>nOky=jAUY}oVfK7nimww;Gf(3EwDw}2P
z_-S+Nm*=Q>WWZ5|DyS~Tc#CWsR_WXRNyVJtg}$6B{(?_A*U30A1o<8kO-<NfNUdK!
zZGp|>V8~bGeO7}6hplP<El9h~o5LnWQ^9rLJ=2#|3`7#Co?E0)Vzk=x;onTz$tV<;
zI=?Jo_8`QDCvYsos`D;GSSzwvl95GkX?$3mg`ugTE2J~lP0HwKTA=##U?uJQx6h&M
z+qIfoB@+&k6wdt(2}_fiKR*+Pf_u@U)rQ6|>#&DIe$;sl5X8Ndp6)}I@<TBo_3#cT
z;QmT^nw16Lj;_3%mXAhus#L<uqjBE%KiYFtc{$fyeq5{(aJqU`PD4In7i03QG@keb
zYh~XO*)|<je}!^|ZPPRbw7UJhHwJnkkJZ%Ik2gE#Ka_HC8VoVA-V1s0x$^LD-lQoE
ze>2o%Ih%Ig1qU1A`|EZ$Ly@J&+=IviJWZLeO+CK@YH2?-{w(rnX2pqJ9`RPKuuse@
z3M28{n;DZ)`mdn;5fsAK5j6D-z3&Jl`8!ksTW|Et(|l6P6q&saVFUzk@rRkUS2})t
z>uwLbCg|a-lzGS%0f)>0@+(SATfT2eRpjb)`k}ey7SuT1;FJc=;>m1h`eNDO?*fB$
zSG6x+7TvR?vo#l0m|SO`BW9NbepK=Mzvu}mV+m_>IemSFn#()|LX^@}*?rlF+s;;f
z8i<PWa+HRaRILN~{;V6_Ou%nXE!cc~iHXpzayr6$aET3yN6JHDzBTF7t_&2N+6g$D
zb3r}~TQ(?Q#=Wtba%Gyw15ptX+E!NUw6+knsrET~ZTFiKruesqQYi}@xx5M}^}uSZ
z1!CGynkqVGZgA``my+jaFLPyv9E18s?;XF3mlqs9oGL+fWmhE!=jB3t4O9nml?6Yq
zuKZ)6-%&4_a3}bkIfnk<SmS8!MK9)34f!cmnhN?V0m1C{fUXfx{SYE~kcung+wp1E
zZRKX%x3>;7!&H<I*rYM_QPp~BaXsw9x1-p;v9S$Zr;r418zGRx5_5P$Zq9^>vu8$s
z+U(^X&tLg<9dTrCaeH}H7;^G4?b0o*xf7z6lyDNo>VCAZ#$KVxpUSc)gWT)#+#h`-
zg`!TnjHZc5^k2QK`)hR;Z*8ltq4kDvR9J7~jUd{TJlY3ARU5c?_^;h|N7(tExXt^<
z9=JzFhe~de7TQ1!$?qz${f(vPV}c18`b%&a)L4)TG;hM(5?k6R^z1IAl^>%%eGp3d
z*3s?raY_uOqtN<Bn)h~J_BAN9*o`M#)RhF79BbutEAgO4K#B2(_1y$Xb`tEF*^De2
zmfBD5`B!1vRut;Yr!e1U->%-6Sni9@=b{v>zp2ZT0$b46sD=Fwox3`Futkc!rZ4Cj
zW}!^c_0=SOMpolU;~S#f>h?D}{9jJ$sBHDR9$nx+yyyKhD8|K}jI%Tk9XH;P<ZL*u
zH~ZmslPs&iK(d{=sYD7YAoEm^^fuYc*T!33Ml^gca8qu$JinM~nEGPN<J=oh-+~R>
zUS)Eh=%^_EiJ>cm8Yb$betv$p`xUrk_MAT9lV)sgc6TdCNJ`@56%{oO4(@qKp>+ff
zm)L=Ltr?h1DYuj^JS!6u@hSBd@!1`ti)?c___Dgc`Qy<o9u<uG=ja0KemtGn$2zO|
z1aJM;R|5KA-yo^~wk$2}n9FN2C^~VQ=f2xU&osHYO}=l^oTDkeQQ4BLkh-(jkT}G?
zZsSni$?o<zYsB*tZLLNt6hQ46S_>lX722vfPvTW3O_?&6ztlHx57g$_aGQGyM~*<E
zO?H;qlx|#oqRG{(ajI@i7HKKExzgGPY4nW#+fa)+D;wKrr!S^&nx(tJe4p2!aNxKl
z^Z0TI<1G#S5BT`|<DZ^Y@!Ij^I65LfZp!Yq#F0bjz<JHpX)1fA9=RK;tRcF&8aXvP
z@${%7T$-!a^%>yGFzmz1N&oZ9swei8q|e&Jo2Qp=P~tui70q&9CW?Rrh9`wUYn)hc
zNc2}>8z^0mh}bKuSnHh+!A1vN`uIz?f7AteHzud#uO2;r)zc&1_Ot$HYrw3G4Bpb9
zcaTTT3!2jpo@GOt>Kj8b&FUPk55&GTJn>w<1g<_^m<0II*C%GagBiltSU$446&j8E
zls59z(?PJVD^~iZC{*D$!QmRnNQ9_p`9vk)Kyqfw-b_?B*zoq4%B4oN2T31&DIl<d
z8X$K%uCYVz#p%`NmEbX2X6{J-ITmX;_+1pk2jza&#idEIm57Q`lhJvqS0i_p_7}d-
z?YXVoyi;{cV!q=xs*r<x-R1F<E;v~8>?_YJ`A%fvsY+j`{8<6Pl(AtzW%zyZp&=zD
zl~+&`f2-%_e*>dES!}&Q!mGc-+3bEH3<=1rJDWEg9A=^(CCMeFW)<O4>reT66z8y{
zHi>?iztR4V<sdinkh`sQAZ*vVx1be!u0|avdvP=}DQY2G;JMH42^<NYTGta^*?$c6
z-yc?od?7(IqEWp6FuTV2R{<_A&T^-`@RO)0l#-eVxEL!01Rqyq|Fs&m60oR9M)aqf
zq*StSBH#A;eUOZzEo(TUaPUl4`Y+6D<?v4oS)aHoa)n%GDBN+sc|sMrZZAz|NNTXZ
z$$3dQ_K{nC7!;H#Gx+#qM4T<-f{+l+EgNsS<ZB8S&NBrBaF9R0An)wnr>r_yoaK;x
zZ6_(7r7ADr9l5+p$`#xj&rPpL-#Yh9LrCb_n=L8BmBY)15ueW`k4nuRKB1laz+QX!
z#mKTrZ?@$tmUhF7HgV#+%W>JJQT6(e=a$7c;u+#wr*c-e{gB0G=oaCe0Sezs$gS&n
z>BG@nJg#ph#<mp0qnpFR>@h_&8RE{%eEXDg+R>amhcV-!9uns@IM8c+E^lQh1VU4O
z#sA%R4KFLx8QYd)BHVnA>YP=0Y*dD3tgHO_la0?c8nw^-7P=jnHRnAm9DCy`oVq$J
zG#615e_IR|YliB2N{+6$KOIKxs8f*B^-)zihS0blFM3OK6fV!(q!<{WjVeoYYAG43
zI_p@JH&44T&@ok2E!|t2SM?bK_L*yzcj^Nb?o){O{i2PN?+#Joyq2A2bg<K%zGHD~
zLi6Z~cBF}^#m061N<&}RRLerA@g?qix++oUU@YWq<LxO}%@S1P<$t51GmBfqdP~(q
zaXU75X0uVrA?}At27GnYTdUpT_&&bHlLNjId@d3e4Gjz&ox@*B#ZjU=)tL-a*mfRX
zs*_iA!mJq>B^%;SDp5hnn5KTuUlEm718-!OCBrIFui5dmN2@YaMm+*!0~eC`#Jnv%
zT`IViWs3fOBFEM6aI0}S>ArZdTdAna+n<54`q~@Uj}PZW-un0VR15nnIf%sfZ|-jP
z=!%UpM4Dh&(H@_AI9QQEr%OX=eU}N*%1U8Pc<(a&!mv{^p-ty3|Ly#06`H?=z9ON|
z;9x72oXE5^&WT*OTff@X_@E9QzG^*RGN-HJsH3y(f05-I#lVrgYKcV>vq`i4cz`mA
zvdl4(M>wf=Pa`#%WVIsnW>&o!zkQGOb8d^o@XG`XEoH?Op97p0yQU7}48|o;I)~#e
z8=EtMBG>`Qso$!id!W~;^S$XfUed>RtNz#GwEk_9;$vU_3R5+E$&?tr45u13Eopl1
zfc5;Lts!<<hD<$S_YJ#5%LDU#nd#|Rr%HHC#<*A<##tul!MOET_}0cxzG#~Ex2Grb
z!C?<BK5n<>x~9Vbim3GmE1eU$V)wUESBDOt&T!?4Uy4&r#%ggooHx(9P+}hA>fBOP
zYCXyB49{HkJfU{-?EJL#s$H3lrsK#qUyEq(fCYM*h`urFL!ZhG8+L4UocNk@FADP?
zHXxTDJ0=c>yn7NP#4P?h_j;UrN<y4R2JE38o9xL|Z`Qf9v1|}hJ=N|x>7LctZ*4Or
zzMkNDS|GM;<eqM_I8{u0tdwtKgp0f7@(Oi{c~5JqRwKvUU$|4|cA8Oha3S4UPrzP&
zH|`sjZ)MUVECM2PJ5uv0ek(i2mOvwzJ`M?T1%>cDqud<x;Z$73bKwK2>WbN}hYtM5
z^<vS*TV~SZdFnmf<|?1!a#^kts95SH1R2ehMOqtQC(VeYmF_j}zjkpX=bdFXKKoej
z_{pknbggOeNsXZ|_8T*_9DqllWaus2Txd7q`PjtM#I73INu%v)KqzU4?Tz|WTe$vO
z>3OkWdVY3A{x2Lkzq%^tptoo_S8}o``$bmMjDz}m<e}^?=Fb-~(|RWxjRL)-8-%oG
zrwe$behH%6i)h_u0x}O})=^@+;-llEtRTEzo{tURwsOT%F(LA$n9x2OjfzJ);`Cxg
zt4e<6ehCk*{|Q*=6$c&71}*$CkT%`&55-S7b|}4PeSiG=(OX&_Lc%oSCbfqZEONs|
zA562-nm%^H(x=zS$FoNb6&&s}Jf?dg5GG%DvApVg3Q}+KLiIx%AE_QCN1d8;$IBb}
z`8UpbG%IoA>XbdvW)ZKLzljcSYR)_6N-ONP!{MZ-s~g8$@e*Ul6Ta(>qtc8OZ$$1h
z7)AP;A1Zk|#3D%OzqJ7ESrVP$C<n)&CVX-gB#W%$16r}b0jpjV{+F%IdS4zSCTyw3
zBLLSbU0`|GWKX%8GtZ1<=&{LvDAE|bd`K)kjYrhrK@u-`Q}2i+_#82uZ0-G#3xGb}
zZGuY5e8CFy_NGmdf!o2duN*d{QV)@WU!>o%L@M<36dPEf3_XfPv^9!*8Bgs>FF7ol
z>b*J}FT(jAyxx$zyMuP{%7I&lhI_eDT$7`*W`YZfy|8H3I)JPnohA@jnmZX*tcx^E
zZK~nWMweKI-d!kh?5fCiC*QUpPVw{hwyn^T_}SjwdhE~P)ZF>IU{zIncTW*H8#KJ#
zOQHWH`af?%@}V#;{<Et;K?n_htz**0S0YWp-F+(V^j|aG$>f(@M@f4MZ8z;Zl4Onh
z`YkH>5)-&&?x2JHbaVf7F%wbVrV~bi3bbMTO>GB)G<fxjPB`uJe2$N{j*c!O?fdVb
zsm{7HArO35F0uxY0u^A+@|g6YtkJwT5!yg!dgt`SAb9q;NK)_aU8Mb%MIS1k>&sD-
z=I`}Ik8O*75gVtG)Sn^OXM(tCK2}#{Pp*_H>*1Ur&ALdUYcZA4SN|b4L<Eb~2i`cb
z!ZeuPAx;f;T<I1b&x&Z6sfb@DKC05(b*onJ(=#5^_f%GU@q&{!TU*6qG_})XPrgHF
zaiQmnA6+kG0-#%Pz^`U%%TF}?8=X&D+dHyzKL4{O>B4~0LOt)S-S>~!p(|eBk_9i(
zW>=bq`<iu>*yahquM>+-LS>`}S}0PIow7=oduvHPDay=uPC(;tab(7+GA<!r)%^15
zU}Cvwsa4VxDJjxwy*o4Ltp6Fpi|1WvR)W&#u!vd12X4-T?<Nw1g{s_*9tp(T!VVn<
zk(lm-?djF2q~0v^Y-xeSjtsH;;vX>s1OCA$&wX9B^lT2IBW)*>`;U(TqrnVWEGIJZ
zb?PD=RF`u250?C?w&&9|(GTzb^EMceu&{*!NtUjI((zwuHiGlpox~<)-2xp$HHD33
zhS#r0b4>H_yKTPcL)V3%9qOHiOC(W#MHNz94e4uqU++=(@)gCFEJ=5_ut(t%W~(Y+
zKWmePkh?6urt+z0A*|s~9*^kTD4Ds7!PU&Eah8D#Y131*Mf5ho-Q7K4&gdBUw3b;d
za(;T2Ni|ueq;B4u&1MdWh<&KdqReqzUsNdeNSE@uP6m@-^3ET8hdrrO0t)`jOZCsf
zqX|ix?T?-A-iS1jpxkp4?R(Tw*m+sjO<TCnChF8(`=R!O&daPuCyqMjpx0Kw=@SwM
zZ<n8D=X0-pqUo_;*Giw#n%h6}Tv@Fx*}ljUi5^@e_NwsrDfO|nvkMOi(R28KyV(i6
z3KnE<`+GjBI<nfh1AVEoNP(8Y364OljRvas@4DLVCuEmmzfv>vYg>Hb%Gh#Czb!`L
zqLZ|-!MKKeU(vE)zww6xm3&Tz#G2sbs}~Is53X{$5L!Q;Y(}?n?ldMQ##G979TyjO
zx}q*L?fjb5ttz5hz_fD-pG=J+>6;b~j*3`1zm($s{ys3HgmyHi2`P$TJAZZCZ8A{I
ziToUTl6ddlbmwY@C=t({m5ruO%gBx<7rJV7`Ic3qQq7nPDz@cXD|0@<!NG3FM@%vo
z7|T6)@3oDO>t;5hq}JTv#@$w84eK8G*hg${hXfN+<t`S~ZMvOwwzVnhj16_#!fO`{
zB<_B4tkf1DI?B3M$G*lYcX7y^YySYbvSO#8RUTE#-2HOUd?RDRV_$fx36=b{SyqKD
zaBK5@C>x$IGt2udYf7aLJ%;JVUkSeG<(*b*{jr+C?>x`y^mBj|pF9_iZ_Y1fb5VW4
zBQ{wx_0ubhBgE46;3ZAQ-Jc+J`?<M)I^rwr=H_OqtXt>J_;j@Sbml41?cQuOUc;{f
z_rsa2#^0h>GsME9!?PKOOXNYdi5~nzkMLAmr2d+cx8ygnbk8{%iH_F(W?m;%=U7#4
zninY1?b?dWf2br8@dQZ|7c+m@-CO~!KW^YjF*MNK+L1)<EVfl2_R>@(tdy_&{SP(u
zAxXdI&I{efFY6{H4LzxF=t2eZ{P~?qZ4zDf`Yp{u_Tj1Ovy+oPqBG`Y1i}GOPwuUu
zfO9Rf)GF;wk$#1#PF8<fas+P;0IXo?NwS%sdUDn||5!?H74F?dO<|LNbZKJ2w9|yb
zo9FL3%YlJY=Vv!{U9@#`hU2}dJX7mEjT7XTlN^)_9|YiDRjKL85GIW}a6G;1zSqP0
z8{B@bhF@l<yy~7cCU5!(|G6K6@(qmr)%Lmb#cNTtoU6miNXs!d_rlZonjz1;_H|V!
zO!`9TszRruh1tXV{{f0~xH|D<gLuL>;cXwCe^q4L#>G6ACtrJqFF7eqre(ZS_9Ho$
zQrylm$5AG<^BL9f(^I1${pZ7!@8Ef&IF6NmJ1ZFp)vg*85Z++F9R9kP8@WO1mehCM
z6hE>p<!H-(Xt6IexVE;>=H*=@m57ba4H|Kr?Zvj{&dy1ZJcU0|?5O1UUvhz=brLN8
z!uzD8C>rX&p9<W)OHAt%j=Q-T9v77{;fHTfXp{5GY8F~hLr+(f66?KhlN@;(^(~<r
z__!$9m+^h_2!u|6Q|dp%Kujdi{g;HbPFVh54nF_Hf5}t{pV$A#ydWq8{v}G~yLIub
zsWsfjnGRRyKX+S`yH)l7ZhSTaYjQOzm)Dd8Pxg|!dwQ~iycu3brvkpttgf;GktKSn
zPUzyriwsOmVKf&nT=-!)W%Y85Nl;K1HNCoOWphKVWyY4Cl~oosy|~C&IcjNhPYn>2
z0WJVCOg6j*%;P*c9i3P$mPi_UI=Ua&HMX?O0oJf-kQg(oS?kBs){kj2`rS)L%I(tn
zt|OG#PcCM9<>ftG{#`5y6wn1cxS8)vqTS?t;T}>wVaXcXet7+KNS3wV6Qfc+_K}bM
z^w#{T)?#r`&YfP9&Yg&iy+xl6L$pJu|7^z>O$wZ+oNVohs(&NaY&-dPV~AOGn8c%b
zopN}bR?KbgAykG73<6im=P1Ryflz3>S5a0y=E_rxhp!s1q$DT*00MP#Zp$Jb0xyl1
zKVzOCSF3kv#5)B~tNidtU%|)6-ZaSlN0R>c<+6c8x(D+YLXMn<mUa=y;!w#z_I5U1
zB8aSruV1ruJ&%=9y?J`VoEE=(Cq)pZzKx9{V1<4>#B*30?7}=on0o6&>-_nyBvi6t
zM>IVXQ#)K%b@hJ;f5g8G#Q#4uMHhV@;icT4rJlq+=A1P0s3%ogak#|7jPAPE@7ck8
zeT@?HsL81*c=#hu)7}^EhdX`613Ks*KVG1xaME#rcQ{G(V?}&E_qbbOQ!tfo$9qIC
zajFS}fPl`p$6@x)(ooXUlKFE{(M;Ff6*9Kh?-COevx2VFNy@FAThmB2X<e|7b;3ie
z>PM#voF3T$-Ba%1Ht_5PeEg_vZk`3i-88zoV-h%H6K<;{0l;9N8XbyAX9n7HO4p@F
z_wU~)suTdmHu%KBMjNXGteWYMPj`v$Uzm{IQ#uFr!cODv6@UE5vx0EqiPg)j6M(;+
zRaaM+|7fkDVxyIj=&!%xzkL&;cw#Fe8Nrm#bI*%xa&9gyx8@+TeK?qcHx<jI2a~_p
zpM5D5C$F0}&Mh%HnFcgv0a)qe4tJJk78lX%Mn8>PQ1@ZbeLy6g%a~Ip^pTvqgonk8
zCy3J-$C>}j$4DW>DVBJ=BMww*3#j249y|~|QypQTY&Y#ne0#E$CTO$8f)G~#1~8fh
z;G%=pxTRcpUSx&lLu}&RATNF8c1GYEv^>P?%hlDLYWk??Pe4H<4wSYkO&2Ea8*_q$
z{>-=jBP4D#&^`T*95M4qS}~BA8AVfrf5G{B4@Wqb0@t)JQ#nH)xsACGC-_=3^R*CF
zBv|>Wvy$rR%~VE1^Kc#4?C&OsUdQ+E-?u_%a%bl<jUr<`hdT()#)o*@^IgKIBG1zk
zg=QKWu-rV3OKSS%2xrzB%`0__*=qShAjZY0(9pj4XXhQ}e?1$}vik@Z(KS5G3c4JB
z$>8<t^!WIvsNOOg{pF!Tv1=qOy*tapM0f6Fv1nHq59a9^P-dU6t909+i9{u9T2+Vd
z;(0|zahRZP+_<qd`^5^gkQS$6Z(jsXNLXlSQf}>u#&D^XDy-3{Bq!pj1Dll0b;3Hy
z<Fy7t<Z|&qYgM&knhbJh$Vgq5aP8h@f;+#23Hg)Xy}R;}lwFkFu$|6fq=S1r>-|mI
ze)GqOVqgUfYWhG*Eq<^!6|9`A!ykAPC*34X=4&@V*Y@%5I7y|z<Hu^H;~td&4)*{&
z1G{n79^9$FR%eD44DZX7a>11Rz3}!By~o7d`QF@Y1~)u_VPh!<7p*)Gzx2uD^nN-z
zauZiTA}b8rKPQAxi(yKJEwPMy;9g=MV>a6!1=Ai<VZV?FERpIlx4o3P%m@D#$Mk2D
z&#+9Sz!YyU=GI1(3Gni2ii=}cQ{nhvoB741y5TeZdcJ7*l@W$|{t})X%#Oq9@va$M
z0DBKpB9dN_bUdw1GqxG3xjGexE0v?Afcx_mi#dvpjwJ$n6X!d=@~UEq)RpCN)R?%q
zRJw<#4{)60h!|CjVPpXxKJ+r?RwV+0wG0fXK6&yaATY2;@N{1`TenuH(6B?n_Y!_q
z(6#uWDH=0`TmYa|`L&7BN+->jOHZY*;^RB$QVO^^XXsxdxI-t-<GdDshd}J(^f|f8
z<m6;JPEKXV%C9%+aCGRv=zxBO(MZP$XoIZ?jf;!JYml?H&a<pq7aesN&~9yO>w%SF
zmbq3(4zK-`G(aXXEH>8C<_3auW+Yc0<}9!GQGI>=O51bV*XJj+RkGCXij#4f-(EC3
zr!zlM?~6)?eHR8kzN+3wFg>C6J5^LM2UngtfEVh;wlrR=i`tVQns|P6eB4)LqA1?I
z(32)f5gHn*XwHXNowVmEe%*3Iyc=%BEOWjN_xbq5BfO;cN6$;kwQ|T>F4R?mAM=#H
zySp3k>w&le7&|7J^YIUeveVR+CXU9UIiX*Fwq|KWE^8V)qN)RQL0L^L9XlijhE~9Q
zCOLbLpy}ypw$s0>9G5l36+oq1nSF5L8L+D?K}t%BwQ~e_9>;?~?QXlcxM<XRRG%Ep
z2}+`J$|i9+O}ZO1KO>la)dH`;n>WTVi_wCfHDLVjf92OO+Q#%n^ShSN5b@%WBUab<
z=i($$q94f&VE$Czzk4Uw7QqaORsNmNFfiJ*9{Po)x%XDJ$0~NS?IDe37rmnOJ~Lf9
zUw7@RfV;-c2U2uSQt*ab->$K)5t-Lr!$K>lA9Kl_hVqrNDFT5!SW{|h;i=}}NSg0W
zO$-ko{sI%0WqYt?!oN4++xp`Not@O_$uXw`Mk#3GqTDh#gwEMOS&EgbFI*PgC&!0C
zl?sU}ax~aNvd7j?$a-Wu_5nwSiJMzhGLoflcic0#!Jm+kiAf3sZUCqqd#haRz+&aq
zO@0sLbloxEUL24O7C1=nvq7k}|FJd(QcsO)WHAbb!XgToHwR3wIF<+Wcae#{161Yv
z{0~pl?EipUD)$Va9tFM9M=b|}D0s7irnXBgM=>Yvpe<??*n~y;b*!%&uLYCe4SbF9
zpFiK`G#``$8xs%^pit{^yaljF4|;r=kTT39T170|2JG)2>mPrtm7Igx+ohc7pBS`;
zF-Xnr-?({m^zgwWXnHi&dF<{ywRs)xKh`pUfgkaXK&IK7E}vB7w!d+sG$kd4o*8O|
z26Fm%SF5PSHS%71Hw95>_B>jAljt+JV6PFA>`2dG9>{{tgwnmoPGgvyoLqb?zpLr)
z%BYIiStJzZd-jJlAiS+?(-ssItUkb?$)4CIM{}6m1A85nngKvc!H-npbzJ&|x-WU_
zufJZ}n8Ga)YXWEQ{RSME2KOg-@50`g1A}9w-fb*bFQBPLv1!T|%ZiLt>a(0zqj*1<
zy4736#F#V9r{qV5(K#_v)6$Ajc^>6kmd_ABfBu~Q;X^r4lOBNK+`YX$+tSer2lU|^
zO?rBIR5Hkv8KyU*VyU3*78tFuYS+c((s30>#~qi#*RNk24;2_h^EhPYRA3mMi-^#{
ztt@^QbtQ|3-XbBHvstWB**=j5W?|r5%>cGZ?BBI3?7JjctCM@uWOd~zdpL+lNYa37
zmhs*_A&}I8wLc9+<m8JwR(k-?p&$X^|BVh~X9My86g>9Ux&sNqmz`=7<G?@Js3K)J
z#3t>%eg{!QEcNse*Dx6C#nn|^gH0Co0+A%KU=<k|nf%3@H*bamQSt7Wt89ZGz9RTr
z=gyy(H1}M&S_(e6VP|Z%D-@#)eg<F06r2qt7_ee6M6WLXDTg!(*f(DVtKmwg+#R4p
z0k-2zFeC?Ho^O$nX@vM|C1u*4MyF<^NJg5@|4R7Yv^1D6j9t7o*4ixa^U&Z+8qBtI
zb`CI?jwHfW#FZZFlixL}obxLj7E{6F0$w?g&Gfvyyuq?WrspsuEf&@M#h*?J6|HCN
zP<1!0FM!`d*JN0j0Ro6hE;bv${7MiGkF^8hWdL}!_Z#cG5C~*F^Su;Kx8bo0?jW=L
zO6+2L6gwEvFA9o^jA5le07*Ds)_G}o(pV|4jsxvyp2#C)4#*q+?T7Iq^qP$v(jRX>
zN)#z~g+^herr@A}$zkM{e6GCn2rGfHXW$UFcC^`9$y#bTsu3)BQdVL<WIJ{)p93TY
zWEe%00b@7-kHlg6$Mq+72+Flsu#ChYuIDckgwbHJ4g4S`r=Q^Bwm>if;MxUdX2!NW
zEG{k==BqL?%=+;iSpUk_7=0MRL8s?Q5*Flu)Y$!vv7uoqm*q%0I=4ond+46+OlvsF
zahaKf!_s(KEK(vNApx9IhgDr56Sf#Gq37YzK-q&&4<yD;DV9_qZD7@3@VKIC>^x<`
zhrzlky`@&V*;?faivzi=`N#JoweIzBY;Sf7R^}dpi7<#Z%#VYyDi7;pr-0AzD(qSX
z_)~cdhAB<bg58RFsOPiaB}Nqm6EOGHXtMErDh)e?D|`U!QByH8X-}NjUMix7N-TzN
zl+x1B+-l1QG*_5|aOfeg)5;R16G+W9&tJ7|pcGV!D3{F+?ns{5I`8u<heI{)u5VZ7
zU=>%MP`a(v2aeV7?BtGGa`$w1OMIOJ?O3q^8{nCtAF!k8a`=70KKjw4M=Lfd%TvH>
zTy8a9Y^E9&oXtat#iR-P$7k4s0UuCFNXX|BA*I&RL2s`TOu8&sDOHGxh;H6Y0-#lx
zgC&UOwu+@$=u_2Ywj0%4SnB?(lW+4wa^YfAgi~Ts<w^y*J+zjH;*rMm7poe2yV#9E
zb1tl1LnY6&L8rwajEeT56IhjsX_TZyH8UHVg8RW%2o0=CcH{EU(9lYfZF3L>8j3+8
zql_SaGB8ZfYURUHf7OM8xkq@hY1@1!(#U`HmW`bqixaB4?#dtss^Ag1wu3Kmi-JPi
z^lEf;G)8B2yXJ635&i&rJt}x|a<Z|zi{;xGwN#ao`rtZzlU61y%Iw>lt8*RNuffb4
z1o3(tm5q6xx`Vt60r9WW%a`%IfdDuZP9t|0mXO&X8#7~RM44u%CpEpN)qn=&tEam`
zcb-hv@Od2V+0MNW3Q_@hFV|<tWDCYZ`x*`89pGoNE@1J&?rJWW@5NltV>Ij`DJY=E
z+%~1F0fA7-3BvxEz4a;FYuD_ooGe}b$r!Ng3A8aPu~u(&+m;p&nVXJQU~P1f)p(6U
zbJg0nYb(?mtZy3rd>VzFf_2V>iXAIsW8?DSOBXJ@K*h)yy&wL`9Hd!fEWg6%r<V@$
z1rx7WKE;su-?4$yjzF+P>)j3r3hJ+1twEFMy2*DBB`)>I3Nmx$RJ5`;JY?XUZ6Fk^
zO1CX7DX=|__|DNmR>{gZi%p!Fl4Da%a`Z5f_PQtQ%~XM-#H6GPHAi4K+BiC{oxpUk
zYzh!k@F-b6fUFu&zDb}UrdFRUNi23af9~8?{bj0okZCMIHKgd<W^9erij2E1E}us$
z4V+|ufFNaeX7e}}*mZw??L%SZb&dGU-YLLDjYitp+6Ih5_|CpNBP+0<+`6(k6)ecI
z3siuPWvnFG#TwI%?{A*5Ut<GLr!ki_h$s?1efn$2yyTUbhe~rdj&cUfngIv+i0&@+
zJ{M<#MnUn*X<d0$yEGc$d58dTgAs`Ejc!TSa#p9+vFbq!92^`nPSY@&1$fT^kh#oh
z0V6OMU>GBjntFN(4`gGPPFBa`%oj%-OauG%m^=P7_-{k(qsnHootbm2NqYD0T})S!
zKg~i?us}C!?U4fJG0czaYIW&I`A>eFVE}_n78-cEod22B-+*{E1>#a^v;Hi3Q`5|i
zjg1jAG}sq#T(I#KfCbixb*+RrD&q5ExT=ngZqWPpU19W!VPNCV9Aij8rpd;sxVcrO
z%0e;lchg7~?X4EtxUyh`1JBnp%mb*okBw;|>8~^?qWh7I(^j`;^CxdH(7OX(S5OT@
zLV(lg=N&LLD~U21DiPomw^^3Lzev|tg#l;p2Cq|aXbXm*=w<!6-Brk(nRLeTlkqv{
zspM#>ukii(^JfK%YN|*ia{B^SwfSyi18_P$fEwg<h}bQFn+MX-;2Hd?cJqcZd}HHu
z<jxY-mR6LOL#*Zip>gDA^z%UCg9%^4k<;UTO{)xmHry*$KIz?tJbz714akQgJa!v`
z*hR>A&CGsbGY9~?ZGX62%1A@g$S(#WRqV29D2{?1iIeSI&+x&1%hJl=JNzyi$**3$
zg3XU<5A*A=@x<p+0C(dA?=WDg<`w)36eKL^OdfMzokk4S_B{fWnoqK`%yP;_&x{!)
ztc*v>Y*xlY;l{s9EYQkm6?#Z_R3o=hG(+b1$0IIXzKn*T;0zNP`S`S8n_;&e%R!$V
zd?w--&P2SM((}j$JacsWFvOuUSYxE8r>AV|shBF&1$hDG*=2}b6aljVlXWo5R#oeM
zm5-UEpFZ^htCA{svK}1%6PX{-I<a{igU;NwGJts8pe-U?jdDs0+@@a}Gi{?{e*S)b
z5j-m+)Qy3+md~g4sB?N_af@))%mFKYhp!C`u*w$@96WHay$I{hvJ4TS(a)F=fKjwc
zb}uAl@JZQ3g#8JWA>{qZI|q(YSVTlBy!FEP=}BpSwr1M#;VyQg2jocs(7g`=G4L&(
z!iq6Fu|Vz0oZzWCTr!ZQUbyw0M0XxCZ<yc3XmF&_;DU+4TQb@m*3~_=7_Csorc2<u
zBAY2uB4Xl9P)Mww4C0df>zCZ}aOrB|?pI~-y;R<Mc_El}YqVgD-B@UY@nk^W1^b*b
zn<);++U)P`mC3#%0fT|zgT?8taX;kHZ^FZ-a3EcmjY?sM8Stw3_Arp(3;dS;9h{m;
zf%n6p%zc14fj1Kh99PG(K?NjH5lrfd<W^N(5X-@5=JeiU<+`^Fl55g2kGqb?0s2Z<
z@ZkR@>d!L#ax<U|8J<T|R0lJWx)60y_i&`_-Y7yKa9)UuJX11=PAw=X0FN0GFFYpT
zexj1EQ;X8^@r_J2VxM623Ll&0S1QlSV&Y$KyosEAPuof{H#L<M#ctFuhb?W2U-@$d
zdyU#Xu(IHK%7J=i2VtWCXC{a&p)ezyT)?di%lOV(vnPaB&F#<8uC$+RC&eZ#b!$Dy
z*bP7U5l}FAdU~Rhu%DyscqrXd#B%|D#ePjp=8ItE75nvn->$><>9_eOLN+&kjbg_q
N{#-^RN9d)`{{n3*VwL~^

literal 123258
zcmd43WmJ@3+%-G~_!mSBLXc1aDPfdu1Ox=6yAh<jI}{Nmbtq}18DbC^Iz@7%OS-!|
zhJp8t;(b5w$M?gt*6Ug<uEm;}>zwQSV*mEu<0l0<Ndi1_JO~6rAocQv5(I+d27z38
z{U0v)i|*w<J@5<HP)70v_znIYi&go8U)SwkYC1w7l!4fvOX2g|#t_H@h|~)a6}N=7
zaaXs9{j=t6{xtLWjd7)vwHo8=|Fu%R^V7DMx05HL|JV{xe*7)_KL?%v+7*Vy4kkRq
zl~OWT#;r%@4zK-C6fuh^H6J-#t?YCzUpquECmxvhj=7)woU-Ggxcp}k5QrN^TGf?{
zYe<4v1OCM|RnR-qi%S6_9O%DSGvHyjf&I6oKvezr3NrPL!R^xD>v}Q8t&5NL3rwBD
zxp)Pk`2Uv=HgUbS#c?xGjNRmyo$C!En!v_yr>9=AuO5?yRxc=iAtoYfwJDQ<{W2-O
zf$JZCw;^$7ch~soy*F&JW0W{HP8-ieVIdMUEn+yPWuvMJ3iU=;Ba@S%fsGHPc6N3g
z9LdY$J4|LPu67Dbykv@K@vp0^t8`gk*|-Oh{@7r@!F`t~sBu~$S1o^x^6!&LQJ9&T
z;c#A2aczMpi1PRkNy!WDM;26V@VT};e^xlUE-fYX1$;_%P0cvvHJivUU%riv=}0LV
z7(6hYQ?WBQH-~jn;b)gyag81uHH?mGpPilA$X#`9@90pKkoX`fD%w64tI437lasU9
z)sK&l?+DhY?ch-4<?TJcy1Kf-V3(Gc_r%!LR7NS*)8=@>bJHAivcC~_Ddno-7GIkc
zS4<AS6;GE<8Orc`>*#3wra6vj@o?vPMMhZ}H;yfi>&BTMIv}XSjHBJ~`^!7wcR$n4
z&(F7kSH3HISFG`55npeEwBGUhOH=6K!4Nv2$VtHN6JyHhEVuH-0sdb+?f+|G^#8XH
zZeQDgB(+c5wNBfW^Wz3JRBX~rDxVU!T<M(V$^LzIj9GNc{(Y2qiL~)M<P#d_6jSRr
zd+E7(+LW`BiAknhve5d*J-PU*zLkytj6+mmx=N*0{`L7Ge4bBUDxD>ti#0iKJcNPw
z?LUi{y`&~pv2pWK+3p#tu0r2$Gx@yhyF$`_cCzXcl8lV3y|tC0fkQkZA|mbGyUW<H
ze#hXJI5#_+=H~}-ZLy1MGIVf#&qfD@%9|8xO3<jY<g{XwtZbA{N>5LZ?cR+)@!KV=
zy%pafprZB$WGh#^EbI*jIWH9xPt|w<NLbA$B(9Qg<A<$EcTE@V1?o+T+upFbpE_QT
zXHl0tuvT_3tln`bCMF`!_+8G2cO5aeVP-7)ygnmYY;j?M3EUbsx4NpG>0`PD()MWf
zZhb~nq9+BR5m)^D`nvaVfUydhN1lC^V_m}TPfC|6j8}st2e~oZw&+W7myIXaHb|LD
z>yp-g%wFxR*)n2_b-P{o4%O8qe9ljMh+jBtMTNQd-FMLgBd|<qt#MXWRNM<aatrlf
z>Cc$dU3s>3$z4<6khd_mx(15DurM(2=x;?MYyDbSQcib&Nj~_K7gHq(c{jYz4jQ%g
zR%xYRZEN|`1eMGWQa}76$t~)nW!SD~qWp~%x#nWs9nxmB_xmGapirEioBkT)^zKDX
zC4T8JV;}03S9WuA{84iKOU~89Ad9jp857&OI;ay*q?21B*Xd5sOj*ZZu2`jcYjARo
z<+<nBT?&YuqR#t2_0<pW5tLi|+#OyAJ|lIhGLge>&SjwPWEvl(E5&w?NH0IiLtwjV
zWL!w@rc`pRPcm2DQBA+&704OozeW5^4^JLCOI#(^+?C#Ua|rV~ZNx_+Y_hHmOm86O
z?RfSkwzMOoVH!WAE<pk=YWZ^<yP4lU9Z76zdJ|`r9<B0hvB3ut2Arc4nS7L+>T1no
z7>;RSy2v{a1(l0bRDn2VhAE4P)RG^#Ne@m*%XOIftp69wQX9v=?l?bhLh2|A@ngSe
zCfSg|C$SHu-~(r@S*9BPhD4*;F%FJJCB_}^45e`)b~qP{pVv_{i3_71PV@U1hxZ?1
zqc=}_ze);ru%V$%(F1|7UUZ?91Q{vm&$Pz@YWD7f6=aJhmmy=pe|wytL#Lj?1MsM1
z3kca0^6yZ%RIL_+cc%UAUnvRr*TcWi8fQ>Ax~sXYY}a*Mg_M*2yL*EE^THo#xhuau
zjkli^r#P$q*l2>KTR6v&h_ONZaxXT+un1a}mh&3D6kp+qcGlvYamLb8v?g0@O@E$s
z_)O3>h~1@shw(i8En%y{Q!~`2EhCh9Acsub$5W<sjl+3^T|=#|-cas8h#l_#ycjtm
z+n;l2FmIADNUeV(x=~Ld9kV^UXf~SAvr}}mlzB~$IrcNe@6o@_yemJ9B_vkPhz_$k
zR+VX<gr8RDiLUckx@VGv*67^~)D_b|(Yf8DD#ODFNh$ny77tFZi0#c>D<-1SRlPQJ
zmQ6;3=>8H~*E(IO@8oT=qsWr%R=42mkZ)B%R%yM&;Jo4J$(V%;(FqzJ|BDQsFMkTn
z&FzU8T&W?o?HNSkBd7Y_7U4EAlqy1xB~@aE4OI_f@A4siGvwCwAM#QZom?|t3Qo@T
zKD<9pNe$T|{1-w`MZSF@f96s<$dZ_0es^+kgN3=o_4qyhAWoq8>PoGvS+{QE9QD1G
z59ts;)VGThybLOq#lZJuZ~|HEZCNFjC_FlPDs>yVXwe_3#M=1<7K2ryxHRgnSN)1Y
zE`kD&m4QRanvW1b5QyL1e-C1+G*2@n!?9L*=2K%ks!h$#EyRM;EJtsM@XG+Sdwd%L
ziFy7P>paH+s3WUu(nlrOH<3YGZv%%~zha7B_G}KPNi8?4Me>zOK5qWl=Q|KP-v8NL
zRarrAu2>Z#H@`BBh(cNad#DHF|JD=xb0*43Uj3Gv8~w$urHYFa{`*I*QZ{loAxHEV
zz>MiL@ouKsUNAT6bs2NDmJG7;bxG=~OXbqY*snL%$Lq4z$%k)A&JRMvC`U5N>h$Pn
zY^!l0#<+*Ae=3^FEy)AhzOOMhQH?h0iD8Gio4crU<X9A~ZJh$La$0xa!y~C0BZ{<b
z(W*J`&CmFH@Bdi+EgI_OjCb$6RBfFjMkdc<Lri489f%LD=M{v%is_llFn09qj^@Od
z{(nUlgrp-byYJKX4nc?-x~lf`jSr93!{Os1vgld{sgpNzf^@fUL#D3&+nygd^JrSm
zb$%UCbwK_2J<F(>_fT5E{ssLkA&IE~#4qB1Qf|$qDvWx9hU^A>;^cAGqTBJ+Y>Q$a
zSUTBc7-Ipz=A;ULv8qC6V#p-4s^)n^&;H!qTg$z9*&~#6QlM-zgqeYtT|ioePD;T-
zG1wb|PtNYqoaHdu<yX?FC{QR0*sJ*e;-n+jn^n4|N_UMvi@a@^X1f>cQxM%Wst$Vt
zEv!gr(fdp<#TH+!lY|4Y`&W96U-h=7X+6a-=j-w*2Oe0df$B&L`ed%h_yRU!D|I~)
zqBKmz7J54Q&9L;4bTL*_lz{gO7!lFRNCp)I(&T>;uWUb}W72F}YEt#<&R>>R>qx=M
ze80XEWkOBcF)+`6Jy_Ss_2wV6Sb5JqusKNYw0i~7wq{&0H}-k!p1Wpsl>JFzuiS>B
zfeK|7d$Vmavy{b1W>p_MxIy5*8$6(66tNX#SMV&&k8VRrhI$C4X{k?q`}|u1)?nxI
zrSaRGQ9RPNiFRF!Lrs~!I*0OpjKyz$WvHiwHR!7)o+#;c&65l*z4KCKGWLjY1|PHz
z*$v$Nisjo`CJ;yvscjK<RD$@$8=GdVA_G0mx=IIvvM02LYQ9V6+qkJejJki00I}#V
z@=Y}^XJ6Kc!>pe*pb6RCDU0IR`99I!(N;OPSiYEZNJmy;)F3G>&$pz~x!?BzrT00D
zFLb1Jg><R^2ITqnMN6pfXb$su4gy~JRZZ>yx`J*8Ki^?k@N9aZ-}JYV_@D*iN)^{q
zaoGnw2k%4)Yn3d;`^3K;O141Znl~Y0oGoaYE%6=jHGWFIzvo*E?bU-kZ@Z{C^`S1-
zn%|5a>taBl6!@40z0wm3n6NC>7}%dS-?{#=KUMQC(s%JhrhxjHOH$sMh_q0pW?fyM
zW-Mb1>$w!gO=}FsI{$QY`10{YmaxK$S32L=FGHkBF3P?Qhx3nkSo*?H*yl1|jW37j
zt1sZ1=jvx%tNKG)nElS5oI6O|rM>C>ljHdhBLbF@4!w7w8kS{?%;1@KcoZiN%6jX;
z`hB<hRy_OC#5@0kFz{Blk^c!|8(2itH`&~Ub`-Y)vY@(gnb&z_GR;{zY2X8K9V$A8
z28yan#-X^cE~$U&^9njKsR+V$@PoeI;D==dgFr#X<wtOo)pw@VU)Sj<b<#lDk%GlE
zNfOp^=p~-qaN(nYq`bLEys5_^l+<cj<n2QdUwW+UJ;K^%a=&`$Iys>bBV|U3E8MGg
zZTo6<&uh-tA6x4Q%Ik;uh;9eEGB0;)(NSul%bd^Nep~6rNQ|S`0@<oQ6G9XmF0$V)
z0@M^(dVQ-FYZu;;qlClEw&<og2x+0XBHe>iFYD357m|b<3IL`wspRLM$nrBhVH=#+
zf%w5L@Qq*m+#BPUuwO!FtX(@h*FV#Os3dhq^jN7y4C)S~JFCYyLLk!r8?;vgB&1;u
zoe53=B#aD`HGETK?YazcyLnN$#?qAr_+U;ANZZ6OPIjTRjF}eJF{YOw0@wbX=^gdl
z?>`<fGjC6k8$YJD_mCUO@y3QlPn2a6H1raL_{zUm8a}<(pVMU^)tebhlbBR&-u$1G
zQkIZEn1O@n?nCO?|2<xbE1Pa=YUdBuH4mb~tTFkm=y0pJ0!;2><_8R~Ft(oXt9LF#
z+S&fCuvPe}3Zam&ngdVC+^)8h;UP?PQ?S;k_UxYT!HucF58)95(86>>@Rolxz*wFU
zLfd)P8_EpjmvwA*>RZrEi%gC`Myy&_kd5%na&}-+VlFTNa@E6|@i8tWk{HbQf6HYr
z3t^~=*_*uPI?P9aa*JG^9BB*`+)D=`@P;7gN!zzyb4Jp?kqUuB@~UzYuwQy4f_uGg
z&g;HqRkK$hWAFbJrqsohYXeI`ohL49GFsoK=-0Ib<f=%*s|*qw3={&XxK{@sBA7F!
zS-5F&!0j)ZZREr6_Rjljqso!6$P{Ok&p^ZY{U})E1F{=<*TWS%z#vf6$ccv+>g1{d
zc~17Psgn`ip-S7hvf&6&!E-Xn3tFQ+F)xIj@3!xAQ<!|)&<dikwH+_KfXTmMYQqeo
zkarQ2&npx@+xGp4)wgWhj6fRovGJ}Q)Oia$dMpQEhQj%1lF|z&Rql+del!2*5`l@Z
zP?`O<)kCJ~Ry`@0{PXuea*>g?29Z|0%hLQv`A?GEmV>U;7M=a}<aX%d%HxKiHym+e
z2^0{We^idZ$mz1a#h5RZ66h?3E4uk67(G>$^_y&`Y1JLCcr$hJg$KfLs4d!K(n@R>
z`2-IKiH<ce<B{MaOr2c+o)g40|Jdk5b6BPRO)2%8RJnTcPz9%PInu}~*c_>nrn=5y
zW`VbJ#Jkj0i`k&zlc}gtXnEunh})fu$eQ{hmmD;74N4FVh?N%fS*}N7;L8BFyU6-%
zz7_PT1I~;7?v<uGv*xnR7*Q<<1KGt9)ze^}`n(5|*+9yw5qx5$`O*6R0ikltx4t(|
z?{Utf6x(0QFn)`KC=^{hVZWBPmokGt$nMU~-_HgcwYQKki^{}~5XPBV7Sq43HznMH
zNdL>YEuQ<)=LG<;G;vZh*q%SlTI{b_(Ua<pqhk+ZCWk~a{ae2Mft5CkRFL)W#th!u
z1EZ}d`sbi<!5XKr_&@>vKR1_>6JV9vAq;jj4{I=k>Ay4{-f@VEP0|8$NQ%loVk|r&
zIT_%9I2P6}K5m-H=+&Jk5C-diIT85>-Nn)MA*~zl&?N_{%eiE}6Qv=(uddF*zo#1P
zQZsL~@9O%>Ed}&#DJ8WP`xsN>462wOGbMwU?*SS~DY>{&knZU>-MGE&U9I9bCpgsn
zWWHe01-+p05Z@xoT;23m^tN%+Fo6zr{e01+h%}**>TMI0QAG|T<+^ws4a-2wPz8S_
z3(driCmA39CYwTDbO%O+I=o!ly<)fsS6wOb<{?w5y)BN!AZ&P4e%PEo3RJh7`SGb@
z_cP6Ze9bDcnC2kxG=+mr03c75I<Gpk6zodf*Wj$|)Rc10<DK)t-?BD_7P;~p!2|e?
zoQn=h6pzFHcq!t`;Y(cqxP4FYP*1{$VMEMloZr?881o8N@@<sbzq6doqWF=HeQY&$
z`qyf9_nD5beW+v<6p&KMSg{zXjORkcA_QgX6l!t<N!%qs4JqxoPm|V@&WhP!I{OS!
zz;lb>#K!pZ@4t(kqs}&k3EK@ga#p*|EqvBL;<u0c;^mCJ@J`mxY!CHh200k&bM_cy
zt&;Q&dG4bx+;mcwz8-RnEu}=}i18$Za5G^v3%~3*?pUt0>T2YR7Z3*fi%^c?U32k1
z7*rsh8|^>Id<Gx!dj9P&t_IV8+(3;5`td`*hI%U|V+@oc>5=3m{j!AB@YAtZpnjIg
zQS0evdmp=No*11PG}$JnwGXVWcD76~PN)BGzhl*|c~g0FZ6dp{oCH2O|K-u<;m&o1
zQElO9T5s+RVqS`{nXr!zE176L#TXw~M&|u;AMad$u8w;TwdGElNck!FO7~qmJkmmM
z7GOlC6;(OBe%;W-_)BwhK!|7x{+2nr;T?D3m$0}!A3a=u$HhBM{hRQQEBVedQ{-5h
zgL741G2ynN99f8j;7+@GP`;Y{WSF<qebeepYexy@QbqXybxXSBdWQMMbmsMbA<EH{
z_+Gj72wG079>uQe{j_&tX=7gakb2I4&x{A)ARS&wiDGdu2UT$7S`=GGx!Hws6#$|-
zSd{(V4RWW&r=YLY)SjPm=IcCF4IE6yObQQVkLnDKv+Jn4RLJ4_HxQU#N^!eL3LRKt
zZ04^$6&&%Qs2F)anPO2U?r7RXX`VTcrz$+4)2@8_GHKSj)m36W{)moZJga{$0zfGS
zj#L<V=~8;D^uQS9T}Tk|-%ealw`?jW1FOdT&`FwKr}l9%v4Ek973|&78qVT@wSb_7
zWW3|n{KF!b|Aanx0gH@@n%T+wgNb|%va!QM_J0@UgLNDKzi0&K^JS@Ps-I)ld-M8d
z`DvxVBpyryfZ06Nzgyz^i^z^TDZgcr+~!}X;8<N+h=rwRT33=vazG&Mf`4n4A8N$8
zN(5Hoe83ONRDXyvT77A5A_FftFK;ANixkXBkam{;^QBR_Lqj><D5PuC&qtfkR@5Ne
zmO6@de6K#2ja;^g@;1cJ;nUv=FvS+pmUk@){Q@^lk-64g;d&Jbu?Ip>SYC#HEWa<1
zJ6mqHbftH*0EUkTdCq&$h8RdfwViu(<wK&u3mqpnzfs}~!N!w(+XTs^0y!2jRD9(J
z*8(Jdt{l5&f$1kk$q)h&pbn42=BA%v-}xZD@%N)Ybz^0dYN$ZmNENYrOVwsf5`cBk
z@r+SY>N5?Td?hwP2dv2tJm1RNBjcBvp0R^&J-ZPPK11?fS;+;Xx?;zP`qC<4QV-~^
z$P+Yu3HSk1#bYQn@tkq8{_47s>Uq@%^KRWPRLf9TXwD_b)~mlU>1U{3_1ZG@#Osrb
zt>(9ZVBen7eVIs@lg_Xv8D3aHs<gdx{|?-;zjXIGPJl!uKPv=6b-^+zJOUKZ3a2%~
zLw2-^AIL=WBo_k@v5rWs>~X1pYYPJxAX0gUV;y)XHo|7?*A6ZNR!raLAtdG3)QvyN
zOSVA~Tjx^v@T5ftHG7F~3tvXYV<{M5XH?Vl9FdjqWWIx1SulIr$Z5IpbF=h;jJ4>4
zSMWvIVP`@x=n(@3r(@(m0d2#bEwUR1h>?vjt<1U6of3y>m3&P$U4AZB)%NAP4!_rn
zPH>n?wFHsXE9A~A-tno1mgW3RkSYDY@bWp%bm1uHE}D>YJ{a4GqiM-PO#r?2<&DB%
z>{V;c;^*3fPrA!y^?FkHll$uVlOo#I<|n3N?B?oSyg5&<TkEl(pZXSSh!Ne>5$7e#
zin4oyxD&QEuqo^VZ)QOZ3d;MYSRmuepV<$82R+0aGhNIFdKe6}{`tH27>k?WYwYwT
zu4wZ0@mdMtFzTF*E1N^sVw0bJ4L-&z19tNwUYY#zesOqj?HJ@^j56Om%{CcrfR9hx
zh!50COY{w29<u3DpE-*fDciZfu6q!5#(zMEU3*upp!g+`*EY2{q*_S3I3c(FoAV~7
zOn{3a-*?kejoXjWe>uNvS)>%c+L}o=B6KKdty}i8al;}POYBtHaF;);@El&piQDUW
zDw}+oM4O}JZv!84&t~FWV};tD7Y9?^%=l?L@D?d-w)ifb6P<P5w=l4~KZ9l`(#}u1
z--q3R<sW|HIj=f^joR{r-6O&#U<Ckc8T<219EKXMIP>anNSk&ogu=a}Mkco_aKpwj
z!8krrY4b%wU?lgKZj3XgNPoYxa5XAEF~<a>Kaeuy3ZEDdJ6GF~hJn#^#o6cO75&L?
z3s)qV0ExUQ<eR1jJ!_I}8{f=5OzPij+iWu=P%D!(&0bXAB^olEI&c5ODJxZcD!6y_
zp}^O%4$iTh`##Nce+H8n>9muL{&5AudV!Yb$I`<;n6FlfgD$Fp(bG!nX6j(AqGzZ~
zI;kxb{lbapKcigT8`c(+d1yYJz*DNfNl+&aGJtY3F?|7}AJJn~F-MiwCeu;iu9--6
zqz2nNL-{+OwkGv{)Xscz;tcb_OdGkQ1DE5bnN>5)>rflp*FP!@NoMA{psSo7&qq#p
zn55w0?Z`Z2fuubBTW3=!f`zNKi)+tc>eaqoIcUA%jt(I4K!9dYc_Z=hWIY2*!Z8w%
zCED+#@NIXp8lw`U3Yu$@R6Ce$p}iC2`L>D?wT;-PX)EMY`~FU}EJ)Y5y=g~SdM@>e
zqDW6bUH`iVGo`9gVdWS5gClj0H6)#16_lP`f{4HP+eYVq_g}Ydw}C1|Omv}ytF^%R
zjw_z4c|JrdCG4~9vHihO6d*S(p-stST!|dbI`zc3o|4UA0JeJJ;=-$~XVy&#g==x=
zC{k%zp&SbqTZLCOtl|^49NPU0(S&qT%C!`9fsKk|w=Y5DQvML2eh%;llbM_{nJBXz
zrJCHtx}Tx|wzYx*KVGdvRZdOf+2bdbjzgXKecL{<HJF6B*QFY&WxHaqbk$%V)>13u
zakzLNtrS6(@bX^BM{;Xo#kBl8@)SL9b?;+=ay`!9;;0};IaaJ0yPOjQ!tqJ1h+NmR
zIF=ymZe37$)+Z{9?%gFBk4ECtNvX!^%=RWQg`%ZlrDwc|>f-r%i<V5|oozz(7ol9N
z9q``0wILtW2u{;&0?5>(zkP}!OD>u3u@qiV!R$_6eReTBOpeOYJ(@Xal?&&r;(V-n
zhZXQsfPcI61^MP_xf;$@jQf)Wm)-9W2+u_cZqET@8jKc7+BqS(I7U_<OP0gh<<N9I
zI(K)emjW7`d_y9Ft#91ZQ(6XS3bYkYj}Lhl!?GMF@gaVmvE!83ZZt1x;sFN;B&`u$
z%0RdL{+9_f;C&0W=~+05n}Py6PI)_90iazvDPHp|)*}rcz34kGtO~u~oZI;#9`I7^
z5Co}d3@pI=3j?dGG3TX1{Ft?nsqSnTq0e=ZhhL=S1PT)NUL4~p*3q&3^8$$T3(|A5
zYu@yHJ8CO$ejV=<5OCA=-JTE)03{;~3Rd?A-_#oe5_D<Q9$|JoMy;ao*|&$Yb`Y{(
z`#KwUg7?+g+lLtmx3a&cG7m-4CJOD;aW}S031`R_OPF`{LRCJ;c=vb%s0BD5v5u2E
zrn=AVu=F{a`(2$O`HDYspe&URvfzBBvcjzt=9|3}k{nOEM(?T)*OX<I0Pdr`Wd+gy
zKC!(E1sgVW_86={j91L!L#FiROQ!^#S0a7T>)iReRccvsiApbCydNpEMQn^$F|)zc
z$$&x-gi6Sn<v1H}QTc2e(L|nSA&X|Ll#|r-H!_A6GVboZZrc2sb+qotVLdY%CybF1
zL>~}wR;CtLrCpiVdlfYj6qNrhLXw+VaNAEQ+;#{C#YIIYK8|Qow`al4QNk4L;O8>6
zrQ}5dI-y3Rg{!MKDMR1veB~V8n5dtv<iEl9eUT97YCw)!qnb^PEX&B@V0(bFvs+^q
zD{*xo_p1n6-JXeKvZSA8969czNm><laG=Pk=>&_&b=d#LcX~4mg{70MdytyG00eZq
zZo{e_57sas8^^J1dye`2-RNXGkb19?P}ppw#GEZ2H7*M%5C-Q-uhZ&GSubsg2u*B_
z1B@ySbP^vCtXa33*8WN;b-z!q|4cPvf24cE!=c}IpcWl!^O8V?0$@N&&v7`TepoNH
z{#b$XH8ts*fh#T^3LOZ_hm-aAP<e&2KL7TPSAz@slidgIo0e@yqeQBUOuTD2E8Vle
zNb#6^40dilzO3cFy<s#`<^YS++l8Gr&IAe2umiFffGB8i_=ppg!evHOW>}5Q0C*lq
z_dy0#YoYoaZf$;`Vr9@%Dbze(C%Ub?hY*CM97R=eYsRw~2R)XJqb(lJGfi@r5I)`V
z&y-It$WhK(nGF|K>q-&?=>dVH=x98hWtNRUDt$6&u<vD28%KE0y~p8@qOKdh;x@@`
z;fm%V{*Avko$G2b8RiURhU~8el&2+Ld6$4h6c8d&%qP5+2o%kY)cFnG37>Bk-m{)7
z2TM0%S=188id9UY!3A}TO-iqP>;V>9ab3Mzw76(~{1g4AO#ig#>3Q>OX!p*i!`%Al
zG)o(wR}HT_^;?GeN|ptlbFLt}muc=Jtvfx%Cz4|wURh=dFiTagPjd(HQ$gyb=HXMa
z($c+iZ4q=*MlE5gEn$?vqQUIn@Bj*hHZ?UZgL7V3(*=csG=rAT?36zoJ5ZuNCll$G
zDSta0T;EI4(mO1*a_2o+Lh>7zkwzLQLG(y$V>|R<=`LmA%KL-dq#m|WkB64vo?U0o
z>1}g#)cSRI+B8)BH$3!IEjYTjDR+^EGgpd8f8!y$^bGof+Qf+JtV<1NS5{6uKg8Je
z7Bv}gSb!O3Zhk)BSvk=B9Nrc%5%2t1WLJ14Ypo<=W#pWSwfG9e_+<r?Th^=CM^fY*
zW}j2jm^*U|!NN1vTcCPaZH(N>`_;4kDiU|!*`zLpA~jE}{t@hp$po|^kLiK@c~(ft
z8&&fWc_jas@*HNMpj-|)GX8BW;S~_o$M3w<a01nbfSO(G?Zew6RbKBFs)uBATFUet
z<Sl;4b&&Omd7j-v_@z80@b&!wDDaHP^0n-Nlng3w>vuO`XGb$G`6nCC=1m@-qj3Yp
zB+GvF{!+}3^m2+I=ctngJY{MB&SeNIVl1J3#g3=Mpy9HV(&5JVcmfa+Qcl~;=FQtS
z&!c**K);qt!k9eGZbH&<plfWjI4n~u_o{L=*e@R?2e8*g(<1LZ?Oi4VH80bf`LTIV
z-lt0E)>QzEn<6{iu_NL5C(FI$EX2UHpzl(QdVIEYpAUTGR4d>o7pzSGs{?yPl!sSU
z7+_Xu?q{K<3v;DIamj^DK7cywbSP^VpX)L4DA)J%#C@M&7tjyT>ON@zMyOe(=>Z;t
zHR2DBjdU}J0+Y9*RN$aA3=840bU6qM?#auLdi|*QcZa`X+1pX52wHJTNlTDMaUjvS
zY$u)w3k!2vkEj4Uka9;H@u3I>*gVhLSjns5SabMTc^Odh6O-=hmKB2wq|vMr;KACV
z56bT;U;do_^RA?m+oT$f#DJRFe5PF2bwI0tc^*4rW1R#R&I!q?36Gdn`}p+{y|%($
zslYt2YxSkd{6rk64E|2zx?Q=O9pP=GEn4kTs3hi*!=4u9(p7YqUf;XeyCs&}Jb;MZ
zNVQ~EM<7N&)PCbiycYS%0f@P-^O;k!va(UM0fB**4exMH<|5=feV@?NOX8EV_WuaL
z=QQhnFz!608<T)EYPso+UN4K$!J1X}_MBoIt7P=35amg;c09CF$H&J-qw6+uS6ok4
z4<}8Eb5i^)Mb+Lch6Or8sZtG3Rx$6!Z5G{*`ld0XV3_6{spM!e;9yLZxKBC0lN&}g
zmO;L+MI>2$XLksVgBKRGKDHUIH0+zeQ@sflMzB(q-R4B7{Fy9m<C#=*^(nT40vpnW
zxP=RBu26TEItQ1y@9WSd@ucptx9t$#Py_@nr}?-p7<PH}*GLu*JEra2)*L_iO}!i!
zw#u`qKIgR7*p7Fx)+?C%IopovP*ZrBZkRhTK)jBIn_!~Z?u8Z&<GJR~?l1@n9>S=*
zy3Qx#G4peCkvVAG_>P4yE%bgrn1Q9sW@DItbZ+jA_f8u<`BMWwoAPPgV>FtXn_FEb
z`bke#vJc$6FEa=HSWHwO*LCuN4s6nu<lWX4lbq~(=ID`n>J?E%6UpmRm3U1sp=I08
z@$3Y}$HcWkdPWaW4Q(4a(d^EX$&}*FcssuBa`>Rd%uaQ0J-**jZqXUfmxQkma@>6N
zA|k?K+D5H8+;gPjsVsUkTj~lJbH<X+?oE^4gXek>464@P>hI0sxKzzGlp%PC=8t_f
znSbEp;hi1)AoAPE8mn%LYjY|sEE!$5=*tY<s92=Jxq20tA6i>lBn3SW1bi1Z)b+jR
zo6|vEO6ouXDK#=VW_~B{bgY>D$Z>h64Tzg8oKnC}H1JI@9x(Y2Bypgh@N!cg=x51O
z_5pzEv^I6qI6Z@hvt?$TCdbW355E&r?XkNH$O%m)o6aeM8#e^bH_tF=eW#8Ab^Uxq
z-I-P(A$4?zNkmM{koXYNtVcBBv25j13&qXAjp|)F&3rYSYn~g<GhSZpvSpiWz?}%(
ze(vt>iN4<;#C`ful)-n6Hyk^TN?#Ju#eBg{GYhnKS#^5R4R6>mtD858WmR4!&@uv~
zf-3HfxG4=wWo~L~X=#~vr^`u6c@At;6B83Usncv&Ps^}!Gj`CA{Azb(OO1cylLG3d
z_ki~l2Hw|-M0u6V*w}0w7f%Q6mOJxMj8uChk#r`c5HOoC^RutpQX_QoYqgwL5JKES
zdEFm>i*|$r<v+$A8;#)9B9dc%VWIsETbmUtw>-lBc0)tVS>jWWg1o%EFxuP!txTW5
z5{)MULpbdf4!i#m5fM3^O@90KjZ}Abanbc`5^VVVgqG#$Qw0KYj`F3JPRt`y4syYh
zO9>MzzMbCk1h`iM^l%P`=vOHSd0YkTC{(O`F^&=-zOqDiZ^F`zMon@<@f{u5<(QYJ
z$N^cx#8psm!}xU7bE69pDnWy@S=ZXp*87t%c_1Mtxy?oxooa6P3fR8XqI%w_nMhYF
z&6EK~AAz)67_#u(e=d0R<D8L)18Bg(%?3|T*WE?zU-DGS8%d5MtXD~s6D>;_fF0r(
zw<)_mI)Cq{Kc9<P7VtpiXij=(zjYPJy}#)z{<tqAIif#rNOVF-EfD}koKlL|>I7yJ
z%7e>*cg6C?C$gjVzDI--X<rah*GKN{1ys?PeLk+0@i|3Io)@VdIa*zw*B`EEvHwz5
zR%U2qL`+SM=<io)YSudna#2$ooE!vXV9FApQWy%<j3_6{Q~`a)l&MFWwXmkghqd42
z^B+KW=UG{yyxZjyZbivA*G3*dC!ce!ABR|!)oJ@Yj5;r61+xHf;iPl`=su8|4BQxe
z*d7WWGGEA?+bDBh7p=jJ?YF;@$IAIwM;V-}4(rnL)sAs-K)67UU-6Urq*B+p>0<pR
zj<)h&285UNAIp%~>T}i=Ee<30+oj;6+n2J~s3wkzfe?Td*LDc2kFEo<yk4+<z_!;P
zmjFiF5y!cjO08e%ltY`Nmak>JG5{z@h&=mJ=WAFxQaLF~ux@FSC@IQAN88s^2JnSL
zH+?G0XTL?xzcF6gZ$U5B7Z3#H4d@Mpj^vvRReOn4FiQb22WXOy1)z(l+Ikm*A#b2n
z(t~k%PLZ3|wa(v@YdMr2&B!h&4yMtZw=`urCZnUQ46byHPvNCn<2+hdB&<|HoM6@B
zH2rcZ9Gn=R*+Ou@^im@nM(;th4o>Kaj{gQ+0Wv3(T|xkb^S7%ujOHlQgZgQCyfYss
z;9gWzTuhf(Q?QUK9?ao$>L%4IcsMrO`4w!AY7?GRmhI9w$mi|mul9=S5%KhR#qMrV
z$yCIpc9c*WmR+<$deD6(%6NTR6~5A))DcuLDpixD`Cu|-3AOP|J8#A7ybSI*P=lS`
zT?5AzrYHPZxannV1qBr<cr}6W@E@|)JR~F&rN7D#Hie~1M0)m1?B^+Dj~)8yhK9_?
z)F?g1@&G}s#j(-+sD@#-kko8B#+7M(P7j=4R-Eo|aku9s$n#|ebsR*KVb4-enlVyq
zA0Y6R%y=86^D~Sy=&s9{!^s$00xAKGP;yQOM_^cobeRAz!x+CXBLl6HXN-4-c=9-{
zwoS9~p>cvx-sHqeO1hY(YL>nBIOgY)(#v3&lU>(iEMEe^h4a3c@*6hNX(cmU|AK1k
zJkeZ^z9j2hq}N$E9cjsplYMl!6Vvo(Chm<iaMU{KH6)gWm4bY_VB!>(_mXV;)B>&W
z!P*<8LqACWu6SNNxr!zDSkb&YMM@l-X_iiIjU3`?@JxEW<X`()mmm!O;M|y{qL@^B
z-(SD$)I#aXz{)Bok*4_i_3M?Mv_K)up7OZcZ0JVVe-MZQ>w8Jn9!sM|MuAE=R-MX$
z_V)tWlaYUY4#%=jZe!pi2fDNvv%0$XB@{kXQlm$~E~ZDaII(rzxCJ_yz_J_<<|D7m
zkAkWv%}TQe2sU5Be3FjjzMC+k0KvrRz2nJtFT}!m)~xN*498P=*or=;5V%|S(UJ1d
zioq4qY;~bC-GNn(zP1_^$9@yv2!~@j8|<cTAr~{9p0;USTC!et#*uX~R_H@`v#Eoz
z1#8QINb6Np-^xcI5CwH-9=g65;Q9vsdW8bj-0>G_^!Wpen*yoCyIE%YuHlyZhw=<O
zl}TRy%x#JW$`U(YcIjZlsMJfKx{ZDa0hcoqTXvd&YhI1PFacT(IXbz%mjK+7Sc@4}
zI|6`^=(aBMo^rHoRau^jmIVw@jEW$efHJyz`nj~rQ{R$A9!s*8&y8$lGo5tBpg5M(
z7CB5;Oh<Sys@4f-@XL=Z+My*ufa_*co{4gC2&Zi`c$zI!SLL+XN3F9HItIvi&nVO%
zEg^#*SnRrL_tOC1mUUd`^$jUw%EbFROX;pllD=|<hc|NLw|<SyJ*|iBRY5aL4T+$n
zsXC2!?(7ga`Fw6j`Tao<X}NuFei%(mJacatX&gYySo+0ssDvN{PIJCx!RKI51?X~}
z(vSPw`+Zj$g;eXble*WE&*Uv*I#qu3cuUKV`TEGdc$J`~J0LHJ@KjK>bZ`e=u6DjZ
zoD)&77J(LA%|F_sd)pUK+-iUqXN#`^q5_S+rC-(Tx2x%=?&Cv1kpvczbI0cS0y=Fy
z+T@@Z6#@M!@YnX2eCYt^1qk(V0k_9ErvD!H$JjVQQCjI8aHom>Fq$L1^$Y0&1cnEH
zI5Y+|Fxms&8Sg(lM^~cA!s==Y;44P$TQbJ8)S8+um{bX%?h-8yb<WL?40Wzh|D?MI
zzKUV+=e_lHFJ)wVPmJp<R!0J`j<{GdQz8oTjHPx?CvmipMQ>%UWQ}4%qs`EUb#BQ)
zcr2fXL6s6rt`#+t-~CY<K6)gL-00U={LYCd2<73BhDAC*ZkV@PG`Simj=k;sQ+NL0
zS<4vAV#w#MucDjwJQRE$c7cPevERJ;hZwQsTcFs(*FA8(tXM4USaFWCe+<4kLgn}8
zu#W;u`e{K0i!?dk^9AFXu2aXRax)W%0PZm~%;Tm<;j23HgvMzs%WVdTSLeO@u;2ns
zK!bX7E=VEQ*<Shw)_(hKvCqoSU8<!ER7ScDea65>R{w@2r)Fhzv<#nT%|g)B*t!7H
zHekiYtA3f+eb_Y8!73NXBSb{xn%RPAm%M=sN}3dN1S|(j=q`Qb#Aa*fmNeVk*TG?O
zi=aQ#h5ub}u%Nd3a)&7{)BfIdKr9SYthlI}JA__NTWz1YpAJa6M4@WEK<?ALG;Nb*
ztfpPWJ8AiQ$$YWPj77%Ez2amo|Do1Xqj$87)mqF0X5Ph*EL}fyGDsY}6ELQwR60H2
z9jS9DQe#QT+?Vkk=#FX0bxa<G@{R%<&VF1;A8@rfIdoBj_A#6jqM6g;?6Ap(gBzGD
zm&PU6vBlAWSfc}o1l*fZ7%F2DJdndH$#wRqN>CrrS6_O}D(AESG)<<jNoz~4egH0k
zKN>!Kzy(xWrt@w%tl=LWc1;bVj&%me5IXhlAKn<sC#+%h?>W9q4S_^b{K<nU`BCmC
zJs$%-+f7ia9i{#QD<53v@9!PT;A;3YGnPmz4{X?l%TZ6H%tvet6|gXKtaX-pc+SAt
zRCh?`M)#T{Fknm)I0h{D_<C}VMW0Wv@Y$LWP%;cxTvdzD5az(FjP1@sUh_AN6Ql+?
z%LIJdx_Xf3f+H2?u~=u&iNIkCPBFewU&w6-z_;yly6TmtKY@aS?LSy~`^WS1_554h
zp8K~}_O)YoY;~p0v8=0tR-IvWzL4{<FTk;wZ$pt^NlE>=XKw#c=e0c+ddPYRBfOdG
zdVs9hC$Epe_M!b&W_|lMz#_Ci(0hk<*wZHR`Mk6h_Dr-Y)j?LDSvyy|_TtQM2m|Y(
zLh@^1nb7nF3UXE5gGn{MxA?DrT<bF1U2(@N=zk70k6d`iiGGnEr*@vj;3l!N@2Pt1
zuZfctOje2cz?h6rSf{tGIsgoO6|S+R9o~pmvT&QEoN<{J)DTo*Xk2#>))z6(H|Go(
zJL@=JCBP4mxSz*SCn<T%T;jAImzC9Wt(hAERtd9a-<H-<@>HS5X!zLxp}xY#SrpJA
z)(>2<>WDExP!HC4v{1H5%1Z-4kNcmJ;n$fo{?%upWJOm{CDc<;eTiMMgCv(va4Yc#
zoNK6{1I6ik5<{0zVuvlzSYIhXXMkpVbsV6=x;OD#qNaZ%tZN>~fEyhT&pjv&rPyMB
zc^J;og5kh&UB1hFXvRL^OrWLotbi_gl8KkX<0xYrls4wQiJb>PSsXuWjb|#HH#?KW
z;OXx0QMgzi9n-Dr?B!g3_k&|DqYGLgM8l$~AL_1z>#wLz7@N9f!GfRotXxUV*?o=_
zJjb}x<zPPV6^(C)mmz+=B3OOJ4{x)6BNWTf=7~JBaBi#u$_TJsf_2hLP5KInmR66=
zglW8)4D}R;^&8f&;(g3koWo#wE6`1x*v|Y&ekIr~BMG#Yx4%TZVT&tq*tx}Wc1i*=
z_GD<kf-DQU(2~V@o@V(5)zNJ&CYb{q>cI)$1<DT&O#6+xue8hY=S)siO2_+piyc?Q
zo>vOZ|5ocjr<8G>pM3E0!f>vQn+L&FFtFs=7+VQu=GhPneP@K>ivG!~$>2}G?MhH7
z$f~Uumg)5j-iihwu5E@K`$fC#t>}(p;Mq(;N66I;tGXf8f~@;xwnt4YM^Bn8e2lwh
zLNsWB9l>%I&I=5t3W>QJjvjImwNNHt`D97NB!UXN1+yG=ty64x-Q(f<n}Dx;!IM(&
zFaji@Vg~N`JgJsAZkpyWC^Uv$cPK}>w88R=+|Vo2G0nL~o1+PxOUt0EQ+d2jFogFH
za#IT<2!!j>*BvL6=O23$Yvv|`2M*|eC5hgL*mu((vP4ka6jv|EA)P>rt6FkZ&Ut|<
z1ibUW@v+W6wwwSWE^FPf#q9W$+WEXt{ter<${mPb(K|_FuwNCo`)a4GQbCr_x-)nI
zi_hdUY6Z+7qxm?SKqs9m997L|;{+}$xvo#}0X?Bhx3Q%HSjFZo3CibAKCtjvtaW+r
zp}5obN5hk`R^ukyx-IfrJrZhXYPUL@%5&8K@n*h6MD<f(Be6rXW6jpJn7SCGM^D#C
zokh_BBF?K}WU8BipnO(yzhe8fyQUW_Kl*5ToX`36Gf0mZUpLMyb`NwQBix3bZcQh}
zdvpkH6TzGa?A^f3{Ku}rR2{K>ZEk)Yf%TzwT+i=aTdOv@7DN->Cr2>~mBPy%8Oz|s
zO_O>U70>W`@iLZ3f^(AVxr7f0zbw1yuFE}s9^_)B1Fsn!V&G*IIe6y<v?hiEm%G^1
zpAhOyWYkS+Y91IUw#8%vSBte4r#kp@fC_*qI|m1<FJI7cnhlB|@0Uh5XdE%GCw>(6
zeDa~v4SZirBgzB|vVmNL74sE7{E_riL{Wf|0*158KcL=}eLuDcK1R}JWF|NOm91AP
zf>sJVWnfx!jqg>u6I{~u=0IsyO}D>Xl{AK3g~R7kLz9c_jN@e43=fIdWmNMv0R(au
z_NS8)#{mN{;N*9Ty2=HBRa6x=M}2E%=ZU%o#-Das-&T3%ttuZdngTimcm&ewv$f!Z
zJCA|*!>ZxA!F7ytHAcsP6nN>J4Tf+Ng`8h$#+u{Z3{`p+r3EKaR|giK@$IpdvBM5E
ziWXP3VV>Gckow6#f>;XHB!m|ZPrci6z)X*c^g0tT8Xz;h<01w7k;+7>E|3Mrf-kg*
zyx&((eNiZ>zMbobD@*6au9J$2y1G{aec2nkJ#8N`rDp&3+|)JdSCQSxB=390{<xWl
zv*sxCE4=xqiYN5Df$hhSz{m4nujl5ku}#9%%@{vCv8XlE{5;|c)6L?|#D|tef{t=a
zTu#PKQHFA7+m|z=!;Zq^C9|GE)lU@^&s^m@Lp|kHKUEO5pG0lmc6!6#6mT3>P*q-L
z*m@K@_G~v}Rd&-i-f}ZwEqzNnq#9|tnbqv$i~sQbt(ndA2Ilv-ME`quwtwnLK+h`_
zRNlQ_Lb+Ag;65W-cBQ*w=wl^Q$Ah6#@sIpfp%bpqhci>I2}c;-y-GGnUc$&!ugil4
zE}JE%jJd3q&rbH#O~=a#TKs@#6*4@{l~iz=pUcp17%jseEDZ~Fm+@6r)B05uo@}(4
zInNOEPQo%sGV8lWRf)Kzg}A*9)ABQs^q&ERCdY;@dB;JT2%nDno<k0oWr70Z(M}Dm
z(Fl*_|1tuG@5SFQy`!zJCpr@#Xcj7KzBp-z-c0(gIbq;P25&9DMUzjG|E2W1%x%&C
zwxM(cdSIjA3Nd7R#BpGy(;Og}Jl%KxPzAn1ZRjj3vY=XUHNWBa&yWpF%b-})z~anA
z(BOL2T}rB7Nzli)d~HZmrh4pejBo0M7@?B3!t?RL-VNUk-l<+D*y&9j-ZPOLJ7ppt
zz<iLkeqWWXr6-5G`78GM+;PNL4Zd#UAD2f6d?;w{2DcmarT#Zu`4!cozP)_v`6;}j
zwMM>rd`q1BQE2l=A+fANTNL4Y951P%Pqz0W>36!>+NF1v*`4q*O`?*K*fS!R?+~uQ
z`?H&#Ury`%aix@=5yaHeM`sAjr00q}$1V6vD4AHN^Icg(e!nW}T(h~R<k3&k`I(y-
zeYTV1L>`Yrc>frVf@Mj`hNG}+wRd#%cvk9ARD|7`{I4f@WWHU+1iC&^PV>fI0-LOl
zx{`n9sM=;$Q%biecBv1K1q4O4`%@i%`O;-0VrkK|vPrtL_+gX0e5ILtT^NztbI;yq
zsAVowufX2xwc6!}+N77xJ@v1H^`<U=#zkOv@cw8>IY&j{q}SApL*i<I&)c7cf&+wV
zMDMukq^!1ZDRgVfcN^4K5Z&Zv<F)AhE6pduTP!|}$vH*e(ehe%Wj=K1$po`TK)TI1
zG_<%w#gKytuVhCN^!MF8(_bj`(=ulS&{aF{k{z#+^pn#>dmr7n{tP)8e&8~*Fc96w
zu9u$26Yi}qm&pUoANigk&bx+Qkqg@Gz303*TVgX=L&Z#tUFE}nS#h<yz)_egxs0f$
zAVDxWoZf1Zl)+<M)<`<ofv0-sb*GKzwY$oW3((<^fT-7=34_RU)~|0<8A>+1scZeI
zauxHWGfY~I>I_WQrG0Eju9cqa$g8U9%b76U_dUCZlt66_ulB@g3^Gh0#}z8&Jn$0>
z67*|*YLehql_mlhR$X-1WB%nw+LO!TBT@C*&2O5LjYg}yUxxDrAbj<ylF{N<RG2_;
ze~g1&Jl|WsWbDo)#f!dycs$2w3a*;*fY#39dhmUvzyB|Q7>ePS-09>4@rTUQ0uM(?
ziZ_s4Izk6`UBYahoN)Q6)#~2QKR8E}lziPPZc)hLbV@!c5aBH`8FY#*f%pxeNCu7F
zafj>H>4Wfhw<%9~23f5M+I#Hxg1>53**!A2LY8ygS?6_MX3xE6o3jLa_k}`FLo`Zj
zMguXE6S6u>E>0QZLoLhgGB3*u<HeY%-t_%oouifP&Y!$DapZMP%cxJH`U*C{$39XN
zJ^oXeex9>?tlV`&yG#etL2upmNJz^~g4DPEylmvC4Nk5*Cgw7+HvbLEa|odT`qF{i
zz$k0V!?p2}%i=w=gEl;BVfi;Y`!c&Q38&sS8Lz7|0HL+DwftsoUTnRXHe_r0{Br|u
zNJz)_hC3<gmFPQ8Z|)mqcEWRaZ16KpNPb}q2z+gh!h~ym<G-dLe_qYcFE?5KkBJzR
zxfBM7hQ_G<5DkKT`D{@!morqe5*y<-qU6=sAi5=b>TjDRR9!d2uqcOPXLyb)Dv$H|
z%N==;8){rP%iT&c(SG(fD4Ypdv|P>B9GI^YjN$vmPi|?=KjSKE{BBIxTX5B3+3SIi
z?zKU2cg{=(JVjFZG*0A?o+{IPhYED)lwrM6Dr-_f+KLU72^sgj&y?Ni^a{mz)d%Ik
zQU~5AAJ;<O5Q#T5v{Bkl7VMD=e0rO<Va}!<hF#oD^S>B>L9D_=oah;z;<$Mm5*ZVt
ztfKNUBV%B{5fumjRYD7SvZmF~7gLc)uXCODti4IXa*Kp`PDxAYeW<0U3VdXp&D3V`
z`g1i8CEZAuQ8v!9H$Uej?>#%d4I!$b$~J@iX)_0pE~@ZDxKvt&u8@$rS5qx3dC!o;
z60vu_Bbl_~%9&5WJP~pFcWgTvtF?)d&;IieJ38kJl_@66lDd`AFJ)?_MfA8pD<RfD
z-xQ8MquKm1jw<_($}8QA<nci-5Qxg1O4+#JR_-{cFm>NEw|6oZshd`=IQ&%hjEfU7
zAdr#Kly~Oyu;t2NM3Pg~&W;k1xQJ1DQ+sn#fwzR?W?n$K+r8Suwe`-<9s<92c}H4{
zmC=pGQTkws?PI;T`QI6FhOZN&8wXQ+JlB2J&OIdlEV7LM*is97;vaEINGY3V7Ms}*
z8z!udmc>{DalXVSzSXi6G4Z%VDi*f2I)v|LQc11P>Q&z68&zFW{`pYv>#w_aYp_XG
z)Em(g74wa=Sk(HBw7Vu_+qR;W&i*s@4<kE$i;e_ItBV<kXGR|h?)#kjrRT;lFf)r-
zzj-(Ki(J>8Y?zFZ;`cUtan@1cQN=b9>B7)-;)D?2J_X)$rK2;gc}AA@hk^?nOi|db
zF(hLAK`}?A#9}C_C{>580C$LK2x;02*Db4Lki~{aYEK*9pT7AK*Ga%YFuCsd0bNso
z24Bff+n>cHO9HnoO{rBZ2t0JjG`APaJxuPB9zy91R>QASk2YB*+x#d~PHtwcgU!8X
zSP<$CJ{^0a!o+EP(wXj@jJDsVd_VZ}4pG3&sneatgIn%yI8xJ>tWJ*ac2Z1sND5OS
z4lE{mdkq(a+5ED%f)<i2s~#TR^tcfomi7$kbU7+$2X<MBmHOey+~y0)^VlgPes?P`
ze4*2A9;%7;)!k<tyhfRVZS8c>FBJBE?|$(-StA$?K5Wk!U^sEm5zu;VWu?`kI{0d{
zA>xyO>pN77+H`kOpa?HOS0KG{QWjG_Icp|i6Zu8j7vjH}OfctG?;+%2%m<!<Q{vuZ
z<;l>LzNMH#d+R{y{7cXCduF<e_{TG!7@nYej`n=9c{lNuGh}INBwr`bl2<e5S49^4
zafRmZO4qrtvM#ox@5$SH)apH;73VxfI;4W9LWbTElIR!?`xGGfJw`0n(bjbDM0;};
zKD6q-U0d)X;F^(@N#jRTF7}OiOvy6;U1jD)@O3?Iw(s?6v-fZXcW9JJ5?eZp*>R4C
ztDmN6N9g0|4ut@0H2R$z9Fi$low-67cyOYnPc9pb`0C}Nc5bkF_4+>LdPZ|PRR%!X
zGX}}uwNS=ia_Qg+@KksV*a!TH?NG6caz-#^v4<n%yDI|uAh3OZd_&`7bL(8@v*v};
z8y;?Sy7YD8=2EveYuv86^O6xL+90cU45+1p@$C=D>VFP}y<Y5Dj((POHXB<qMP$Nd
zL)#*`y<NRr7IJ*r5lj11!oXtwS31sHH|Inzl%R(7`03ZJ5k;L*f{eWMuN8!Q@B?p7
zQfT!qbGC{O-wBN(y-$v2VXECx;JL!rM8ALxg0u~<YCVg*>->`3kGtG%d%H8D>nehG
zdVi2NWl2!dyY*e}w;*6o_CX}UiLm=JbL*Y@)9cM3!afHMP3_}twmp@H-UEaxIZIs2
zEL!`O@ge<ggP3j(a^+EVUnvWv8w;gWA#NU5wAfcITC{kPCO_HZ6>>s(+zswte>Tlc
z5ub3YX<GRFd4>}E%rb#SmCAWHZ~iB`MpH!NkBQ)rlB-+=C)p$4J48gS+?w9Lss9oc
zZnxY@I1tnqMPhH(<o`24?^^crjLfIPSyasVwN=!f{s)QZh})iVg2|z20I_J<e|qgX
zj6S})H=iM}Gjm7l<Y>q-<IgD(NM4-M{*#?ksZNT~fQ&Tm#FqP8s;*H%6NjgyG)C+$
zd79ETYWs0o`^({*6PHt3`dyhzrLrGpsnwWsAK4R$Pon||C+)_1#luq>@T{*1g%o%B
zk#`vxE<f*DriK){G*xki)g|VG0t#6V?37U4qd4AHyj+_k?P91AHA^hsHSoE^N^{IV
z`)-|q&7w#!>s`iHlwnI8g1l(&elL0a)6EEilAU#Fhq;Gbb>jga#U|6B(0m6hSZwi)
z?bzsQR#N)f6nSlt?JUXke!0b;ih4p9n1{uXqW#6Q=)p_fuYV?J+o0@d7D_itHrV}J
z8&TK!iV~+I&Xfa*2VaR=(Ske!V5B>9^OgUu6FGm()|JUy&fCqN_>xKdiqBkr>G&_2
z53EwX8|+AXz#jIs^;Wfa6?wv{%a`c?hpev*i)wA#CKOZz6i^yeq@|@xk&y0?4(aX=
z3F+<@M!I9@?(XjH?&iH=@BMr~zF+FWfitt#6=&URoFauW;fgGb`0BCM7H2KlyrX(Q
zGT9LoHR1w%+1{!Pqf@rG(Jj=`@AH1KiNNnz>X+ZQ9ncKy#P<3viL+gYZl@xyKV69C
zWuGcJV$La<dz#M0x27c4TNsUg63es4K1YER+MIng)De|5d4Ubc4u3KXUt`aVU;CQN
zU}in@<E=d9``@knY(@w}RV9)uoAg$N<h8vWt1kRGlL5rwylwxN->p>vK8o%$%62(0
zRz9*H!o6M>8|bUwpU-bsMM=voeF8?bk@7S)N3c*@I^X(pPr{3+S=_pHU-^0_mhs)C
zDaXE&ZeEEjGehnK{e88yr8*j4j$*L%ECI~CRtgYGYfIaGchlaN#<%XrVreM<311Nn
zpFjlg8vtJd&DkSk;(i=7EvdzroipSrT@t(}wL!<NNf}}RJ%|uoo1BH(+dS2m^_t?b
z&oW*AZMaNb{Z4sF!%_uyXqGI25;Hdpz0}fH>L(9kwc00cd5+$S>&4c`MBoF_g}#r(
zsTo|<-7+GGuTN63;IYcHoJAQ5FMw0uKC9Q&{b>|*Ut#~ig~cqMO!{hItbR;kqHmlT
z!v{4@S9Bu)FQ3-b)*B2kD0q!{5sz~mwfnjrwPLTOCMQ5|J`z#=N_AH@+Ep<#Ui_9*
zrXrE0EzkB)Y@bd4d$$69sUgYHUcO%NnKIBWZ6giDFREfXDiJz`H8XyffxX6tiQ7}0
z+J8x8s-j|}6m*YcV&r?mljCK#mCHSDYW%K0S2mWEm>bLQ$_|bvtEdh;+oZ2WS?zJA
z7Ii;TB0Y@%y%a^{3qvIVixLKjlPX1HeTf8l^NKi8BG=KIsT+Vep?%i7NeQ!xa#d?5
z<Nbce6#Wn~2YO8pL4U5V8a5xUj-+~wW(zI%eGIEp^h(oZ`?Na7Po&XOSusnKNB$}t
zSrnU@*))$rkO8k~8k(wWyToya<`GE<6iB*E&0EMtr;b~0^1O>~5(|xu@==u!N`&B2
zP1~evl{aGP2O6rxKfd38sT^;urMW2|pfZz!m_^4MKh2VIFC<_RW^1=I>6`7!_F%4=
z!`@&jR`dE<+$K%gY^iKoOb6Av)$b`ueo5nzm+NW0RWsK(M|)9%o+D!TQ}dU2J>n+)
zCUTGQ3{1z3Awh6d-RLLL#g)6j+#hu@l%elsg8;_1BBTwv)}=|8MQ5+*ZLNKCP=C!7
zk-E|8#|<`d+u&%9B3&!rQ=|K3q_W#P5UwYe4WeTx*IbDZh`l>-n#7fV1UFItVqeMr
zsx#bgfT=5UmSo8x&rLAI!)%P18mUC;gcJ?yl$f+uw<OQ|=I2%n8BA_<39*Q)Dt>Mw
z(J4LIRNkyo@-K^t`5_VVM!aoQ9e^QNHBn-k+ZUNKEajWtj_rd0E%D~jrPH#9j&uur
ze>VN?;+JWK(+=qs_XU2Q!<?woh7C;H{Ou0ac2}?1`taqsDtBhjRBJTob<O1^|Dssr
zR(t)Q%E%;pJ525<@;8?4J<)axlc)oFac%9})Xw}VQdvNic~e`52c0Uc8cG$Hh=_Mr
z8~+ufq%d!BlZN!miNzKNJL_;9)!1^M77fXf25{mRC1~G+ICzIzosDjH8O}@;8!b)m
zzt1m9Q?9aUmCXE!!{rGTeC?+S+b}az6Pg1wl=m(5D#I5~9*E|%!2o1HSPY}6J^(6g
zg~f`Pmn9o$xsuL{hF<!X0m-oH)N}@+z@r2a5Xzrvq_15Tl-PnC=KgT#+SVX`@iy@=
zVM!M*c>RQ6s;QGZiIj*b=4gnAgDo&^T!nD{&kwF>01A1^>xC!gXjn>bVDq2kr?YV5
z8?uuE<ThKHUb>P~bJ!vwg;t=xy(Uml&D~fsTtRUzJ{B4M%KU)6&&b8Ux86y}Nra`q
zx%hQpXc>Qh3fII0`*7v>vp$5q)uTbeL;RDNul79U@-_VM0`r#LMZdhu4mxY9$tfut
zbGGqM67Do2uWzovOpz3$2O0+W2)JBM+Bm576`q~v_39FVk5l^I#^P`8BTD2n9IS%(
zXrB`|SH;N~t5IIItIZF9<mjYqSc^DsL>lD%%{2|~y@hiNp)w7Rcc^STG*6kXM7ux)
zl#LZhe%5Se*fjr3-dKpmCfRD5msPdad&v4u#io#+w6K7~{IME8S2=QC)*J9!Fm@|m
z#LRhF6`5}5YBe4aEtGcF@l7oNHeChRFD}~Q5TUaL*aZm|AZqR|b#D?)|2R9e9Id$i
zZ9X6F55rQc!SyuOoG~c)xuP)O;y`Ty=b+b+mts0fKE=8cYHJgpR9&O<bJQNDdsoCx
zRt%Z0jlXm?fah=1$yGl3@C%W7yn_E>IzD^<51sQtIBX~(R5xDdZAsaGK70AKHK`aJ
z5{fhj`!0+KWnid1+4mf~OtCEgl#niw$%T`Qs!L-ZGn}l$ufWDg@#Hy*iz^Kb`ku>M
zCB$glAZW4q#dekZ{X``I*=xwS`S&ZE+(9WNcQBDJA{9lW>mVj<E~n?Fmc;Z=<mVe-
z#zS%Pe6^u#MK=1~F_?EPBSWFw<F4JYTGiQa<macX_u4nI#3d4<+vaFUr6ZJ1T`FV+
zgIF3C2?==8r05Li<n;ngYmmKF3zkcm3IF`92r-d<JE(smJn9mtXj32lHTdR9111t7
z2{&`j6yu}r=JVozs4<9|D&~C9p;QrFtiUx8Xc>Q5Xin4VD#@cKZs+Ng*{aY$)v78L
zV@osmBV=EU0XXpFVYr7iPgYfsf}3eNRWY{RRZ+1oY~#e@S-j+QQ}uFTFq*r16AxVJ
za4>-rGq^d^>(rKUIB(mBH7-iX95@Z5XzCU#05B~pxI+sZi)akK_nYdd;6GMia~R(L
zWeRg)ziCJ|YP2w3-oEZ4*uLk{8Gxj*&0hQc<+S7QVbu{@-|kfmcXU8%O{q_;c)dXX
zS9DVxOTBT^7F8~9>1{5n#KJ#qj1K2D{$=n4>F=2R19P~t&1)X+c}sobXFIQlzu|lb
zoouH&`PB$t*_Z7QX<OB}`A{wC!`+P7(NKYeVvMLs7qO;T!SX^ewvkf_yx`*Xp~0|k
zunU{7|8?(qu(4iSTe?#T>kH<BGK|&q!gGILqkHvtKz|7Xa;CWpn-{C#U%upzKc_XB
z>IQ$I*K>QT$_pi@NQ2i*QR*wwm2Um2G)qFmn>L0eqZ{puG=LRrran#4)h(2RXKA)x
z5D^(Q;Id)#X@>(juD$yD^Ny0$s?g78KtI8afaGKA2ih<DC%(jUrUCtMQ~48votr=i
zOlPo01O#39TNkY)+gVChjhQ~_i4RoCOTYp_t$pOa#-W=P3{2QB=j8r5wHD{eRkMdo
zfKa<pQA(aTcvw8_7M>fW&i96oBLcajT$+&U`u0;!G>zu@)&4W?1LQ@-x-5&_*vdE3
z1PR?M*fdO8k=6Ifm0pa(<6g(zL-LbX*Xw`z@rf9>CXm~e6ebThgzvQ{TD(nSt4sg-
z3^X|FiGDiV;h0;%5zR|5JH{2esZ3O6vGk#?6_CS0x2n)go|CKHXS5@bp?F|0Jr%FS
z@aO!*y_0^gXMZC}U#q#;Gd&@ok5#Kz{bJGb6*TFbnN#_lmi<EPx&G?(X;kI0Wl05=
z#os0DazYccz_(|rEY~Nbt3$7;y%GVdC-ks<1mic;-#z}>rN1U8*pFK%g)6s1$Dgy+
z1eW}A6cqO*;$}v|1yg%QY0$ZWrRto}yE08wE80=m02+sweC^|ytlW~kUB9|;mMkdh
zmsXJp0~PYhsw7r5W}|bM`g9dsfds8`Q2QV(=98XVidBBrJct!LlFg5q{HrB-TF}W#
zArg76Vgo5NFu?pIUqyN>ZTAbz5~SH6G9y1rh5{k5GWd=t-ID9#T?UFVF&A4v?#ai1
z0MSBQi;P59ZU@xZ0Qq=aLiKZT-=fzWr)&w+C(L$Ur;OBG@Pd_Ap5Q<J50k;_A3S#c
zXEcCn`DZU#9*Aa<a27hKV3DeQdy>(?uV#~*oiH)eRikXtv3cw;v73UC`H_)PA<6Mx
zUbUDMvb#344E3EHqQBa{3rEip<rrDuexQWxsX`3<skaNpRUYd`Civgxx+BKmOUZl}
zozNrG^C_>W$eKsCceFIH@w<}wKx}@^mb0ZHkD4l++gQbska}R~YYHe%ZT{KuMwbbP
zxTV!D*Y*<Q(Ci9hHVR*XL`*EGXS%mRaxn@qI5nh&k8Qlm`!B)hjY%c*Oql2e3<b@}
zh_rHYbG!ib#^!|cv)kX|4`m1B0mBv{$`adbg~8Zl0_kn#af=D|D*38(Im1Gz?R!e1
zSg^l=e2Sb*x?EtNK+!*pr{b7oeTg&4D(A{*_Cu2V{dO=~-TF&;aTNgiTyEnkYu^}q
zUoE=|s41Qs()2!xOrMqh8>{I{sm&LA9+S$?xSqHP<u$fJ@zdq{BHpZrMZfKx40mL4
zv_LhKujzJ8x{vt<F*NT`zy~nYE3Xy_yC{RaPrQ9(XmCOKxkfdt-gu>2<oGmL0={m@
zA5~@qfN$_JdA&*P_(Gmj+;Vxc3l-oMAS9*aek9EEP1Eqj@bW}*g&U}|yf^Sln09in
zCCnqGVZ?>Wi)CcW&l`wSl#GZMQdPvzWyV|H`K@JlZrp$2#b9Q%YoOr#*zlq;|K7s5
z6d=GBubf_!Q+Kt4kDx+QBL-fcmZ`j!i@$)92((kvnM)s_%T8?fT+}Kot=}xUcvQOK
zOSasBTQj~OxHDOrAS~e0;JQPJu-fUvDavm|+qXBz%<(1nJcOO)b>S}`k9pjg1uDu;
zQnmwI7zbpg8bK#^x*?oltp>ZGojZ?AtsFh>9deJOj{*cAm`~}coNvr+&g(W6YQ_5Z
z&gIZ6C}7-x`!l4`V#!jD@1}TLe_u;Z-mzlP@s`K7Mxe*nOisixSU;KbSAWPB;Rn)(
zwQtU@R~4WCH~^k`Byc;sp{|RiRgfi9qhr!)586+iI4+p>s=z?`{gtjY0_=3qApkuW
zRZ=Ors5H?CuQN-nMvB(l+sha*uG_eLku(ts1ge4fLR>U#4E?dT4D&_YU4x|U^1_aw
z@&;2_6a35U)0rStvOZDmg$+j_=4wa;O%LKTs6BZHHiHmSi2+^#02>CoGk(gbHBQ}!
zj&o|?7P4VPKJ8#1FA@Cah7ye1e7hT)c}iPUn6c}2Z?>L_^uQJcU0%C#R7O+`M;j_b
z9(saTn&e+Zqw|rJAu?DsHb_if49Eb3&Rt7Ffd$!iRyf_B^1+%$I=yYOgyx#}Z3vlX
zV>ODlKUeUBrKCdm8pekQBHxUY-JSft{3w9e0MTX#)8YcTT|ZDX_Oizj2JhmI>xe<q
z@U!^LWM0DTA%UWn^j>mp1%FXFdooHb*5@@YKZ~%*`(4frYwZV54DR0U*GY^t&}akS
zFYC-$U3l$guv`vvflmjtF}fM{fSOedgTSbyVXi;tocDZc`C7fh^AZt>YPUF#TRM*A
zzu83uTIL(z?Z${4%F6}a(^a<W0}yWb10X7cheYNDqZ#fDMsq=At4;o@a&_g4^Gig8
zJ9T3L9E)Dh)m)iKsHtPpnv-nQ^4-vBI0n2M`mMfpa&TL&5#hfFKhKtPjt;u{0xrQ9
zNxQw>$O`B|mYlNX(aMeKq0y><<^+Mwz3nIBOYHIhAa9$kv<p;L+*9Tn;l-REUd5Xr
zOQ}k=%>|Fa(ykoN15H&Pg2Nt^@GZWMfvlrrwWFNf5aZFK9*hwDcjv@*^!NT8zcQtx
zFZ$D7dXC(+IBHG5htD5|N7!0aYq(v((;O2MSF%uITza_u)nJ5<sb*1N1G&tIjD~7O
zXn_!5lHt>~Mfd)4Fb4Lcjx%K_lI*m}zqR%MrKS93ynnhnk^KYsQZhI>8`j`%{v6LB
zVm(@h{MoSAoRb&b3aR@2(?p{;UTvo(N0by4szUuA{B>Jb09_o?inNI@mAT6N0+DVi
ztgML3Q1WXITT#F!oV5x%o`&lb>}eTx<se9G8>{&q=?Cr}ua*)2LO*Q83g#$>_hU7p
zfBaYT`i$NvV4F%w12Im<2BH>LLc8h2@W<b~t1FGYFY8+6*Kqx6!`b(#*;|0#UF<ka
z|A0k@ao4o=C!Y#?N9mQ4VCayB5ujT+<9Y|pdLQgeQ<_e)+}sD0w)Om|ctKd7baS>F
z&x?1J1jIP4)e!_-d9{FEBl0d`V#>Xp)3vl5UECyxTSt%$D2R5fS7E7yu6RVYE8HDg
z?#!Q|1G*zJFeP7jzV+M78EM1-;x|sl^vv15b?YGCS>U)Qp3=<htVa9Jhf^2k0QWAS
zkuh%CexP_8BNg72bn817e%l!>OY}bt_#dUvS!#;?1_Sy{K8FL`(iADfXGmN!nA)S*
zz-diNxTiwjk8dvHu3H#Mq59e86@rav;T!;vLoK4H)<{JA#*}FfGOBDfcBD{i%k>(q
z-ONh>mBY|saVv1FlWIycx3z!}oapdU`(WZ4n`@I>FQC-gBDkEL!ICn`*^oGV&-K@?
ztJBwkMdP0>NhCyPC--d%3d7D57;PenvXFGUoh;!8&|%1(E63<?6-)%CuO>WE+GC8u
zN9XH*tCg4E01jrX&)6Wc9e9-txbnpe=mU|_jFoF?x|G~WPP@1|wO5Rr<mP}SOUfy_
zgWD*UuUbB1SYDNDj9+Ow#<Ld>4VlUPEET5N*(uaMnfDa@JSJ9Yw3DMZ)ONa$<F?HL
zqc#Z`U-ZLhgPUCi_56&^;*prqk_k#ak<x~r6^9i^bxz0k1zdz|@{S%c#^*Qx?O75b
zzZY(P2*$?xqMs-&-ePK+=u2-qoqkP|T3p%k3eS<R?bkM5_pFS>PfJUSiH*fBrN5Z}
zo+h`vGKNN49tG`^R+NDo*>$WgE&MxH`A7-l;Qkvo=w&ExuJ_)qe$-l?s*|e(V{H&8
zUQ<Uy1v725(lf@4>GtON>y-A+?%pp2!x|de{@%8ZZL_zZXRoo%o<-{w_ZWSW&>sO}
z-Z-25q8jU`gt#?FrAOf={a@i`P^6;V2=U8Q<QHHOFtUgP`45=+cICS>D{|TV3G9vQ
zrd(g7W>$s9ElP$3{i;R4?F6dG@j^l{7K61Mkqr4(oKRj<N*crFVCO?ZrK{))as5_$
zDl!~~#SYiyRmwj_`LAv1qm#Pc@I3cNnGZ05FM?x{5SVne#0qjcnDsPOqU4QrjiTf)
zj{#m-o;aHQLxvSU{U9F`;0tJN4xmBXVw!e%9~`T-5gZF?n&VwZ@nYk4ft&U5%FIB+
z`=WAXc`<!Ka-B7uOg+WGeOHgNX0~vGMCHRff@?nwK@6{$KYr)s`v-m23W<+zYgv-n
z2KEtgq?%k@oVJui5`yGROX4)xk3*?=)H|x~LTQ=@6WG?uDT_?CHSSv7cTf0IfMZo5
z%xIqUUlw57Ll1%Df^&{DeR*RR>H4k8!3m*2F#G&!pDi{0LdvS1Xr2$M?^?HH0fc6p
z`Had(af>6324?;K2C!8Q4~v$h$?tMbWYoKy=KAYDqE{br!DpQRRRPd9u*GYU8V-|_
z|B^AM={(2q(#lEt^Y>kN3Od1GzHO9K3GLpeccsAty`;cL!D4d6uid+QqGdUa8zi3-
zcz}OGhOWl#3nq&N8pGx4A~CW;Q6?(&uF`<HCJnIuS?;=}|Bx~I(e#L|v=sEi(B7o6
zBdab_KUCx3$bb9<`yoODlpf))avY9agyI@vQqZN7D+{rkJ3M)@yvD&%fA6I0%p}UZ
zq@yLNGpCBTZY~T_OS4*vP%U6XVO{RX0etJj`ugk0;_<0$y52jjwRw-o5-Q+9y*JzK
zCUa1MF5TVle_Ln29<=ds<&^(NC^P%7R0AHii@oL3UH0<_y(`b((PDKZE<WF;$B(;x
zj~G&kRfvH^sq_#c39Pk*mvh#dp1XLe5{m*z5AbDtaoVTLH+Z6-Bf(5uUG-b5B<ZO~
zi80L_?+(W(#62jeHm*!jJ1t1ecm-q*V6f@}w%-xJF5InJC!&1gmOJ+#zw?%S_-~~!
z02=t|Mu$eN{`v``?M>6K{k_GWwQ{}5TQNMvvxc6j*r8Nq3u)@tt|sihnCHJz7bd%c
zFOyB$dUK4m=9{cTuX$iL*e<C&;w|v```%lvd#{jahb1TZr$MY$qR4P!s_U79^OO6x
zeh3S`obFC9XIq}r_z0Z8B8#)eiJrLB8H^z<cOT@r?<PVYVN^1Z)tgdd^|lroI3MV<
zPdYEx>_5Zm8&F!omhwwkCR_l<_iNxiY=DOf;~5&Sx0HITY0aEWg7g<|#Sm_{XxCqs
zf@ji{y2?%nU`c|BD}HaT%nFTbNIsL@si!b849Zi9g`QH0-J&&w?%RZ#X)r&>w2@~W
zt1V$BQjHVJxpP5QPJ8UG=Kln4hBT?HjVasD4=J4YOzy)1wjp?g4~qBN+cB03yQO(c
ztljFDUm*Gpgh<QZQen*wwtfo%J(SuzinGGPfKfjhZ9Dbz;1jG^gU#~^tg9Q{kMbVW
zuPy1rY9c-f9r+iTmA2KTF&KMaXzkB_F2jA0T0JvqMEAoV<rdFOO9HCt#ld0~t!PQ2
z*rF_$7{&3ruev(Ba9>;;#PTeC)A@}5HddQD)?AmF?npfjVH~f2DF|=)PZQIpUw)XY
z=2=F*S%*{C)M$8%;m4z6gcJK*ALHk$#`>CDZ8QTYC1`$55ytx4u-D-{E{fDRc?guc
zN(oLg61bx#X{8UB<o}tauSCOg)R166b&Rsfn`~*r7;)fb6NX207OlFX)m^`=?EC?O
zhoz>!?pNkc;#gni#R}#b=9>D6^<NtQO8DXEIq^rBjO}(lKVBggGz<4F0e+xTMpU58
z<E8h^B>#7K_>gqgLNE=8kSBdHf!)?R)Yy;?Nqt3Fh3`f^Y~rjCpDNFCGRhR>R1L49
z&k7SX$5~`^*gDbkd#UC#y{33Lnu}M&0gG7fOt&W=gFbU{j>ajYTf~7KQ?JTbAv+Ej
zkf3CNZhBV+;yS31+i%hV$z}Pb>`R|1Qj4nvA<3sW-=g=FNslO+*W<rzm0sM4f%2_l
zh#mFsLT9ys>pgK~A#_(4(#ovp!8n7%twqb>qSaRI-1uYsP*~kTJ<&tLb2QK*jWcb#
z`|xkPay>0QjZ{Q>Hm!4$Zr?PkRfYc&4`R-2yTZUWi#qM6f#haB0lw8HlQ!L+3lRmr
zvb<bE`|5HIOSLuKdRfB9OHI6Unq-f_YG)$Z$07GV``hfq)5m6cW=@1YXAdJoLA|FW
zWV7Uj(;gj-VP>$a4CBy;=W^WW*R(b+hlXwT7MOx0_1Gf)`h+qjdvKXD6ZSSb6FbnC
z@wF&KT9$BJv)>Q4ni=zpxQIPfHFitprQ~FI%Gux2o8LX>U<=9z=|Mn*7nyhT_muH|
z$Im|-q9#7NURIM(l$d!I`H=PD@!9&{1uLFmvm4A-b3Cb%I&)!-O`VA_Xy3Bd#=Bd5
zUyR;N3uV$BnaN_tcXc>@Pi@5R<&{XfypG>C)?1)>Y<C{h*0EleDrGovcvkRNMJ(4Q
z@H8to(-SK0>XBcXpla(ZGA?N9E@a)!f{x_K&s~oW=H>T55%uvUZk#`lX`E*oWdExt
z<2JwbBCF!}G*oeU`zIeb*G8zc{4At_R$Byo_h3cH;^Jt5`&ApEvoMk{%faPanBQEL
z2n@L6IHMb5E-bVA3Z8Y!?RUiS1{g|eVqj^_tbD4e5=#e5)c_-BOb`2N1qU#*=aXgu
zBORU&*;0?UWhCl<?U>Fab-Z#`USv~}0c_{I{j<e^qu7sL`vL`&*YjT<^LUN8n1y?c
zgn&=b(vQ(wjv}#K$`tiB-CJrXy?eS@2`)*HWxNZ*=%o1w5|EqFfvobwTpwXIlhRLh
zaYg)EB5UNqzJSW@-GyhR5hb?QBlz(wWILZq{M%kg79DZ_H1tu=?kD@ZhU;8M4Qx!?
zM3SwR0^Y=nH-Td>EtPCPE20Bej-0CJj*`dAl3jdH?GSH_+NQjRuxS4i3<x$dR`3TF
zVdq`X^j~swzo&l}!G#(~%OY<KY9{3!CuVe+2cU7LW(3K1sOAQ<l)MO@sT*%(w&Ixm
z{%NV(|0<+{Gjmkr-sx>6`HvkJQ)IgKCh{KgcQjO6SFiH2==Kt8vFUZ`=W3^A*{zo<
zu|S3>fA{xTHt#tK5$pXr>BodxsT>YyB$mUf11D6&VAaz@o*DR|ouXYWZOtjvUFeS>
z$@=@hpnR!Jb62zdq+eI%-qOW^TAPn}!k%TuDtQ)naQkE>V_nl^*N@*&F=y)2PA!Fj
zP;Xsp$5mvs6x}`4!Qe5%F`<4mWbxl#fATLI<(V1#gr8^lreNdr4_W4C=)GR<m>%rF
zwiUEb7T2K`&aH}iKTBC{Tlw+Lr0)IuYf!W|=5d5v_MzRF;*xSw?x`Cfkd--#vy;e)
z<vPwFVt=rz*e|Ur*A=_$YO!@7r}?2j40r;HkjA-Ad_dezaOkn`lbR)R7@=_jk{BQQ
z6W<*y=rr*Jg~UdE|DjI@c&8n7^Lzg<^i5VXIAA$L>oV4WmKp>07@CzGWl?m0Xk=1A
zTgNj;0b3{b)0FX$POBXztVq}#qwR2hL(kYeVm{VRG6(!@#Tswm8p0G>mF49ey|l>N
zREbGc08-#yXV+x<Smo9?175gAM|WT3fRD_q{$V$opYU>vOgDL_Q|?M}C*LqAT{=Hv
zkmG-|*jOJKs6sD;kv|YGzW8&q8<$G9VYT{~xb<il)AI5(e%$`zQXC6pkr|V~2SeGv
z3Vcwj4WvZUw)F*zcs&VpueWF(`~)B*hY-NG;aZ6UHgS1)T<Mw2sko^q=YS^vq0h{$
zokn4r)%;w`l-B0T=(BYp&yh{W#_1EI>(KdnXa9ha3!!mobggHooNZC_cf#9ev-TU6
z!5cHxhbZrI{Dj$2HO2Tt-a5nSCc`4E3kGpKb1Nwj0M78M=(8jkAPTfRXmji2K!D?9
zdWJ5{U^KJ4on6BgN|4x77d)oI{ciuhTfF-{F-MTUD}q*y9V1wFeE03`e;66`J%t_{
zVJrHE(D$Js4L%(VyyddH6a$&bNW}-~<8Y4-Szuo^Mv+xgRq;DmwM+fG@0J$MF@CCQ
zPqte=<0y?Rub)!U?qgu;ro7FqsR+m`xNoVXBK_qD51Bei^v&6S)nhy>iNX2w$Kiuw
za8yKCiDMmRyC(8yUP}w>@<>J|&LAH{7UN+{CZUdl4n3gvFz*`}eyc3n@crozZL{~N
zC^{jI9|Knjxdj@EOk?jZKZzLkAiG4BR}VFzG4<d6@~`%Q1*}1Qo9ldei~Sh`E;EJ-
z>t(Dw0DjGBMBJ}HhF(cL@aF+TIM&h5M;QZIqWP~+=NDCcSh2s3bk4stIiPUB2PnKQ
zself;v+7^uMJ|YW4s=%cSg8M3drfiEFKVhn@r|640E}2Wsl1qe-I8ukNlruE+SG?W
z|6Y8c$rWy_j|(eSK_Vep%t7Q;QN}pB^%5qEOJ(E%=lhxK-QnWkN7*|L|DRou%HQ|&
z=|Q8gv;@Y}b*5YFla63;0O58T3s|fus<QyZ@n;OunBVn_KbyND(G}lYgbE-W+@VIw
zcie@mX6T;o6!@@Nz*@4SB5+r|06T{=u*Bx9aca{0?3>PO^qM3HH&@%?s;g8?Y&$;T
z4WdlB$vc7@3rfWLnHP$T9LvbdiHpgB0s>lCOA?3ut8f<!^n(J;7y4`M95q+60loLm
zl<?MRt2bvGy2{8^0U*Zpjo>=X!cqp(3Y27~{M`ptH&y(uttdHpWnd@|V-<sTulE`!
z<HGeIdB*$pewKG7McWGJ7EMkp&oF@qO!l9ozZVD7-sHRH{W1Q8qeWCDhWiDq1pr2P
zfy6n=;+<dP8Ip$tyJ5Z`Qbj$eo)RBaqQ8d<^AL4RY_Iieo7gec>d95n8?=X`ViW(<
z-SQOg$0cp&qPK$i_%%nJIjkIYV71rpNBFE_poshDfeNMyQc`rS?`H5r$X)=B$zWsh
zC2ehwTfa2jDz?JywYVuy+Sn@shk)ehI-BmwGndk!Od`xe>tn4UcWGHB+k5`Brs{Sp
z5~^n{Gi`w-xIwQkEE}lSpU*p|r&bzf{|dfUPZeGT|KX)V17_l%@Hga(heomGXc((&
zbiD(8t@W0S8}w&XRrH&u4NCb=vRNY_HW}_)fZyKIlqAiK_^NpCJG`u;<=db!tGW_G
z5R$yEuf<i-mkeC83~OdRoMwP|>F6Q^z0*voLv}ELYkQmSxqly*qsuh<NkNk3?$`(%
z{?KPAB+9#j79@W#Bs)}osibxfbf@0J$?XlB_9mg;S=V9<gWed-^8t`7T*+81yviyV
zh>T>GXnouTr{e!Crpz)g@j#w}@x_JFz{cQw9xa($*&J0B(&6hQVi*treqA**gc~GU
z9nsNAcn&0JkOM7OH(8e3wa-;?!oEA<AHkuz=9>Ke)nO00Jqk0r$nbFgjcd#}0-J)Y
zBS8=2fgHOT(|e<F{nQ{&S?)6I?=J)2Uj%;xvxV+xk_!AG%RbgOoUBnY4iY7kw|8uQ
zmPp%0LtEW7VQ^pgcl6JkSbf9Xd$w9zy$;sH27wy`8-ig2oda=8Bt%M*`ukTfh|#_8
znC!m&k6A|t<+n5q#tFKv>RI_U<@41;D5)UpC#Ybo<xQm+5NRQB-N1|GImp-8$`j_d
z^E{42oPW`hBvo1(A8$gz1XB5YPl0KrK|TdI1Jc--gX^c-AI7XRV>Z*6rX+dtB9XKv
z2wEZ@xDHT;x6L;lNjaG&z(nTEW@4e<ouYm2bvCpOZt#DN0v60^U{IXOD%TFni{RB{
zW=sL5)zFwvZ96pC^#-SbzJiYo#Pzx=>Vmd>1eLWeUo#MQ(*z-thoH=lkwlPbOeMRb
z&;n<2o=;Z!?CzufdB=d#--ixB7v8N2sWu<;9a6yW5e;Q={@!>od;RyYYU0Zq;=z(3
zvX}2_fYF4n(~b!LK=v7Mk@jx)uya**aSE4ktDP?34&M(>*juG8XJP_JV=(R^cL9TU
zWq^BSp`sE^_NZr&{X~7V%4E9t=WCw(x##SwCn%mXl6ixe9#}*V?t**pezt{(y*YL$
zYzi2N3(PnA9^0pc%)$P!!q2Cz<0OyM<u4;Ik=WfS?p9T8J=Qe};cv_$JqIXPKI&*L
zn|y&>Fi9s!e;)~@75Rh@<d@7w`<wRKt~3ZYS{Z}S`+l+PYB=OqWhG0s1NuA4PFGwF
zi;<`x^7rdow=Sr2VUR*;pBlx~VgLs-q&j_&N$8AFL82vdl;5iU7{4kYS-gRH*cW;3
z%Sn-cDehneVxG$Iei7Y-J!{D-l+c%>#$8yzwFNl23C;-WCfh-Ext=uvSZFjmB&4ix
zF!J?pXHG|rb`)h?wQZb{uD8#j!!ebaBfs%4)&`0ZfmR%fRpy=ZHTqz0`g7T1g4;6_
z`QM?h7_|2T-c5d-9q6&mu6PvC*WiqttZKc(720ZRTImiKEMPw#>qUjfEh*Zz4;a!%
zurVcadp<w<ZPY7FCv2nA1$mD=>^nie))xzboxFsVU&^Zyv6Hr$0HKHD%V-ChzCqyF
z0tcATg*J!8yHX$R$tMFH4rEV4kznksV*|!|;ugj8NXa?eLh}3teK}gvAe5A?SlIEy
zAgWp==CXYyWLc0O{&0PO(j=PRh&140#jJRCxVm9m{p^Qp#EOpxfE~G?`zFUdT!Mqr
zkMW%)_2b5(Z&QPmLFjlm_1IH!^>B+V796VL#jkt!kE8YZUn`pzy0WeVtG+#$C1(#L
z{z9vR4}c5xl=AK*-qkrUk-mqAVLzYhg;!r6^bMF$FGi=QV6;1yDwa5TVHYp^S>Ld^
zI`wtpt^97UiVY8&#G#X6cOL}3l05V$>Cexq9`ceMVjT?=J!_`Zq7_iWAf@X?jP;Cg
zJoAvITsvUZ(Lrq*yRsPbjS*$1vR9@UIt4T5?rM=cd)nMX3q-V9gl)rKIe3+J3xYBE
zhmNVKYw~TsNnZSj-|W4(T0RjCF)k9jI4Dbl$n7b^h{TKii+?dC=|3M+M)=2%)}+*Q
zs8K#li~>stu!|#o7(jfKkM-#rtt~wQx6~bXq4E6to+=bBl-o#1wtXDL@Z5g)a;yzH
zUP)HH^=1}1rRHzKQ5b!0LUtXYv~art;l^qGx$^RCe_rNi1CDBFV~J9H8+F=LK7NaQ
z#=T0%mF-R4LpmV!GW_xU)nLGzwRN-i7q9d?i@?>^&|Y>csb<EnMs6FKDdw*%Rn!6r
z6|%m9+wAf%cRxIkt<1OT?P*KZ6v6M)(lL~@Pcr20rYc}#Nqf?C^bYb;=v2MnDGh_3
z>GLhuPnWlp0<y=S1;+P|r;cy$G5ExRC;a32I$4(>e;QNML+EwPIh%uae`?!EN2%mo
zONE}EuBn4pseQk)tvdN(Uiz+Uey@)B@~_H+sr2xX#)gZEx^u#^tXnBA*_;D+?(pji
z9=4O--9JR<^S9_`_lxE0^5?8YxGMHP(zX?~&>dbTG&q;MT^dbyP%9}fErVpz2Jd(t
z+gafmn;C2KRj%kNKW!LH<4fCSB6yRtU5wU`%hAsG100>no~n>nJ=I;+1bIy!Un)2+
zKVuvWbyv1rLC=k*?C>c)jZ&hS<-_{X`XWA<35O9E>&4xLE5nF``ti(MvgqS#ZenmU
z@`Ytsc0h^sF$9dv1MCZ01MAk3Zn0duK%CRbESZT4_7HR8Sod7JhX9b6^)2W6;0ek!
zWgq#<MzM-#E<4Xxe$_~=1mH#857J$T?+yPwTad-oFSIBl=B5e_a~Bc$^wPENhimP+
z(aisMJ&3f^h(F!0u-cej5Wt{YvMbsU`sWD<i1h4|QFpwfyu*9GZfg#X!jYbv{v5ic
zIashHOdXuL(fV5ddU*0^8!bO+mZyD?21i-z<}wkSNfGJIR^=Q|tw!ZfZ~Le`5tv_^
zIYf{#DhqO@VD|!6VQ#c4wOLhE`JAfi+Kh1nO}PgiQOj-}jaSAAI@gtjoRrw6xw#R@
zZz!Z8hL#l_9A>fktZVAyL?{AH+3#OxW;L4iqhs)FSTftY>9WaZpp*fZ_yS@GU=)4p
zCj7%fgDgV<5-{zpwy)ln5diynMYkmk%+j5|{&*^<PcT7L;UHrSJ|rCPS9q`^grE>%
zKt}?RKK~;+X0g{<a}qrZ@U*GkTq4-y<_vA|w#>a^s;>%LO5l;mYM94iQ<TjWp^>MM
z%SF@<#|dpnmv>oM5qGRK+}t67rk3~q3G&S47Mhtf1}A3*7C6k`E9I|05_dh6e=883
z&&&RkgALxIZm_1oDlb^wOY8@dBm8ZVZChP_&*$$MA5Jo^juW@=sVzXonNt{<YBlb#
z#p|1+)b52r%n3we7e=Z~0}W*`I6?B@*_ZRX<!n3%BSV|5f^L%yg1Py<ycN!|Mc_It
zOZhet3|qkVkHLX#9&d9_N}L+X<#*}NEV|6}CKFYMxtsEC4=9!aA7&}Bejn-`)wuK>
zYwwvLv7MEPwX?RNrLKXNSUfC#h0_LyfQwzs1~^QP!3m@iaS$3SvM^B5RyG(h7t2c&
z-dcqiIhLdLaoxgh(B=z#i&9HNS$06Hpl~~`j$UBjD57vRH6bQtO_*2+i~(|~gPR(s
zjD8F{_m0N-wIRhFZQFAKBUmx!iuwU%1)IpfNb;B5XM5}%*#?FS2d0V+`?Vnh?@vf-
zpQXZm5UTW3!Prr)cO3Pv-G(YZb%e(i{rV;DM3v`V>8oddL&Y5WOZS7}UweLJ2!8^X
zdv5Q<(YaI7&qmdgJE^-{&{p4vKhGg3A<{zZ59Pw=KmU1OFwIhW#hH3F_e&_Rs4Dno
zr1mGCsz*sV?qaPX^S6jcd>gIWZNDe<AcQe7#EUq6J{AjVGi_$CVb@X%fXs$$T9agb
zW<z8Pa*Cs4IA3@@k>t!Scf7ehOdH>8ZPMJ=7s{0A0AIMAAwWAOqlmvgcv{jaJ`Lu^
z+UB@RcFF7?0aJuXzl+rZ={Vb_t=$#keG>EK^$on;%DrEb)@9pV=I`=d_Fk#IXmyx+
z=@1g1JpEPudeNH1E0M!v({()hrK{C*MMqr)R~)xS@c<}WALZoaEoad-sSFMkJTRFT
z5M8qwDv=g2M}u+ip}u|0H-wCltqE)uFB~yRuVy%xq`Xuz3X}4DYw|hM>`7S)m#A{a
znXylIF3b^hN_=5a1NqnW=3h!G(r8#Ggv~(Zn~{tck|nK4ZDBQM7Mg=-$CA5JE!H@1
zo)b@cMN=<0@S<*g|2{5<p@X+{FcqAuI<J?h(=HwiL%R_gFJYJ`{_0U*E#(b+Jh_w)
zXZY5&bFWv}{4`K=MB4CE{`;*p%-+n5`<715kU&Zo_331!E#vzmT|*F@U&&}Ix}2`2
z!#4O7M0eX@VbkmSLV|9u&o#9uhrM%UoFu01Jm61yaSSv$Ag!2XdTBRe1yA=jSBVz;
z^rhp~+XUyCEu$Rkfkm8!=P|xpj)=DzE*gg&y@mP9SVsdK5?K<4BTq_8YE8ByJ4oqa
z@@i4sV9%Yj4vCH_pudyKYlW{hg5h)$YE{h4O7dO`NaLO+3vS@;Oq8E!i|yPDE}iy}
z5V5-?d>tD^z9>UUJy&UtrxWUqT#r7wuCJGNv17tPP*f<F%W)U@pz)VR+n)H7ql?(_
zQ+?z}KU~!qcm51wo5lL?fl5Y)*(8vpy!lyyxlWZQ)<J)+SHAXQWj;{TmNeq$4;Xss
zg}m&Kz8V@eSoMR_<RaUmd=a;BU0V3PYXY}q?s)q_!O1=gE33t{A30yrJ`mfy&zU0>
z6Eu;ky~RZyZMi>0*yh6ap8WaOSVV2l<z^OP8Ru|+T5q%-o-bBF;mm=i7@c{n-hEQU
zV8-;5BZT=)>yMUzeSzfaZSm5yebEocUtaVjy;rs!kG2k)$`jV>>9xYTV<Ca~ik3tv
zLH8m}uE%reWVU^`m?JamY!z!=_B#oeUXT28A5lgdC)E4>EN>m$Eu}k0?o1h@I|quQ
zdM^Kzs+u|LE~KH$<K=kPvQ<MoPUV=JR>%fk9?kv$UYW}oBIBwoZ5!d05-WM^3%d_!
za67C;tF+|o!A!ydV=w%C)q8MlHJIWR?kAs|(KtVa)Md|E3yG7KTzP~h%$zb%*X@bi
zSRvpf%49!`=9bZKXb&eNTi#IHbkqACl|0|JL8=R_HQpksXg+;XiZ22*IK;Euf^uV#
z{ns4lcW5B;#oAteG~IJ%bi1cIg!(BX`!i<Aq-R}KfcuFZO0J#%{M~R=MJoece6Q6e
zr+az*tcj$EL10{GGPh6$oiiB?M{C#l@8Y#)$&bd#!uDTrNFW4|l(?Z9^p7T2?yIIM
z?N{QY64H{q_16OQdwGsHA`^Rw_1JFITB=_R*MCYXyN<7}#p>-Q_^8!@?>m|8Jk0T(
z)cas^rvp^w+4fokX+i^N)LwZF!k3vEQBiDFNGaiHqF^$odtx{|qI_L*L4Uc<{g6CM
zBAS0)@sl2%>@Klzb30^Evwh%4B+CO7HAL_gri~?bzpqsHRA3r1nF~z^nFg{JQE&yK
zw?}AYf7tHV^onAsEQ4}r_Y6)?fsO-}wEN~NndzabF7>Ym<Rdki;xj3FS4SG(7{wpM
z)Zb=MP5R;<nqoFNBPBOeH5^;)yFbb-;b8^&fHeD1&VVP|!uV`CdhaG?%J806dhE1_
z?o8H1nA~pVD<KUy4k*+qM>X{zB8G>@HXRzzTRKgj>sCpDFT&27nE?-c{^z*>eTD_V
zRyarBH$Pr!&wC-f{bzaXd_UuKK`i&)>^Zjgc!;^3sBzCto=&0Z_B;L?L%2Nemd_LA
zlX^=RzdbS;6+bwNl~UwevQ}KY8ZrN<e;pN4!_b&&thPyyk}i>?CiEr#YsCfc#Ekg&
z9?=avW!6egSZ8f(|170x8#~PO+dH`JyKBnlIm-_N;f%@|yW-$#Y-)SMbsqU;j0@2j
zH5Fv!$(OC9<*tp{If8|>l3EC?>cOI57!!WWPujAWDk#3`mY3SJ`dIBbZu?j}dbu5m
zxxX$PW)e{%gcE*`@Y}sUM6rkQ@tFdH5g#ApI?9VuL6pJeh^6|<qMdGK9<bj<+4LkG
z3GXGi)EvV85;=k-s%u!pEC0Im7ob*!(s=U57FIEPGL;B5n`-+f31Wo!*3F$aX{$&D
z<5A6xoH<wp#3kdDJQX2rt>cM*`d*h}&w0t+PC<qg&fKK*O<qUxY1?_9H}rKjpIx}Q
zF?k41mwV2%1Q>p3jCVZ2fVm5-S^KQ$c$0b+3QFr~R^>%GlCj@GoTt1eVR?U?n5q<s
z!1xZ4_PV5B$%`zCtcKQc<bDaYR7Bg<Ss#<rG5b1Y3$&p~ufD*5UP*Xk<xDOqA#)wr
zBn>JpPkJ{~t5@Omb`c4XO}tx6?yPDw@=o%<Rk<%P37Kn}ZhQ!aV#|oCmR2?`Dx3c~
z8Os#50F?x%LiqUjqS;@@%gVAU@4ZsQt5pqdRIixP@agVw>ihl${8QD{n8SdI8YIRf
zWMy5A+oS$0Sdkb6q98`M5v$^Hx<}j{(dSL=hsY(`YmCJ?Oe+kjMSg2GZbB?0a#F8m
zQ!8K^ar1KRI%YVAF*|!%X5{6-lE`Y~_2W`Tw*1C0z34^B5v<vfUX@m~D+TI#M^rcm
zM>pe$zv?+^#`tYBUA#z4?;oigiA3o;2yxQ!(hi(<W5)9>SVu}Cw_^qi^AlTRCs)xP
z&Y@_o;rxvL((P@d;g$7`g!7kFhd2Jh-9`u=s=<SLH8UwbIwJ$?uHl;nq=*z06q&iX
zMtdQj<Wy9gRh@Wza)LkF8UK~a8qia0BQhwH$Vt07`k@=n(xu11N<9v~*vd*ZOg`V+
zaC#C|zt<Um!!SEDKf7Ua^ti?|l<_kTl}`9!Psfkxdt-WBWazC*(DwY;ZTaY=YrAM$
zn`@;mQzEAgPW*ZEPg*GzIWLbKNGl|S2D3L77*>0>S_zzYIC;D6dts6}bhx6KvIg2{
z#cxHN&v5ADwoUbX8CY{D7xAzSUyjJgEHA8i3e7qdB{|9{UD3Vn6;FeURHPbjlFd))
zzjwXI8w}$TF8!<E`KlL8aV85hvxZUVYM^>ygi3eJOeG&cUovuXsiOvi2CGpOQKKUx
zBQ7O*!<+dVH0Hd^vhwoq>f<f0w5N=+MyjVs)y&k39d~&Vkz|maQZ{z?=IJ4=2GN(5
z<`gM;Yzi|O(c-UKpH|9XXfN$(UlZ-EKJ6l!RjvCfS~nk6SYBNUrK@$1986t(svo$w
zy;F4j?4;O^yN84Uv~YMKThPkx^N6)K9+Lq=kT#-@_{^{Pn^u;lO%7!yR!*Y5L<2~A
zsao=OYgI^RwJYOp(3kx1jBPkOJG&ClgO>#YA?3v(O0ySzM-f~#?uv7)^w^?!Kd`$8
z8Lz+3CVsbfA^-ESg0uNXkMl1>V*A<qLTS5jlk(#1VjB!oVRn+Gp`6^@FOb1M!%V4j
zP`yK(s~a^*$*+8TPyehS?d&S`%_Ag=kqWB5&$kHcMFZQAI<z;(MQ5-5o`hRIOD%hB
z!|c!(>>ge|$+ua8h=QVQ9ZRhHv@7Y=73;p|X2(Vsp{HYs4&hEl+_BFM^=I!@<CnWR
zj!0n!@f{{-+9O+9+mXw^(7a$kk(5mE3U~a|>#Uu?YV(yY^et4}pR+Gp-u3;pNTN)n
zP++*gE1Mt!ndroHT2AkYWG1xhVSdHa{RU(jvBnHpx86fWXvuY~{ZC`a)u4Q2n4>1*
zS$>cm-vuL|-crDnAzeP!-pis|h$iPpa%=C|zK>^0e#^M1g=@V}S4h{%>IX6##!(VJ
z$M~JBkm`mfC-nC1-~GMq*b9hb&1&8Cc-%+QAx|Ej@H5yGzz;uPcaK5&wg>Lr=QW)v
z<ixM);5e$4>nJZxCo8|j4i8F!eKK<Qx*3ZJdcSL7#KA`2>DfHkQ)4%0#O>uS0^8er
zl1Xo%VUtEn7iasCcfSKl9BwJYg!rZxRVI#?5yHxx#d`wFM4!mV*@h)(#hY2#l9UQq
zdF-CwRQ!@H=xsKlUql}a4<bq?`8&LEOyZD3*7w19ejiWql$x|0e{oxBeqjui@Ne%x
z6y2|Qwe5EtJWd_r(*j?wTxxi#r$+iaXY1AJM_zGeV&Wv$P#Fs~(uJ~IS4g^4V2H``
z_&|*DZ|!k<bQDp+*X$9a=4vvPY=&zaUHyzbvJAitKy};O>-uY^yDHN>2B#3i*A&Mx
zbVi`Nw1fS2HUm_N=PP+Mdr~UTSK}4VdunQ6p#Ci(#=*I@OPuT>q`7OEdm3M8d#KY<
zs`hLP6p-7@&Gf)(`aL+#MD&0N$Q@l*E4{A{8<y@Duoe5i&*6&L2_S@6Iw9#F3P0aA
zaCT?XzgB)Bj{lK`B&-Y8doqJf)0$8|RaM&i_4P47O1_gf6a0RA<7k)UUmb=A?AWRI
z=UfGHpY5pVU0ZLYxk||m8Qa)kK9!zAiL)8>tO<8~xBHtzBp&ws0j=OE9KJm!OzKUX
z=jQuaUqu>Q{ZoeGYL49<TzIYirhBaw=^V+dT*nh>wI5A)X$3F?>q)PBU#~KXrpXPi
z=P%~otJ66pBG5pgbtY!Muxr|GLYp~h6Bmc{O#O-oC^&ZMZ(g3%UDRDpH+z1m=p^J4
zPjTX-=AJu#e@<-|<#tvUUGkWd($oL@%>SX_0?zKM0&9<%78ka<WA1kVV{@ctCq@0u
zQR8OwjUc_aBH?maS8HUPGd5BYqSI1O0#kTe?MU7!T1?Rfpr_5<$uk(DJLh;&BDVq=
zt~|I6F7sD_=H=d>Fp2!e$r%%$ppB21@B#%ecwTqaZ_MvRPpNkK6g<4CgH(RV=HMGL
zfgyA(h@iYqu%4Y8)Lzj>Q9pTamh0Pn;k5RaUzv;s^uf8rE|Z3Gd4^2blP9CxnPgop
zox@%s`FU~VU4}}@j=z_;_F<=y7Rx_v%Q3|)o@z;G{^`DOxqu>@B)!YCGHQk;uH4T2
zh050Ohb^pn-9n-nJx|@>*;O{q8aIdk)=(5fb?svpY6-eOoXXN80dPOpd7l~Px0sq|
zj%h^ecD9VcmB=O5{TO_QXpj25d){f8DSIlV^B#eiwf%d_d2C+QBNP9!aBD00oICcH
z-$b8&OTrSi+h=;)ZU@D>yxDzHib}kG@m9{2#V@;f7`fmllcm`z`_9d(`G_KNxhp-s
zW3&UoFXs3k$7@a^{6w{+&fnD?xB0)7;U1Y8H@Pmi3+E}8R~lVvz9E4;2PRov8coUh
z_&RNT0CIjt!6%K`ox{I`7x0lnqI?>x8E1(#PhP7zS_PenByhq9Ccdrv?v>2YFmvD#
zfup`AIicpR7O&1<&yqH&1@-Jy|JfUV97&ku>#nEUEiacc@}aKJf2jw~#;GLzhvwt+
zJ}U73!F->?D=w8X$;i@V_sO4>V*fn7G6)Rf3g?DX@6K>lXBth8t-P!BwW>U2qwp(H
zIJ9vE1wsba6Q;3JSbWRa{=V+%HBlxt4d|I>ES6=R9IYB0AUVX|tZISVMZ#`Q<gd2z
zrKw6BV0qPnF(al>a7A5f7(yo0lKBh^w3-BlpKbqe7445u@%`JL*g%v?wEJ76oV;&u
zdA^}<F!9#EtjtHUhP_@E?Wan6ja%PfAUx%~&LnOvtF_rvl3_2RK4iu>c+*G7{%Skl
z9d77nZh#AU@U;O9k6SxmNWIMyTi=zkdrr^3MwB{nZ&Flp^a7QgeZjQXQL7Cehp*%V
zOs(EY*Ezjp*Y!r4Rz(nyYm*8ILsFShcS8<O%X0H>{+grNUhbM&kZW>;_2|n{BW69~
z<<H>|*S3cza4a*OFXGR^gJSArt@8?i?>q&kM#eI+&(AoYSUuBfMil-Iy2cZ^<nYC=
zyMQ|4PGW5)1-qz5+gCsONStaE6EWrcon5PHXSZHDRpc2sj!M%;%UrulQj||~jeTka
zy5~*02mVzk^2yJDKt$nJ+`(Dokv*gg;{@@jSgpN&wu3RJe*fI?40ga}0+KlrhoGpI
zP+Nj2IX=q=E<H;bR}z6V!t{u^bO&1)zbGGLpQ?rx{xRC?T9Ts6?W~oIz19BvSx@jA
zE`zt`N6(smj!3J9@b?|&VV%mtc-q!S-r2f(NviQJt<iL_N5PVyXKL?!wbTO*b57+H
z)!Og$&4dSmBOYdtZDRp$&gKvER~tq=YFQx$n8db|#dP`oOxK@R#5N~A&Mg-(+uP5R
z->9SBatxZ1FRYY3nF=a2cK-qv0gq1<?<0@qWuP6y`X0>w%fla)E3_7ORQ`KU;0oS6
zPgfrVd7MHvYs2!~jBR!%9)CVplD{UcT{?uoRh41C1BPvGBE0TgHQ1wC8nRTaIKet;
z*<F1;xW4y#=Rya$otfYrsK`lQSEyu`tmmnbX!$haD~5cUYy=Wnc1r8$Awj(!(Lg`C
z(&NC7x!0a6i2^i+?zZ=-#zGu)%<e*70_u)!Ju{+0J3MI2u6$xkBty&5iRD3kb6x`k
z=o>6Y`ZuZhIk&a83d~2tXQ9Q{EU%rXRfldXq5yGOO_^6)StwT7QL2b6trd*1#Yh@&
zn``|Ap2WUsd;i^?O+Y#6%XVnZF`1t1z%0ZZ|Mbk0>gnGd4vI&WhDZ6$za8sQT%!pf
zVZMuGrc5Ai<i^@KPD6J4QDUtM_F=*EfQx!1auIG(a~-b8);BEnRSN_=fdYlL5Y4K%
zr#3H_$506xe$s|ne)4b+;0PE+0M&M2iti4oeq+tuPu5Az3xT#zGW?Y+s=QQkrTXlh
zdcv_#V2B=3H5gh;@_Nz`4J-o;T~etZ30{hj=0VO^!GBQnO3yG)R@dg^Zc3e32>L^w
zm@-J~kF?AT1;;2|Ny;jJbrC{ItK`$$wW_$PUPgc8!#_^XO@NhsBJ(!VqrR42Us3A`
zB{8W`F%qucqrGdctrxq$@K6zy%GR+*WL15Qd4tZAtW}BQ8L~1I9L-LLFOQA?kEXAV
zs`B~120@S(5RjG+NH@}@f^>H`(%pHbK}zZF?(UG1?(XjHhIjD$d;ef9)?)EI_s*O-
zXYYOX%-9o5<5es90G7Kz>JBxsS2eSHP;}g5p3_i-;bufEGkd2|zjkn`ps6&Q=ZUv}
zaODt&)7o-Kuy|T+y>U0|Q9wy)mayXcjbI`9%@Kx6eMgy!WLJDEWT+;ZgEKsGOjxph
z;-9qGIr-)P;?nv`HbN=-BWLJQVFlTB6JMzzps-{XT!L4l@=;sw(x>ulF6}7F>|jS>
zxY=e6Wc;ald$}=1f+aIJ8leQLiIn-B9_D?=T#b!19*dj>HL}}h=R?oNZ{sB3CQRp-
zJ(G){_tz{QtAC{?35;zCyvy2>OVRdj4kopPP7%DY5I4Ryp1_Vtgt9($fKS-E<7<>o
zN#_ea60f3V#hOf-ag7^*p^DnV)@6>1y3ezCGb+HfzOk!kqaAiMls7>=;CJ^IUMcHh
zb>ndErsbuNPND`S`fo$;;Gf>{*KRn%KP$sCO6diRtX%$EKR%P!hdycW7hc8A+NxO1
z{79gR9z1K%VvHXuB|eSti2HC8HwT+lT_8q1bG1qBIIJ1oQgJluG3vv`bdb7^u=Do2
zY^&C)XVfN^{I43~)W8(BE;wthZ}r$b9S^kXZRT5@HKgV@ItcOb^7L~C<J~I<heFQ*
zmTA+30;4mm4_70tnv}vyM<4keO0(0aahG>lJbdkgO*OPLx+wWI+qY)>gVf7j>xZsp
zSSXlc9SUFS?8w)y5biGPq*90C0VM!Y&TiEn<z|L4;i*V}^M|eWzrD*EDF3&TV4~H&
zO{9f^uhpLq19>}Z79sf<Cn+>`^@vI-+GQrY@Id#Hg(cuf>r7X<7y@lxYr8SVla%7E
zi7_708XPxC5@<9;o*wh-TyBW&x<#s->=;F*(tjzgCN0|M@a7I}pQOMD9rF6QMr#{x
zMe4Au7r{pSpUF8196w*om+*xUF-^#>A4mWq!Msv?OWOF+E0Ur2lDk$S+i>%E1(&pt
zRd`SWpntf1R&Q?%?dHyaM=kN(DHNmZ<rBK|;_ne#_>;hnQ)bnsMDFn<;uBP=<%De5
zA>8_+#N8Fh*`wBk`VNRM0SE!-D0PMX@m8?k$$cH#(`}C?J1fKo4$e)$$?2pLK2%I}
zh={hWsrd&fjXq}sd6+B<#rS?02}%rO5Ya;XqTO`c)T_FWe`=rx|J1yKQbV)6;d`NY
zzI7NP|A(-c*6(Z$zR{lcYg1%&TjWhlO@~R$ox)`2Mk4Ib;$53r>s6r;)9V6MK^c^u
zPhze9fkSJ9Bkasa59Y!CXTrzXfmFW4M6KpXP^r|bk~<wvY27baz|a@FN~dpISgjv8
ztnMr*HRq0)jorU!oXxGg!#Uxxa&a-;bjM5TR<;G}lFc(WqjDH*13Bw?khK*(4^Hog
z2|za(x_bMgm8PppYm3vxhS9FwKi@5pB`v`0^4<#k=ep#%>I-2SvDp}Zm(^WqwuEUw
zVYvoK(Z*5gQxK$agzMhP+NqtiGcn_0oIvXwTC=;w;PMITK>G?2N_KiD7DqV|pp=S>
z+j<?gj`xtwl*<v#8EaoOiSQrG6^xJ0Kch<@By5EE0IF7Resm(578X6B@9`b=)UoO?
zFf0G|OU2(Qi{C4`y;xiOg%4%CHr;qv^)kA7;)&@@zV;U1bpMj9DL0v1-WlHJchcYC
zKy^I1n8^oBuMw(L2#6kWnXMjPiB;_A^)6N}R5Oyk(lUloW2Mo$jR%4OiX{a*lh1O>
z9b#mgNGjE{{)rF2=BIkM7$Gz1fYorjb31MzGRdwzn15@YKS93Cp6`SPAvVi*O)~IZ
z-p3}nUq-&Ih7;{y+hvq<>fE_uc3^C~g~G!MV`+5}_xvj@s0^BM3%mi!#yCQcc65gU
zykd%?oaaOd-@bh@X=V!bI46QaEpkr;lTmmD{{av(DR5ARTA}WO`n5`Pi?OPl6S4m7
z0CAKw&hWPm{ogp!M=6_W{t(!XM#4Ye`mairE3H<iTx-z@F!MtO_%&g)bJU0pf}r_C
z;dFM!*YOEl7+#W@P4@qhZVd%$OIHhOE_m)MI+x%2iLDjRlV{%cQ=aZDirFp~aXfKl
zDWik$jr-+gW%_=`LX?EMo+$7dRG9K-J~KG~_o{NV^LW@ZJiQ@BG=AVaxMdyQ!;lqB
zyC4`Z_M5mH)fAFs+|wK8Dn}{vhvnmqkA_q8o-e5f{XhV8ISvx8$-YO9W*qJm_Rrkh
zk^wA$KE;v%0()})X=m<{K&=JeX!vaj<N+;-$D4L8OszQkI4V47$A!Bq{`JG{P=@Qg
zFi7?jKf(b1Dta`a{T3fUH;4Kez}uQZm$yGiq(i!fvQ5alF=GQ{t#o!Pm-U$wRz}~v
zH|=_JDOGZ!hc8p8ayiEU67Tf9`@}#mhbu=%1;Guc09CzC5C%+B)SKB!U_}l^&1_I7
z`rnRkB#M79BLnxg7D@Nv9t{YHvusXy^IJS3iYAteG;xDj2b!G~({8eHHi12!hjpu6
zy`V{lt+$UulTTIZo%ooR8*oV&a^}NF6lH)_b{#9}z{cI+$d3yjaf%$2z8{BEycD8r
ztlSXhf?f%HE@%+293yj)fCl>xo}B_?%A-9YTu44CS>>3ko}%WwTm4a#zl>m*QQI<=
z6(+Rg2vPleWc=kZ16<nj#IUt}ZG}|6*th4#md>T#NEmU)PjOW(>xtc`Ac8X}eGIKG
zNNmny<0NoW6_7py?!Ba)s()By6Boi-)tU|uL)0?b@cvsch$|(;@$)FPM)p}5@%lpi
zkaJ;e)y$B&(g5Kyc1)w;&t|1Q$K!y4{!+j84UCiRW9D&ugBpu5LK33B)VRFHVEeC(
zmcYvW+ly`d{ZPnpedst<Z&W|k&b)t`>^Mc(IUK>)z5yC%Sqb!8@@R+1uQyQT^l)UI
z_n@FOt?>UymM+LzeLXfcU`yaAGoZ**v=$CLBwWNoZ2SqjGd_IRA$7))m6*RA@bBfC
zKO*mqd*0|1Xk5|<?XSIv-fe^VLp7cjKJ}2%^8q(w?0DfPgVSI_gwXo<5qkZTH|7D!
zkrW{A-zEK|V6|id{S)Xn$BX{+Mv~pV<Jn6p%xKxxcY@e_=)4Gr?vaxB`r8{{x_?V!
z)^5B;rC;nS%k7?CR4Lv9H;l75?YU)jk(YZM2@q)9(t~A>G4lPb6w#Ns5ZAu5@gANE
z{|#Owa1Atk^iMt~G^2?Dv*Z5YLfHqa`DnjJJe$@h()Tf&*4DH1LW4;mYVy~Q@0S`;
zt{=`Awxm-fcVk}7$i9JTN~@)Soux6<M7*>vl8FCxXp9SWSaSTA?zx$mo5F2={U3id
z_9`is4gpEB(-32}l#oTKxi4n;>BH5wB<P0;j9qc!Sm2NFMd=PNh=HqbrZ*@n2L$rP
z9;e^4pA_W!8wq42w=Jg}keUF{wz>y-64A+A`V&MKFgnMHK4wxinE3Jwh8-q_<HMic
zV&NTY{bAR6H{m;$!)u%wUME8JRu$2?r)AN3$mwV+z-GuUdbCXIrG@OP4pXNn@X&!f
zHtLxr$$=@g=wKMbtT<F4-Z(O5lMCnltpKz5Ra-G+eL9APS8D3=!m`q=Y8K8zt8KC)
z)@@dsCrCBsq(&6$<@oiA&v9gF@+xe>;qAJRq6X8hji~Rnsl3%WXr)MKpCZn;Y7Leg
zPiHL0q2Zr1%uR8RrKU?Seht8P)rf@1zi?HX%gBE^q)iYIY-~D^<lmp&Z~zcO<}hs2
zXB#}CJhgJLA0}N0N_r?snr07cvTqN6d@*fRA*9OGN8z&@PES5L|Ho}_SSu4LbRe)i
zWVoi4+nG<~Jg8l6c7_r6NcWcKH4lx3J)wo>3@?(3EXDunwLkzFKt@r{D>9abh=flN
z?MM&+*I<lBqs?5(?x9*m;>4?UTBqll8O{cpdwn)_dtS#>j$cQZ@V0vWBv%h*UgH#%
z`P3z!vt`#;js!(wMedf^F`gYm-)~Z>B|SuGY~q=0XDrx-jt;$E`$;7-WWv~nw6e0*
z31XFHC4CeCVPBvpfVq8X9RL#s0HIZ&p96-CIsM;`pAysmnFi>Y;GHPALU#pnnL@?e
zHH-Vs#H0NFhAi>+_oMzeYl=XswNtxIo^&u}STiRE6h^5y*I{JTRRsB^-NyxYW5$o)
zEKQ9Sb_D&xiNMR63pWu?^1PA*@n-UE=5drm&RDL|QzV^QQ<~Z747U$H-^(tVFkt4s
zIKtQP{?~y2T1Dv?X0)fT6py#8nvD*XeD24}K5W@7qq(#*xKkK~m+B~m%2uo$mm)ls
zt^lUZ2JMQRgB$~g->L9^ly#F&G1g1WZ)kLA&p{H76qFE<_cNzVtuO7mL5DPfD1K?|
z{iQ$4b{=eKGMq9$+QI{G5uY$uF3Yqq=29tLvyi2O>c?F93Lpb@v<)IW_k18&*e_`Y
znk>*07ygHp1F^H&%HvfhZ-y$O_%t4Q=$(5fwP&S7kAj4(axN}<Sv<7rk{jL`ZB!oI
zasnaev$mpL+dgVZ7?}Kl!BWYtp{|O~a<d;5j}Z%)pyyY{5!1?|5G${m6j~XJWLo@p
z?J}a4qRi>(yOcSfpu_C#_VDDe4_ikA*jK1~BLqK7kZ25B8NZ&0S`JxNil`wF(pL3Y
z>bLaLsv=zS<8uoCL(@K-L2q_kT(s4!&A={Do#KI1W9{5UG`23K!xlg6##<)OL1?-2
zWtiJv=Xk6xg5d>Wy`75qw?0VJqRNOy`zYII{$td4(;P-5C7|wFnnZuDX~Stc*je=k
z%|aJZ3d&wi|3pX!EB}6kjrB*URctAOnvLg{&6Q(rro>^gpoc?}3G^gZ8NDBoSPEFm
zw#K)XVMRX;PxSS1Io5OF?op0s6Y5+NdpW9p_Gn>RETs<kRs+HjyqEUsoe_2!7`mhd
zN5NY^DHKADHg$L*mW>G+*NQ+O=c3o@uq;{<|H*rnC?K?<P0-d=BN9)2Oqt;Aj?$hf
zOwTNoL+sM=Pcjvg!Y<qHM(KR22m)k4x3kq!z)Tym`+7O=a}tkc9)}#9A)Iyj>AD~8
zOlA~^i3#Qj;k@+^EQUuxQ6;Z4Bs4-nF0VwT+3hN^3e@Zf2g>L}?b7cXc9^@i`wp?}
z#xF9BhW!7Ji5FZA#x1LhD4_T?<4Au^n9`<qo5Vf|lS9~73pWpC;2ss7Iid0(hZu~1
zm&!dLVQ^tT`l1EI)8g9V{J(#2*0o3(sN_X8w2zT$Uw`!-zYQ{v#0`sTW5A8OdRM;M
z@zJ1@g5@??c>j>!Rm2ELtG(q&J1+C{9Cd1Ji08)`Q^Rm+hk+$1HN4f&WI%}j2HIG#
za~%^hJI4rN5bmvSt}4j^E}rBz=e56FKKS)&^W{WYz8$1CRQ?a?yE}lXuYEb%HAzN>
z<weg@Xj<WvHg&b)dXG2={!smYFF^f%wAs<Qu+5cCVWm~HkYu8+=FIax+YKoW`78&Q
zn=;A9B;}^4dohFbm`069d4$6;gA#ef{yKuZcUtrQ?R6opNXZYsc!d)F!Q<zrFi?jU
z80{(~j$V52f#P3*vY^p%RS)r~7ciDJy<mNR0S9$(csOcI3!zngGg@`W|4B{s#4{0&
zcacE?n=R`z8o$vG0DMf?CuWPTCv*?)Ea`;VJ%#4m&{olBmI?-emf$!HX<*y(-;4F+
zz9j@&d5KT#+-HjEV0MMVsOcqR_<iY)dkRh<3uX^{cF)uv7gs`ros3*1a0gO|rz)>&
zw|616myqr8_1NDo8`3-i9=rbvAa>Tq**VN}W<lL6q5HN>bYMORJ{_UzZIx{~*<-3e
z%RcY-A+|eyFj99PoTO1bDo5Uuy!TD`Y5JZRYP|i6e7D1%1-uh1&Gt^T{>!``uc|IQ
zjz3RwM_f0+3!}F>b8p?oGcLfp7ZuHGSa_F?L+w1D81GM62$l0&87x8|@))bbf*m@?
z;EO;PC#@`xTKf<;Ifn4^7j*+AO;<X+HHzfl0sDB0-Q3nMu^QO_V%iWy*?EJjcyffW
z$pjHsOkA0K=_t$}!kk#)$QjAwXgk44w4z6*`#S)$SHVHm<~&p4Ebp_4ZJ1S8sU+hf
z1!__4?cFs_qzmIWs*V%g?uN)l2>SrMmKBmI);(l#Xq5kYa^8J}lzn|?wY{8f+1h^H
z2>U1`gObK;P+R)7PG@Cp`sgGt_8)^!<twAo(Hlp_UMB{H;4#~0BW;a@AO40WW#+2f
z_6;j_QPLU}F_r23%@wn~4%3Mvl6A7Z*5GqISw<b1@(Z~!cyc$AeH_euwaa8!CRehu
zDY#Fr!4kjPO7WpMB9PC*#HcOrRGfsHBw4ZaDTAPteQq*&;jxk4^2AfvdaUfS5{qEj
zxy%LTpQYjRKUvGyI7or|FXRLY)atTgs3$vxe98G=3#QI@j0994^Xxb>wd2NCs9}ik
zBmp0d+8{4~2<3a1V#-0HpGR+W&hGTP><E)!Au0Z1mh@6e=zAVh*?Qt+AKLRz|F`|*
zl7*9lgh#R%laQtp{borZQ`UP-b>y+)IX-asc!+Ct8kpY97!qi8{H(o~mscHdKnuwL
z&D6`9@Mj17`XF4f;zjt5qQdAR#Wp01Kbpc1YlIsFlC^q<y7@Dy>~g;nkQNtLOYZsu
zH2!FzFw{a*gpNszSvC&}^RfR1Eu|F|P9B#x6@~j2Xp{)6I2gL}e4uDZ+$;;Agn73r
zOf}!rfCdcF8=traaRm*9I#QlR#2?ki^zG(V|E>Njfx0mNlgKi@*`C?!?j_}(1|j;^
z>D3vv$~mqc2a&S;Qp)F9^U&rYV4$dSdne!G`l?vq5mV4hPT*oHJZIEVP+9C;>g>L7
zaRzYeGx_zXhQ#d`8lbyK@o?vE2K;{L$`I+CAs579#%0&q@?A)){seyXsE3C|5vQBk
zhnqRs6zz6^PWLSEMsw+Ehf~6aMpGX9wb!p>^*GAUg-xl0hs}(baeJf|IvZx!RnJ!`
z3;L0E?S|j@^V^yEtOD&jw1Satph7ksvR$IM7oRYAaC0TSKaF$Ak~qvMyjmlvv||BZ
z3$M$TFvDtO(JWbg#YGOejbmn7?nSKQCu))Kn-JS1;#HPn9q9LM1Rm%Ann)kwxwcJ`
zn&}88mj-MxSvbls8DVES>Y(PiaCboc>g7%mm4mr1d{G$%^;y+Xv@%C~{)XbKwzzZJ
z*w`2t8a}eh+w+DI^Dk+T{DVV?>$qF}kHIPH!?idSfS3&|=XPB0u2_|lrAmD7DU>A-
zc=gFq?T&{dB3(L6K;P(go&XR6@}ZQrFstb@!z*Gk%cqFWLG87bz2IOc4!mR-tZ%AO
zTgiEvxsQ2AA)dR$*phFe%ebUe-ya&OXQ2F%LLqfujYS5+rS!xfsY^C1?k1#b2G_8c
zYN%Eu=lTWQWUz-HnwbcUt{<aP_=14b4GK(E$@4WV0FC21l()UJNqq6eAqod0fmsV8
z+Bejp>-eZT%n34P1_xi{d$}T{L>)=0>}2;k=1i-I2Sh1sG<h<$ljM%ys8dMe93`E<
zOXO4iJ7o3TSw@M3ZRiz?>V6o_@aoJg3YGy-@EzuKe$v#`Kp%BdJ#)RF{VN(IQjh4-
zts0qhf=1G8vXbE3%U^Ni!O|~SX6-oPW1HK-43j@H?pGxxB|G~UStQUb5a`G3t)4g@
z$M~WP4}%2i#HBp+Dd*E7TCFv006s+%MZ&LQ?--6d0u^<dm+_}%?WGa$<`Jml4Cmh%
z6o{Rh&ey$m9{a9>(;&4YEhFLE<$+_gMT5l4WB#*PN!sb=qYt1`*sRxPm;=20&%~nY
zsCL%7^bu^}5ZHUlu32Gd88g<Zh7w*tj&t{YpRhxn;W(;~vnobr9&JXya^p~=#e$|=
zX2=w_<eGv;M8!zvD#M4p&1@efsi-8{G_uCBwtMl+apr^F(~AauN;k-CXZhu+0Yw%K
zCUMFt?v^;H$vb=|A6iDgga()5Ce1l*1t;ayX?w}rP~iP*Da?@5(S`2p7(?ml>jwq>
zW_VYsD4DWCM^B&Qc6|hHd6|AH5e_H1*cpX!bBiZYeHqT0QxmK|J=^5WXZ1m#R{@MS
z=}Lt*rw+BIRh9z4QTqPmBox{CM(ARa^_@Y=Ulof#bR(EJPV+*Acql^FGBkz=cpcI%
z42qgchPPbQSU?p==gq^68&rr^0;2xU=3tm1t&%+R_pIs(>RGg~pW58GA#NNRVkVa5
zWX_2)CP(qM1d<M6Qc5xj%I2pylqF9@$S$@70x1MQ^hu+weIGgM8(hlGCEWP8dm=>4
zZb)PKDU`=i%P~#y(V-Q51LG=RUs!B9wa!$?Q*XLJ0Pt^hc69jU<xw^`9ZP|Wt9R#X
z8Ndb4d1@k>*INc`iKF|MDD3Z)o5`h5hEjMN+3M#D6lrQK=7m202(2yzu>?##U-#2G
zSgM{HY9+rEPp<5hhOCY6nY+&$FjKLLI7rHP{*l`4FxbzMq{aem%}`H*B^A5xV{Dq3
z?$%s(8pxApt65aIrb7h7<V{DW!;R6Oxaf}BmO{hVF_@ATr5lB`mZ?AI{*|$qw*ki5
z)7Y^wIdZ9_o&v=}dS^Or?!<!wbMSfoQlz@S`ZLV`JP<{NB3(n%IhoOm*>t`=T&k&6
zW&E4W=^ypA0L}}ui8wKoH1_0FYnq9Up|3N_$n~r;mQd`3uyX<oXq)^LgW88kQEDBI
zL$TNSd2<aNuO)v)@m5Ngc!tt`9FY*S3+AYyPXT>Esv+96f~;2nDgZPTL{t<KQe$WO
zaGOU}`l9Cp+hqnLZ}?AdW`YF*8>EZ>Dy+EYkdO6bo3|@X=eO9~IJxB4=elwYQ!;b5
z*7k>XF;)A@?Mynmv&Bdi@N7_8Z6oFZ*TMkivXMq7t1u#7*LndHoA!<lx`bmfjmHF@
zmvvn}w>gx|&F*mhkGUul0M(2dv42Gb*WSiQkM|)C8nW?Y_1LP9X7xiExwPn!6tXvh
zmrkj_KsHnzyylYcZ|RG8Im7^Q{OR9q$Owjc)p$+qk9fo?nAH6=-+>_^y7%#kO^@r@
zfDM=-XIqVFP{AWWfV)Z)`cExY;QifZR@j}reM~jKx5b*wj>a*u@$P^ee0e;Rwom;O
z*cKkzHrrj?=4oa&PekuGk1u<dF$de-Kgv0`PGq3&u}>2UTJ6ANP2EkuFnSd>v{2eI
zD(5N)&QD%HALcKWJleD8T5>z)wXEO7_V)H3F|Cdp{0@AFJ-xlFf4pF5wOa__;RE^!
zs^7*cMN&8meEljSCP(f7E<Z!5!hRW(y1Q=|X5GKB%F9-(BJw;qWpN+((){rs(!wcx
zIlY77exg`s5Y*tb=eXbSY2H5fTSPy@BzIpn7u{pd@!|m`pkPfr>hYw4uN;ArG`1gE
z@Z$JKvR(h}hnR-@_ejK#5~6+FCM$PPJL}LSVJk(l7#H3By6Pl~>WC`$L;r;p&2(f9
z6BX0#>IMO`1+b!s$NNfTd$%o(EF)zk|Ivis6cQ=*P4<hMN+G}61|8<5{!VFPk}5Za
zW)I;ji-4g~)W@a&C|Q3LhFaY_VSlX-+WAeGST*G!J_S<m)k7ULTZ%tE{%Kc7{C)Lc
z8NKAYctFMy`EFKx<~Ez%uC*BBWWa%SYir)Y!AP>QvLVF$PjUg2orcq}-|B~)<}GR`
z>0@KqY!&{Us@SQh)3dH0?V410wzAtHCTAaqHn0m9E==9q@!9Po0?t|**<DJoe`zBn
zsb?h3%5NX*ymz5e0>cpG`l9>Hsk~iRGYDhCYPWoigPxG67R;h0{)l}W>sYUa@Xcz;
zHNd#W{BiRrL_t4Ni28gDOVmD&a!<g2ukmNlPisa3Ex+5X8g{xAnL@I{A6Axb@}ftM
zwtl7hEBwW}fwg0;g-Wwp3~0`@7+a2h`yN!o42DO^$5f;e`{_V{F_9oPggt6Po=5c`
zMdqj6uH8R;wAf!qd^cK4N7S}d)(X2OZoW^}j4_q+(qebaY+$$e0n{)QruWEMIXSOJ
zDvgGPc&S>fmiZ124$3MkNl686YfpDZGkG2_#svihR}Ky&mRme-Z`Q-jwqq&;fy<Hq
z@d6W*T;c;G<6B{2;kjzF?hss-nM$MZu1DYxt*~DGnj;baRZtN2>go#i^=l^A3*+yD
z35xjV^<!gWrq{b5f(P+(wMvan*SaL)nf;WC)pV__!W$d87aJV(=c-I-)oTU{*2xoI
zV%umSPPVbh^>Q!gNb6r|{|VaWU^5IQV|;W4?6rS;Yi3)e*Y8pi)DtDQ_q#}jeS(s_
z<S1X~UkS;f{xgW9ip2*J`b{}8vJ-~IV$Lvypt4_?-$7Z(fPPs<H@`A@c_KQrboHxR
zxG;x0(uZ0Y<g*;I#$T_;`MdH;A#!v;KmkTE`55IQvb*I&sC1O5SJu;QEV3s$974^%
z{lYz^WQ9DN3z0p|L*&@jrp*njs;XPWe5%CSU4_P7PpaizB|*Wy*k#FVUE+rRrZmM=
z!q%oPs>3`EnKLF>Oittz0Si_rZxWJSd~k@be7<DPmN~r-N>H2xdCT1%&0MY3Z~(EV
zh^Z+x78cgQTy-!-inDiG+6Po(z6G7s{#b_Y>FHSR<EF@E&-+kN7zzpsL?k3ODu?+p
zosO462)SeW`huf%@A}8G#ZOO9b=NwB1`^rBCngk3O-+0H`t&9X6nU;@^nHDOHGX~O
zbxKWO87eBGZEtU{yPQ&wXSX8i?d=6_Ul}>M_NT}D+r7fl_uP)OnL?qaCvibrqZ<fI
zBO@{~w5qTERGEzT^!J<X{ktpdtA-6QRYd#YKkC?)F=tF;sA2KUBI-po6i!K0O(p>q
z=KcBK#XgaR+6bLkKT$0fMt)4sK45}Ru{mA}5VP2AGuAR&N|II-{kVYlrdEYXwoD4;
z`KNacy#ay9_hW{{eT$aufJ*~HDIe4lu`wmH)u%$4z1^_OTww02KBg_Gtylj+!QNsU
zaMsmc*i9>amf%&NyHZOFd=BsLr<%nOg3Ls=nVMgG`Ws3e=fbP@p@uXuxGUL$BGeQ`
zMxP#u-p@99A(C_|N);=OSDPtAy<h+}HMPi#``z4BW{bHhE^P)Xs&w^QO9101XSe39
zS`7!{m=qKgCK`(k`(v2gZ>%NA$G>T(^189D{|z-OBUWc+ld7B@2t+6C85**Txe**M
z<7`Wej#k`;fYh}W%D$AAmls%8#t0V8x36Dc#mC2AmMAZWlD@zWo9sUEzKx!O=9LmZ
zlhqgsr{S@Jm0l5<!kMc{>Vp0~;@7tM_I2eH9jWzL1tBMT|G6i~JsWpdiG94il*ChW
zBgA}SM`>kii=|z<GawQn0U-Qlj*7IX*f7_h;y>sa9yd<^y<lR{=0j2E*Ur~~krq-*
zS>T<9SzXCr@99&q4a=FHuL8&wImsqcZuq)B^F*;gF0z)Mf@b9w_U7~*moT^GvjLYD
zI-S|8Ch@P?dJEjwZ>L!BBKm;IOC`E6s<`ewOIh}3s%#A95NU-EVE2NTkkhU3JQ7J~
z^wFfUcCltySlHIAQR+u}`Vv`$(b{EC{7|6|_GNbgSy^=Q$e_JFQvhAf`akbnx*j!O
zF9nyR{{G$ec)g7FBOn0YZfk&qg{9Kb-Wu#r0EpIbXCxg8AO;^lBpj^~X<J+7aK1a^
z!{PPra0))Yl)gSO335z0AFqc8_o-6t;f>zN;-%r7Ko$V8NQJ|oM(XX1z}u+k=(-G;
zC$D}-M6{#vKYW&u2$^L1nWKBo9d>eZ0vGTDMnO^WbVfg1b)j@tqdSrOpW;F@0V*(P
zh0}TUWog#DvGMiNQn2dV?u_Z^+iAy6dltw9`t|qObGs3SUVp6mkwHydn}cF`483~R
zvRcqv^Tu6x)f~s+#`Ptcl-}v48nu)|-p%rB8Z&wdT}E<xI6PM4Pm$Spk&6{6oKKwf
zhi{_0Hv0@r(WPoEy1$^T!>rDKLM;vKYSny1*ahtnZ=*ak8m%YXj&@p=1Q=ePsJYf9
zu8~BNTs*zt59!EIputG)areg4LS5+9lkEuS9E-yczv}6kG%#vm@p0r&04F{|aZ)sc
zxfn_x4h}aGNh6*Nxg!4a@_o6JMcTp3F)zWO6!-&B8UaT6O|#Z8$VwgH2j;Vm&WCAj
z2=B{Jhd&A_C1uZ==k?*dla{15H#fKI#ZKE&v)k@$B?UYxfzIXLWL9o2CZ6k4Yc-3}
z5bU4v@yJLD>9tjxC$d6RR8;%prQ~X}X=YGarixTM#2K19vqU3@%&x)K)G2z%NgF)R
z<%IlpbIGXj$=R762!U!*6|+y@lD1>~R4NP*iq)zc(!x(ptO2g9KOf>Q4wL;9O3pC)
z;UBUeimkf+RbicMw9d76?2yyyZ`Q6UmTY-6sK39IHJvOlYGN-|?_VmndPtV-znf&w
zq3{!QRyv-oUWFHV7JvKbw7OuoF<BQ@!tEA@%378zq=f_%bs5Q~>CO$AzN~@fX}CgZ
zkj4vC%Jw>ZePUQ3r?uWn5!E|^(Pw>mt}_84XY%NxnYOpRrVUMwVX|5pY(7u@SfR~L
z`H50nZ$QkSJ*=a<w~P$Gq7dO{&<e3Of7`||@$Zcia5+^u2+m(1i!j@6+-6%<R8+)q
zJJHMMN`2(ui2LFL_d6<TLI-la<YtwOS+i@;)pEO)IF~wIq-wXc@7kPFZ@2yG1mI51
zU!Qnm)rz;eaKM4;OiZUu1||*;tVr0jP-$k<B|NtqktlyHLUVqyeERe*R0ub%46HNq
zP}xs}iNYtA^EJg(>n{Y!x6TwRnG3}<`jC`u*mUzk8A2BVSX!p`9sD%E8#h%WMlW~F
z7ZWWeaIho}Kbm)=AO->Etlr*XO1HUESbiqlD?D~Dae?i;t83ZNBxjl#bbHgnlGn9R
z(w3f&^p<g)ejCXMz;PfOqaN~CD!4Eeo@7xCeTJzCaEpH07yW7bDe`RmBWNV}Hbt3J
zBlcMX_X@>8c+Wo93eiLl13Z}Ri9+z%i?o^(Dk_+-ju!h<c->|y43YMyiVG!0EN3hH
zri#@oZ8ya%>sI}_?i=<?8u1pURDo(!ZHW3=lsdIop4;Qjo(S@AdcHT6qRDsf+c0We
z<Z*bhJI=vPa}3SQ!eRz_o{*PLH(hhRmc+_!Zsz5GIDvV->E@4PHbgvbedcpoa-s0J
zzxv3<g=b}TOdhH}^LGo@suX-EX4HRqq%d+;;+rmRooR7t^S#p%7ZhlU;u9;gjLz)`
z5={i{e8GvKFwRGwgu94l4V6B6<valiqehIzcvHt(!B&W9JuT<Z@CtBbHT_#37d3c7
z(6SqG4V>t$jHwLA2e3c8q5z{(^!D??1{f7fFe)lt;6$G`A9xn$6P*;7E!=_cETW>K
z?l-Fe?pL!$Pz&zogBzQhtGh0z6yE_#*jw~hD<<dSO32D0lS<;SSu)!mBF>)Pr=!+(
zRgxUkj&$7G*qETNHJct<<D;EauzB?Y`${w$&yE_8nVpU_Uh=v2`PJ1{es0Q|7d+9Q
z<>l5>FQi0~NeGExupwJQ4^Nbs?41cUqPUH<(_h}6;FUO$PmX8X`w4n%k#0PZ4;uu1
z(y?iFRUMPsxah{mwuXFg>Y8f3`8SbKo7kD9MSzc?q~-ye@T)}o8rRIF&qD5qAHg;<
z?CABKU2ow)G>~G8#NKEI{~VLH)0vezi?kb2o?Jm+6PR!~4BA_Uo=_s-jz)x6rT`Q%
zDku)m&b1#LZc4CFBuyDF`Mm;jj7p+Eo2T`i8xB0cJP1c)h{>6nq}~ngfSC61{Q3&^
z_>udd8xgZV%)zc>SuSSjsJYSMant=}*6#?#xBdQ#1^zlXHFM5R*%CHY$g*(YM{=dl
zLO)nXB_`C-KS3Ct1H^$rKrk>6&%%TkN~==#BY?;?Z^(gm_gZg=aCh$g_RyEfiA<;T
z-8G&<5Kr#ci%#y3m(w7V+HLfp1C-qbYH2_C_jpb_@@2>~uhX)JQ>3M`N=`2e8AI!X
z?cukJIkQsMc`mS}O<7b4TAfe5?RT_WJ}q_QiUSZB0g>-Z6Si!ppgv(1q@<7D8xwPM
zzo$42u9Z*v4QbmqE%pa~n4RYl<Nxlenvv1f7+kt#YasPMHwY2pLUNF_uHRhc^!q+t
zG&>1qMoi!Ka+=>>pkYiOWi}M;c~28ABpdS=!25h(@O3NWQBv$T&^yninYmp|RfyK$
zU^#!TbXt~`)i{LdZtbV1ZT3+>2YaqQm$NGzFbPRcbQ)vGvodu>u;)CDxk;Q(zK>+b
zi!?sKlhixbG^czMnDJxGZTaqA{~haCj3(@*=g?}BSF_!Ae8v9VK=L-xoe7ey9VcDG
zYo@!x+%m_SLit)$t)Zx7I<G3*lZqmUd#~Y-Pok=t(OcN%#{m$I%2mcLRlnLMPm4~A
z<NS_94b27W-OEl3?%wp&+6Y^oa@%*Cp&IJ0BQ_eu4H5_8Urn^!Qkij4Q-(IQ`g}E5
zIi_`!_a<z^2udD#>yaNq>b^v0d)pGCdS%f5`0&R2gT{F$rnG_-ZH?7(iz#Lk0N^+G
z_sa9(Qn~93^>$FT64YBqo{vIjM~jU>ys~Q6YqqH}b9Zm*O^sq#OnN4d?Kc_Ccul8P
zrE_z#3f3z-nD+o>ZI9>|?$F|KKjm!h=dXJPd)!S7(cnC2Ea{!QD<XZ<t|?Nw{pY!+
z?0nzBSP{?QyDg8;-TDTmcXlyN!I=BmxSnY6d~ZBTK+x^}o5hs(J%65g$b=Xc{nl!d
zbUG4SWnv)Je;_pYZz^6JAyH0$ReRmA{yJ$HzvH%<2^M9qr$;(9Z@TC2e0ouu@8z?q
zV0)PS7_m9l+auT%F_}IR<<~~Rz0w1^*G>@I3lI>`y_i)&$gIZPuEk-5c2l-15cLr=
zhqPZ`jD*4YLb*S_qL4+)5oz`WVYvlqx&3GvKgzw1TD^8(94zYmgw208ducYL1)WUr
zrvro5KyHm8=Z%FN)%JDQiDGy>avW!Z)-Y;ewbirMp$LECtAI0|5Onq&)J^EOBaNEL
z38^ae;e*6aUIkTO@i-TTP}Sx|3}MBptm}~nPrt6!q<pa4M`$NH=sIyC{-$dp=v@ho
zWw=p!7wtzTEc(3KKB~bbJQy`ros#Jt_g@7=R^{b*G{KW0xBeE7<oQss>EV2<kvdhI
z8wZ*{F59aca^+#!el5k+%$VjmTiZ*|d`)7~h)U(!_2o*7$NRUiU*7sQh6rWmYtYj*
zUw(gNZGC>?Mn^}VY#|~f)U019*Y8f$IzK(l*U+jlS8c}3U0Rx^V0pZnw{p1{6_(_R
zm{zcv`a3WH@Y(khK;T}LSuS$rNT)7+n8zrfNU_0&O&0}2TjXYeo;ywawM1r2cSeE>
z*##^6u$UBS60588Fi!YlIn{b+yV#&J`|QLUhq0FV%b{Sp=uE3M=6tIP-6O^>V)(6C
z&<JFFM^_n@R*D1+P{oj@?%LX}q@|##rzu-2?`d@uk7<r?Z_C@&S7+2Me>ABAKITUq
z^?bjba6xuGByr_oy_s@U9EmDaJ#BZ#xK4sL;|w|X8zma+lzmnz@=`Q0R?aO4(Fr~T
z^Y@|gII03;H#Cmg`%nF;_obf-=H}(NgqxGIgGQg|>XAJ~zqGQ)9%>iWyB?(Emp?lS
zba38`e^-e`CFRXB$V#_WC?mzMuTN5XEJ(SmkI)+43RkvvK4mi@KC*(tY--E?DTho=
zXg=hvEzeo+3bteB^W2@&sBq5V5$VJ91&Xg;Q?zDt6n<{dJvFe~IZ|-w4Mxf?qx5U<
zsHcs;`AEv|;{9us*Yu$EaKt){ZXU&KOeau%a=V4i1gnez$BxFZTJ+{m3!#ikLgs8V
zO!)l|H{Peiv7KxVuNndF$uX6eX+I()_PCJUv~_lNu0Hg9I2Ha5BoV_ybwG;7rTH~)
z?^oK`Avm>FRR#-nHbpAs-mj6c8FkxX?=E)rt&7C6BS2_W00@(jkpT)gF+d$Z+a49V
z6>Nz3J(Jqozy9~%e+81Cm*XU)rKL-q2EIPB2EiUzk2Y>?N!nHQEbjPrNQV-BjZ7Do
z&bMA3`Tt&k_;0w_Vo2}K$WLSHC@k)6Q9#tR|LqOvG1<31Eqo%@x67s@^lugl9?}Qy
zNitD*J(cIwZ#)-^cI3sH)Stq}Aw)3|jbbb}(mI2r3-6fU?;tJ`u0{&lhJ8Qjj(TQx
zxizOPN)&3>R+Jm6FxnRDU9^-saZQVIFAj_SIDqO)Bz3asQsdcx;f6*rx2)Vxs_x2*
zzx41;?C4IV?cU-~uD={q4We^9&RxTw=Pt~9>321jSI8dszwZur2x%EU25-u5iS{nG
z&@<9N(V96^r`2OME#lRDZ=1#ex&6(M=ituuS&ymmDe5IfnzY-{ai|wReTQJG>AaFC
z8f_gc7FUA=k%E~?pbx)QbVQ>gBGnY*{T{x0oG8MI*3gL-kFu1u-exdAqs=IdB>Ad#
zgl12L(H8g`Uo1~5THiT2sHt_pTj&VDHQeH;KHupb@Fbla%Ra)A)49Q};uDGVl?$pQ
zqN3q(jH#gww7$XP<;_7Zsp<Tg?ZdiA6m0O^E_w3!(K)wz)gfLvyKU1}%^t126g4yQ
z^_RD>t5N_iu>TDs6tA^he1nYK+1rbdEf*v~E>Joxuc0BKp+Rumd`$#-yj0$F1u9NK
z`7+RNuo<)ldVZ6ZmzRGjoW#J#?~jz`>jkj`6&MoIIXrwE<G}y?V5d~10{s~Xf-alU
z$}a#GRG82CQBTBRTR0Up^%|&6fNaC~eQg8v#%gaHJVST0FB(cBnJXIblCvpwEA;p8
z1MBNK0pG~T_)jX8Z}6pk0iZ}gXsBdzcrw)*EiG*p<oU@rAmAL}!?9*JCk+jaU%duE
zRC!BGJPc$lw{aUR%T^&s4krUcscpR?&FO2!w@2S_uDWecatBpZI2rAyuJ9ro*(>bK
zddrlZ^!C?;4|4Pgv|DPuNU`wL+M+esnofhG@*uemi^SSBihtZJK6Ivp)G5?>)D-B#
zu>Jvg;FrVE&x`lBjm2hCM@y9Lub0eUGvAzFcD*y)n4aNrT_lluc6qNsx6<#(O5tQR
zF^0!y1%HJYMmuZM?Ays}(A&u{YEmU9_c*BZ^7l_Yk_`yZ2l2JPld?h*5{HRsqWM%J
z5F4$EsWj<$PZF!bG+Bi2c=((CZtWV$&s^P7GqU@H`<Q}v_skOTSs*=4Y4E)3ZXtfk
zcMCx=@4#=BiA#Zh)Xau;m+W1{UtC7kIukCPEC=OB;a#lavJ$y|PvZLT#QxQ8gM0-6
z&0Vil`nFY`B=^Og(}^^^#A{c}xQnc|Ck>UP|BAJh@9lJDnQf6?`z*hwurx>X{_w`B
z8R#D`m13;cO1r74sjQ-+eP94NUcw(-`T~?)`@^}*oQ)$e!J>_hld)9^d>&i?s3KsJ
z>zJ7ZgIG(Kd5eMZRYBq1hsgxZDr4E`{rlV7a_7^vzU9{2Equ_|D66UxB#%_Q+TPd@
zoG#If2NIcbiAH^P>W?2#K=>@|TaKmI6l-)mEGsL^Ti}O>hj+a`dI=Zvopct}B&hB%
z!u6;7<L<%1;<FodH8l)8yuQ_r0B9g%7qXVqe?rd0XbK7nO5%1p%1OmH+6@j5Z{55!
z;|Ch(!F=s;&g|tA?|~+$tEu4l-udmZ%o|>Ek8qF5DY#-QlWn2S%6_Ps%f-~Zeg*Py
z**XtLk47R%KfJ%LUvY#xl$3f_ij{Qewr9?GH?#)K<36DL^56@NEW8p$*qK&W)Qm_c
zUrg2sq1a=_YTSNOwO}b0>lm_{5^Zh5Hok$CO%aPDkt8aA>}QobTypE&y$qnNi$Idv
z8Y^e{q&zkJ+P~39YdB$-e`;o;xtcwb4<dKBnQ3=oMxAHIIrUL3&W^|vY-TYT>Ojhw
zRO8aU$0^JX>P|WHV}pxPEsGE*_1t7erCks-vHY#O&;CPP7*k{hmoAz;est`_?$ZF_
zU87urdS~0SMR9q)asYP7%?Z=U?rJiuip@9b0GY_IDDLj4N;$f^`n?_e5`_-lg5qxb
zh1q_zI<tm1JhIe@a+sl7ruUw&c62u=q*q3#YPTFfse)Z4`s8+Hj!mzjqMx`so(o{-
zvci2saxx*9s*dT}JBS%E<iQ4rsQ!_W^3D3M<Cyf}`eW$K%5vvILo20I`2ekfGTTqs
z=CW~e+U^X(tggYt^OqgY1l+NtqUoz*L4l!MB#xGITbwoQmpYj2l%~(F=FCGax;IX^
z+NFF8Fs>s_>y#1FMWmkd2;RrX-`6Pp=F>CsLdN&eyg~_xt_rj%!nRI+QQUpZCs1z*
z_)^MiF!uq)-h9@QYLsmXlVI;w`UuwQh9Xkl43+uj(ne)wfs2`4*QhEg>tO9ox&uM+
ztrc$)Nl#2s9gPN0gv@&8fm0;!Ihz<#xbw7!{bj4<nOCtY@P=nn({;Bu+>MSO`a5i;
z*E3Hw1OHlUPJ0aP`%RLM=gqX)8`(-|IO4hxq9W;l|4=}A3Q4@?NolL)`_DPEX@fQ&
z*|$-y;8LBKFNSe|aPlv5#x!BZeI#qRkcF`@WS)vn;^Mbh?L@bM_Dr(C#F(a_1in|1
ze-NiY_XSv0<*S3*BTrGt5F@eeyhKcA%51x=QE>;)>&^DmU73QerSbRxoTeY@g$*8_
ztVm~@r^{M~0D`@<820bDWvqx-xSw5CYaq#jAM49Tn4HQLP<xG-%}L7d;k|;cx$i{#
zZ?;iEH350vh{N@Q5+03c<6yRu?BhpS6_r3gzt_;v(4e8f2n2AUkoR7h`iC&z4482N
z&twL$l~$wfm&5+l@A!C86_xklJZEo)`R-{Q9a&>LU*-D>xVmxyQnMY-|Fi<SMc|p~
z3e+}scBDYY3JiR6ceyXDqM}m$$n$X0u>eX(ky`an@HCs1KhS4qXV$j1|IyNh=yv!s
zeII!5)5j7=Emf~EWH-#)y&ld#1pfQQc>~f)Mq1hr;0hQ473Oa^@o<@Lhv4p{&;%J`
z7~JdErvO}mQZ+M^{>IxnQPYipr>vXIpyG~S;q8omMAXWnrx=;@u}qNLkO-?xW1|?U
z);`S`N=#Nn*`q-wg)^q3MzRO{SoKr#3{_e%>D(e`(lU<tJV~uoo`R}UTijvK4=>BN
z$`b8Yk=KIW%DJPW)xQlz!k?yNgJ0X*{~*03P_6J;k~gJKAF(v%0LFUrNBe9AmfHaF
zv2YFY#TCji4Tym@IjOybG~56|XtLYkvub`5_J10C?7!z(f0vXo`P>{*;Ip<BDw=V$
zZiiGcc<ifGWK$D~8CwbO?tYb%{<u0%mG*awB-3qW>2G3jY6;n{PLhJU(^D7q@NWrA
zq3(eDvFB*nnYxK5;x>F2zp9mZ2di@ZAug%n*>KNYLlsxm2J*vqV{a@EFVQ>c+^Y3Q
z4$40vF_-yQz7W`jHgj|HovC8{|7+@zA9909xjo-zfjm99UIVv98yN8SZ02#Y<ukLh
zOfF{z<kBe`+YS#84<q~gO8dPt9CUPaxcr_x>)q2<Ix!<<Az~*fJT7N3NK|BGu*p)c
z3nvr#@;#f1;r_pS4OG<C*VoqMavT2x?@k`mkAOaTvuCBipYjDJ>q##%G)&Z(AbOjA
zpsg;9CwfFeR7wtRk2(6{jXh}|B|SJsRva&S*K)(xijYvp33a!;SCQOV((#~V%2bj+
zQBQ;4DKMn&or$kQn&hLfiZ&+sN9V}!6dx`L&x5JHV1#G4`2JDRP7A&vzsK@-Ioq;I
zGCxBTNyGT)VTJ6|7R#(P9`^Y#*o)89^0q~e%;x62$BIVz<<I3S^C{^Abhz`~eVN)1
zR_#f^i>t|<p?T(5jm8*QL^y4{>gc*fxvXQ_FL?Tb`^s#bI(w%>l^(f9ELDt>5*jDD
zY1P@`g*?wVnp5$G4)|rsjyNT?Y}r26e^cR_(K}uh>`$>kz#BtIEsMEa3VWFt5YE@k
z4T2r5y_;u!uvQ`>flP}?-qvW8eG%kg6*kP4pCbIKNfEPAzQw=Jc_VHN8@)X=2#cVq
zjMS5t#!>huuQQlQqJ^LR>h{>ea6_CFDLKTQY|pf)OZEEiJtfWQ9xO%)Ifw8ikK+=_
zlXJig=Ix%(-)^XP+KC~<Gy6jla6?-PS<&o!p%Hz(uNOm_L|<KU3p3-kB)_7@YzJDf
zq8}oS8I1uP4opeB9g-99Tu2?~vxY<Ayrdk!f*~v}LH^>g^zjYifE?*D%MU&kPmA-n
zC-2W?>--#7xpi(t*W9@fl9XaF`|+61Miki`E-^lsoN;#Eoi&|g66pzN7U;p86Kr;z
zc}Eo6_sH~(@voB&_pXDh!FLo0FRn+}|Gi1b=nS?rH#axG@g3|c{rCO7&v5xs8#y{A
z%Fn-J3M$+`7078cCB`#ytvXiNHL7WSXlw_&0u+_+i<-ahvSM`^c0SpK_#RS-28+fB
z-S!&HG|a8z9tbD)$jqVaYiyDkDV7&65J$JIF0bt|<sY1DA1n}34$^t^uyWut{bUzc
zTz5QTD}h#Stu%;VN$78I9ArHzuXJc#Z8&GwgVRqYvYn(nEJDU>R!>lt7|<3Low4}|
zFsztuFk@t*?*7&fX%+D~L4S--b(n~TizV{hPVz0`B`KL)6U>&#szJ(uxBI4GC{?4d
zqI7iigM4yXo_KxU;4K7UYpUZX2h#P@Z-A$A)~>&c-!74fkPlqYAW6;)gN8*x&Kyi*
z+}M)TD?t4>DX{6No$Tyi-r_|Y!SfFt2b0iQ!qa=!PGJ`p(iqi>KUSBz_`9U<o*}om
z9!b4xe|vrBWS>pIOk-`a>(yRyMHZozyS-z@J#)Ss3fyJdkT$J5iVEN(0#UzKn`|(T
zV-^_4<~<aCRq9}nJT`@B#P4K+Q5u41{JoQOivr1OuheSz$`eQe?Y*zw>vS6EQXrZx
z3@zrU{$gNe!M)3pqm=RSLGL|{94j|lDh>_GC?qb`;m?>Ridx&%6a0^*=Dh+J5BKDu
z+DHK_!Ho^eJ&xs9MLGIMSvZck%*h?9x+H<NO#(e|0KQKw3(WVBoQ@vLYI2%n_eGgG
zRatapWJ9&}y%VkG7T!7bQV20h8YO9}MQZ8Vg+J-EFf~mU25tcc78vm!v8MLlAGwW}
z>ahft2q2H=V^9gUtZuE(!2-cG8T5?_jIFL=pYV8Z^Jh15zuSH9&;@`!@L|}e3Lnq1
zfe`C|fr%F)a)RhRlBb6D#&G!>66tJ>a0oyAnVh9inLp?r^7}3rnbpv_-QwLyF7J0l
z&ISROG2~0f%LfWmGwctV>eV3U=pPO||1MjPirU#ZMugEM=_HC$F6}}jOrrKJR*gCu
zT1~x<{cX<}CacY(x^KKR1@o1&%9zHy-(R?Szn1?s9!FC6!q+eTOSMxitu@J!h1VT1
zMRyd=mw##syN`%mh~udStW8Ek4v==sc=7*EoKt0)M@R8Q_3oS*`<nQClG%q&H@{m!
zn~4XahxfNfKKattaMM)jbh6yQ=<%&cf@sRgBf>3u&49zhEWwYT^RdHl{^Up?hHX?h
zPCeH-bEK>u^Zi`NxVvtM|0<;;$jF2r-d;kGe|%RbhsxA7Q#Z3pxTf98hxfVf^t{Mf
zmk%7II)>seOEl7t@KBUJOvXEQ=|G7#l@m$e5SH5Q(H@9K3`*%RmsKv9xlQ@!{^;@8
z&$@8e?2&xO-sYvLnR=o2VNXM3$IjS=S{k3HyWtpbLQms*_gx4PKuZ0JIhHR^agkA!
zY8}h$;_Wn+C2IT$&u0qMB@Pabw{PEm4nQS*DTZNTlIN$KU*1aeQ%n$-qSDpb-hIMB
zeU$nfE(&-3+s2ZCY81cx+WJ3Zoo={5xjMqA8YSPb-Cm$?>h2mkmk?!&*iDv^6SDXv
zoglDM%vG79KzQ!8=?NvF7{r|mv0&#E(Bp9J9ilw{y0!%%bXxGX5zB8e0SyA^WW=)X
zpG05-V$FqH&kz^3QNu3ZliaycN15$%q!VLLDTcqY1ZS0oEzxff5C6val)NWPu49Et
z7BxcK1K&ztjH3OFI(ZxouW~ttHpa?6gJQmaI=K`3Ze-^a^%Ou|%a1&GKQqH=itqfa
zEtNkN4gV;=6j7syZQG)as)_LpJCfCx2b-Z=`L3|y5n{8ihw+R5<o?vxp$k|V8XNBY
zdUHcSp>!mmsMhAA%`d+$vQ!fJUpDNqRlc0uSOrn<-&Fa}rMdlC(6!km7MfYf^_`{R
z5WgG^I^Jr=Nog)A_6QP5R3`|_`y*7dvC*fffxBsJ=n1M`buF1bhHy_3UdfRuY%QzT
zM2hY)Gt_CN!54O9#C+l#JrRU?Vu~RTH%zjb$}a^5HXVTlOyTX>X5OS}=Do{KdcY0P
zAp|A0ztqk58VI)BeR%l}@DK<zPOCLRa{*DdN_2&{y(67F>@N&f2NyDJ<A%HDPZ2Em
zsF_-k<e?dds|wfAlU&aH(0x_*p0pSg%L!&&RshdMEm)YKXjc`^2d+-Bfqe~(G%>*e
zYk>Tl!nHD$%x%#>4eop@LF<k=^hX<R%e5#Ck0}iHs4t@@L^Av6ksF#<6ThI*_9H%1
zm1Ug*@Uowy)khBU^%|(lWYBLp_uI%Lp&AK6tqI9IHa(NC0P@;RTIeUl>Jp%HZP5ZW
zuHWL4v-*nl^F|R9*%7agrz~Pst)HW#gLb_FO>?-ttE)6_SahaE=-_iwvx?Y}|6B+B
z7YFec|JjZv)RR8kmxO3kM$DT^Bxhl_ECmX~p2)%!=E=AJnkMJU5Z<vH{7E4&Up(05
zbNK~L!<JOxjS(ZAO!5DS`U<Ejx2|iXLqI{gL>i<^T2Vo|yQRCkyFsN}q~p+?0@B^x
z-AFh8KHm2m{~bCmB6v9a*?aA^=9+UZDrd4YUr|r9Wtw_P@fq&gfF}vvbqxs{>q9O3
zdc@{*PIj-@AB~H|t7%iZ+QIe{lP|XrJI?W?OXh-Akc4<;c8q(Npb!Q4Rr|d+qjqny
z2(Q`g;+OsQ5ck7kd%Is03Ex<`zUW&<Ytq@5q@>cI#Ms)J#%3>4hO`e4N7vQ8QwXL7
zhjx9LB4kM+IE#t->$u8y1Fd(8c^9bcSl8R(Ua?zR)Yp$z0R$&?g(DoZoc0^O!_*Sd
zRNf94PDay~oo~>aa4^e1yCB}Z<T00w9Aw94vfr3a8w;tvn#WhhXn)+}<+8(gs*rMP
zYDM(dA(N<lO%}ow;JlX8{d528-DCg-%U*b%S=#kmmrp97i$E3mH5~^np>MAyEda}f
z6E}?W+{KK=_hp--#xc|Bb!lZbN>0?4$`WOMOe?$I5CY8YFuu~hRN#DPb0qi2C45Xm
zk|qL7xf@3fRt+jGv|dB_b1Ru>!XG6tVo2*Diw@5h;!d2*J4B`$)V>^31V$__KhQ~;
zsrtVplWrfKN4VT-r`gUPqpKc;Gap-N#bZ_IaouSwm;}m@#Y@j8?!aBwF_YN_N&s{X
zqNH0zDMJj;MEgsp4<W+$mkJ>1gj<Rq>sbU}Q2Mtb;3<aMasKb5dLNoNI<ow~L-nC9
zO)`CDvEH5r6n5Q{N`=p6id2}P)EZ{y?d!km!~l#29Ot$5_1)=GG8n)CBIbAb78Zv3
z_3Kyqm=3e*WWYSjl26jzGWhT!x$3br;@A4jQY=pP*_E*;&wV#TERJH@JB7F{d%KRi
z`x~!!2)lpbMc~;_*SkB{vz$YnI(XSP2LJw?l0X^492Sd?Q27k;_hwt2HF+oJqrt%3
z<^2S^rOX+vy?FPaBQI~@R~wtk78{J!<(@UYJInV}N|=(YLFalMjAv`POHMVH9vL*4
zB!8e(uZT!S`P(?Or^+p;`3(7k|Guo-C+hW<E1EHCHpN12-oq+I-k<r&0d_5fUQO`h
z`mZ%U2O~AGpG+c|FAr(;w9N;i2sKm2&A<DGH2hft#Yu`YwDOGV#EIX-{#lh9*coES
zbRE{#Bh@>+1wzf~M2J*`biI7rejKwpN_~!Hyi#+=_Egsn&sW;4(;?MId+ASB-*D5<
zA*rz=ub*+T{2~AVRIt|)7%FdW%tw0jm+C+uHCw$&;g_BahDp&oQMcDE_}W5fXiVfP
ziLAHbc_vFdBPaMo@6|q_h<YuOPl~3m8LQe|C6AJ>n18mL;9(*a%db(}<Z-JBJH>Py
z9l}6tl+#x2`@ZZnK_UBJGPvsWZC6TDPrYao3f_Xg1E7;40iMNYrqWxq9O*m?HntyH
zjZQDIu-XB>EKB<e1X^B)o%bFd{4nNI#d<(|ko&xH>D)~~tHEJa=;;P?y-Vpy)k@*Y
zJ5-3%30W$LRKPIDP?eTZwx_enKT^DOyQh?I=CPA-Vw3(`uZ{IAobmFDx0Z|fw1X5v
zhi?f`8b?hz6MHw$J&ceabX;MQP^f)?@0H1fJey>JqVnA=)dDFBU)JnN7xUQ0M;ON9
z92DQw!JGg^^?Jqrvptjk%PU8{Kfy7wbLx{brMbd`cv`txQw4g4+qaWv@TOowLuOTl
z!PF|x81e|A;;Xpw)kvM}m(eiFb~`4Rr1A34>&@P2DDs?6O%0}h&yeJD`2+!o<JAKC
zfyN0pZ=|wYG2aZveR1~pzCgk+^nr<ANA~4t2Y0Xf;LbhFP6WJTD6ixZ!@Int#fB!$
zheR+)NPZ^Ci+61ng82jT{m($Q&uVk2+91SQ_A;wP?YzO?;#QYnGf>6)M^(jlFXF)*
zyXOg|EzuLa-+Ga~>zF?-yVZ4t4!=)UJuD(+Zxegg-@c!a4NterzY8xtxaeX#FIuhb
z4^pP$v>R+ZjMTVC-J)+x&}Qb&1Tju3-481}80i+Ph{ruT4_?Fx_TMR{RfLK-^u@)H
zZ<F?UGNnR)=G312$*ow%L+ZGUemjdfBysrY9rN4u_unMgUzaRfm{K17D!a|dk}y0O
zpyQvrJw3=co)B7UlvHyia7#O1rZqR;7l{%vu`o2up&b}xRSA*TbFRFC%_#YM0m6j1
zL{1(FAS}z$r*tV$z+`w{5{O0->jB<aBuUt7Wb{89v&EA|D%dzU2)_X(N><<R4`_XY
z#@LHjuRb|AZ~(pz5;l$WpM`}~fJk5y5TN|7Fzip3^iz2efJhlLdQtubT$=-rWRK)q
z$Fyw2WUtTeeN&RvcvZ1E%7~itFuG?;)7cBrq_p=R$(!mP^rP-8(iF@wzs1Bp!?X7K
zq?*Y=^O`MDdl<dQ9(KELH$1<eKHDw!Lh8sb_9uFE-kRz&Q)8yd2G5){9L&KqVbJZl
zWM23KzylMEjm5E{>Kvp@eYdfT@*Z{3H!7cvsI!*)DNx+YXzp<D2WJPn>G+h8^H8nL
zjMvR%8uik49~M{+l0!qKo~1q{v!0Ep5Z}5o+<9FGk1f0@z&LHCjopGjHywt{km?$1
zF{N=57P<K8_#}&QUzIdYgrQ)=gftYlFwxbnE>=WNe(q&erzo9`!35~iB5lt0h2Co~
zvi}sfa=MDCZRqsI^|4I+?wU*w4bO|I;|}`xG~eFY@ptEP$t?;z$1y;6;Eg<1*dI)o
zhp_Q14*C$TEA1Y^FH?Q55~}G-h$*qmYy7^QUytKLD<+fzSP`sK(wx(Vxc^l)Rzfk}
z>koj-$&ldh=VxX~hVSxVTMxJ@X`$FOup@x;*VlSHUppM|64XBvvkg2f8<1dv`z{x@
zHHBW8!Q8i{_!3A&2Lfp)7RM_4Ssm?*ZQxA-e#WkV<GXr1_TKtLX<KD-TxD^K6rL5F
zh4vLn6TP`fgKsi}6rTHkrctACzEpGUH;B8u#(ZKycT_`A8PfdqE(^xd6@jbdvhLbF
z@gO%4B)JrV9ucd@Qj`MJNq*;=eP<G7c2%9D-GN3QTCOGx%z}0vH(oag@q_^RgS0ij
z<YlJ=hoKp<rxwc2Xw;ULI;T?zV!1<^(eHuL=ZUOgfUIyTUqRbTCv0jSs>M#z<o94k
zQhsNRkn2Hqqbe*<7ij7Vg@wTJJs^_QUDl-!>&w29Q+}pijR@c$z!J4NvEvdfL6pC7
zT{Yh*x37kgs5BGuF7cjHQ@(La>;3jss`Mv|fDEMzv_A+g(AR)k?|eAt*lhp#aB=xv
z`&E~trG*830y`U<BvuOO{@^AhJ&&$wajCknJdh2b{#_r)H^dkr<E1QEXl3r_4O@CC
zm(d$M0ExYJG_~}2UJYW<D$W>e={H>T7D!_x1k7J1i_ZQfT&PvVqN1y3b)COQ{gJu@
z5{G96X^J>1Y&zNFv^MvX=Tf2JdCUB^AZHa4>!f^rmGAlfgMz&4frX^afdx)}1;dD`
zW9`EATV^w#;Ra#W(d2(=pRj@ffM)7%E*d@t1_r)+H^IN}EY^F$e8yekPT}tEJ`PU<
zg&@z0(@LWR8Dxo;)mkQktHaa&dZGx!G4#@{-KVo}P#enuOaq_^poOgNVD0bD!0d0v
zGOyXVC3P)WVas}Om#$qzvH1r<UM3xsg=A?W43R6Ao1go&K|{)lw;I#hVBMpm+@5nS
z$d|N7MSoXUYg2(Sj>7Ux$|Wkb(>U$_|ER}4&)`(+?WkY?v2guwH(6=<0}y~(t6u@$
z;XZqAt>ygGA311Bdqq#9!);1FdNS%)`%cHMT1I944l4lBj_UReU#oWa%|p5T;sX7<
zTARG#i92N1pYCbY5QnBqL0;>P5b1{t!9B(FXR>k1P!0nI^OA_Ewr!l{1}>_fs6)=O
z1chzBers*`di8MlS<VYx1^e~n>{wrs=op#F3?b=3DTgQrw>sG{Nvo-B37vmP;p>Hd
zkhP_;8HyYFT_qW(XMz;F3NVuTYBF2Dmi`|X00jkw|L-5{xK_|0K8x$eW7aO#j0y>1
zaZCj7s9m5h3&PCB<P`hol|<GU@jlPGV-)IISh>CPpJvBJ>RhaVd9l37@{1T8z>aA=
zM>4iIZHG+NtmOM(F={decie2G4;ws~hli*)gxj8_mpZOB<1Ws<=2rPBkc^4LD?3%m
zbQ_4(lmENY)80bUoUF~Ahc^^wco_P~1`J*pVXj7v8Ccp8Y!t3YzMRXtNmaM^VJC`a
zxqqSt<gAR_&S@CfJ4Eb@JEoZoH2=lD<x#JKE-T$n!Sp&-vqdCO{HVkA=cxa$bk17+
z@XYs8XC;vGGS-Up(tpp_Z{#&tqQm7@p^w2BhU2+XAer&=_fO(5#R4}>w(Hqej{%#!
zyu50I16?>SW1%K9$bNj`kdz-Ew==wznW9Mr<|g$1J$u@ug$R&otb&sCq#Gx3O{F#J
zu@2VF#9IfCRV`^`!(@^$UM5L&EuCHdVWPK%WuSSFav8hjg%&azlkJTnmQd8=w{CB5
z9K0u^$xCZJNVb##ZFx#mTgEO+O}e9Jf;QsMhmewLjsW{c>B_ah8$j^d*@c(6Qxfo<
zfg_bI_Qg>dDfyz`ks{F}F=+HVBKNOnh2gN1D&MTU8!aYNtgwW62Or&dd$I$}9OkjS
zA+|Y<2CQ3cLWwU(q#yT&jUxfLxWb!C0gW*%REP1cgnaP+y-3T`Gvvruv@tRtnPYlY
zLE}ZLcTUIqA&&fG&`6D?2IvIibC`^PHnyPWE$9g#4h)n8kAwORa1GkRa2R$L>MIL6
zUV*T6q}QDHynCTfSX{Cbmtd(a>CH@{Z(udx?Vz)J`uOQ)OOf4^Q(bx#ue7$w+m+_N
zOeMcOk#Y}l03013q4Yaz2)5bhCVu&b`oDLE1)4sYm)q@2nih-10G4_pOfgDpY~M@p
z{`RABm?QYfT*1-qickZA|6lvP1{GQc(Wo4A^LHWsEaS)<KD!dzq-$w`p-<Bz+3_V~
z`rw;<Nznj$DUEZv587Vu)m2krcX5J{8U=`co!Xw;QXOP!2bD7a?_A*mX(_C#x;jWZ
zaBHj<#s9H;0GIX}Y>Js?cW$6s=5FRP=y?HX9Q`xp`qtw2|4QC+@Y9B*Ui>_tIdw<*
zX7E0;?~H1US6#s0D7sm2Hug2gOF(b40qsm1mS7pb-$hZ;lNyO_MbAb$0r9}W$)mca
zYYWK!&)uBxW6f3%T{^HrMgvb~U15S+@A4fwatPhh(l1IGg_YEwMY~C41@7~j2o8`g
zSs@s)J$hX^hq0aQ1M1Tre<m3~_h0Bd&!CzT=q*Bu*9vN7FL$cZAMqwFEU#Djg4Q6&
z1rx?`;zVpvv`1>7C^%5>Ga%XJc`WnE%IXy9>A(9P9^15Rpg@vZAGG;^v<!3(zW{QI
z`{jY&aJoRMfP1{3A1q)Lw~b_e{7%4P{;#o=C_F9pu*=bcxlypXi9ZudCq=DD0wH}I
z3Q+m_^S#uz=-tc5ko=d-9IS2vfR;wVC}U1OnadfF^!yPUX38##dHJKZopR*hUF#`~
zC1OxpoL!<5;pLXNSkGR*W7!&NlM`-V^?8P1*v~c+^M$Ti#_oE%+c)rYzx7*!4l&=0
zBH<tU=4uY6*AtmrdlU6AzKXbebcGgG`dsYF^wY92#5f*y7`S;qWs}*hmd<9}rv!Ir
zSEjp1D~vVj6sgm)KDu&dWTBGW$D>+z`Kxz>zj`^6w|B$w-;xC6uGTDs)mBz!Nu_WB
zIn|bvg*5caHlvBLUA~Jcvv)QCg#*hrY_X*O7ZuZqjH19;_+I6<Ug4Q^gn~^N=ua-F
z5X^0L?<X60UZ*&wAJeu>c-t^hij8ECzv*kWYxrxL?9E!|U6Ca{C>)%QUU7L|2XolY
zAfd>=sc8*Yylm?fCZx+yd)AGbE&9mL04*=hn^7FnM@pw`00=|F$iJm<^}X@wGKAZr
zXXQL282|RNqlb~uo1-wTkwOOI<-e)0Z40g}$;cp5>5+Z{Sc`0w*Ag7x<e8`posB*Q
zV(>X5b$JV+gKWWZE-FcQh(f64Pgkw?=0q~wi8OyzdEAhkJKrvG!NoT@^2oWxbaWp2
zjV-&QryW~ikt>st_5UL8Yy8r`t&7lOAx!Z-sW-&Ml=4wvzWgCvDVtb`d9tur<ZQ?n
z;B_xRZ^=qO)=$=X+et(M?IqKNOpvyQ_pTw9Q)(e=8=5?<?R8`HSc5n1e5<CEh$cN{
zyyAR|QY!|ahz1#b5UdW$WK#I<@EW6db7XdOWij8N)F%ofA7k70?-%}O%_&_WihRzZ
z)XeUY`8ka<?A8>9NT)0Bwc{)|x9k8*>=Vd8%bh9!ixm)TFg<SJ!LBP`FRC`^&J)`t
zFMP>SM*&(2U7Wwie;)o99grhG{>#(?+_k|wkbGR{99x%%c6ZJb4N%#$DZ@S4Y54sg
zOaxvB7+HF}kHx`CE-SWI$qsZ9nF_MfLszz|7|!$6*2&u!WO<Y5BX|hi!GGsveh_#k
zfgvaxs5~`R7%sup&<ld;k<<|b^gq!c>TqosoHAG{c!!58Hl(xL<0+DUc-b4~wKRxh
zrXD&!Fw!>{O2#+x3>1E71mrYxUXn^z9n8l6_9A&MeX#V2n!-d@bnZDa)x`z^SlBoF
z3pinwXx}wx%k1rc=PAHgzUi<rJ2Gk@fktxhi!c@1#dwVj>mwN?s-eR_C(bFzK%b40
z6)O&Bo`@<;K|jP4i<>$ax&MXwBC|y^s;~+cEu}$ff|@;Mm1Hm4Ilt9*zO~Qb&i*4=
zry4@n)g;US<Skf}SkSE?a^PdDCumvqy$(X++Hv~d(VEX|!)j88!~^i5+;&|AIBrO*
zs}s6;O0n~P0fO>rS@M5ie#|dA6@}Qi=myC^NRfbYR87x-q>!+xLdl!61)Njn?Ujx7
zFb#9-S5iL_O`CTwV)5eSQk8rFIf~CYn3v}Du@ma%<XPbu!Sd{+AYr98q`mz<S2e<}
z8Z&SVzW3{Y7|90{VZsd}ccAZkxV@lYWQ346KpUI^e9CHGta1|0*sX_@fI`@`%L%<?
z?;KxSNXh_?ADbCp;h5c{t25oUQ(7*^@sxTB7t<v?I;wupeDnfq@6y0^Ka|%JK$1sW
zmuf$54jwOHP*Zsb!%MeZ9j$LV9jWY3<f?p$Nt;X%N2)g~PP6!)!o75%H$Ct{{YxoJ
zOXc%|=stbQ|7}EPlo)S)W?I^xKtvQ9tC^um!4HUvCpLc*i@+JQ86Mj|2el@e`<7+)
zuR6`!tV$jib-%Iac9XHx2``#2Uyqt&j#a4e9o^}EZT=D0*zyqLiS`KQsEd1-#rv1y
z%=Bf-!WC@BZoINPh!#lI^z?bgE;_zm&_=zWF;34+Yd9Y4{i6Btn<6Ie?`|wj#`j1D
zN}$Fb^353Isd6FGB0Ut{C`OV~U2>=3^z4A+DYqm0wE59`8?+ZIUR2$Rct<Rc&f|a*
zDGVF!wGMieNa}dj4a{{=xER)dg-}3aIMgs<Yp_W2sK#RU$HU##`l{Sj@4sNYYL&&L
zP;NE#^saI9U1G0p9bd#bjJJ8g)GlWy;Y<cnYo48+_gkUJ$B^=~+_7I8DGioi^46z&
zm%llz6ra1-!9|?;VO(;{$mUG7g$iG<v(2%81~d+;0vcu(5O!k@D+@L#b1DyJ>j8rd
z-W?(0%I{gp%2!+ajw;ld)KDua-eB)p>Aui2{w7^QaAG5|T_{W^nc+cc_K>dW(!et_
zbyl>bLjY41YBA$hRt;jH2{u7jxsX$bP|`_=^8+uh#&bMzTeG7e4WB$y|NkAl01+dj
zJ&Xiq9qsxivq}2z3(&UWktw+N9ZR*Pj<^)s?=eoqB}{HgH<-Tk){$GC*yBf>{sfGD
zLm6*T5l!*5n(F(&<D89>vFCsknNbxOg~1JG7Y+>mETZt3BsHpm4~(cxiGreFl9l7s
z@__padNqtx&W$|nYb^Ntj^=iFzZFnquG@z*Ul{bq?S#Me+|!saq;{!#!uY!WRdgp&
zdu$MJuJ@;V3xwY#1~&&|w^1}Cas74!b@h7#J9bqs1Utnb=bw`u?GiBOZbq~{k~-V!
z(5MoS%TSvi@;RlJ?Z{sSIfV_q^e?vQ!=Lf<lpDh1w~uk_+gL<=pFXf8gU1f)mbiWr
zUI!M(D-ehLY(#m^&v1vVT9Oyih~82;nP+`#J7d)Sfw^ag)1No+QA^qW>;BA`)*ov=
zlr6UFV#ZH(6D=eJEPqIrtl^B<q>JZ77q-?NXCXm()0IYD(XVnh#qUnFQyL;g9IJhp
z=RcH@0BYy;-C_<pnK}0O@#gojY<u%@J>wcZ24ltF!D9ina7Iy9ScXiR_X0cjU*JK4
zwg{ikqT9|XaH7xl=hf=cTuNDaGtpb=|BCqd3RIlqr<&OK80xcuRBN?LYa!zwOh2?_
ze+O1MT&9Jtt{e96@*d&@wLEf-;{71xGx(6t4GMox4-fi;O}qMn#{WVNC=*%136AY!
zYntvlxy{7({Fy=lfVy`T2JRP!4!kv9E#m54EeKVW>=t}!Bsrs{ZrA%yUhEc&h(F5M
zle7J8<qucXo{g~<GoZ0ww@$6-4i#`BeTkN`s(YXE8dTR^9uE;5@)zT0Wz<sQ4pUne
ztWCsl2xjeMQLi_AkHDN1E%amo++$hq5aysH)^A$D-<#-32I^tV$I1XmvT(o68k2Hg
z^k2OtB`dDX<!Rj1ch}04(<I!Py;4)`dEzCXCm{8=7_TQyFfP9r`<mq!4I}ZvtkXUG
zqE7124+y8nb_*JS2hqTPY;0_Ve6QNRP(Q`rjeNvZV#340NosJ1u4$b<u||zXvaDsk
z#-hm4W6E=vA<O{-zz8_I5F3Bo>^W|fjZ=J&W{$t5-KtA=4&zKk#q(L}>#O6(z?AI3
zJdd0o*g<wI!J;{CkUQ@MgZ7@I=a9Q#mc7c}9vq{x6|5!?M2v0!EuAE70h`QMExeee
zf8Ljb0f_+q?NEnQVXUt(U~~#Kg#hBl7g}E$nYJH_J;lG^MMH79jN26QbnU4D3MF~H
zYnb}5ce4cD@9?WpKfWi(Ebmsn`|grXx|E*kX7)x1gRG~_2e(dLLytN}&$8t8%wV9}
zVtc5U+Z!^G^G{imryB`6XB|3E9G6TBT!fK`)#{5gddSqer(#p--jA+=59bOnT-Hki
zpzx4tM)|ig6kyxaF-CCwB5jVw37A!IYVz684^@(`b27VlX1Xfq+#%Ai)UzNd{)t&^
zb7xEo9(`BwDKS#<hr%b&gO`Uak`7%Qn}iU(ZThrUG7d`FjV<b@x$*44h$G)hZ+5Ph
z>VSc|eo!x{_U8~ii`tSc^zGmLMim)UT~(X=bS~%MTBQ>sex$9x{<T}bb@1z5g894K
zd1;5%oObJiG1_Az@dAFaC~<LB9Vr8(OLQXu*Oy&!tnp&~w@8_zX0AIA(O-J|j2!l1
zaF>PP4H;z;G<r4%_Hd;`8X5GJtv5DZ(M*n~Q<vHftraHxbopbVS1v9t3;}@+hX3xM
z5-9wW+_0gMVr0X#Z&H)T^JVH_TPtZ6Xo^j~^x4R_S87R3i)CVTu;sAgQ<)wPfy`%H
zp;K+DzS+5u-$m9xjZn5jF3<Z_mkMra>hNRfHSDP$+f%I_i99mng*wQn5Q0CrZN8fC
z)R5KaL4LO*<oSyE@H&7A`;NPJ<A@UwFac|17)?VEI=uAVO~VIqf=`<rZTw$o+s3-W
zFfbjq%a@H<SG}3TH%g!>^1{b>i^ZX~naTi8`}OUY{^n9gOD1KlWXsPZ-h)jr)`YNg
z?pR<;g)HUY8(fi{lQRlD4g_y^D3(0Vq>{O<hwyYfJ{TBK0?Boc*^P`plneC{x}X8`
zCA17W>L+C!OhE6D8GAd>Q+jGBRj)oL!k%M0|F;VMy^?aL{&B6G+}7m7IftG1%58c7
zKsUC~`duSj?}icl@^a+LU1O$_DQ(|C9%UYK4us0-C<@S-2>Cww^V~GBUPszK_wsy8
zRS$-2Y-TB$mODEC9vXa>lwZvN%}uU|A=2GQtzy^*UVQmv%BUR*+A@I%XZ^dXQQBX=
zE<VCV>@N0C=uL?iO;&(Q4r>1NokBD}KmYaX*FL}kiVg4$=K%O{*kC-As`q$*V=-41
zT2;jX3kwVNwqS0#els*QZS6)T;1;Xh?4}fNd~$i&HE?iyl?PC0yJgQSrN@%#{hR&L
zR@E}yXF#?7{=aq?R4>~VL0~`L&~*^54lx46b|^hHjnCQc=G0JAODi-YVq-Hwhnj<<
zO1t}9TNePC85zq`7kqvF{jPsI@%Ecqv=_)<f>l0(-K!{yd9RlIdNey$gfzduC`nbk
z6e&>cp`TsRN;FI<Z<1g-`aAvAIAyo=Ue>9{!aBhEh&?p{+5k{nNr1O@bz%_V!O`IU
zyZTp4>^TH|V~Z`^Ze2#fNII%9gWnNWtNay5*2;&o&hyyN+!*j{FOQeK$Cm*cdk}EA
zZ`Q(@f$vZ~of?`vlO>Sb1H$iiZfp0SKQFMcJAis{05oQIf$n3|oO|o2N}W+1a+&*B
zB3GdJoPaF=WZH`pSMiJ*P*pWp1EZ^>MOBUE`-!ifmxfydNdswoSGku!-S}3$D&%uG
z?%!UT3>HxTYec1tz)%rF&+Z2pkiVIdcf*HvY8GH>r~q7!OQGij#2;&LkXdnO<vr;=
zH_B3I@Gk3#_g|K$H^hTTldV|wu44)Ofny6lPtMgzdgVebp6t*W@;ryvL<Nm^$CoQW
zZt|}A@r5A*72<{Q{#RzkjYQq{VDJJ&D((Ps{rAEzl8YI;vqSwv;ctm1hkxP@mVfJK
zxNA%UcugO8)27oUBtR{buSoB}OR~AS`KPOk4cJf6XGCNG)jjwq6TEw97aM5H!B-$4
zAOP0$i#YMmz#0b#tYI8>#}Jyeb&^h+?KV-}np;kd|Af4Z^K3n7M*}7=XqcFo7$m#_
z>JX8vQJd2ZvHORI(6F$t?Exs%vv}U#-to*j(tsyzR^4Dn2J}8aK7(G``p65Cnq2-3
zXLJy}9-NJ|p?Ad5=cy<zD2kUX58Fvt#<`ZUP&tOjIvqb)xD+1)<$20CGDXRMb(SJ5
zu7jnTqo3B(;(bTE^3uAlHHo8F-Qp!ojiY4Ho}=5>N5Q_K(&_IBnvR1H*bG<X+zHr^
zUxaS?y>-$=)Ef{r8K)g;@}+GR$zF34^HESYVE=C+n>HdOVdfee@F$}ofjonUhhGNz
z^YV+I^A*jj4DEsbFbrV+ujHO00KFQJ2b_WfvQ!~$Ys)&GC({kY=)kJ2*eYatX(`<}
z-4zj{+z*()Y5-9MtZ|FY{b_LqMU}7Q6cqj$Ubh8oTx4V!ebd4pKZ5mYUtCPP*q@F`
zO#Gs&`%~_Y&GYtL<I%|4dKeCq|2F`vKLRn6TAhtv-2T!&6v!RR35W7Mp`}=~q=;@S
z#-~yN<hwJV+k3c`+s{TR_xT-yp~Phhpy9kJ(f|RT6+^nnF)5}PajtVG0C%hyqu1Ec
zj=h&Jcs2f=U)2Vs#<0WPGW)_83MZjU06E)SlF-Z6huHJrOQ(@;g1H&L7@k<uRdEF3
zp^uGx!4wiT_D1V3y9V9Ux6<Iu59ZltpnFpMYuIZIgc4jt;xZ!2S09ZK(%ZbIMHSJY
z*AZyf<Se||?P{rv%Nh0L(NGg6-FSt&GzDaQt_8ocvcM^pNt>Dl+zZ~csA+BYL!jG<
zB=tD?kSm?m2Q2Yi75r+s04)P)1?U4?=<3}gD=XQVo6oTTNXzN{^uP^1{vmfK9|I)&
zuw|y>)ZbSiezc?iUKyPd6sD%DSBfJOm0kBbN#Ba7S>DvmQ5r7e9U~<@uQBSV8ke`L
zje)y=1$$f-^ucOmPDK2%raUn7=;Q;@wIKZ5?1!%?04RcU?HMot4SJJE@&a)NRsK~p
zy83-Xl#J@Y()kX28M;yMhi6>!?as<D_hT;s=>X_geu(M(sN0r|eCQuA2#DGvpTvD+
ztoG_gU%@k7>^uAS^MOmSI`z=vSq~!vcSEVUXWh*55}S^GS411={hE&FvI2+|NaH_?
zi=*;tUDz9F64CJi2@BAm72em!TSP`ia?++l?Gl5~iIgo<0B{d$1B@lJt~S*pDvbu;
z01FHajrl&$vhwn0ubG&JetOoaguwqxRXM)lF8ZHfn4|3Cl%4!oCDEk!&xm=3&tflG
zJqT!>C;P<6yv>R~Rv(Q{<X&7)u+9C<;q(~AMXVFfRBWvs{22v~ta59V_VBHWH?;tl
zn)ucbdQen#M3f^}0#@^I?fbmZirY=H9x?G!+c*G$e>OEncrs{TZA!i??dy6T($_sY
zA1J+kN7&Nx-%{9!{yqEFmz37q$6A>2CN4|@A@JSo+|tx&<Lk%}DRJ?yfVOA>JTCw#
zW30*btnaomDJiM8I3}IRvvg|LA%Aj@-n&HVr_=f;Gso{S6eJQV(|a^K&0J#YPEIwW
zcc;KQWVFT89Z0trz^<dCqhoa=JHn}@YSU0w76A*4$E3N^9*9<}AM-DXx;}T~Fr+u-
zU2@z#F)viat6^e_Z%?t6Wm;TiVtxxLuh(F(HqA`ImOMkcfy6K?_JZWHCFn~Q_JW4{
z^vwOi6n5&O9dR<<>LQn}@%!pLIZ~FX@US(2j##TjN<oM%3I0A(?z(vXK<-A}Ye^%S
zWjTIhWXDc=eJ{3wOI`Yf>OYKwgVmo;%&?1#nlQn6?ni8ddY-<`wsbCXzwXk~(p$BP
z01&>R7&KTBWK`5OAPed2@86GaN{)%ahG9@Ht~32rRp`Z}#bz<mTfK2u{mEMC^z7`J
zjctAPWjXoXk8j_8Hy*cy=j2cU4HFgcCbIj*2}UD>j#_Wg@mKWtmC5}JXoj$|qFW4s
zm@YN(0ObrkJp97M5!6lU@?dr#U4ST3L_x|Lx@<Pvq4RD(BCZurs^D{yVplwAFy#9R
zb;Kb4)w?Hl&kdB{YnXBz4)K^5Y2^SeD+to!^NCCH;<QA<<o*PVADGrhhBfGz^Qn`H
zrz`D;cuifCTRFs1GO6$U@_wM*>DN1CZxPtcNAC!?s*w7~Gd8`Xi8Nv=S#1<9A#ACK
z$)NVb5b=12(pmyRkCyvafqbZ2cR(@wXz$F8p)I}selO1LB6EEIX0J@qY$da$S&AUN
z?JXlC3L4t_&2|P<)CDemJP^Wxr<sdFBLI_NxU3dTPVspizHPzGf}E_{(&=F4+57IO
z$>DXm1kc%Ob37PmC<0Q|_Q65)Kw)1=>W6J=Ztm%s#rh)7FMhR6O{pMnfht}lB_$&*
zB_t$nfECa+`27y#c^?BPppQCoZ6K&)tX*MnbzawWVt>`c<@VKpAX?Fwt~V1%qo>|p
zbD>bD`5ZHg$HYpV1eN-?_5X-n-|lVIyhuB%Oo`4^GS^FNRK&;>Y<ip7idOi;Zs_pV
z`|^m7zmUy+<=WXup1S#az>C)b0r9%XN?~K<Z|FZOPT}Avekwb{#C4dgKKlCJK<Py%
z>aP%FKc6rM_+5ebgj50F#NE;^@7AfmCknEQy-poI+gLT)YTi!gW2a1FKNdI+r4Ode
zsQ=`4kAv*wCwR0)(k+63Nn05p6E7gh2_{W4b%{uS&@b|h6FV7^%kgQX!@)3UYmvOB
zL(q=)6^&}oRXorlcCpz7-(fN*)c0?2<EM4GBE=@&-D4R3t9j{D#|F3V>x~_;D9Su6
z(us7|hdVXC&*30{gO@dG&^F)b`3?g-`68~buiZKwoK+rkhx_}BP9K57Dy~~vOsv4c
z+sTOw4UY+47f`nx1+A2}<>UP!mXpWEOP*JYyxpw?oL_vMf$iMEQZtckdKE|ne<`bY
z=2zgM-K#<HfXRuK19ett-J+a0TENngc<a1OV_R&ExQ0K)thmn7w0`Vt!P%OgKB~c9
z>|e4)*ytKLFNG@wrpB;&*NQ&<&cnA_RR>%${Xdu%g`7l5q7;_Z+NZ(fRQiyjt#{ry
zTqR^O)?VRO>vvg9A57cIPKcX4aJ3_YFo!zQi)QjU7H)f3(q~V(+WZ^Tzy18ryXd79
zp<wKMLy#RCV_Gi}mqK=_d+U_cU%bS6OKvVgbTa;jS7vck-Y3c1hqcQ_onbhX)YS1h
zn^qizVqd=K?oAZHMBuajY8N~@I)ZL@AFIWNa!`lZ+1vLYGXtSC$fCjL6M!aKd4`*x
zpV+6g<rY;72f<cINQl&Jha3jT_d5XTrB8$Ulc?x8(BOg;5(wuU4pjg4)!(K|WC(h~
zByyVL0%awGP6Q*!eR%A*Sge<tc!i?4DT=sr_&lyz0Nnsi7xdOZR7a>*j|BD+b~#Pv
zIxu`r2f85po{|;=K20VL2rBk;z0C5cjdNp|JcpAk<v0u5NOJWy^muIIv4V}HH6vXg
zk_h(yCdfeeXc@YUsI!$^i;GYSaJrHh>*Cjgirn1)j|-qF31JzsII3B(l##Zh4#M(G
zbK7XRacA91>s|QU_i4>p^t&C`_k{@y;IPcJWL1S~E_(M!K4UVN+V=LJ`<9gpVRf=n
zY15Y{<EOUU8T*K9C-jU+QfX0{XtOIiE__v+6=ds!rC=1*(|h)FwB0cmF&zrvJ(Qs1
zRX=y<@9$q!obqs%@k9x1j)62V0g2S*=iGz$-C-^2ve0P{#U~Mw)2$S1=<i>nCbC77
z#afLu9wJA;8!<F8(h!&dJAhS_9{`GY<x*|SzxB`YS$x2?Ujd>w{k#CQP5~AgibTgo
z#uo%l(Cq2v5?C%2rqye7&E5>hHOQp#8Xh`;_Yy+=Jwxucpxzw^K{QfaZQ|gAw+r<P
zBZRhfOY-(cJKL*deAiu`NwR)vbPPhUcLk)qL?*lG>eM@UCht_m<+z{BtQg@KDx95O
z8z?ddHL=9IKi(lbw?44=XOMmD&Roo{WB9jZC$`SU)9z=xljU$orIh0jl{k`1bx9ZH
z+jTz@qL)l48|wnH=u8J5)?aHSFl}Ef-j!8W8_pgi@e9)b>a}~?h>@8o(}M#BYt+Ef
zGg8DKIB|9aV;av^7=oVc>V~3gMWA@jS5PDW`SXXxeCjRKmX?DMor#I*d~Y&*w$j)h
z2rq*$NYwROzkPcFwb2r|TJQ&Ee2D(f(xiiX&$s(VKxGN`dWF#-p-2>QD7ZSoCxF9L
z<=0MB(1y{@tT8^S2bC1GwEFcq92p(`Gemj)_;@%gTCfis1DDfY8p!n*ZXxB2HjODw
zf>?Sl0wqp`c}t5&NpeQW9>GTtL_X>F*4xF+9MXK@b4F04Z-Ubp)BA#7;&ox}t$V!o
z(_&-#p`y$zu4|-$SJ}08s?&FELD<ljO9M@?kx?M3Sz@xg+@6Uq&kUDWg2A`{^S2y0
zBu<yv4?nZyM@qI-=PJ5)d9#_PY&RuglM=>96LxGb9J@>PiBi?;)GywF_*GR$X;)$(
zAFJV$70iAF(O8+CB1&Ev`){Hvkro59j8oGc2HJG7{!-!b4Htc~b=I#U5^v*-rTeRL
zQ4~wO5f?FKV7>r+?RMNQDw~yoXatxqP_nUA4BegYPnX_{-|qo>37`&j*V(S`5l^Su
zg920!gbK+F!REfZQbzvpuww0%X5hhkQ`6k+*?b9%y=&~apo>F-A8Da=LIK7|{$k;4
zJ$-o6_VsVX79b-Y7}-AZRWFf#W^?KEIY&-koE|m{Wdirr9upqy%3Q^sL_hE``j+OP
zL|0N%)bLIN=)or4h+(k8e>SF6XjueY^p;VP4B;s*_0Edyl6!bFAlqEqXH;Pwq*c%G
zm<|rt`cjheb(3X0iZ7JVIt0c&as9<aqr7i<=8$1SWJYvmEM=@Sq&qyO^RDp*+0``8
zSjxTXB6nZQbBEiIQk@$uUFDw(TFSSnp(q~gWBoZT=L)?iFRnwOF9(~H@@(w3wl;r;
z`(u3dv<T)~$@C84Qh2q<-+sM-Ed@InU<)w#&x1}08YZjFXhBlCjDc)2CnF9{&VmOP
zM%U>e_A$eQ-3K+8LtK@U8if`a248HRS4Nv>cWaTvnl*8)-j~D>clc?16@BxM^Q3K>
z{P=LjXRrTs-cw!~KN~5r&Z9rJnyzI}i2B2@{+90KXc+!=na(!-F~-4kQnQ4&FI<vM
zEi-YCAHrXoYKIR&%IZ4@#Z@-6Y_tgL`I^^^=r!4GS}iP$Fj1TDL%;w2AT>T;%-dNX
zcz=G%CT-1|y#S1>wU0)|v`u}#%fZUkYYN(?_tOlWb?2oJ@IA4r2iKNP+e(UHH{x2Z
zTt#zz9U+KSYanM5$22lV-`F7+%owmYJ<x9Y?yqC5zafG7!t2-v@tyg<CEXfaoDB^Q
zLf_2P`)E$~_78gB9>S8>!)>|Kq56B%`jzc%mGx_L<9iHz-3M+aPz*p_Vu34dYFZkQ
zP}fgJgMVeM%fV%niR<CcTv8-P9zShCL~Y}slo8=Mb<XuoCsi0T?rG=poBF#$QREES
zvt!86OMzI}<79&~w5Czl7jfa{5}4|amMRyvBA2>XYmsH-2wk-trBWY#pLvPF#f3?C
zbQBSf9ZgA_XpzwIenZco`dvuL@~qai{UapK>A81WeyX*P6y1p>)_pAOUj*H(=X?CW
zPS;SRsP60#8A$KQcK@oxPGQBUca-6Zf4@?vUU6XZa11T#mcQ9+`_lDEIumTPJqKdX
z@+h+Or+^*CZgITDGnMqp!dpTboVGs7FfS?nhg4><$d-W~yyKV}ba0lS!5y`@rDqNZ
z7y4^~$%^{^HPEWpdANfdM|lufg}x*p2-Pm_)@9LUQO(5s1-{;BwY|lJt65!rp}89m
zcllQpR(_#Qu?WJ0z-kr?{$Y5p$y%Pb=jZm-I>PAD^~d>GFGwR>tXk1U4tLuc8q9*A
zK1ZS19RGD-4V8<m$``tz;~#t8#6E_h-yrMpjd#F<p|OvgkdNqA=;br&7&~GO&ZdcH
z+K0%8MYKn-?@U)(luyL)9c<*ovS{p1OxC5Q8~2r8S(;GYeD>hGfXk_yA4iW5Y6I<)
z@W!91`uS*0*s$ZUEIX0+@6Onn4oX|{G<ae{(V3Z5!Wtn}C##GsIZ_wlQ`KYbLUSDp
zLn{^?EzDl};1DzN{yVN9;GF4Sz#fA*HDVU$<p!1a&(6;F`4*atmN_45<Gxjef%>%q
zGtf#irbg#uYI^!GP<X0y0w2o|q7*snjroFO2C3Wbnt(jGe&e{Q9ub6L#P&t-5L4*T
z+DwX_>LC6C6h<<#Ev^iRSZMp{*(feg{HeWJhF>%qWa=PYs}x4d!I4wA@igs#4D3bh
zqobfVhy#*g{L|=UJn0?&UyT(ug{Uq$6wfaU=ULTf+U~!d3N}41!ZPmZm`k;`JdhNy
zCS;~CG!?v3FwcSQx@)ud#YY!W7P>317|}wHDt++B7JAGac;R(5-2F>=ad9P$NjJ}E
z|DF>BJeAsqf0uK>3vPrWJeRc8R44)TXO$>22F6xVX^U{)cwF>sqcig`|CwZ9{Bs7C
zlH#jh7M#8P{raozfzQbVQ(sbus51Urj^(YV5@8ip*f!}C1A)8NBA>ju5ALGksA03j
zHy<NAI=U{ihC4pSwG<AE#9h@x<Z4JCNzhxf*LHT>;m8K_tMT_1Gja5{qybKD-9^uC
zfnHtAR>a+wkg&3755^%MX}!a7eo98GX=T1qql;KxMzwM*eb6k~{OLQ*>l+!i1L^LO
z;nr6M_);ge?E#o-DtEOR<;ZZZvoB4CKj-IHRDbYLbvs*q`;}yV`A+U;D@Qk{M}v$c
zHTJQS8e4<HTh>SpI*tgO9e8xv6$E^CBT3i&!1HqlQqS{IRD`8xaEK=We2VvOb=Z{$
z8G`8~F>nzHfj}4m@j%Il5IPnmC+gv(vEr}le4z&0M3)GH!W^v4d81Y}MzqWBvs(c!
ztkd%oagG5VIQJ6aRZfV%<nMaeoR!-ixtWAeV;{_DFLtzA<#b1J1_uZ0uI9UUgc^}!
z6wZ4nGqJbZZ%howBy>GdC3z5+HM_u2nL9iw8W-JZ!gV}@EnB+>!5n<v(N<|5rM1#P
zKF<-HcpHa{<E_rJUvT}kbe6o-HL%?53Dk)Th#d47Ud<#~EqhL~#@uA&G(6Z=7qYm2
z*}G}UOdy+UZTLTisgv<O{)-{VqX(*<->cccTQr==muos7lYkmFDn&s}4*(}2%k%CQ
zBB%v5BDMh;mbT}`<f+q3=o-FPA21X@4Uda|Ca^GoIWbSdZvo~qyjk}^)%I<f#85Jw
zh)A?2gES(QZ>CKekcxpis?S0fQv;ZCDv`V*>hHcvd>4|*-o3iArC3^JRllv=<o|HJ
zIT%Mt92OXulW^8JL#xB%mNTMBzjG=3=J5iC(a%i#70zp`Zm}GEx0JjwjcLu7f@Ok%
ziIDDk%el0z?(*2)PXjgUeP=Ac_F6g|*UT{&HR*VNNVR9G&Rra8ntnN-TJIAHK<FOL
z*G=RGAsBq8$!t#KP5J+h$)Jcoj>|J(|K4%B*$*w>zle*kfe-PYBYJpxn$FiSVqiEP
z2&iCw{`|QKnD#mV6JXGpNy4Gv;YkAds)}ny!TFCb&Dt`yw&k;~T%64xbnbqpe44d4
z;~*TRi0&U8aFJqG${GzpsfTai3A?R6-*$JdEyP18VgLq&i1Xt-5Bub|0x#wEnMi_y
zN4;c`jlBiL2d36Fd0&ZoiC}=~NPMF1O${}aE+>ouON&IqFl4`*GNu`MtJg9&nQipE
zpM!5`zY#F*k!}cMtwoMZL$^BgKxq&NV9IhEYY>%9i;i(=s#3KO(<6?Z!XM(IOl$Iy
z&MR{ITNf0c5HDpV9$)c3Rg@DgZBhx7``r(zwa~MwBNENsqwE;G*vKPXL8nEnRfal`
zHNSf%Ygofpnsc{iO>(Eo#$2hed!1LZx7xzmXoKpO$4?h-o^kWjaQ<zwk1Z<5qx_sE
zzO@m+hv#svItz3ql$MS+HWZiIGqbb*{L*IxAQ$lBdi)#hok+3{OkD2omY+&&C4t{|
zeSN(VSyn;ZD~-!{+@fxa7Z(z$91!#QzK|`db>XmzRjpytd?mWb@3!RC+CnOP5j!Hm
zFBf502qW+v+$^eb&dR%nca<06OeY0oZ@LUX4D*BdlV~a=77p9m$&241Y23~f#Nv0I
zAsFNtt?jH@sPVHKY7fbB6<@3ONVllpuyQ8F3Rh%xRU{3=2==)l)EB6oHa4N`if=dv
zT*rVn4TISH>?5b8+ov<xjH{=>zzbE?+Lt>H4X80{0WSt+vR=8EXN~U`%S-aOU%s8M
zu>{aBL0!GtyN@E^imEhVi+!{;k5tRbVeQjT#um;Jfs_mTmUGeL)I`NI8mT{a{SJ#J
z>B>!!`fBEL#;R0Vmey-fv1c2B+qCt6`?lXvZlxMAO}ZPvbyDx&`*fn!ARr<_8x8cw
z>`9Uyz<&`KoU8&6dDJ%f2LuwB@QaIrT5`wH4HQvkj{e1bv{o+=vof1ZewB~@R9^Hc
zMT!aiqrV-x=$0Drv3w)fAd`W%t00ge{m=uMa3W5+qj7y_mV=$T6e-;@ZVD+|O|MHS
z2&;qQcixeZiWD^v?$8+RvqJyJ&57ozha$aX<$`dF#h=eWA{sE=cx9z;bMaS}YmCw3
zKvMj9oJeF*Eg!wq_`Qz()tNkW<!vq;OHZqzL+CFzpXUr$G$U+-=hS$u=t(?U79m0b
z>N+9~B1PN_mzUhU&DttUIUgRGQ9-Gvo+Xl1b3%`9t5tOP@KlXG{cnC>NyfB40MeVL
zaegJ4b0I8Qx7-AQF$HfZC>W1IabLXn9ZRQB*d&A6G$`ZhJ1ycLnLB%*@WO~?S8@|U
zn<#S7@j_Hh3!IhT*7mL2YrctL&NGE{ycxn>H?O|dW*F@A35__PB3q?!p=u!Y8#ye6
z@M7$C9g|st#)^36=z<Ob^@<AyOj?jr(fIJBgy`Ch<<S)-Y2?$H2##)6?v5rK!S;{P
zS+`WkccGD|P`;h3nJ*RxpeZ};m1!9liOplvn0Mogg-WgwY@NV0W14crxwE<*2x>fp
z?Nf4<(zn9KC1caVMI%m1Ex77VreMp-Y|x^etGN)jeyf4*)*W=W=ZJ7MH8t`p&Aq$m
zw;;<n^_JVY-~}6?hU-+_WXR0@Nk>*zR(qwova)08@%DyqzN<^bts8B0fHTExYBWpq
zSz)THhLf17wl*o4BLKGoobQWjXv#npV-i)o(3v}}OpCBLrC!&McU$*Sg3&}#R@Y~T
ze<se?%AT+h&zjO`-j<rlm5|%JG*`M+uKH++l*nD<T<o(A7iL2uPYMq`o_u(@V_64t
zw&kF{6!oQNec<wzS9Gt59}HkT7rnX#<G8kIbuf)jb&U}_HO6Kim!khz2n;=tNp*L2
z`}zIEaU>p#jPyWnWMBDPvo-u_F}dUq(cECz>|};`R&K{PmZ;KMG(NT!E1fe~!`&4d
zt{RNXs{Q$bK5~*^CknTI*3E@M#*<{9L&IpTL|4{-r_1U^^7pJ1UIcP0{K4Gc8eO{i
z_2r(HS725^<ks&B$o=V-c+Qu1j4)kQiZ!L1*}kM!ebdYL<ZS7&F`*uhzctfU@7RWx
zxK)-<hbS8D_wwLQTwy;2O?cy$q8RD`RHals_^D;XxQ0;`jmyj>7Yyw#ScsmS6`{Mo
za0(86Jdq|&kV)O{5B<*HvPDI_kkSGP0v6_JL5>t|`?jhH&#!6=y}uRls=6aoM+TkT
zSyg)dxR^Mz_6HRXcY<C3x5*%_dTHHrDv@@wJ)Ogd{cmytvsm@X?T5R4>r+b+MLtf3
z?==M(8>`w{HSTVY{~O*8cjLT}CzU4L+)R4!qLBC6N`5dGr5`j-)LOkf<8*+C1on4t
zZ^1oRQZqw<2?o;vUw^5!cit<Er*VR=J0+&0$GB2~rf~#B*Y7DmItzEhts4+hj&6o2
zOe{G2HjfyXr-gn`hXQa9+)r7{sfz0(GbA*zYbDKuTh4MuM^iO3%I1tsM6jL$Jx#G)
zd&WY_MvvM^h(g>VbRpiZVo5;a=Z4*c1J^0Si;mg-z?&$r#M>m0ChDZ(w$nmX`s#RJ
ze^^O1AtEhH0M2+p>5#?U7|AL_P(AcX`R&H;zH@5!o9ksDI2dfei{r$#0OPHonIQ)n
z^a;mQXtR&miy)7tCO#MO@><u(Il0p0+tH3ztK*_iB!F*|hGGfjlkS9XhxflnPDQbP
zSBMqf#7odMlwU-3r;bfZ@5latu)gF;Ow_o?9K^bFet&;k+sDoB|N2kPk=R_pYNsx;
z&B{HQF;DZZ?BbQ>`-G@i+z%JjE;kmgGBfI%M9^WH^mslI_Px_mqi*L!F*fF#D17X?
zUi`5Wf8nCF2w1o=e89`vn60X>pJ<Sgik$o|`KF4@oDRAB>bQ-ndr8nUJ`MtspA{x#
z-s^wmVoJM|+zO6x$kxs|bFf&rlzdnah36g_nA9j!3#U6BETq@aYxLKtxd^jXfK#3B
z)GHl^Io-GJe$P3iH+VKLE~$tf|NMM0U}3bfOFvch4oNL^SKK_}gVBJp<H&D?LztPr
zj}4~?p2=aOx(3s~(EGSQYS5y^3<|=9Gr!L*Nlue2V<gE<^YK^fJJ`64%I?zKI*8AT
zjCYqxZPL7+rx~V)h<i!+=gGAB7Vm{W(si;hYbS+jSQu=CUyW1!5`}cjfuz#=7zN-B
zDsEqe9I?E_<-#_V=vaPNd~!K@mHi6a&X_r9I9>NK6!pJMKRL%}8afN!3OCQ2DQ(oX
zRp-IICmzMtw}fdUwQu7Tn%s~YRNq^{f7HPbcfN}+PC29MqL^iQ$gY4<uOq4|Q$Y3D
z?X-D|goXNSWX24i@0e9s^s0pOaaRX}foSWzka4^N-7l>tM1q`?8YM&eajUyxB!|Lw
zBggO~w@M;*@qObp?0w?%eT5ZVyq%5bq&4EUr{XV(2Z9d9)~tmv6LH{0D^?+@L-!tu
zOcdtK5nPDiAK@>*G*lM4Vokv4rmKw6nEAl$Y+{O8V^*jB>um{M8WOdQjs04Bee{^A
zi%VK`mZRgwYN_$*8PT$x!=%;q3cpAEPvX2tTuqDD5}4bkt0ovb5&4{Eh=M{v^BIJU
z5Q6`u4RNZ9D-YX3Po{C7`yCq0j2=|Q0PYMu;{K$dJRpqg{ZlWH7xzm+#<~CN#aVzt
zed@gZZ7#zJ>;hXZb6QYf<xwXyxV4BoI)&?0RZUa7rh=p?+UvJlqjk6QA)F<nx*Gs-
zu$g$TI+$=YU6_INXVczlc$b>-B>nE;JmbcvLUjQ`Do(a;xhvin)LLU9TNEpN;{WJf
z^K$nPgNuI6JKC&iWB+&ObXxE7V+?yrbcORSv!<XIsh1dpf2cd{1O;yA$p!wB3FAUA
zDzcdoicAa8U>HmX6NN031nQ}X<k&h%t2Znx`uoZ(2Z|C|R}Mk@4WJ4fy;oCQ(E_Xs
ze>`K*kWy4mRrzGA{j$ndR-IAx&3igoLmI7WVIyv8K4Zo!PZo5a>qWUghx}qjUwL8D
zEt_^(C?kG9YVF~$3ke|kOJn*<ix2BdR57|o(c?WMQn)eCPFv65cAINIcLKQD&{o^#
zo1QVcLn`%|5BcG23>WGcgeT>BEOcaAdMGs*PLWO+)$#>9Sfl<-F9mue`l(lKh?B&2
z@x{;Ut{c{;3(!G;e#UTqVeKfvN9}db0WcFv;wXx3aq7=JH}Bkl=Bok0(hWYdJT4sa
zzo*L?;|zczTR8dwrbC>&>GmWZas5|DlOx`j)L0Jgv>$4N<ux^*`!qK<tJ(IE$r$eo
zh$2Yc6HmHGo5hWtW|g4z#x);RoRyZs9&@HPFYa)>X}Yi0)i?J~gIw5G+mF$QmqJTo
zXmBBT>o%PfA8!QV2Y;UM_iv1;_^uyVL@TcU&gGp3={}+{Tk;oPN8gQc6ay+lAMwD(
z3>q);XV;GC`6^EeNYNGBU#wlf=CBf{T=U$Xr52wwFdX*p?=Y3}{qF_J0zM;Xb8~Z>
zZ-P(!B0Wf}GFc?16K4t#_P6V8tRKRISvfj3%sH^)r$XpFxLf1KfXqm1t4qh}Lw|C1
zQ`%ZV-sc*@bH*3EuJvE_t-}S&ksU;GbSmmgVzzs-9MKHdJu)k36YEdUzbX60B%pDr
zYwT0f{BI?>>ClL{K}0dZqQ?Qt=C=foRQWc%nj2eI9eW{yi8NxhW0kpFf0g%i%swu?
z9P_rr-nVbs|MA6|u(li5JZ|wO_<Q?Z8%>DqBq8VFowM&7IO3uP6HOXw-O{oD$KG2w
zWw}P}qnIdZ5Yh%n2?)}uBHi63A>G}Ih>8f(9n!6Ir?hl;cXz{CFSx(onKS470cYl8
z#@TG<eV^yfmDjq~F>}fnx@kH$pA#*u%DHPwn$MXI4=>N?qc>I$Z>U2XlUJT&gDQiH
zVF_4QS}iz{+vX-a>j^45kkV&u<5}}3ky^;h)Bg^YkKV6XxVi3G6K|rKm%X*CPoVe!
zCC0PXBzh)EZ?l(o8=-G9H+k1q5{;dm22bYo-JYsG`;#2NLvz)o740q2nb!NF{4KcW
z%^_^xo*vNNcK4Gi&j^jW*kmyCB6WT+Ax2|^kMmtaUnT1eSyhS7gW(+4D=6q2lMYBX
zdF&+0k6M_KL2zwOXSCCYi*ATM98nY5%*)47U%a&(Ku-%HOQo2}xphys!YV_%MaTV2
zhzx)ECe!qb4g0b4Qix78aj@!j&Bkr*SG{@dKZN7;I&$}axweHg8@<y~=;+(donmS-
zuRuEg^^zT<BJtH0)`fce`K>}w;3T^H&nl}>y(iXIx79DJyoSU?jw|IHASPMYbTb=a
z<V|Q7%bD6k#b3T>*TamTc&n(8IHVPqV3Oyz`ES1}*QiBDg_utwJM20Xow;-8+;u4|
zReA11-le*-RiVMp^E2;FdnP`x?xp*g)R$7cy;~XIl``puJ7I1+8t{y<ih$lv_D#bh
zEo=O~%(Y|YSX1reyB(!ZBB|Yre>WSYX4z_ZCf?RxF@CVSjvP^@dC}DPQ8-<Ebz>Yh
z$HBNCn={KQFGluAlc&o&A#af>P*IVi&Vd@^XW#d|${NGnoQ37zI?5zJ`#2b0R+Zkr
zOzjM}Fr~C_a<zClQ>a;YJ-eZ5>(3CX={)KLiPSG1RNsOzQTubDL?nm%Y&)$OIdtz{
zn%O04s%T7~#(EKig><oQjjrK)-P~Ii{Ud_uK6dMzJw30dqj!t2Zniczc|s9mj7XW@
zw0gQ<j9xZT%4*!JJ5UM3>4nyu0SNtV@jCv4t$Ss7*Ph--NZ-twAtvF9i!GU&lL-xK
z==ZH?0MEcLCi+^TV^zk(gFJ$h4%CV@jP2JG+&t#Fvpo{WN(@I#ZwrXbmU<$B07L<-
z8by|~&0<l!+WK2K=s?5+&h*y0*i)+rEa-&RN_jGNWx#M%k!2orr{(_Sylhb|@==jT
z3=UzBU#ds0!^W}PK+u(^`rR)=k9nL?knQG^xIOt5)B_<i5Ei7Ok=Etnt7A1&I=bVk
zFX&Ty%*lxbtwoqCatLeZZ~vm_bNKF9Oxga!Wzou-9oJTkCdP+opc3`|nA<%tBVYI%
zvwn}J6;)#LlIz9|uwNk*<+^f>D~P*3Dzds(KJ0y@EE+Py<UC`M@`ud6@4vT6;zi#0
zCUM`m)H<TrvsfE-*Vg@9Y4L{GYW_Z4Sdg{w^F}`y0DHkcg~>TeuxVdGk<H}pKCS(S
zj#5V!v8~o`5TTLWE>*J9M-x*A6YiE)aYtwQ=ITgz=63&`^4SN@Q7O)H)zz(EhNDXY
zxG^#Lw4Jt3@xgZHS6023XW_1yJY<(5F7Qr2k!*H_EGHm~%G<a1AuQv8=gnU>XPBVa
zZfK}USENnM%+^%l*>!GLSEpom5B9c=nR4d5#W@Z!ncXLgpco3f!to-FXk1qZ^HUq4
zmqIMwi=+!Q$H$aw0iA!saf7MqMu|-~*1u20raF)#qjTOxdL<4%%<`P``**Q);LQ?{
z8$8O0Z5kgeH0b76egT3GJMY%k^m|`DKb(*^JHgtjoOgIl4ROdE_*k7vCkNX?zVp`H
zBJRQFAUBao9@dct=35^X)n}*)S4&D>O5Zj3EtHRAf5n-jaIRV+=E<z_<JDvLmH)5+
zohrUR%%2C3^gOfnBttw8dqD)!*0+cse9iPI*F9vq3LeNQiEr%upZ6d-`2K#suS%m*
zRHNjbo{fCPCD4;v`T8M~P06K<3G*JZ_A63Lj?MSTd1+}<H)@1#awU~)fk$FuzF+>i
zZg1prlGu=sA;fsZqH4tAn^4Pu=vyr61NauyW7U}(7^Zb5-D}26vRNVy21?DJLZW7?
zg|!14GM<dLWx{?AZB1*CuS5~uDo-y|ZGL*8f5N{v{rT>>9Jrc1{9cCZYpL$qzK6*o
z_#gr+Z2{H~2=#igjQnF_d?Awo3GH-+?6=x7Hch6$X+7dFA0-;j_yrOl8C|Ez;!$s*
z-Q~$v3!Q4Qr6)9R1WGw>1Xi4rD?K_8zz4WMeI+C!@`Re&8z^aVp*->4{riW2n8IT=
zegF;XfX-$c8gd~5!C$?-ZvztTJI9*l)=U$?h!dqp;N}*Xjknc#<8E%Jb51^iYzkzg
zLR%wPLVWR<46!7*4@tGOf=HYKn*Jbqk3$zUU}UP(eBJ@jl4$#>+rWVAtE_gOPSW~C
zDWoC&H<QDAu-!lEi_V%*?lA2*vs_VP*A32!Fxho+6ww~}mPvc5po^kazX(q()ye9L
zeI_=YyCIdX`Ib7AK%l3d*GGB6#?taF^g~2+*n;#w1B{-kfd){}5G?})FZGBiLs%$u
zBZA&yDdtYV6%Wl?PrxWflz#)ZAYED%kaFV5+JLWtr^MH|K;+4zx+?aS2lN}2@H`fe
z=4@&TAV;*Q#KXf2LTT*Xtz^RY;RGHEXXpZlZOY?z#_5eqrw40KN=`n#ysYA?AtZDg
zfHcsiAijn>g2hY^BJBtq1AvmI78Vp2?R)82^h|seWQw>6juA!{`k%T6wLwA3EPDaR
zFI|Q9^+H8fJe=ka)_5|Ed0g(vtA6_C|MfV5iaxTu_2ioLrS+hNZBZ&tNpQ;b47I0$
zh%t?4n(1e9j4Qr7fZXLA+{DE(2v|)20J3?koLZf0zZ#k3muB_V*RP>W(3gke*|R_T
zjQ&wkt<XyEHH_EeCi-J&bucvD9!l>aK;jYz*&<CweEKlaZ-i|A<Iwh#%V`hj_G<v@
z3n}hueMV_n*+1|<hn45YJhHp`?;s)BgiPY>4l!UjoKHA72;w^h;mP#$^Z-Rl?)p0j
zsFq*a4H(J=#!b*4sAJI@`YQsF9zS&M102E>j4;CrqwllQ=yWfZPV*1Q(nWmL=tlw^
zPg49~v#46sT}ZSI=}^#9d8is!mF9W!D&PJyL#;tgSIf#1I)jOiP1wqxDK(9Y-=Q=A
z<nr*iXZ*3&jW54}N_mZS*gc!;hEVYWiZ_9c$u~#{#r7I1<r`uH#D+~=9J^f?QiT9l
z8?nMhiFI&rC<B<hy@P|rLdPRQLc->CD;t~HFXYixn$I_&-P$ICc|QD9MGw*a=<IMl
z=xaU@lwa3u@+ZSqrXLs{?uHvT*B+xWZ}qpW?V{5if@vH9{W;-QQBeUpvMRUp<8hSZ
zgoK0*=)4s?-&L|3v$|Gdvs@45BOB0073zXu1Md<Nsu~ghw=OrcFV+QOgWwSGCSepL
zRcCZ>s4dnQ?H^YAKUW+TIouyw-E`G_L%G$)4gi%$0usYxM4ctQrwkAO+1tC=I>BwC
zHn}yb^<57R71Co?hwcEEb|^GhK&m|15=lIsZ;e2%dg99|J-N3!T&T^653LhyJ*U?4
zp&zC?67;duSl!fcTFs~O4$?ifl89uL^2$vagenZA2V9QUo>Qh`fU`lV0w^Xm*4q|)
zN5?Hlw;+(+N&May`}^0m7w~2H0jZzPJoQk}(eoWeVQ^E*F3H~>PD(y|#+m(hXmtiH
z+Zri`iaSc-$|{kTGSWAr+hG1Is(x-kjvd#2g3sb`IiWXX%KcqJxDJjj9<x*~vP7x`
zZjT$ZeYxrZHWkWZrtTnx^N^A<walhU<6S^->#wL6?=a4f`YY4E88K2RP)BpSlniOx
zK@-R3;Rc{-ht_%yM~lheDRW71ZeA}ObEG~yJKI$sSb^>(NZ4(e%(|YlQt|w99Ei;?
zgKmd}+|EUvZ{_#xnQr8wmQ7Ah$}tNU$ns+OfcE@sWJ65+Js=tNP{HcaU+l3pv^|@H
z+CTloW2m0E?NzC&2p{J8LhvUE#mkE=^@OFe>qf;rl(>-6!}ni({G`7{1>iWn%e3)N
z_o7uR0nkJP6o&|jiPbeUAiJ+u9AmfxL`q?ZoQ(}n@@tW<=g3k|-mTE6D7BzMCr3xi
zvl9n(=S@B+=V+U2jkp5Mz7gUJalB_8`kvCzpvTr+HOy~nY($C$+0px*Qe}dMi>v(K
zgbWZ(E2okI>`%(1M|Sf(71;YSi&W~>(X(x@W2$U*<p{Cc{GPlAW(#67dW0t2yT*PU
zGti(~lM-0-f4KSa<$YDxZ0S@4ln@E@*Ui8tKiyTlwD7~F$+&0F3IVey`c%e81D8(z
z{46oa_8P1;0t*U_gP=J(0;>h|EH8i-C{Zq59y)DgiO<i^&fYjtTFCUyd2OU5RY0Yr
zq=Y@i!vA+Uk?WeG<b==n?@zh8$Cfl;NN;$LPvg6Yg)s~1vjsv*X*d&h431ocMjbi-
zdAIoD5yaJCI@J6o{M^by_xbjLx>8Jh3lUdFh$@#@B<o%G#zsM;wn(<xAZmpoXm^z!
zX)y$<VWORH0N-Ceoeav=K@3{$EJ#6j@M{o-H!I(jkJ?-B<aJQ(&s0&E*B-O<w(QkW
zIv-0j9xgz4hlXk_N`I4dY`s5KRlPKtLo6F3ikcSvW~IRU_?!HW5hN5f1i?@56fR>O
zJtji<s$`Ux(kulxpTBig!e?%trNWaBSJU;@K#$d@*#c|UAQA69n%l*>b{;cIxUR0Q
zJ1c_$kg~iFU3LJsc)JQja5^E6<%}Q_T+(_#2M9wlKs&M7?-4K3=PzGUyASR@31kxO
z%TjBBb_)w&E@U2~5ip}kwgH<^j<k~(q*EYYev87pu$PbbkFjvq&sVn|{Slkp+T4aP
zM#pI8ncIdM01YRz7QC%Vqe?XMPcFxNXKs<+;vo(15S901I!@$z$Bb=jD%>#+WGZaC
zvLlwB@FbeBpnQW+9ye}8;Q8#?_p=~CRl$beh=s1l2;ln2$jH{Q_t&qiR&!|Zcir`|
z^4+%C{{xHt*-z~~fak-1@7}#ob{j*VV!5DY0<*F!FX4lSko2dZpg;;ds{-1n8Ial;
z9W~X!(F(l{Z@B{>konY~+mLaBzQ2$^JK9_I2Kvayj}Z*7eU!eRJ`V=|jjl$ZAV)r1
z`inSO=I7LWxSdFE8Ib$Tm+W$ezdl1`44FH;I(d5hv6~gX&eFvBhroQ6G;PM}Bw0o0
zyF2_4wtRVf<&jT_HB+;mk1Jpek-&7jjg0JN*l_;;#FAQ6p6;oXTYZ5p#ZaY`wRbi#
zF=6-)S7BD;{0JASube?RiH1IpYo^!roMr%`dNO9_VBi(#&QvKBp62D{ZE?csmxE2B
zVn9Sk|A-|f#3~M&|EkHZ>UvOj$QRsj@A9%L{>=;`4^U`}<2w#Ee~wDbh9W{s^}^bQ
zU)S%pAk8Nq%8ec&K)g`Jf8kIM6FhI=XEr#~1!xvE7EiK7jQX^FkZ|3};Wl)@(SowQ
z5z7`D;G~9DN((@j3vJ)$K;NMevFD8yJHv#H2Y$T=OX@x}XX1~qhN3BGq?2nr$N=n}
z&`%@HiZfri6^<R7^}-zhiY7s~4R{z7P_`giY5<UvI$snd*UD&V2$NCYb2!P7k&*BR
zG-A$BY-)M)lLHGF*|Lo~`rX`A#vwHIC~*aDpVFsHF8GKGrr6xnA~)FBqePs+_(=w}
z-*Gi*=L&agVy(qYg3eF6XVp!^J~*oY=QMu+0Ec{J1PB7YCfk%7h3T8=`X!3w<WF^W
zp_Zc6jXqVWHR}Qh@+_Pik57=`G&Y8;C5yFBJZ&yA9Ctf+Dc@i0h`aXV$B$Xmyk0)g
zt3LZi`O5sh#A@s6dAi<#a>?)7qY);9&G%&abUYSQLP)peDr3z3$ynVqL_wFIU<wm8
z9<PW>6bZ?(n65`D=P?|XfZ^y<L|@@<=KmH(y{t~H{AIGYEm^$QI|udPbUKFZmoxRz
z&otD(5$CyI`$jz;Fk5k6amhMYm$;mWsxJbTbl7#e6UFeMn}_7FZG%vjy5J)^8f1=l
z0;<lVw->s<wAI$$Ug^h^#mQ)Y^ya>i-&HeHQ(BNB^RSWVchzAC&m+vp-4I~S`%$+k
z>n4oM)MKgZ?!@>yE#oxLYUCGLlSa$EzzKk8K!M?dg&XCgv57c&hK(K7gQ^2AL4V>m
zop<*gcA^;rKf7^t%%E~!Eb=XjcuSMTSL01j7C4Gcz1?qHJjprY;16he13wNHg$mi~
zpb*gVkJ@<w>~HrsoC(#3|60{{J5T#izL8R?ZDxKqV!x=QQ*RLrOAM-844$xRkYByS
zoMe_6v)8c3+LYrpM_D%-M>CObZeM)u&B03+1qj!la%l>y9ApsuMY)PgfI&N6Z`X~Z
z#rUALUFLFJkiORggc%g2A#7%(2hTN;j`vozp@?a=PSyQ*y)p)L19(-PhWUsx9?;Ft
zZC4bHk`cD2Y}fhbt-P%7&`^o6G0n}Lhe1mOA7f2?CtdgMr=0`plZ_Md2i7-3agcI7
z;yl%MIa1`F*AQ@*-B33AGHUT*e6Fu@Kloo{?sJ4kFKQWB)zstBJZ`R@HES9-u3z7P
z&Qvn8vTy9|S=rdw`oslb&?#@0=NhB3&()N;_95a`MqKIkU9Wb8^?CII3vta@;-|V-
zVo#Z&n8p9(ap(dj#E7yo%THD&GZFZ@GrB_a!-K<;d5TN923lVWrVsk6r4Xb_I-%9-
zYeU`i9&BuEs3O)-<RF48@>5uCeW|u8dHvJXYk%3|^Nc9c;;>)GT>(?w`(0c|)x><m
z9fmXopN7xFM0^0%(Vs^0&C7E02Z*1&3Q|FNY0}y8Kj79!{XGQa5srOxVN_I9OQBr<
z4mo3(=<K+YgG12x8SKH<h*i512ie7YJz`3c6*TOa%g@m3vuUeKbc$N<*QlKv$NR}&
zZLpl!M{DehFX%MyAI7}PJz1bU9dlq2le1~())87vW%iLJLvRLibWK8SfASN(94!7#
z`8A`6JlzI68=Dj~gIlzPS_xY9vRf}-y?W)axj$614vtGk)OXkysd(c1$oI^R1PbU5
zesot7enO{g?$=0_{Wvn{9Y^=AHaAKl(?X2@z6FQG2OpFpWy!$a6is*XlTgu-Hklp!
z3EOS$|KjWVwt%Qc$G~{{<jFPnsj0V6;*12{Yi4hZfw>fnYY&VPE&FomnPc>byPOiF
zpegdy`MFP7p7C3|$Qj`tK1=@>SIoD!|Ip>p5Z&~e?5W9bqeso%qG+ace<@&|Q9<Go
zWi9RcPQ6@`Z&K;s(f%XW3Q%0IBXix(U6FvXVq<?Fc>+BC32-8>YR<PIVhgw30T~v!
zKAbE;a(SwI!k`iP=z6Ko+Wsg%6we|-pEkX^W)V8IqSo5Jtv`rp$7RG8hAOP&(!2t6
zd(DNIMC`H^Tv|0Bw^qRg`0aY?=z2Jd3)NIAuE+N!IJFSf3_ryIZHNHDdh)dS+mlBB
z{m3ZFFSE*lB8ARDK8u2bf*y=a{XDRqF?S#`xOH|Ks?==0CfvtuY;VlxBu5Bq#LW6{
zq_<BHJ=c7BxChh%)zJP62|TIEAO-bFQ2hz9v9Tw63~_OBoEufwF0M?PE<-qL`Qn`h
z*MZtYmv{6_)HqX-<!Xzp&mvuA_$ROja=zqXu~6Zfu}G#qOno5g9%htr#hh=^WPM3x
zU%i5l?Z(QvA@w^~Hc`XBv%>p@Xa*t+QRNF<1{Aj6=dizDhR?vj;{u50=;&y=st)?Z
zZofvd$#4O}3ZI#u4;kmyZN@?vkBs=JP7nfpLzEaPrZc)iC&(QYJ2GK>jcBy01p^!b
z)i>(K7yz%ITG+4bUN-Z1KtIoIypm~NTwX*8L1%rUv1kyncOo!}qW)sqK16kQ=O$Wa
zJXuH^J;5ms|G9E?7O>&^^RZ=ZRy*bu?vSq`jS<%z$E~zzXmNY8lQlLs|6fw%XOm)i
z6>9H3wQQ;^=h?wDqMvq8kF2n;@W$>gQ)y+~o!6lVh(cf4#_f`tdx;Bs<^&6qH<{hS
z-iBp*h^2<glUZymuOcute|%%An_3bL?%}dTwmqo{y1HJX6N;^)?q_ww{4sJePF5a<
znZ~qaKy_ognR82=nJq=G*ORiz3i*z{NxEzAw{QH~VmRbrIGFU55af3HqIJ$9K8R3<
z;95Q1bKICl1mIY{-K5;;<Ij}u&Ph~Vd-hS^`HR4`(&=9T5V7_}Oef_wyCH3*gEiwP
zS_v#=IJIzml$c!ZV$#KYsm~H}b^b~q+Wy63V0)7e%T&mt)H%-E>h$UZv0pP+LnjKz
z1>(ATM%dkkop`9~)L2yP^RJCeDeu0-?ijnPzJvs0T~*w3K6!zV?rxkvUD+<v)164J
zoFhGWJ>vt>Vo~pll#~F83ANEa3hEA?@GtSz{@vG~kEojH-u8a$np(?4vQPO09(LIZ
znDy5=EEWYhRmRo2CyKw$RgcxZe?<39PVa98@wL=CoAJTx&i+WyiW0?GsJ#+B&B7jJ
z%F5xvJM$ij?Ea;%rEa~$L}QCrZ4C?FoK+=0*|cew`giT$1mrrt-$yW>gPVeba8*3k
zKwa(bHoS{y5^4D~%%v5nacWzjP`eyerQZ5pceqhl)GU;gFSmQS`qf9DWn=}Bi9wk=
zK`O`MeEgxf$VORE`2u_Vu4mwUQ0Mv9FvmFccv%PWRl+nPl|=Ej4Oyg6QR6SF=uc%%
z_Yj46h>O~(0}>Cg3d=vF%rwt)GGnExwM~|pt@g9aRwvYXzByh)t{gMnCFFaNC~dJn
zjCtR69jPeI1NY~|sv_%;Yo98yq`O5lTe0nDD2#;w9ONUUae6JgDts8fZk<g`+1S79
z>f6{52ZyoAYbHcR)~w%K$mv4=!^A#%6JJUjF1TdPj$~}2YvcInO|2j+p@5X_`<zf^
z>deFUYa8#)XzxSZ9kAbiH&Wk3=^L9ALDst?N#G8o`^enh-CX&LgbYD_$`^mKmjbVv
zr9vX@v;E-}Rp${SAClbNq8tk^Uk4I(_pQkWL9+wK_kjA}z(i$y;rE|LukV?;G4}WG
z->p@kU|`gG3b>7xt>0G7{Ov>Dx=e-!m}kma>Y|CyIY^NA9cjbKOx3BdN7pN(I!ces
zPrSI^n6dHKOtS9aqPBS1=WQauPbEIfUK;)G>|PblWjnj0wXHdXPQ|s!x82U%{u#5k
zEJkxvst9m#y%y<Qj?ixM9@+v^xuH$3+Cy-F3=9mQaxHkA8BA-DE;yQovq^fC1QoWy
z$lJfYTKFKzDDqyRCjvk><>csz`w4NH3{s&o?peRClhffc7R;S0w~4jN>#aGe931Wa
zK7JAX-&_ek@ZA{m^Yh!<Ly<%9c$81IhSJyiBV@kvO|?ZY31)eHF4Oq52Qs&;Qe|0-
z*Z?%FBlUpyi@L#YO@kJ{u5*%`QDZr*{|?V1{8k&FEe8#C2t3}mF)&P*WC4g7m}|F~
zxQs`}gqh*p^?;TeM;(LiV-V|c(fhIMY&bZ2v-y|VIDpF&dh2ZvbJ%7`dw=@%>p9FF
z3Bo=C@tv~8*}{o6ot<~-=;*+Wc_suf3B%d$CXGSXAXK~Ka1sg$Ok5O;9rOC8mR|Oz
z+gguezMnxS#_CkU2NMRau`+B(_Q=O7lrR3%M(C|mIOuMPo%+qFWuVc6Kf(n_zm0$A
zLw(=P+B4Bc7I&Fm67F8Ze!1u23(rZ>f`)2ZCOv=oLr&!Lko}#4cZ%{8qtS7H2>M!u
z!~Vxy(3dysG~b-!^Y`}$1apq9KEf`K3J8XD5)`#mux!uE{JtwsdbQ)UVmK4hyTOZ$
zoP+|H?;vYtaX5E_E&7^3j>Z1|&r^5WJ~7Gv1cS$K{hMZT&>Js<2qMfGZ#S&8KM$EB
z)!GSljG~wH#@BtgeK5`@NF6ur_k3v2KYDWNNNYYpccR;Fvn7IhveIJ1wg0G9y1lEq
z%pOw3JU4wZd(Sy9&%jTi&~N;2#aHJyqQywb{`dv`_|VshM@QoJx)d{^!Ci{ARM{Lm
z@=pJIeg|X)plAiw5%wqycZfZ*)N6U(`FYjOjgG(k=pli+5(Y#7sfhw#o0*_JRkXZY
zkTAWnTE9^El*u{gbq_Ro!h8gt4HSV^SGLtua(wVIS>;H=sZH<nQx<JZQUB<(IK0vH
zPd-<PKE3Z|tk(LqjxHeZj?8Rms}Xl~ec$)bPRSPM+H})PlcSAFC|X3Y+>-S=48I+B
z4NGiSzsAQu0p(3bLGc-^36$xx{mOItcWs1F?dqX!Hn}towxKuJh71!;Qq6A-=%PJ+
z>0XPY^EmyqoP>-mEgmupIRYn%OR`@tqWGCz+Nfvbcue-B<R(NmLtXE=id>6NT_W0f
zI^cKpg?M%G)vJznI)6EQKL|UDTo}rq2~?!CxVQDj8_L-BVc;Hi$1001<DC4bR^-#;
z>@8f`gl{(j#_x`t-<2HvWiflqY2`-o?)&LjPih3s<HuJyI5?mOZr8K#qk;2QPYnI=
z$^h?<i1Lq?+V!&+K+8(IF$ZLK=(F1a_U9)-K-4s&N3aoR+j1K_o)YZvsJ@SpjxDAB
zSmd5egB9^G@fL<&t>;xFM2_yPu6tAeNzB66<&eJC9|Dz!<OJ&K^-a6SK?SH#Sd~e9
zX99_q*)P*;`-b^LnZcWm;)H*Qvc&`Y&@L|b)F%bm&~|DGj7MubuOe0Q;?EPv4!pi)
zF#f2fjwOFzzMkA)ZpF4!OE|UXT1VaeX=|#Kr^~7;N-VJa(3)NCFlf7VazM<1o8{+C
z2XYpRsXtp&YY^eGU(N4-z@T%BKmK>u0HUVcY<=K2FG|_#&-~yTFn_pds|%1K)?!l<
zvg2G$H~B;hhykF5IwEULAI0}E4)9^VR`dZ>Ie`8mVP|u;Ph*OH@!-3PXEKT>N<qU0
zYL6HK-t=F#J~1hm2VP!}A?#qohw@Jp_h;B1ytBU&`f7&J3DXmO{=>=6Lf-DtE600E
zV=jwke~%4$&Q{r3r2Aegwt-0w{&ueEC_B<k48lf;)j|ERK4Ul)Y8erbMvagC)<%Bu
zt}V}27)}`(E{Pp~<>no`GR_~~$l8Km#7N26P*pN%OkLIUmb>p}R|ShjCEd}xy)gR2
zE99Ky5U*JEVEWmh0t3Zgpzf%}E9|4W1^zxVOLCI<KbJS1t>~D1@1y{QyuAR;P0uA0
zvY<A2duTV5gdz_dXk*@Rw=R1ZGHNFE0i}F_p3O=UXAHrQ-sE@}2x!a>Lz?YnX4c`I
zwzjs}VMs~l*s{YwsUZuyiu|!-d)=e>bg*=RgXsf6yP&u@{069w*Xv7S=j3ae0~t<>
z`}Wr#+3f&S^kNzH<>7{oKXb)%ufVME?W;>3qc8Dwxuy`KDNZ=h+{8Pa8THl&s4#v5
zq`9{@zn&=#7&bpZ0_wbd%*BgEY0r)1Ji(w#3^`qQ{ZZ~SzEAEbEVQ4Ba8p>$u&Td&
zy^qT(Df{tQ>aJ>vD7MA>oL4%5kDS`2^%vC#+es3!jZsYk{FBNw8YZ;_vz9q4g}U<&
zcmD2|LBPGO<V=;zNTtA-GIuZoIK6=4Q4l;OS>TV*VUE}Z&MGXzclpY@kkU$lKo-jF
z+fuL?j22UGl%C;;MX}52<to88K4WyXu!><bb9OG{_rXtuzOmG#q;DXN*#|sJ$}6YO
zgfPcol?|YD$FTFKC(xL~Z%=$4(7#5!+B&0EuN%VSJ5|>(Q<QS`5N|m>+O>hs>xPhV
ztZSR3=B)Uz|Mz1#1kP`Spw}4>O|73x+=g&nG%d0_rq#d-dx=G*wC__&YBt_ilaElc
zu@g&(@LRC3CW^nP(vq#P{`{yW5a`MvcXs6|Ute01BT+a-f--h{>op|!p}_UpH|dQl
zccab`SLrpf;|;6dcoE*S7f!EyIOw8$)`FJ5*fU%!heHRGx^@$-Ba+Q34`k>r)PFQi
z1C<NcpfNK#3HM3Ha(i!XpJ`!>UakSuPT<qO{i9UKMu;IKt*4j7ZnKmC^(5FM1;E>G
zGVtpaf`AEDiaerKeungZz4?DwfaGoO2oATgVyk&W$jaQX5^{){(kQV`g<Nzpy>@+|
zy{M2Sdz2&8N-!C8CFExDm`}JTiH7~8lu4p^^yn1?6`vKYLVl0J`UE1AOfE;(Vlmum
z(+$4;FWNa}W}GXkrmw6+91@il`FbF&`)|c(#924p>4e!LQ)UBj^&8AAa`a|#N=A9A
z5}qpL$+vrDD_FOb=th&XaH+VtmPkg-Wl7h6sX4XNeV$@!{6y_+{5<6_XcdKN78$Xc
zvS<YIu;FRrT7D|)qe}&qub`<{wLNZEEhbL}5HILMBA)MN+2Z0NM1NzjaBz}<PdLQ!
z1WNeyA#;V#<636em*LR&_OH`=xdKq&OG7lB(QZwZ$!tuOHbex-1$xn0Mq+`npz!bh
za>emlw+AM73KT@9IqojyC_W2kG74TRo+su4B0@(*YWwb8sj1B|$7LF^a7H0Guuw$S
z3tcAQ)lW-DvpkS74WX3ElO2W8ct8>ka}TzHYFwIJdb2(w)srW@BEio(iR?z=VCnTY
zCaT}%`|pwwW7<P#J<!kZA?12D(jm3Aw^6TVzA`Fpf40VW_bk%JBidvU>ZE5KTbQ_w
zi{`>lc~V6__E0FMz4d2U?V2PUWJxfXukjfwHjif?9TPc?uH@sFSyXScJ*^PhjyV)%
zGf?t>CEX&Ksv9cXOR5wp{ef@DFX@TJG2?c1lcBkR6?*i6hNnDueH@q<Cph<G{4uFe
z?~<LhQDjwzAxfsQ2Bi?PPX0_I;hmW;TfzyDfIvY*>&Z~cWi%OL^!yk(DsNK#;f8;L
z0I|bN8I2SqcH|Tc5bMCLOF}}D4SBd09R`Eea3&PY^LdwbMEo9ZUn-w>jv2e%ut}jo
zb0Fo7a}0RqsNHxb<kiK1cP1bBp%RU<%M0FUCnhFhXB@zR&xFvXf_X%fjx=<9U~)Y<
zcs<Vrwb7ng>C%Z`Z=yYj)o?xJgPeR12%8Mh)q#6!a<_=rtA~L+@IwokL9zMV*Kyzk
zaVQ}~;;u2U(_ec+Vz79;J-ZWPtw3a8E3lZ4(^bBk@QA*vqom8bbt}s)<-Kg3#&V^4
z8r`~UVPjB5-J#WzS@_|^Oqu!q(oz@h@2=y_#Pf8zQCX3a-0wR&+KbB=T`jGNj<47c
z`f0_8PD<|W*N61;B9<13B0g^>z+X0Ltn?g5F|l4rTb%+DdydfSkTkXNEYGa+w19Lm
z+G&ll`#`(lJ57$~l0TWl=_i424=m)1BC?^Y#k*OT#9c@m<2ancu_F}%Gzrb7V&!>s
z;o&I4k(Mu7{th8mL8>GAeQY>dkbYM+Km5zzT6ee*?~6h^?zo(Z5OltZtggWY=OwSw
z@E*E!u+vSTGvm1Y8kE3)KqF@Ggz0wY<gU}s=zW=4%aS&Z_JWu`gU$%61SFnCPra^%
zdIu|<x`(Mo35>-)J*DAY_oa7MH!y<&!j<+oP!RLnD4OsqJC!jmUNrU9ZD}HSLNQab
z-sE5DpzInNLC4Ll9>HuP4Zwol1fc*1pHbl3f1G^|p<^;31yu{a-}7exMxpGE^p}O!
zfseC&zkLfN)qVFa#6AOh2nu!iu#71#u><ZV3ve3JP+KRXub=W)FZZvWRkiRU07`-#
z|9-6j!32O6tOpM$gz%XR1G~Dqww_YGT|5AM3#y|%cy&w7)Y&`MY)H;k(TQ$GtGM2Y
z&)pU?MG`($lX3j!ci-U`p?oS?dNz=OKK_hjelqNfK{(G`ZrQ|Magwxeiht|<339IP
zHzl0Xlmhp5w_7wqRyYxJueu>c!*1ploOi!;6u&{MkYc;%(T(djsYx?tT(r0nbz6eJ
ze)-Y~2Rt!={9Nf3r`4ROP#4f!sm=Xv)wBB1PfkuQazt)UBkY5(tAS_2h*NH#p3?hg
zK46??;xBrS@DyAhd03a7HjZ_D2!A&>H&YEX(ztCGzF>L3%><{N;NHETaFpb)A!9&n
z>T5Wo!B4mgk@?%BF!x=E_$`3=d{%_uxuV%kGDdQaX<6(dxj#2|I^D@HElR0@oOU46
z&isSlv^Pg+yVq7yA<V;@SbH#mM7QO)u*c8Dmapg4a-<#!$>D;$EC$0LY)VclT)!HK
zjf37niY+TI;;A1d6Vz3C{slF~CV4FUg-PUh(?cN|Ctr$f*4w4QO{T5#$He_Xl<w9c
z)tVi^L%0jM1S7~NBd)@XWig8UPO$FmQkXU_8P9;27|q-5&2HhYPrI*El=IUPx_Wb+
z_Dy|=ION~Ie}4pwGCi3pEPVX@AY_|0pEv5|ViUL?qxfb|1Z_F$&lIYIKiCDWEZR}*
zOKp~I<%^Di?J@=0>C8%@xTCBIj&Mhf*oYe9Jhq{m{5s@jL`=2&S-LM3ZB_Mw!px%r
z!gOZ0LUuEzwN;yDW{F=?CFKRA2DOFr>NIGG?JhD`uG^E<527F%K8lc@E(~#qH1U2a
z>`%MQ)hLq~r0j~?4Oj^g|2zQs5b?OGayjlu<{S2Ma%;dHAQgg*q}mGwgn@(eSPrNx
z57I>-=%UzUgA8U6DWuw^5W5Hz%Cqb18h7mHvo-g0XYXwYcWq48+8k_XjRM}Sj`)TB
zh9*E#B=*PM+#G_g#C{k)ZAdCZRgiC2Aybiu{}7mM20u3A-k8T2v@($nxkWkRgGslO
zMujKdKF#jdu7Q}z%foG!Fs6m*<QYr0-Tp$~#@T154Xc=D;n~_MZgM`MH<u1$IGW>{
z{lQj}0{$Z#zyPfd<_bU?xG>U1Skq@AhoCjdgdERL_k+Te?Aya2|7rvvvRMW@12rh4
z_D=vaQ5;y1C_32=_nUMWupN#$*2|{Drpi)AuBr>AsT5{a0MrKTK_XW3mw>kFfkrKP
zR`cR;`;FE{iVeZka|lL?{U9=6LAw5$gC(LrhMPyLVbZF-yRX&Zv8eQ^rtZ@u%4OM4
zPTE%xlj58kQsGaNFW37zepcGoO01T~$3`hosDIc~#`m?B*h%1NuHEowz*H#vGiudb
z7mMap>f)J9O72aS5dibduh&oxKymqy4k%WoWs4FA2GLZIxw8yrN|1+JfDrl*Qk%6A
zCCI?j3?#Hpn!_IQI_~x=v>Q%#0N6>$!FMr9EMmr*olFR<-7bXLq|eKtF|$cr#z!B~
zYS<C`9-mR){87@i30{VPE#|kfaxBNAA0H$>jAxKY!se2gT3k+dh|Lza$+dp2g7?J$
zCDnG)PVNR`a#%Qb9%Q^DZ|j%YS51RcM@4@6v1E*mZ!UIYO2V?(`nhSeyf^_>1euJ7
z7k7Ni5n6>gAMW&!QFaii%gJVge?sXRbaQ|8YL<B%@DQmGSuldEtKmq|rcxooLgM{|
z)(H}NW?mu7h)-mucs{xtVFlVO$)e*iBs6M&!X!~QF9vg9X)re#kTh8X0|T!c03E3G
zgg>y{%!+TKGrJj#sXSm8XpP~~fHU0-bikZ!obJLYu}!6=s~D>Ok+gUxU)Q!JpZj)~
z8;}ZN^bcxZMNE$B<Irxza_@PeqMT7r=UHV%UnIRx{$X-|Q1GwijA`RyJoX~3EZ49O
z0g*G7H_ReQrX}kRqlX>K>S-s&kj!}J8G^~RBMF5PU>3sVFaio!!?CjBZ~XCKn#4m6
z8L}>7U%xaBuRM1Pcwq)EQ#a_7pFmq`1jI~fGRfcJqaZPpK{06nLQMt?7}W19573Sv
z=f(&GcW{k`0acO)Ojba{lt!acu<#z4cw?%36U{ec0<>CyDTI<iKc#n=_~0{jv5Y&*
z{366hYQs|IQYE(~4!W1=u!`ACRZFWBiIRyewZiG2@J3{CzA;2Uh}5%}wL#<k#Ti!b
zOyNToJ1{Kgb~DQhKHUnMs&G=b80Vssk;J|s_a&H%g_muSe8zG%6DdtDJM0rT`>QA?
z_4V22c4px8y+=}Nzd7~RY#8u4U+Qn&y43;6r8x3v`=1>qhY^T=joE*!V$8<Uvr2@s
z-r_h#EuYOS14<K4t?k$}Y18<)24Wq5zjCm!IY22?*q~L;69oPvzE7pn>QClG=*ndw
z!Czh~b0y`YVZMQ%otAcICbsa(&SCx#GlgUH$udh)DT{=^9J_z~#*iE~A>PuMB4UHx
zE$`lylkqLv@*ImeEHou#=0^`N2&y)aIr=i%EF>XmFN<!o?&KZ)VBd#3v^=xMvlP22
zn#(Cqr|Di;>)h*A0!jL!{0}VT<N|=xq&l!OH~&TKdZ3lB-}xGdP5Ne<0)l@0D7@WC
zQ#kSy%yde|z?avnc)~RX)v1$CNNI!vKm8G`%Tr{{_H;n+iP$$#m}x~XVGLm-mFz=L
zb$Dc=<g`iK9FeRPV|$MVA2HzA*XjK!L2HEc;&$I3kcYG+My{ZNW^d6idpIn8`0dZ1
zHwdQy`jN>&TjZJb_3V)mRT{Mt*}`$xQc{DHgG~l}AN4FOJ`1B8>wpjm3JM~lG;1h7
zIIu52oQuF`HVy?ea5flzsc<s=Nz;3k(-LQqXfg7KdK97t>bv41Kll~&Fei#EOJl2-
zriE+tRy6W{({r;aSw(4D^XI14cH5ARV5C;R$>YI3K3Hn3jTkjjS&im%3f2UaNz=QL
z;>kc~A;)AoqWCmG=%xqU{Z~ulm6gya&AOGTf8mD!@uqDU#V3_3>kyJgr3Er-waLkM
z`|)U~s2u6I_28_IgkikY)p@><mTdaBl=Cn)kEhgno?I!BW$Wl@$H6%E+uW3!q}_4l
z843z!6RN)m!P8;2E+eOeXI^j3*5V9(eu~&}UknRd7VcBN!?vCeA)7QbT?&eciAj^s
zOnc>VbFp8;truX5isdTri?KIM2Y$s(SOTNBH)j$80TlwC(R!OT5>={WRCrqd6Ba?7
z<eeRi#r1o8O<bWgToq!%1*xMGExmJ}MA{wfCG9`otlZbYJ3;FzJx(N%5F}fSkJbu_
z9;Y-BQfS4o=~*9)5RoUj7!>zU2rV2<cwpv%g#q3+85tQ+gE}nrXUCxBQDk(Cpx8s_
zIAwD-R@S0MZD1{*nGa!U1q9D1EDKrRblmdR5OKm_LMKPUKBge~Gd_|?y?`X5V1Cn}
zd(R9lY%;Q(cC+s7{xKOT54u@K|7_jHpzF%wAW>J=-=7}p-ZDpr2|MF7(yHNUEMGb^
zNqhn+DXPjjP%p+d@yL{k$fEd6=uKPD3^0>|HnK~)D%k`RTRQJrO&cuxj#JL25*#Bg
z$P?uk{!tL!Bk`6SW$h!PwjZ`S!n7f9RCeum9Cz(EhRWoM$$G%B<Lk-HS{HCj>p>Ul
z`ox&3=WZ<&ie=!R(N<>eAmXUrNS)ei6J;}+Zb$i$F{O<7LNFt_OUI`(x}zl4W^(oZ
z)IdIE+hs91Hi(kmkM`fwi?R_F%%?IOEc7Vh>d=_X|JlqaQ7$gV_;xb=>#=PybE`_`
zv<fDmfsIj6Bm*iktq(WOQ*28==~hTw%o1tV(;fGv=Y2xFG-`qI6zUkl3>jnzirocD
zUGjGn&pF>r(o3a{<;kbYui!N3Zz;@|9;Ixnxdd7;$lWgWrQ&+Tm-I6Cv7ww?ru^<K
ziRM-n*X5uoNQhDGNt{-Ge^4(y*UKFgwvHq%2R@T%s@v|_FCNI|YvrXef1}+`tucH<
z@sL5$HbPBW{oLI4)`g};!j5XVItHND_9qQ@H!h!u^s))7G7#r-)JT$X#UPC6={sd?
z{{00>xGB>eNG3=A=P&*V?}^IQu1s*~cmDSo?yv5+|9*GIx<dEw*8_?|N5osItkg$H
zrdRKoW7vA}%6s#@s(!`LII(Iu!N*oyY{?eop8)i=bPCtZTH2a^H0k(5SN47+q;<_3
z|K~wh)&D(nL+X0jzh5Wf_3vFr!u4-S%$N}cd#QQWnljI^=yjU>k7{4_?-!qn-MM9d
zW#X#f2toGli(tckl87Kz&clr1*XasU-D9ggiv`qJ-6a(vRM#&)Uhx^`(fosnTA{L>
zTL!{9{w<5mr8%yCvQHkaTl&j|A$R9n^r4tU=D*I&TFo%3AIpmUEV+)2e(`PRn=tjj
zfIRKwoz@K@_TuAEtR=cpRgonXwXY1P%P%AHW&(G1=@F-Fs*te9B<7oJem(sdCE>-l
z573yEuDM%kHZ^vBa~f|!wHucEkyIFBnkmOR5|m&b!1!GGD5KokAwl{(=~BAZ{h**g
z*Gbfi022~7oHnH9YTtI3xP;T&(#f|BoJA<hu9Jk`Viu-j3UO4BIoiZ(by>OJG)r~<
z_Cg)Iw?Ce^frQC%uTWyxO^HtK1;ST4K_}xGe7LyDFFpE@NHtfcG{l&)kp1%W>rW7$
zKZdx01MXT{K)+LA-qn+Y<0WB#dAF@}Wa^8+AyN=Jtbx5BSASADb-|oxo`O8}@k_P%
z1d4#~e^zB|EkDhpTrPDG3gUy3HQ#G<=7+YJ<zJL;-ZS#+p??i=A_11weMOhc4YZ`J
zedqN^NIqQ-7{JD8<Nq2nM-+;rJXdWV{?n{w*Ep-go9-Rfz4)xi*lwWx$WAD+KBl?!
z&iw_~SDjb<6Y{2;oZB7WcL(k7&i}3~n&rVvDC_^c<;uO`oIbGSMmu24j3@i`Y$?TB
z^_}sD=6xE{rbY9{`njT@@LA6F^LA&NU&@)5tc0^_(q9Hw_j1(>kz~2Am+9&h4hKa<
zEUz*&|23V*?R2wZC$eOxZo|~F59yxtlSy>rR<snaqz&cTs6XF9(G;=87|&JzqwVWV
zn=G@(a59<9Xl(0L{2jCX*YKgsmGg;X9LwSvbM{$^`Ix&Kw8IW#=P`J(%9-+bXxQ_Z
z?=ieQ+;LN8ZrQg`u(B#boeN?iKKOf-tG)J|^cP_pWMc?x-D26Iv7pZ#%E%es8cPUk
zy#k6lk@!iD{B9X)GUpwXrz)uRXed)od#`g9pZO=ayCcylTs4Ea-DtlNUs?J0sTEk6
zgjlj!)3sdxF6yiI858A=vomuzAIY4@A9IikfWm$1HYOsgSHE3$-rm+mdk|=*udnZU
zM8xp$;Y(Ln>xJu7FDh!URC%N^eQhiK#l>s6?>jU6V!Pg$aCPqG<(}qT9KE!AM5zlR
zOG3AC4D9p<k=adu*y^SUiv}7pXHOSjy*P9)6JJ-MA`$nU)tfrtR<J7Xy{cgKXX%@0
zNkr+XywoSM&p;qdXc3vMrhXohOSDPt@+BAMsmhV*A`wFQdP}PR_{g#1^1YG<cQ&rN
zd*QX7;j>P1AL5-RG$AMQeIes<u*qT6iqUzlUlmb{b*H|b=5m<!@MKAhB^$%E<%ZFK
z!iJ;a<E_`jq(+<Mpc^TX6MvVIdCRyoj3nr}45`H$US5;zt$o`o8{cvF8|+IzqSTUw
zSms#iVBGSn>|8)dJoyMYkjeg7ckVolFKzeQV=?SB^07WXU<t~3E~OkvYICdoINR~%
z7=3a^*POM1OPY=a73=et-7(IpEPD%2SM9bjs%db%`J~6JuHe#n)k<xsZm=A-7&AB(
zKSqq2l>i}w;?p_m4f}T_cJ?yZj>dg?sI5%W@|g2GEM8sFLobcXjxIfw@BQ{|<8|W2
zOsjA~8@y!mMnUa9IR0d?GHPdd7@Rru>s|Ku7<RJcmc-CBDZ~DhxT8F?>{ulEn-`kH
zz3b7$4R=)1DxI!gB5fO+21T=wOO--(8zkAqgOQL(_+SGsfBwIe#Q(c&2f2OuwJL%C
zQC04~h=X(a^S=-N-=(-b3IBU4{(o{oPRiBX$rF>35Gl-|Us@hO)E^fZh~??!l@6X0
z!aITb!Nt{k|6P1-DzU1F=;$oyfKb}83-LPTGW*{+&>$RA1Nu<XiEu<y?);MnoatTH
z%fJ_e!a5nK4e2ei(1u<)jqU9cU^)Q(asS8L_w#_7)&<DCOQyebl&R7F`({yQ*Ukhw
zDrz@`uQTD%DTcji@MwkyXehUCNdiIZ&dGMWh`(NIIGI=N<m{!K2dma4=;;Jk41}9R
z{2>bmFxxWKOFIz4sp~&)+eeANtDZ$@S@}z`h=dd^W#G6G-45A3G9jC#9w|x54mRI>
z<;>>4f1hnJhqTCToWJ1J)|mEh877l`AU@BdAwwME9kKk4u2RG~Ms)>6ecgdrwWsZ5
zsZ^Ddgx2;j8&y16#1*%d;dUfuHmV=#ruxq=j`3r+emt|MRLbh+dL~Fz$=;7A;2q)5
zwLXv{(#`W+t`LZIgtCox2T^(XVE=#sF6kJS>pJrYh1H;!->kQ5iw!MvaJ&meUf%do
zyG$l-HL%+Jd$?TfCKZy1X%^zoHSM<r{af_{TJ^SV@hAj>rOVFgOshi+6`$ekJ-$j+
zEL^!)wP+1NI16DRp-w=>W<wV(`%=<>YrgSWxco!^tfc{jv!Ehb23FY`;;kVIWzf<A
zLb7oK!_!k!I*j<b3}KvdOE=~l=26<Gtl4!pwgak%Jr9sdYvY(WYz+?kGQUc&C+<WN
z_I~~>IUTCMyFBp9=k3pOwYOnxYFQt+(PuZE8ynj?*9llCP1m`}5)v90t*O|lq)iBz
zWyin&{)&dayYbE3YkSA4y1k0Jwk~VpN$|~U?OxuIB4GT_>{CSUa$IXi-cX;KGOyS_
z$MA}uu)7)GsTikD0Dvl;3A+ce(&>3uvRh`m;GCvIC0QB-2=Sp*@{2~tTW#&#%?1)&
z4pV@%VPjz>!0aS~onNtCz4m`M-9)!o@XOW2J%fz$m)v;HG79zD<libQ|7iGjrW0P0
zr=QjDHJUlLI;=K<GWsrGNw;v;<z&d1p=lv*zV>&5jH5|C%aQTq<G~d^H^%0L`D7Kx
z-JXHc^_@k(GKSxgrrp)`k+>|Ew!%J3^M$QoA22%U*ZWrLrKUUD8#dhh$dQo`(ZTOe
z)Ll9eURY&irRZ7E-YOF*Ui<bL4+n<|Sl<XgZ31PNBVo`uOY;2-r1tbcp7&d|qmT?r
zh0qlK3q3SCJ$C|I60W3bZH?ZxCjz>DJ*BQ{yJjOV#U4MAtuyZWvO;Jg%9^@vx86Mz
zRcgK4-EN>*nZ7&O=!G>=OFhb^HP$SLzqj{BcXp2IuW_GG2&4S^kxlIkm+*gB0J%cp
zP&EIk7rA*W8&ao5?<y<Q%Ve#BvwLQwo-k&N_m0sLU7WE4cVXO(XuP%h`uZgZ`ECzT
zaZbF=vsmX=JMGC8Zf%y%KC)qqlaR>IQU|+;h1u+<f}?HkN|e)1&lGm}+3@344p&v4
z%CoOmTXE;V7E*5G>i5i?lMCSwrk2-RuV$pg)}%YH>G!@~R8VHx%w{dm*DF6eUOQ$F
z4@|&aD_VUWa*<d;au!Wg?pv}fzDRTdE52cWRurOSvqr57om^U)X(3S|C_E_fN7T@A
zas)CC-rC&uiwD*sEqcl>`!t3s@!`Q{gK5a)=f5VNe8Ua!T%O)p2=v9$WHg+MHQKR_
zA%+0vVOnUbV`_u0viUZWXIQJ_p*LQ5vGw7Nj0cx1@l*Kz#++AaO9-6;WNR3qjs;#W
zCgI8`ClKI|c$H(i1SX!yKtqYGVoC*xW%*BD-ksTX@7<P?-C0$GI)9Vj3ukuY0o}7*
z=V#pl*TR`;H;c=c`+Mj6_XA#a@sc>F%zXQ{MU8d&x-#MJpOeu1QMrQCEPoifi!tdm
z`n~l%g?N-K+;dbp{E2`9V%8M1;Ae8Zu@A+T`b&FMntD*sdz}>2;#4ZnmD<IQRZltW
zO7v|_Pr-VnJd@2a)gKZpEHso_+Rio@dC5Cm(6vJ|j4*HU-Nt`errf)Dwd~Cxy&LOc
zAdLwz1b&~ZeIvm|7G6&02^RDxUcQ%3rMnWx5v!z~pD88gM#7pxbBt-9Q--qY&F3cQ
zufE`F5Y}LyrGkq`>au%$kIkX=Qf$|yFjCtu_oT)Hvw~A8T5Q+BVYZEbLUn6IpaYsm
z!mHK@{gx(DnggTEAtuwAxFxDAUUCV;;c8w(eC#;p-t-`cNOrBbC24A9w>onD?TXW{
zCemBX7a#?si76>cg^jD`zOb}w+6`FPSSkE>Cry293^MhFx}xJZqbV|LhOKr&NmH#v
z^M<u&d2CB1d~aV&cEb$Z6BMDQ#=y}@No&Dh!`i)^T>P(vzYc33ez;GW>OT|LK>Nnu
znv-pC_4ma$Ue3O$LXGvCa7Pl74sKoBQ&f_VAf+UIUbXt$`IWimnf5y?r;KM2a{gHO
zSTlPU(QTwyOE9@51tr~;e#FJGM~5X6mO4FJKkKm;lFm$%<C0SLR5;!_9f=8_+fzhF
zAugpTY`9#up!-+o+#m1icPq<#@wQ-)*ce5Q?~w$O?ivpdU6Ye@Ud3mT)|ZTxC;Vb(
zgq>YG+f{STOLu2mre8$a@z?8%@k)k)zedsJ-a;2ZdRkdj*}}>HS}(X_yj%iE%8zVU
zCQ`eo@{j_A;Hb!Fjb_oEjY-W}%&m>Y%doL6btdIM=H=y5e01@s6jAUeuoi>Wx#Y6>
zZB9$hV{wRCWs~TmHO`VPS8TcLPxDcD%C!dSCQ?;C3HsM(s;4KcL`uiRnC34fVjGbP
zwU0Ok2pIi~aiAimb|bviar(@OOlbe)AYzX+Ie6*hyw0RkLB%e`(Y`3Ucz14#;&;tg
zlP*O&4VT1wxmMOK`#jYBYZRNs%*itQw2CR$QVmLyMBb_OLVA|^(yyub4q?OJrWHTE
z_|Z#VQ5NH<(Q6k36S7;5S*<|STlp-R_}C>(^o+mNVUtY`HQVw^TBhPtu9H|edkJy^
z(>A*-RyNO`Y?sZG>dr454f5z$k6kmArsj{~+C8|D^6!Q+fXg+%)91F{{h1(3#!@`%
zsQpB}(Dgi5o|x@vX}3e4MJkTBp7C(%fWeEMx$s>SRFtlfkZ6U%z7YFV%XO67Xgw0y
zl^L`&;YOpI_YF^*LL9XY4Ohj*DXmxD_b~dXl#7QrZ+Qk2v!}_UFD-{`W>Zo<`->WM
zxqKQ^>#J9hWR1`G<Htxg-IhjvhSiHkWg&FJe(63XXH~|{$w?H{=hXBZD#y}YD6hwf
zO_YtAQ|=lnWoLz??`JJE3evbbm8qYsh`1co`Z_wb-YhTI^Dy-MlqMb|iAP%ZPH$1r
z@e^qisKUwqz5a;vas{nduf}MIFV4cjb(lVx<$RC}W6JO>D9-gGt=$;q&3$<j?l47j
zr(P;8#I}r%B^PEbb*SI;sP^uaCw#A$JBIHX!;e;`F5`A4w|#yhGK@hi?L5arLlEAZ
zmAOp2{PQ?#F?r5r$^N(U#fv74;Lrss3TnpkkC;ax{t=2N-|OxQj1I~S|8Y0%9-(N8
z6UMGlPe(fMdLR4?uG8z@oY7&6qOE(w2Ql-N)vb#MU3fT>Px*Ox_yqW=*crNp6AvQc
z;G-M%?bza7d?fbXLxeY81tt|hS&uf`EAw32iHRh{e{?U(ipmBKv6v<xhA5&^;75`k
ziz?LK)_1(^+A)GnZL`utsXMPcW5$>!EclC^#VF~AhpFVctx36>o@ZcSa8}Hja{Bl*
z^I}I+fN*Q0tZdM8rW7R^7Y-ZAbJ{2@Jp83qh6|J36fp$L0>fonE-st!zLf!`g#wVV
z3x&G`J9B>r4f~6J)Q@w>!WTDQ2Vwj6n)YA2&(qQG9nx&!c&*YICqhiAxH$ZL9>4Y6
z%4bPFJ4+(QS=ME!a9cVbTSQndY3x8RnEPZ>K*_k^bz$~MiI~ReZ^cLN-#f0<zjvHU
zA{ZRFkg69E=)n7s+3K%2lc#f|=k@RTX&7i~)~gCiORk(si$Rb*I4p72%XcYawKUIQ
z+*NIR#M{w&<7K(U`SU{ppRw|5>edT?gw)u#q{tEzMf{JLGqMR(7^UVz_4&N6Q@7}K
zzwfa=wVjBqIB`n;PWm)${6+inCIRwd>-pI7>{7WAh<SX*Jvk!ou#}f>Zs;aXU$Tt4
z1Vd!KrQc|}`&(J&Gi?UWtq*1oa@a2C<Bc2c<pdqA&EocCh?|&&T+D)k3{EKpxv%J<
zpv)Yd<gX@;bto=cGwhbGcgxpIxqB~TAiS3;%_?JVWhDLwYRE#l!~aFrS4UO#b<u(d
ziV}(f(h`Ew-Jx_hTpHmL(%mIUNOzY=cXvv62wb|myX$R!-}`>w8{>`P@E<+rp0oE}
zYt1>=JbN$fI9iuAIUAe`)mqJgbyQjR@DOEEnWw6l!CSjdsi{VFT>$;<_2dykIm7Y`
z88quR3t4u~U<qf=hJ)OhdIsAoUBXVe&IbWA<&efyS79(8xRV8?^No4g#&`LM0*YUM
zB?t<C`&!=VTzZp<TWjeLEDi<cq1<Vh2U4<ub#@JkJdexq7xis}CNMq<r{3P?XZ36U
z%KP#~&ACeo7kpj7zqB8Z5@BvUqDXa6r)t;A&tkJ>4YtJEeF?EdT(adH*KbentyCK?
zl%}C+-pK<0z1b&0$~n^z;>Po<ISA=T)<d1_sQcC46D19+BuT09#nR$WRBR9#O_<mM
zN<)@x&5Z?WYyRd9G21ssQ#|li#={cDA8Dv|o+$kT$qpam|F9&Z!j|XkD4P)%C4r-)
zc^a0Zylj&QUT!~l;#2p6cCq_eC0Yu*Tu8JsH=P@UKv!hs`1^(c1N1uO*YNNkcpWTH
z$eqh&&}+$fj+;aEA0P|!z2>{vOt#FQ#Mr7$vtnKBmBc38@lcVZ8n$1YnJ8+zgmmoe
zsoL~bN<R+KcSFG1zxy?E8Ky8=TiaUHD{pI6pYlk}jOP@IaCY??SMQwREW`--U`CE^
zjkVV<h((ZVOex5H@ysM#NkcOmCkE$>QDX(ibIy%^P*!;+lQf*&e49lHZy?o>XXxKH
z=PLZ;3(@%=P5DU9#m1fN@XbGHQ1<BYXP*HZ$bd$jM#SluesVAOV7jKRJ}WMZ#hUTd
zuDwh&B<-%kKH0iZxiU{#X@b>AC5zz}#n-FL4OX_6LS4+CSZ!SiQrW)6gqqS1uU=;1
zIgm0Nm2B2j3e^rL)X*W%AK&mJ-J>%5g)0<)JS~y}Q@y-yTN#<^gdGguwrX|N-?Fj_
z`weq+Wfd!_8~rr0E#ItQ@x~`-k`T6nY&ehfWtk7!%x1Xm>eXN5=b;-R@KSgZ5$>AR
z^Lq)8cZb7%?6x=#|M6}}sWmfEbZ{xSoo{?5O;*DbL)U%4Hk<X9Com90moZmN&>JAa
z^Dd$bK_g79`;)=BuA)VbSWBxjccjM~X$wDU%u!J(6ZI8pmRmPn_v)344WZK%Sz7pa
zvZG~Xz((3Tsk<n76*`v!Ju(kndl~U}6m!!;@n_E}jQuKf=F>lIF1YVQhd}#<<KECS
z&Rag;9k}^c*OnHS^Rd!=@-u3>iL-4=`xx_pN`qj1D4~+LaMreBanayFqLF2&`SHa`
zV>2FT)k@u%g}bZPON(GczBOgd7fV(HdFg$n(c~{hsF1I(c!a!7q0oFvSbSzM1x9pO
z^rJev8TJZ5+QM#owA|X528K3%38f#!FL(%cv(%`~L!%!i$4^?sjkRCB#`rCus@hx0
z^>)*O`UE_<db`kZ(qmEn-uwMK_MsK<65Buu{<@qrO=xpD^sHH9a;`^QpE2J~-{ZQQ
zdBorK-J3mJwJw~^o(|cOa0vMrh6lO3KSCUB?z@bH@m$SyC)%jHt~X;3Q`i-J_l{{d
z<;FK?ZcT^iAgk7$^n&BTNBs1eQ~22P)p-s`_qk*~U~6pH)-v1Nm`nN36xekI-fz0h
z;EpplxZ%x>o@2yI`>obXXC|>2=tJt&Sj_lrRlqg%;}WoK%WvAod1SdxbDSxZ#r*l|
z>2+wy+_{=PJ;octImr(VjVdWNkJBrQ15hOIE+}4gduMEw_8YWf^0wc8UXK;$J1e!W
z9vI5H8pJ5$b7bxLq3w)(BHH?K;65D0Ad_ozVvw1zq3B#!7vijO6S{Hcg~||btN)|U
z7y~D*1g>1#K%a_sDFfC?@$GJ($Jp-SVQc>3bX+7f<n+?*JW9txW(?B&Aopsw3h<3s
zSb#G4RqH)_0!gU}BEOK292AgL8jXx%Y`T#MhxxHgnuJuFSr{tbBYM2&w%6$4s1+X^
z7*9U&7|(j+ACWN)NQ7o{(Y6wD^cN(8f$p(m(yod%y$C|B8bP9hg^XhJ%g_WNq3uus
zKg@5e`8ywI`Y!(Dyf@z)R|sTn-`a3{cJ`lO?@d4It(aD%X$VmF7(V>qraNP_1~a>M
zTKH1^WIpwrFpLZwg*C5``A<#no$5Fm#Ln@C9n7ciq!Q(~S1>=)r(#X3+9G3F8FwUs
zH0TP3^mfOZ$M(e!zwsUa?6WHcnQx*)`yUbvH^ND`8JFes)v5D9!NRW7%&VWNQ2hKl
zUo});Z#$T8Vrz6@=5&j$dFue-S%vlT(<`NtD%50Q?%z@_7h(~<72oo(>$UR?<idSI
zY-%(Ts!fU?G4OeIR_;x@jPdqL=ypmI1dFR`WJu;mopw>E$F-0QjaC|>3;IVOZe+(L
zI9D1%&hoS&5BD!>$qIvIhEMvGO*+D%+bqy3^ZE=Q#zI!qZTng;geTmN2UJB|Mce7m
z<`Xkmx<X6_N>}O%>?Pwn<7-%@BgUF-rUZ1gZJ-zC&TD<eXCs}>EO`nyjAzSIV3syK
z4#HCZ{KBi#QbtqyhXC~obck!hMRkX_lc72u>3_(|+`2z6tUNAC9Iw{7BWcqV+kT}f
zz&FqR+2VYkHgT?~ej(`>khD@9Qu=2#G<mm?!KiJkaS7x3&J{gWS!n9jP^}e8Cy!NE
zyEhnpZ$)_Wm_6r?ukp<1AToHzy*twt4ypaZ=~>z{)n6-3<v^u;EV3qk^I0Hu_N`fO
z0S-U9@C+@*CM4S&hr}aYj;ZQEKCk1qT=KNZ7J<N6{ZC6s1WE*LA!eaZH7ZPkhgP9#
zpvXwQN-QF!c}4kRzXbj0;$(Es%~>qsk%n0Bg~*2<&F#CX*)Eny3a66?yLOi&Dt|4+
z9h+F(pNj%XYC^e?=In}AY=~AflQ~eR?SAy)_m<{AWjGoho^BBeGP7^vU6FJ~hVi!<
z3Fcf2i+G7)JlOlK@WIkYSweNse(ri2uebDV<7<^BycDAR(_Pcbd^ydR=7hht|9tt7
zchrky$MTScr^k4`RUe&y*0#+jVXNh&UTc@=t`ZID10uz!S>fQEIk9%mR7UZ$b4auB
z*_w6j*h=Zh26vNtFmB%K*R%jFK1wi!ygcF+S4m&~#uX)>$%pm<7w=m_jNcO56I61+
z(H&88q<|aefYi%!8@r%y!ztmqU$;VGyCC!OgP6=bw_uv5=bDm%5A%Lg8i4l&{w6D@
zR|_)|cKgqEShve*W|VliHR$H-M6TZcr$k~uZTKE~Wp;-|GuI#_l}>VcKi(Li@!FY|
zl(TBIQVMtbn|uWM&_n6u$~~{8ELD|xWY3Vy0?MxTcBrei8kHK|#U`bg^X8A9%z`pk
z6sI*i07qWU2>(c6p&hRU03cO!LIyyF-u|8uuJR2`m)%*+{I?#I7UlUzjdjs=gNAkX
zFwC=UfQ}?m@9Rp&Hg3v3m#a#UH+)J@O*Pn?EZi};`vt{`-{9`rvfk<|<gy)~u!%M+
z-wCicJKY|xr<Lu%6p2)0xHj(pc7=oDyco-OdRsOQ#Nr_*>8$ZA5iFYVQ3>VVRg1A*
z>{O+~JUQ$f-N&r_o#HVo8}#0$l8MpTH5W`OBX8)2olaxMrV^TXYS2Q$yLe8vP8(vL
z?y2X?2|C6byX6RQb$I)33W^*1tMNx=1BpXiTmtf|xjCi2K_b5MCko2P$=jXa`0@ZA
z^WM;x;Jz0P1%gx#4|@yR>48$Fk)79fg-0)#6CagB&Eq3%=UZ?Pw;i(<mWb92Qf$U$
zzogVcnbcf!Qs%?`Z9O%b-?iO-j*vye{v6k$(1h()!GX4u93YYjhZ9JZ_p1_(Cy*rj
zWbyEo(~CMFV8GfHa33}}D=#l?+$xSNlnODjw7Z(Kz*&w?SN~z2;(<~yI3ggYeLKn;
zc)^lr>ozK^PnfMg{?YDdQCf|6<n>L<mrAp-l3|AY>b)?sx%oVy_zmP!He=Ditb_{p
z$oa&J#wNG@c*fARq7fnwO0n&kk`!a_r@*c~Lu*uI7*n8VES!ap<`?wvVk(k}u>x>0
zrI%=b$5D4;ol*uFSjL$t`Df4xVD~G?#&6#Mv%afbC~INGLEsjkE0~2dwdxAXtR~{K
z24YTMT&f1NUY(1}TYmnoh2tDcNWeX;{sHptaUhy`U1f#<XTS0&8nNk9-m2Bl*JFP7
zmhj!LF^#|^!&{vVptP_&27z|E??u2rFe?fN#Jhk!0Y{!jzLFarjBa#$j_E89B4UM1
zQE7G1#1CH#sH(7i75@OaU64K9>6TJ)2zlW`2B45NEFtx<-R_qKF5|^3Fm@IJFFUcy
zk7hFBuv9L4EVZMWvle2GBxy~#>Z+g)Q)h^L<M#MS%BrH-boIgSmyRED4fMQ~n;(i;
z_vf-Fnp1Mu)bEv@?PKJvX&$>g^zmaE%0cFPa}wzmIPzKJf1{PkIV)K#c2A~j98%W{
zH8OqhShu?Y`yK374F36iJhUj+oq%hzlW=Y<@5|J~tA8rxA@38d7Qc4TFc-@sz}h}n
zxq{-$07Yb`bo2y>o6)I>#QOK#pW-+6)79qXnaatW_Dg3{q6(VrcPZZ<63R%QZ*n_9
zvF>p-C3O>-VM*AjXZAR}pQ^;59jrzL5Wo4oJ^9DTfm`!llWsv=wXF@<b5h6f9o5N~
z&?h3*zJD&XR_%_&7do^;TR*xr50`zAjfd{xX%l=jG|`Q4vKqS8`A#}}^scKY!^Ua(
z4-I_%=%)?DI+^GJpKIKIUV#>0`S8gah_E~Z8ucPX6PWQ$S{rA01W9>MMqG}lCaf&e
zyyjLOwWcAz|Ng^`a*%)T%-p?-?@hkWNmn<~$*vQLe%Ra8Z~#izV3hN>P9y>n)xYh4
zrtfSqe*x-Kvvq-Ttwxbo=vs9F1afh4ko~NvKC692$lLoP2akXV&*h}%beU7y%x_ZC
z4`L#tOx&z8lH;w~>C2PJ(!q&Gg$A_hMF?GQ^vY+}B32FQKQez5jn(|iKQ!+iSXXWB
z!Bkko<HF){^@3)$M9xnpNfD?9B>V!3#lonXbHvI5&i0Uv^?}(G!-0@wXSdF3wMI8T
zM(eUQ+d93lN2@Cmi-|EnF3wAXGvc>ywhUJ$UD5-ZrQ(6htlNh@f|>gIpBR}dehIZn
z@zm$LMRv~9zImLHMWlIDgthk*#^ZPkgMaI7niQ%!NIqF$8dNU6pTd_W5M@t8K&Chy
zh{yy%S7>#CHXfETH^gOrQfK$}8>`9L$jU#UT>G52K?0osSn)rS%pRSbD5zML{4Fo9
zBwq67tRH!>>)CReW}}_TvdphO7~Wf%9~)`U@Y?AY<`Y6hq@tPlji0ZCcd<YEQ%vEs
z)lhpoT|~rK_M#8`4$xGMAC8{V@Yo8XJD5VMxJW+{Gf8ksf9M0BKlQcZ4ObeCeFo|h
z4VSefT5Nv5o1^CSMrk084jh}+zJv2oIbAfg5=#TyfR$=;wK!uJKNEb&uyoB%qd}An
z5bI`rDP4|^?STA572ADH%JroJ5OAda*cqhDjieEa_w<a%+&!z%e}HdfGEBL4*vhE(
zTPVh|lUTU96xIEW)Oec5dN8A1k}x&&RdyiktNzSxZ;4vv#w0`VgFzPgU~)F-r)bJ7
zI4k6C<OWp{@fM5_x#cauEI!uI4nI7UJH#%@X{sZ1Juq20kRn?#fByc|!C?Yxvru;W
zuY9`rNS)lYybqQ|V14~FsVB&|4i3pd4MgNih>Y=Mk?(14F1RlwhNi8lFLVZYP4W-q
z?8T7N!sWb$g^$**gM+rG<BPNS9t2@dSWmzF=fwI;x>v|rU%q|qQJo6XyhO$}pKtw4
zxz<rt7&qJ>8d602PSNy9G=zLb)|A`B*T;sP)l)3h{ZIZ<+m%7*%T1Fh?t!uHpYD54
zCNW)8_Nvq4CN-+Z!(Yd%dV?JowpLa4kR;}A)k`*Rg<#$b!yJFqNn*>dgvsD6f^amU
zx9IBCOgGhQAC)Na2PB;e-QY-X8;o?bwd4w(ET0@t5n#6c4b&0bSZ7#@9ExLcCR-Sc
zE2=yIuP{{CCc^O!PsUj1_29YzXYy#>&qTOMmALfmC9SEr#L>oRzOL4y5<1G5v?eVj
zUE>v3Yf3o|r;5|lmQgVm4CP!!M%Q`%b8T6QlP8)<j=5%3NX{<HUyYx9=g%Zzruk9G
zMtZe&jM)m+H9ERQVI4?p8xHwWnKE%780)xQPS-<4nH65kQH)rnQ<KWZ%L;SYmy<Ln
ztS5n99_fB`8PZ-|9k+xvT%fgaITW0+`_Y29oagWD?PD03*vPam#2FonwI2*|h*($s
z@L@MJWMv{OKS@Sfe=tA^9|d`-Wh3>yn-`nLjQutBkn0(ZP-dp1it4fP4J*pYS?lQZ
z_8?C{|Db-?-nSU77y2_><@hffTlt6@Wt#2Hh|W$C|L@ek^5%55en6ytj{aF%huCew
zDfK0r#p~7O{_!n^t27U~y4w1AvHd41I5><|xpeguLAp0JuP1Tw%587Jp%ur_U>0Y?
zkauV;zZL#O^<vtkq`!!>*$t{I4kz6o3u;`zrsMV#Bv~3Nd2VjF0dvCto{8NiN<-P&
z0gFK!sGvyqn5p!u_R#c|&s+wtTzkLFJij%Tl7-o$DO>DyUQ)IiZEyD6?cQkJ;ovFd
zrF)<K{JmK`>^|AFz^bC^w`9@c%IJn2XX<W*d=sz3>+iJ+DS52s>)kqV{d`U_)&#~K
zluDHA_>hHkhzMFSgZtt$n<<Ya3}I(yYNEfPkqQ*_9nDwCtF+%!1o>SIMR>QSeDIi3
zbbf|XtXXrUZ_q6(3JMmk+p&{Hy|%%QfGGAS5=_57T2`Fj9Y$LSSf$N52(PZhEPsMf
zbKR;qzxP&DN9+?6%e^(_1lyQp&_})c9j76o{@U8<g>io{&d{M#anaLS_PKJB>A`mI
zY&F)E%c!FQeAZWQ+}uKZw9+PaFTiCUBO&E~f*vlJ0>mY*N+e!EwBlM#JmDRc$Q-`7
zh2EMLI@LqDNU~&BDqG;J0|VZjk30$S(R7L$#r|)+zOsFf(cm@wS&<S`q50BxMyv<r
z^n7t~mBCR>6*g)9tNK}XmQ>3~e^m4{w4<FO9<`O5%9p<^X5fBXAIH@`h$ge0s9E1L
z-u#9hS=z3yEsal<m@SROrI@J^Lnz}L+|Yf>jyZ#ED2e@jvd2;nA8xND3ritTU8lWl
zZ8`YVi|yVBcOC{N`r+wLS(#9D&S#|t?eF2-C(DwP4Tp@(%AJ-p$XbYWmI|8P^I;oL
zwx_EU>~~44D^T7D2x#H4{GOha!?cTTV=8mp=<eYRMmXHsI!O{2<(HvcoASJ??4(sY
z!4?r{qX&U9tN@K$UTEl)n!=Zqnq)V4H&kWrFQ2J9d}Gr?M+=)K=jHh!mg?%spS#)x
z-+0vkfk+zXIGGd_cl{<zzH+nVdmJO)hlHG{w}ocn`i$3&C&Vu@Bf#81cP5wE<KvCZ
zNa4vVzqN*Ecj=gaYl;84)6ned17m#<s>X<nJ6~hVQKVExA!unZPp1ESEClyG*OOQK
zVry{OohOXytt5cP7)(r+Rj>b@8$vpgKUB|A^&p2%=bAM?LxVB%{m}Ci+%zmqs8~Y?
z0~ASg_5;RO-zqo1juRx&>EMFxZ4&?23t+LNYtmQu6R&^L<tL-i0F{$c3*-;>)#0U4
z5DJP4-vvSt*nPTcs??Gfw!ZFdsK?-Vr!4<6(AbU_OeRjF6Sj3`gny$meM98A%t`9~
zEmf5G2Zg)D6JbC5O5F6nH3sgkW0j<oU#}e&OoD@@#2O=CdAMJEQPyl!&SyB*XCybB
z8wd|bsL2xceH#;{jQ21xveLb3O@@Kotku+t=NB?z{Gk{O|5DFaULG65TtP1|@cACB
z%6QruYMmPorf^@U!rbjBGt(%-58J+L*;<(h=1O*+EYObN@;PK)*Sgkfnq)n@oN&hf
zZd90XV&G@LijB>*F{GphHKe`@|NU3qOC#x0UzX@}N7gv4FcN{v4;AcF9%IHrUie@#
zuifO;8X^CfLwacP&EawesP?Amd0J9;y`EqdsNH-w`1suG{eS*&KmTVQR#)si5DtZk
z%~f!WWRi^kmWDHps<BX1=lmq9-W$f9=^l(pCK~FR>hv9l6ptFM@$EJc1oI0xJ;Su5
z<%srtFx8p&FMo&k?s&rQ;PqsCaVJMSF5c_gHitMIJR9N<!RJ)>x?7p_0ex$@E6|f;
zC#wTuVT)-oPf206hnRw!Ma@E}x1g{U4K1OK07BE|TxI8Yc0$8wE)St{jNbZM`Sqes
zFau?DfDA0}K#qwEZ&myQ(>bBREo=@<uaPOK!pBRhWh76J`;L?x6*7!7w_eJRdQ&NG
z<Aa8<zuIt)SxPphQbAW|$C!0RT0{4n>`~*h>5<4{r&`tiLql3Ev%`zh7-!`HPm9Pc
z?y7>UYm?>)_k&E7PwpC0Jr#SGn|R<R-Pfrz9n1#ptGid<-g8=`aJ_khb-t;wVFky#
zJq=I6q`;cY-cfn!7|Z~cmq!TUv#(aoge*n8<9cz$*gqP0!l~XiC|ePwJOw#o%8E5k
zczxY<GV_PD8};@Gq6w5#nTT7`R80+GNcKN?1N`kM+fHucH4L}>ceK`(3j@T4@0HWe
z>=51;GgYTGTy>dbos<~dVmSCF-lueTA)it`267LdcUsH^C;KBe^2}ORSG0Vl{E+)A
z47_Xm1~%Mq`*^e+QImwacB<M?&&0S-ja~lcdFRZmsHoog(wlw`_k{W2xueq3*1yAH
zr^^_$*;%h6V&K0kk}O3p30%3aOR$_bJDw=RWnHs%4Yp_aej5lnJ#_@jb)L^TV8P))
zU+tkuNw<iC%#w-09~N?+^su0(HRs{Y*Ge+!#Fda-oo-}oD3C2o%bpl7PZIkk7Sa-4
z^YjH}L<%TJNlBcZj6PobCafiAr#SQYTzS3h9~g0*goHN5$XsFhR0j@dy4xXc8lqy*
z)hq(`g!sSprFiBs>!M*{iYndymt+wy#`3Gwr)M5ht*dE)3ZDB9lK7yS3R~_Z{QRKG
z{NH39J73^AHQBzyQ9N^Ysj*?E_2fI>H_Y-)Xr38L63Uc6(n!Qoft766a~dEaNuq!2
zpvWF7-8)rPV>Z|LU@-8f?PHWUXMNpfLn%?tLb}^X>2Cg{?Y>yaSyY|DNub%C33aj@
z0Ung}ej*Q1g4VlarJutfxTp!m|NjQubFah8e{dw6h`>1B8`SfoF(X7{ut|EAqpG?U
z3wcaX12=v_K}Bc%6fl4!(mkTZ59(D`l-yoYY~~PulbxNmK!Il`&G2{h!^BrmGjnj7
zkM@-1E!hF>&5w+J>T{tkW&hEM(R65eR)M=kRi}3bIYp5M;@6*{kmL=!>NH-h29MLX
z5Z%1Nmy3sM!!DhwoF(Wsb@pEmugzxrduvacZ*aD^j`;CBur3seZI$vA+2ax{&!_D`
zC1zk^ZKAmMqnUx<+=V??heAz*gE<)No~WA(ACG3zdqv7K!yZtrt(E~ScFfZ+=fd+p
z1699DUCy;`pUB<A1jchDo*7Y*gfDq|wlzN4hz7~aBj!!P50>~k+M|H`ra^?im2!F8
zA=LNM@wjKp2EPfv*DsPnnWO1`(jN&YJKu$r8U?kdcSwXh7McTXJy}?_+-w}GF7my0
zE?wn+?kB)6<OTU}04-@RH$N0>eQ|t_!8k?*Nt~V@%O^q}Zigkr&2mRDGMY#pdexC1
z91arT%-51mAtD0!ADNWwqmh;oUPuFW%(twp&Yx(Yl-it1M&_xHR99)ghIZ9PM+w@y
z!v_-(Ae(M?4WoS#-H~O&d|xc8IZ>Ufc~_KN9gHdO?=7zy-?`$CvmKni-y}i{3K1_K
z;`g%&^D;W}kjYuseiYQh$|?h8S-5%F#XJHi5~+o#IQ~6khWaZT)0-(P#By><o#AlM
z#Op`U3E;1HreTD<inn}bYsV~#hYMHIz(8Jqt(Xu|bworSexStAXG8#gVY&0=<k2e`
z1`sJrfi|9n=oK=ha({?*Or4^zTPh+#?pu&XTK_WD#D|Ge(G5ADU08(06*Xl^;cgQ8
z2gClY(DyuoU?6cnYjvzcl{<yc_WQDaqnqIsDs{nHy?D)L#$3=?N?F;Uoh8Jme`6(!
ziZ)F26`M`6DZ;<%>xoL|u0OTVC#kJ-VJ;O?Ag{`q>r{-s$SE7D*I(N}_-a;A8|4Xd
zlpD8k^)jsPSoPz_9P0u`okEGmjT@yD89Tf#rIc)ao-Rfv=Fi%ch38YW)Suv6>QaWD
z+&)wW$_zP@U^LVU3BB9d8RwePcA7gju1)*>zYyE>e_DK~BA2VEAtSn|vWbbo#0Z#3
z>!0Zt7TDNGPpjIdvfT4W3E$oXL|4Ukd{7F>#Kb&l>v%<*Z4f4>5#*^D5Lk=z!>z$F
z%kCx0f%;K`R>pqY$n^Ce8I5OP{q7lhUENGtfxCXB{?zfOH^5+jzk$`6ePDJ+nYLxJ
z<&|n}QvY%UWlwL07e87@=bOYt8DTNxs?N?Db+zG{HdO6Im*P;#wD;K+D2>L-uV&O$
zpClZ}X~hS%l^)G@_YxQ8JO^l4qkmv?c^zPKii`-Cmllfi>Ak!Mux{x_HBNyxb~~jh
zzewMW`H6AF@BI<ip^;h)A<@1{?Z4ccA{ooODL-}e^eb*ySxfYVQALmQ&Hyabo^2R5
z0=E$*r-AZWfvuXel8TCM!*NoH|MzcE$)V3A;fvc}Ww^Apoh#0}x?6tD+qk^KFQE39
zr2Q?c98#OMZ&Tp9&df9OU-0m}bw!NK(WlD`V4b_tmB$b_CnHnFo5(i9<8?QnBuHd)
z`B|PH@6h2ZtFwjUa6c2@uBSC=^K)~yf%1sh2xr$nVD0Qj*XGLBV4e;_DHDTmuDlfs
z=^GFX_33Q_i5_MERvFwMzA-^3NYhBmvA(^G&cCg4K|fcHalJT3A-~vH&3E4`qy{LL
zoSFxL2+!{Kt)MH$OKC(Ka)#Pz`$GKPs~q)b!k1U?;foA%_zu{L4boZZHtG-LY#wpG
ziX&wlV6dMclq^*?tZXX-zpDDb&EP`ac=TQ%vXzxFFw(blz^cSuFJSitT&_`6TW1gL
zJO#R=byZDDZ2LB)l|q!^uVNve%~e$!l^d0LtAq2M)*F(Okplz31O0&4BIFFU$<)P{
zF{B9YYD`M5#UN;lu0ZKzJsWpQ;#cJe=Wr4w!)VZC4z+9UZ*Xn8G(_mT*w45BQ10W8
zJnDzC+$i>@q`YF-$sOG*HjpX2=aYum(62z?55-eVKaS6L)&4hNy}MFH4_}08s7cel
z;k^nPU7CP9bSQ!T48@i*Ei7ebBq^`)4y&-W7wWVmBMTxW)@k5&{TQg7P&DeRgU-KD
z1uyym<6N+VMB$^O)qxI7yWal=Jsb1al4=UgS==R4b6>DtY3Wi0N{;{wS(laMCgi*O
zb4+}9R8sany~f-8o~D#}+$~d2px$x^>u&xk?6<t#GF3~D=b!nM%V{;J&NoqKIfhuO
zeK}_>Uckr_4ysL8+8wkfYW|Fep?#sAYl<YOPJg$&($Rk>qJnLO@Vzqz3()wi=yU>m
za`M&aXe%D3L>{?Pje%~xWf;#Ro!nmfF=-{Cc~b?o8PLZ<LaKcBbz|>a)9z?;DxJ^K
zQt4=Sw0J>q(7W9gu`c03_QXUvfH@LlhyKJl4w93f%#SGC5;{i*r~8Gg$bR;e>|8|i
z#D&=K&NLIzd0ddD8R4|At*xZ3h<7LbWt1i-BZlX7H7@IXCqqH;TR~w852tT6rjm-z
ziVt(!<jLV|3p_3qJcHrR7L#c+re^WegOC6#Bt%4G?)%m-Pg6dd81RCs!|CONcvyr#
z{ROgQ7xx!s^Am>ahmsAaV;908ib3HigaWSITT_#$ke4=}8zXL2TuJ<(Z&Vtbs66($
zTBgzGbI-X6eLOw)I2=ED10886dU4KO{f-bd&+-3V3kxgZ1aDGQ7;=qH<L?NOapEtz
zCXeh=_fKIesTCg<&USjX^Czx0+i~2dY7-Ov{daF2|3Hs=Dv~o1@4)du4;>N}YE3aU
z4X&q|@&*p3LxA%e6+~W}XMZ*~SUdbvs_+=QQ<^siQ3P?5`CV<iKE`{z4xS7zd4<yy
z|ISZ^7vI<xvwBu!q<ZnSk;FM1Q)64qb(-`()HEKWOID7NySv=JbDu5^44$yCp<zm-
z+KLC$bL2EsB>8RRDUaAV^FnrGT^Z1&970o8ze;jzDi>3fFGt=^t#ldVcMZ&^6y4LX
z6SxPz#;dX+qlG8s;o;qd`Kx-u<MNq2{IYqqgrnNva7qToZnM-LfiPy<+sy7lR7NIL
z_}rdM-{469P{^HdGJD!)!r<siB8n<nB{5z{NlMk*2;sWy>lr7IQc9;ylp^m&O1-o6
zzkd5Q6lbo<+!?ev8_wo_A6>aLarH~$l{u#WP@O0ZwQfiBbCk0)obk-^&N$D#86Dkk
z-Dd@+_Ol_n8L2YKU2=Jfu1@D9aFNEP0k_q2E>34%sh~V@Ph@>Ksk0xeS(@AVVkSmQ
zt+&);Q&8s|43w7RzMbOZ+eJ^il$9U!S?Z8STF2Mg8b4RcgO-&5DKmK!8#}b{nvH>l
zQCB591pRBw|MKz=+3LN0#58!!s3>~ldlC#rMx2?k*f$E~ijlKQqp{*nMN$%j(utxk
zUDQLoy^ng)`O@Tb36~4<t2g66ediR@H8X9W_fJ<DZj6LUl_X40l{oEM0=5(M;%9$P
zN6&P^29lAM*i}hKMiztlNH)O4S(j`Z-J>fNMQF@pqboqh;^46(X<~(CtgZe%3$CsV
zS=wLxT>KsOQIWDJ!|-u|EE>IupDi=0M3z9NpX`bp9VKV%o{0p&?1kRiC98Fs%;&Sj
zez)6xX1I{{4~t)oThp>T4|QVS7qUp)5$sNKxRU`o01DG!Wj{487YAc5<gBNy@Z|;x
zys}npXKBs0M2O;syfNp3_H&^pw+uG}#w)x@+^=okyeZN*II%nI;0%q_RO?mulZHE-
z?tVR9WN%kdWdO?wkNWhsf7auv5WxaRrg(?Rvrst=+Ksg(0jtXD(RUQl;!cLG>~ID|
zyt$=X)>2K);3N`fq^!<P*V|p@!cw4M9!`hyLElH0IbVJTD@+7xT`f;)W8;oD8;dFM
z>=&T?V7RTAnF|JOTwmdGanieWzki{>*3W4fRd=rh7Ggu$!}B%q4;CQ=cazb-{)=V5
zyvn1%XlNxZw>;w{(OubGTk7mxLSSIYne(znr;@M_A@Yxm|CEo0uM0W6dd59E{R#3h
z49;2p6P_??W_>-dsf&HT;6!Ashq@?!cDG_ca&|_y&Z}hIO}~ojTSGZITD|z=#*nwx
z0JOC!4)>?5ox&Qy{%iFE>Xm?tOd)K$)I)Ed7m8KYtK#AwU*!CGsywDznX993WK@eH
zOYyrph(I_@VsAT0`~wD-?Lm7`RiP|=Euzn+NiCMt?CfWfFS+szMmL0jQcf7-f5!tV
z`a;>ucz$>cWKXHvywq4laT7W^vKFt%KLtQX*|D<j>y4|i8%o9<Nv}87Qc~Y)6k9P`
zUhf6$eV*MB#fgO_@E}_@7pP-m_VXc~eN+A%GBSU0A4WzaEp19Dh?~Ecs@sQa=G77w
zk?#~Q=nft+iGI&4ZVGCOIYt3OG55`xZWJXZANuZq_qvoZ&){l|)RoR>y0_vVhTQK6
zKIoUJRl9|)%eW<UhM<`?Vs<<ekx47iC($ZWJf8N1;)h86a)LS-$P72U(E4|h9JwEn
zrzikghbJMLak!i1)|+rB#&Q%V=0#ZR)C9wL-#Mt1$2r4~-t$}xIOKe1^Lny5Qe)pU
zS5LD$5+8GJAJwhbuU=WTM4=+`j8=2qfS)8r*cS-3L`DRB4;N)sx>nj}l;BK+iL?T8
zVigs&-UX3R_!3h~zTLbjvO-Ss>+1qs)>m%1QGwMYwb5~kX1hTL?oStg(0-|u-k;yw
z;hj4wt0HNxiL+oc&t)~gu9lF^ReZO-6|Sx(d|;P&NwaEPiC63C9KECcnr*YmfV9U)
zXj38+2z*&?S5lQ`t-pVB*zDnA3Ht)PkafV(4gx4lT{-KFdyqVH7&;&f>f}fcNjz`0
zui;>Ce-MaBU8wnB*|4Rs`Sce>>2cJj2Uws4#|>~@ZOOk=HpntnXGd3mxkn|CDgOyO
z#yvDQ{PjkJ!d8_t0>zuW=4`ktMw|m8o(?9ii*&`@I0n~`qt_SX(*|4T$r*YZ^m+d1
z3b%<S!Q-n_FdCfo9U)x<a=wFErHylRug2&P>e&4qt?k|)u=(FL+p!&8A*#*H4i4&1
zcJ39_V0eORr%pNj^54YyKnUb8ZVV419$!gXL*W@%rB<O+vyS_Gg@gU7ng%zv#k}z%
z4ou4a3vu<ENFco$+KbGwYW<BAjhm6uys%!o|AETEll+K;;S3O~I$-T8*}UXc-}n`{
z@Bn?Da=Z8Hj3cVv|6C*sXn<yRJNtJaeS+r)>G#rcZUGm%{P(X|cX!M5<j;xeXPe9+
zX(@y@bx+|B`-<kqJNJ0Ac_$ELf;0UKQT1OS787v`w-A|Jt)6lWl99hj)&9)y4iARz
zit>`=3)TI4?cB3ZVLg_W^c$_^+ri_~B7%yE_}Wg3ui$LzY`@ObEP;l`@yQx}!b4R+
zHCpoGXG#2Uj@k??tk(%Ke~yPOea$Bymqbl;a@%JMbXxA4x4|?t?6|nX8g=K<ol<kN
z>?duy3AOcBpx4FfQLq-Dd^6>^!U}p?8JL)lMSXyc)XDw=UX4b*4Mgp&2A<XRO-IFP
z!5oqq8ccasqZY*w5AV0l$>zfa7Serr^nySUMnudPah=!TIq-x04S^=bd}D?Qgp~K;
zqH!l<WHC+GWw3iCN=&MA0=Td47t)mpk>t^4kUhQ(8zR)g;83>g<_9bbY%njKi_NB}
zLwmg4F^S<CL%rmMIdMPxsbSj99b*JH&o7}gUCS6oK+rGebM;Hj-^Z&!{&yLK`cCRe
zIg(Mn`MQ#z^l;@QMb<KD;I1JqWeD_YPG*zo&*k~`sZ|Gph$x50Ey8`ix-2-Wd;Yne
z@J1XQt5{e%h0x=AVvRZi<d#e4S<+m=)|xDLSLb!^=H1x=OH@3zLcXFRc&@bC{gs><
zHl4L-6QFbC9bHDy%)q_dUdgOnH<$9gx%&E^hhIaDld%?G;LhbCNUM%O%t{p}MLEXS
z)@r%;rGkp%Hoo>VZ8A!VP#NjejmXi2zJSM|e0F)__Mip#xMo59n;A^fm#i*QT6RcP
zxL_)ZCjyB`!+GBbEBS?i*(Iub6MpAVEX#zST_wGtR&|Kvkwr*~U~3LJ@2w?6)ZCv9
zC;fK%!i|HC>7R>4u_bzatW8rj{$~>wrT2B~L@ND3!A=&tALnt|kR%ia@H&*bKv!$B
zS((9PYv{K2;C<K18xBHkHy)^kq2ayjg;ZUosVa1#Ix@mvQu@!rpOAI?D!YV<IW>;S
z(Xl~alcOWS?2P5Yw!Fh<X+9!Q+S^l}1@qZfFffk-uZPg=@H-PD<KA6lsFT~NXn&C1
z4YA1ol5Yi&EF8&-x;mga*&^x`e$Hp$L}Z~kZC8E)NeM%2i`+LF$r?Mciwv-d^yG5V
zY>bHKO;$g?c<$lpI@@I_Qhti=DPds!?Vl{wc)hJ?6-p8r!9pn0-`zkVy<vr$J7b9b
z?VgN>niBW&!B7~YEbLohTS5w&p!p1?6~M{6;kcHio~)^vLv75^tS-X44GT}D^*uJ}
z#*BFRBK*@>zBOuwd{V!p6HpcA5B_@AQ{C8G+bw}Pq9_(t6g!OR)q8w<b0bd^p*<b8
zmp1N1v!PdWoTU_$F+jj2!^V>P7`dcp5KtV<BH-Gvxj+BBt={;#1GtA@VA`x`&YZ91
zI(et-=_#2(FySB2f@KFB0jQshj~i-N{3`u{@QCfde!X$N6HTLwK0YacT&;I(-Im#~
zJ>YXmgUspkA@kklj*h;u@CMd^Beee;Cv(0V>RfL#_L<xI=eyat{@x{F{#Iz@BBGsG
zu+T$dulh=7$w8|Snid){=rVy8!!8ZYk*&S%=No&J=cW|mjQ{o;Zu=f<<8X9zY@~w9
z&-zTE<ej?tp()8mN_rNGY~rk}^08eFxa2;m*@4XEBy$RKp0cMFl4@YfIRk^S12oU_
z+H2S2fB<Gfy4)P>9X;i0%#;fzmf<cSRlCNoqZTtBA5*R?Jh-fy8X)OeGV!K~d$F<7
zXs)n>Q_NdHI)*iIdpqNc=7@SN(9p15q{P+Ns>LyZzQ`8*eI;&JrfyO0l$}c`wUiXj
zLOR!qi*2z&eH|kU>+;?a)Y0WO4r#0S_`oO21Olnks`u-do(&RfbYaYO7!u*Z{BI-q
z=@R^FWUZw%rr`+@8OlsW+G5r$S!HQK1FTOzLYLDjqwRx80;qVU(UfoD-r?bHGh_S6
zi#co2`sZ#<&Zw*yTWGIE-b~XMU9L7u77ZH_I^=Ux*bqR*F=L`4B9enO(tsR+Ma85*
zW+$1U4k2rz+>i(RnoqR<O4!MRSZVLL>?!7KsL$2n;=EJr&%WL$suIWC=A~vP%Nz&_
z!Zl4Q3$T(5%Hv(=MDo389}`*A2;PyaTh;*(KyM8@EGF_tZYcw86)R`sO(*7%SCvUD
zU1YettheBAX_&e))8K<P*W5kQ{}Ls?+doq(_5mZw%uIB!AHZ?<Y(isYu}{T>?5!g;
zf<S%RHzNJgC~qzwm<1XgIClXVUp2IN;exkiMHF)7OV7q|?m3*35MpT+6xEzJ;!+gF
zK}#rTYd+fW*?HGkj{TXf=H4FM&%!tvH^meMiO^{?QN&lVDT5YV7bQ%W8qwC6`)KDV
zdy}5nNa$oO6WPjlQ>tnt;i8n{$RGEwlA4Pshk#<qzhVV|e?VXF&%VCqa$I?ABOjse
zw*g%Kvp2Pq(-^pzbDu9nHIZ5uM$3B{_}I>btxy@HO*`1<2bK8l97?MAwPivt8=gc_
zDm5y%K~iMEVaUlhj$wd2m4e>Xc~Y}%=2H>bQF1^J{c+lmAU8pC#mC%Gn}bFXwsLC~
zLaf1}?svM@C;J@hW@a7q+H_gZDq4qN*MlAtYZhVu3YYJhGx8)z5SNHI|Ncg#T(s2S
zhO@xumR`y1n{s9fnElIW8WfCdZS}i2o68KJfPQR(w6^4D#|GWX<()-A{v`e8=ZZSF
z*ccjPmVKYZ>P`9+J97794t1H-FT`}&!RZKYJagc3(A@dH0x3GamTCUGJGE87Xa8II
zP8lY`^X?6#i^{4vs+9zAwBHob7?bVQotI_D){oqCeW<A$g2!sb!|Mrz8CyFO!J$rb
z=F&=HV$w)$hdpA;^&QE(WxTRumWiCkQ~<pILZNuWJTX{Uh=4hfQQy374G-Tm4+rXc
z@do?$`S|!30$8rVkDxd4krKcYB8-IO$ibz=!CPlyNY-+6nZcDGYnP^&yU#5Xd82sO
zphX!zoa@8{XhBU}MAixW;lo9aMBn<^c6OQ&c{csx9S4e|2bu{w5pi?QPXMG(k1VsH
zq=vvRQ)_vGWdkYdJc!>=Ez`%NjLz$HM@YO*E6(m(s(JdB5)+XTr+Al-e}-@3Xr)6I
zOw(ai)$jUxs<}DMqpB(rPPR}YT3?LAE;TJK%P64Q0d^*`i~)u}+b#0i4C8<8)aP{=
zRE|C(x`MGGrrcH$3j%FdxBZCxql|Fx=Gy&KznE<ecjGkHE0bZx_Hn#=(}is~UP`yi
z+r8DUSG5ozOcG9C0hJAg76GXdyZOcwjQlhQ*Rj^d=W7?|uf~TD#h$=*!~V&Kh}OI8
z&i<{8$314O`P_Qo){v1VF-Mj<_UYMaO;p0*+L$f4aNAb%z5xUNvrTBHC`(jEjk>cr
zEfm|mcgkeq(Oxqr?I~$8j<1ck<FX9e<CB&vtlEg$s|V>_&zszn_lSnFg;qM%Zit4P
z2zJnj^qn@9uAS?4Y<WQ|0_<Ok+>!_tF$r>Vx|BuJXz0(&pK#1pc;9(#r%hAvB*w=V
zfj-R;J(yt#4Y$Qa!edE~9|P#AMp-KXeSEEB7dd7u`DuI9G*)Gm#4KEX9z}Dj9)%_s
zD{Xf`f{ajB4D!kSG%(Ae{qS0o^Rf4(pSEQCEn@~ryT`|87gvnw?!R6D|A3D6;jkt%
z6K$f)FI>2|o!+5*6xk)OK$C-a7o(k?&!XT!O{w@v>t9AdZ7B1Lw4mn}eH;#V<Q{|W
z3Q*NRan|+q({Fo2;bHXd=(W87_L4A`moPG}7KZFO<q0Xw(i~`_;-_l2OdsM=b9lsq
zCwWSddE5-6>7k1B5AOY+{wg9~u!u?M&oEcFytUfC6z2%$y@ZujcfvdTmdE%x00#SF
z`(Dln;5!Oeb|y2D)N{q+m^&tFuYw2+pk(_eHa5bJ>{EEU9-6cyYc&S38|$r%x1S1M
zU(H6&A`VB2Zf?VVo~8Sh03Kct6!a(wzEmz8pKey9)K{DsgadNd+W@r!k_K)0FT{aE
znn+)dBt|PcW1@QAF*pz#7Heh$%=74@0j7_9V58CGvsL~IH=&i(X~}1yD?j?~^j-vX
zpG*i=60x;E%TVg?6n>PZ*scn27|FGH+uLh2KS`hs!476<<Faf2Gab<G#qVD8?)NA=
zUh%>*zc^*5FH!@WQA6iD0XMhnN{c{XXZiun&b=Y<%1PpOuY2qJ^b$0X$2$kyoEhPW
zQh7zR;@71XDn5nxU2R_Cz!`+RQ#PBmL3s8R&_&?x@K(GPm511brh+Q^D8GQ-BP=)6
z7YhTY<7e&5o=?=P!1}|V>XGp2e&7HMY@Myd_eg|hl<l@Hqc>LN^oA?`k=ze^4lG5h
z$vo!Na0i4uVxSK5;MKooaKk;Ydr=TEi}=md=#QVK*H1?)ToCNF3E@3AUD8hu2QN!e
zZI=J#175!n!ok!clOiBYE4n6F=zQxQ$?h5%Qn(3_m-UFbx9_sQ<*<eOC4C4Amy$_h
z&8cmc`xOf?EMT@F9SY0NsEtz*EUx`o%I?$!*BW{h8(!3&sI9~{-ywmtmn12wyMM%F
zC;lEQI+n?O$2_375BcyYOXf3P;gQ8dRZw8H`Dj`=lOK8LcwAUrYGxYojtHubbzc^k
zE~c$fx(f0?_yvWIbru$IDbr1{?Ov^%Nw04)t$r-<7m-AZs@%<k4gg;>>l@^fLp^#R
ziB@#Z({Ymq#$FJ>FShkWR+vXr?+#3ai<0dI2dteiVQ(`_gJzVKx;?SN$8WovqAtzb
z65smYu|9>ka7Da$5nl%-vd%1ntpGow*1SU;5OY-G<4b{dINqt2n8@JJs{sB5*kq)f
z<*bqiw=Tl+3j%9A8#6x~NF~T-SD}mlV-a(e^q1_s-A}%(lethE|8Bms+LD|5#u7|j
z+fn3gjafGA@_uh-`J0(gEq~8@3mJQ<cm9n=1NoNVTdGMO9kRPdT!kXpWm6M@1KfGY
zzYy$;W7t1DAV8vT(ndP@>y6DfkA#UMo2wq0#7Apj&Uf^f;$d7ak2SRy65>vzIW%t)
z*>XqO@6qgUc{7OM4({)tSqq)9C=CP$PceI-VfcH45_>Z3nciK|4iWFB`*on5Seuj(
z;**Ok(sadd;wos0-*>*-);akTHQiTH)u$jB7$#yp-Hg`04#od8G3BrmHn}~n*6OO3
zBmEl?uvKb+5Gy}wf-itbUHhu_r3)H7y~Vb|cnnhxbDcVN7V4%Sf2Qu~dh)H39JH3)
zPj_YOX2Jk0immy)4@A5yawfN(ZzzC#3~Uk9oS6OfBGNF*E8OHnkpnxt+LY*EK@wt8
zy3mRwjY2bN>U&%ek7(`f46Z9^WfJ0*od|Ec^(5!3VnNd7*QQ*Xtk#C9<gHZ7wc$*p
z;~`%m+G^K_=i|h)gaEnaa<?DRfPmY#DTNSk!|kUgjXRm+yQZ14Xzw6N-y5eB_l*RJ
z;@N?z*`Q#kFELnZ6$S(xKI0D`t!{gj{@R=zmXzt8{q%e{W|@3okdxh&@cIXsUkZ@&
zL4RtSS>}I}_D%F3>BR<8n599R9qWhJcH3aLy^?NDP|C`>Y%kBKGctk;1vWBOht-c&
zfaMGW@61Y&@v;D$1H_!1ymZC11UC8Sizl_)nSO>ma-Tq((McC4%vYO2TltLX&inQW
zRtiei0Feo@R9{Q$SX4P5b?4tP#hrJ7Y!;8k59#dBprgL+%G-JE><W`;+`bC*VEq;x
zEuNY5V1!I4Ee-hB2y_4mEO_sq{RNa%QGIV;fNJ&EZ2$?lUtWNdv-?PWkL84Fy%KJZ
z&hY>gM2_eGV^fdLsCUO2(nX;x)&h)}v{@NPr{|3*GbPClWsK7|SJsF5LzU;nV5|+=
z49HqS5nP~VdhO`g0!9}b+Q1wb$gk=Y&B{>g<m$M~ymznBRoQO}tScw3i7{_uqmO{h
zsMd92X}#J4N!8W++A>n}RE<`BVEH7LrKoIx_S4!#ALLNG$``bX=!Zw4D-9eK$HxU*
zWyIm4mp12#LHpA_pv%KqtTqIUYc@<&*}XiJ;eAu?Tmggmu<}hs`Yd7LTc!=Sy6V3T
z@bhitk$tLVPE^<OpR`Z6g$H(a^Nm5+Dn6{QVqun=4mBCCG)<Qve@C3-)G<<VYn<O!
z!NXx_cA<c4(rA|Ri+}NNPUUZVl~{lr+Fmtk2=B=7D_U9%UzS3JoYkV$f*5jTe!{BU
zbfO#`r}<|CEObfpk+)PdiL^$f@LDK`8;blGT!TqaNw&QY#0DsM{8Scn!UNiZ07TO5
za(Mq~BCW8m=<1ja0%oGi+b5u5HET3FKNQ-w;2PqD$|8#?{-NG$=#c=M?!<5=qkrZr
z?gXY#3=AA0eqr)`wDAf0DrT0o^Im<)+tZg-$-I35RIeZ9!um`^URGrlT-*pPjn?r~
z$$;K*F?nFubWdQ3G`hQi8EL6wxwgkXnCksWu*3J4SQOQG!rIx%TwUGqr`<cLOddQ+
zm~uGTc%*nduPwcVwp(py6Wwf$n6WgP<{a*k#!k$srNi2qn+SV+SUV6rp&8jCCL%6C
zGXPc6zmMvDnQn9uXc2<W$8vfR!Oz|vsM9z0nIWWdS&6Hs-_2>%F>0^TulwCof!+0S
zz1V~#XJ!5%Fvr3YB@5G#Z66Pki~=Vb%;N8I9Cl<gTiFzz*)f3z(%mfjrNN2;@68U8
zd!wt5Hl6mL(cVYHSwAW&uqu;tlSVFEbHME-UY70-s8*bx=bf>mPy16rfUUwHteOH0
z3+Kg&?u_zd^NPUG%gLd&hO+A<FpvWm*EhKI^-99c7WZKLx|HM2Ffy8-x4fd-_2d}+
zzvlW^Db)XjzdvSb_{B+H?S2s}LId|BVl&LU60AEO-2$!jGrTVKxqC%N!NTSGif_`_
zoy}MFD_HLHwTYLfr<7}`D5t02*4BEe=^8~Zlq)4*^OB-uRL$RM&4YoelUMtNg)C?2
zu@)tvg?^PkNKsn>f-~NRyQrf03#ORc?%EwTv2MEe#LbH-yaa`X@x$d@3k}Q!cbu-r
z?$sCM!G`NOOAXg6L%X~VDM?=ZgR|8=iJ(i*GgGl#&1SHap#$ZRl2)Fr?{(QSfev`~
zK-u5D=>lEWDI$U{pQs9#X+V~AfESQyruxr#tH#{|IN#UQ=!>HyHE_78#2aY~?Ezow
z8J~yYO?-L3X02}`vOlmlzD^0ZqY$PHUjzyqI=M`Njr{FC?|d?S$OGzT*ASO|>7FG0
z5<_YG`r3m3V=quI_H)YqOyY=U{rg-Ynr*+W!5i=aaq~;FWc(5)hNaoy#O~E&O1pTv
z*#@CnjW6JcBhF}pnp7Y(1IOT`6<5L|+9<cncOFpxS6O>yOALBX0u&HRxsTD4i3yvh
zMyGQPVEg@B>j377Y02K+f}iCD6YMZy;mLP+;Yrax;+c7Ny1V~TPTm1aagkt`H4iT#
z6BK&e$o*}3TMg<qe)B;;FVag(`HU;I{zi0gu*eRN2pb#C?Ft9Z%4kc&-=CP20-KMG
zZTix2Tt(}5Rex~sK|%K*?ME_Z%7D!%MX-vER`z@X&E||*5i<00hel_vA0WQsvJ?-`
z9Gzr~6eHAB1dZihlV2m%#3kw)%G-_7Aa8Hb@-}~84G11KhU&!*2JQaE$&?op)Zkt!
zwfgw>g_^VTaPN%7hWF(ieqo_1z1fSGSyIH=C1FtorY`~l0bfZAPfmtOZ4y>y6!dPW
zUyRrPlD2nIRgZ|4=U0<9kH*|~m?z!24z_V1>QPl?5&U1qYyEL_c`u=TIYddTWWlOG
z4d<WLi`uzrJ&+iW85^V2jx**${l*Q!S{xX5-o6{N+VAxBwfzS3MjP0!>Y2F<=%XCU
zCN4_8PEgcJuRFH1Zg1-;oj=L(FZAJXdon#t@e8PywKjWswGK$~pY~(0Y9@5cOM7>;
z<&`f?K-<!2ZA~@*y6#LLY910?dxb!hoyNR+VkU2u8rYX-ZVurH5ff+-iC)>fB>evb
z9@T^dAgPE_Q3@k+PAnqskLj|o71sU5Xqkc`eFt*EEVZqS?I$J5zkz|J)fELc*44xu
zkSJ2o@e(<RCh%6}-W1dbKiEk-6luye-sHLN<C~}w^5+m3CiY%b5~mz2yIhYkOB?Il
zrTX+b9D!p;^3r7~k(=y)G-A_QQZdAz&T8~aNdh-sx(JmAlEd?M*za{Bng3C*x;H>{
z3#Vsd%__<8E~!9yl7w%&Pr?2P>|8A}KqZz~TVGCf$-NP&0Mw7aPDy>aTfi8SP*cA(
zD<W`cXD37wZpq7mnJMm5LCB)`Bl7>0_tsHWuHC!vLPeA`KoA7B3Q8kLC{lue3y|&(
z>Fy3eKuIa-l5S~fBt$^Eq+4Lo(tYM~?{~kyKfmvcamM-M>>&fjUhAp*nfIL6yzY6=
zO9sPG#jmsGtP>VDE}|=(7sn1(#wDR#*-`k`$BYcpa6hm0ScMu*IAhF!!2aMm<isz2
z&CkBp%`Y<jejXC8u%M)#Af|ARXFR)Sc+SC1CmQLG*DCSIV^HOjj11eilOhK`-=iad
z)Nf%<H}tBUJi7NEqO|7&X~t6OUjx#5*<OsC_wCjNF-;2$`N}jYgIEN)E`3$Es>+F9
zgl#rJeblWoFfI+$+%^(wiGS4>FP<F~IC}$D;IwtkC4j*tv-Axq#CWsT6D~SJep~Fl
zT7eB;^?lt?T0&fG<B+;Ujf-7UB;$9lc>+41QKYk5HyF#OfkjclXZAfx%+)c~WEx!(
zP<MDOCOmih2_*29A7ftKR8)keoaOQEdim$btKfj!Bbe(R&{6wz6)?FDW<C>1Bo#Z@
z$-cz*_`ThItjB}(PmgXKSdHP2);!tn(s%x!2mt-o(91dqNDqaDt=>zx-FX*`-hu>U
z!~M+A^IJ3%W8B=_Y$Fto;kQ0mVv|sjW~uj8`^v8w0^}=dKt#}4pPvtLD-l~bU6(Q`
zsgl1^rmf%2g^vX(s)~c->#+V(0eQ7vt?rX7WGrFbUGZ32P@)WrbQqQ;6`8Qx1JW7g
z>$g&5DS@uijIX^S=Me&h{QY#3sUeaH{bt1tpZv>ZlNt<CKBq-YTGY3~wiKw;Ei#y;
z3FhSqhI$jI-hT^(Jca~*pc$&8`0nYKv)RY6iXtJbL2){H?0HJe%!88s*OeZ0L;3Ga
z=+a0P(V0dfAtZcm-FFFds?zED1N>tu&#xdJi{Q6@j;;D~<m`;DDL|kztRGwt!hv-H
zc=OV#GCs-|FTb3yxoJoIyzZcqV-`x@_%yL|ZH}Qh>x9%A6TB>MH1`(9kIk*JB@bE1
z>s(z?cc28%&YW+*S=n4~@!IrUXG6mtftGJ$W;f5p29bwdmVo`b)(j|lpTjDm_p?iV
zo%MQUR<Qld$2*b;y?{O!G-gM!xM9LMV*U6AEZo5wgY@Fwxjv-12K5@^hH`8`aXjI%
zyWqw>URG|!`dANkIk=8hU@cf<hVz|<L6s4K(YJqJ|G61D;<JtscQr|ACP5`NZ0Cq(
z_4%Z*CW9S7QfY+s9@FQ`vnQJiINioUlvAl>y44QU;JQQO><656(*fHpyAW7CdXT8U
zZq5+BXPZCDV|W=>za1ADp%TXl$^wSoY8Kj!-P}TQ*Sb-{1f<k-JM8+VE9qa;cu$Uw
zh#Gb`ajwmZI_u|;W_uQTG5?-8?0D%pTg;lED2X$Q$+0yzadqxHb2y7mKfd{Ez2Te$
zD0KYVrUmSlu7y!s->^n6sfD&qV{wIs$i2|b*+jur|G@0sI~7yGv+1>dFzgH-$9Awn
zW-_r?uQe7^V#?!o8EQ2+f^PfgBiFh#FZ?i_@~z-1y${Rr!yZR#UnS)JV<$6L(GmCl
z+|2y<k3KfN;iAf9?g{Q5X1`pmjmx&+VIX_^w4LK|1&K~(-Uje_gvjNbvV43=b@KH1
z6&Hs|d-{MaSD{9??@EP~u`;_Wcro4;jxC4gSMg@1KdUS$6!i7El1v0wz<YE|I+J@Y
zyWq3vZfMJtQWb3Lv7%Q8zx$d&)!o+D`Kywz-@kFn9lFrdHs~|oDBzqw_I;mNfF@G$
zd`>`4g5qYyx;}aX$jZu$fk{(-GZhB~-{I$s)8@m25u?E6p-9Luobx8W7e*ES)&Kiq
zPf(tzu-RxEu>rG|xz3M_LmT6=QFo<Gd{_2UC((K9Zx$mhkkf{#%cz*S!t}ME<o(*5
z??NL>qjb_|3k7v3+u(BrFgqvuVZ`S&P03YpanyLL*LuQcartb|!ZiuCs}@VN|3+SJ
zD=Ip9^7<f;#%wt{kX&`#pKyqIq}`aB7~DmOWzUDA9k)9B#&}eFjM1W}hp;uB|8Y%|
z`NT92GS7RpS#NuwYEoGEW@I>)XGOr{#VGfXGN-lh0%|>0jiWP>A#z@9PjCQDr!_`%
zOP;AQS1Z%?#@oBJCFNEVUf5{e<m2c*zoO_;eVKQ~*J{c73VJTYHwpwox!CkzG%v+E
z6a{M~Kcucqes}XK5*V1fP!<eSy}Ro!kp>C(rnKZ~ida{B#+Gb9dM*E3puxefaq3$M
za>Fe4Y>)N66mll0C%|+|eZlk)3N$aWSV1}civqb>W9_bQy@E(aI{hEN5^z@r-k9Mm
z@}T;Ux@Bl0f2daaG7no^YJfE9y8EQ9;d~&J4+$+SG|&Gk;o}OP+j<WaPK|o}2<!QM
zfNU#lL%=yWoyAc@(uv*!rez`N?DS>ESUXr06N%|_CqVMGg&YH%0Ji*;%qDQ&?$0-A
zN;=SGVq^7hh$*}|J>M{5@IcxQr+V}dI8VhrO6{W$&rg0I<>GO*`esPc(BvMRKXkk3
zhKq+wxfS?U1X&>U7d?Gb^T4I|T^633no=?<{y92=QnIrBBHUy%()3KUUmF?_-p1|U
zuAbvLU-UKR(NI$px3XeE+`4s(<o0czK%>ESkOpgm4xS@5PTQ%8{(oH|c=vL1DD>qf
zFG)J<ZDx=9i24u;kpjK+#LD+qw=;QLfYe^$9-8?e#j8>j3%fw&$>wI3zQ<=iWXboK
zYIVO^oOg1ePNhCn(|)}!oT@%4QLepNgGUK);v`cSfs=2DA9_1sV})3BZ#mJ_{styL
zK!$C}L_NvZ;$_stP>kwWE2yerLg`tVf&n2X$X&@-=H6YC6<^~OJ{-Kc-)k><RVrgI
z5;H53=^7Y)UEP{<w41jS5407lqlU7@wyJt%yXkAJm1??yp>HtG$$COUGQ_{N3@Kz<
zovrbnyIh2o4t*$$aP4{ea?zdDU(B$^_G!W1UUPx!h}O2tHK4}6#6h8O<UfD#-iAM}
zy={#-@c3({NXyjlyT8Mv&d8X>W_PIWiQDRvwispD<)&q#{RWJm8IMx5dng~pqW}fN
zJ7#43kUJeStKH*bkBs=iV9Lm71Z?3{xtZb0iua4(8wY-IP-3Gg;Sr44pJB<y5QxKR
z9vY8c$4m}QU_47q#Kkq$%pB}G7m{vkLzLQ_olEYci~G3sNp81LuLcA!=qupE$|)35
z&6%1DD-H|W+F88)s;<Wm+l~yiEjoN6*kp=JC<B7_&pJjA4~rN3h0<Odl<c1fnLnfs
z4U6=ye9yvUUk-VpB*3ASQ^CO8TJrjuGdj)M18%6*yJr2&zZrK|*mXTPxO8_nMYFm*
zdboIU)kA*4Wnic+MdK7Uzf9{JJr~e_09#l|Zj-Pz@BE896aR2l3Q!`Ov9r<?;qFFp
zpv23>RKu!6LMInN7^P+Sdgo{1(BQY`?;q}xdy(`#-<P*L*SlSb`b~jPsxSiQel%F`
zlz(jfh`q8u6Rh$MQ+YWzr(wfI%2udaf*kHql}XgITn@_2LE;b<VGZ8oLmrS8lD&E3
zH_rF(9Zj8+b(5$6!gX#aoL9^$3uC=KQ&a@p@cr<pIUN%*SqDzoQbe&ka4ALjBhl6D
z6ISKg<Myb?ss7BrxM}H<4!tfrg$%fHOOPr3>oZTtU}aY-Rd(h&!(?fsH!RQItKO&F
z1uitWGxm4mhv(wUW5=4c7ZeUGZ0w{8uR+bbY@xdx|7!^*M9?Zlcjx}4ndeqEQheJL
z#Wql^m|?@0ZjE{M&_^jYR!3W>Ppbdse}>WV9?}H=dZHZczd>)Tubp%cAvf?2&5D5i
zx?0Yje2_0|i_2;P339d0-J7^hU0^IqgPM&MB`v9XjSY3P=dx#QZ_zI5Ayi@vG~R>y
zoTKJpU7|$p+A+2tF1E16OQte@5+k2XOW|MeA<0qXrPxEm`qW3;=GXO@c(1s0Oen;C
zEwKH7WM!&aKoK?~aIwT3KJ<QiGl_qnb4q&+YlF2|EK`Kd=6TPLss4+QsWN4zSstB_
zR&Q@+w7)%_+eePtrc0^N=RV<Aeog%PI>RG6y5h>BcYk5?Y1)0moV2=Kaof(e;Z#Nn
zOM^tHIh!W6=^o8XB4x3W7b#M&O^ilp&HAZdb!`xs`i^2Q*~k7mt}M_{6i?iq?=r$|
zn+RV_qfnUAe?`aEJr-B~GE8`9O*V21k<?Q!_K^JwPXBLBe44{%+Oi=3ydG`~i%~ix
zY41KNC-9F{ko)`lj?#_UtEPA^dxaoyT@a^wE)lm;D?9hBj9GeJG_3QH?W^q!6{kfr
zXWoL4-)fju+Y3gcU+71?=J`i(n|fp&g2SkI_STFSh^m;KgUho0b_<ijh=`;m_X%#R
z4PJkv!c6l>%7fIYnmZ%=T>Y?iJj1Ei=EP-RBcIFiWj-nA&=Q8#;D<jAn@?~snMDug
zo^2RA7{{6$+g0+`F6awAxwAO&?ZeKC4T@NfP11#GwW>LvsL1dr#Z&D!0hYsd^~3UJ
zV}>^0q<KEOE(k@+*2i}ZWHHx;(Jou;k9S@jny#Rz(~H)L<LLE2#oD!PgoY|D)eGNQ
zPV3XBpwIJh5)vDCI^YGmafg&LVGYd#qkW2&hpvTUO#-`LKajoV3=9}OOQBF_XDd(d
zlZfwmrpb_#5y9W`(~QGyZ@k}l{BsKLA+2(SYx&F8gv!MU^OiGj)P8R!*DqV+zE_-v
z@H4=#l5CHShuzm$&zt+MN2)gRi_kuUM!}C0FY5{xYQ~>sBEy~MY9p9TJdf>5efO5-
zAO8_aOb$++7Ge=fUrjnzOFw$P$Y;G@fs6DAs8;b!yRFF-mU%&Q3NuHxe~BjFb$7#f
zEH9>R(t&?X-#x3|wO(Q@go4?g>QldbRtxPh+eyi};rC2U)NAv^vqz@yCJ*N>vNHW9
zpyXHYKP&#Fven*Ue2YBa<FrO~&A>e?d*w#i@qD(<To)nX()K)|`ZMUkdfd&r@I~#Y
z<DlrM<4e&kYP1D;R{3oRH|F-N4fO@R$NoIRmLXV;-QCYYCsP~CF?3XFOT9Lknmbya
z-59~YP-ZkFX9De2yWKZK^_dPu`-d$0N_XQ3NlB8$ni?Zim9li7N^IP~XVlTz`Q3b#
zw8c5ab2zdnzs&t8?r8h`RCZgTpifh=a2S3RdHe)fwC}dlSuQID_aFO(??JIz?*6NR
zAjRF!Y@F!VIlxX>vpMQ9p9QV?vJJw8iMX8i)rQLr3wDi`M_+smP|0yuik>`cI(6O|
zx*mIO=5?!ZDtTTod(?qT)>nH8S{)7+>SEzlS4FMAs}_F#=8b~LW#a_l(9okrUD=(V
z-^X^_axO8Be8`F@Z5nl<ljUfV3SXjW0U7%l#r>gob#8=O%udCB+I#XZNNqOKpEh4b
z_UT>UrQ=a7d1r}~`DRg9fHT0sNXFUs81O|*QL6kjT~AEj9S{T{yhTd)Opnm=d4Zu-
zj7Iz;4Q2@v=6*IX{K5}^`K@v&c%l2sd7^~B;xJdV-bWI<>6G>T#2YKLsiNm0mI@c}
z_Rb4N9I!P#f{taTc7L}zC4&JmzB4El77-h%zFgT?+}+iirQIajz0lG1BtfOr+#7}<
zD+&KjPpE9>{FN(mb6LKq{<;KT^$FFB)%Kx5;n2_${a^A08poO4YT+O_d0tvzs&y4M
z5vkHvX6rsyKNV74<*XoQ_Ao-ALbbnMEv??;M&V6^;+kQsQB%4d#|Zb|X&f8K-oSGP
zg(wBa-S+cp#bNsuW9>hkzz~Rod)OL|PcK}s6%1KFa@{3$t8^*+W^1UL;v`d2Ty`WC
zOFERB$*WjeI2%zm)t1K)<YW}9U_L@8*hjvmV<&RVff~yzLJaTxTMIzy`}X_V77aVS
zDpZFK#INxgZBJ_2HidUDRaUPLJa!10G5kkQPVG^byO+*7C96`{PGc-yyMv<X={T{w
zQlUVI$W05<c?SdW?yk;$yAZzgNY!$b|B~<MBke$5Qu|cjmU20{m+x?q7#wj-g_J2s
z^=HPN2hvB2AD(05;11wy|619KDam*L)8`Ge#P5Qq(bpPgs;YD0Tzi&~&Bvrky?w9Z
ze7L)seZMuqAD3WwE_FTYV_?}x8s}u+p5|V*MacuvFDv+V!S9(IYVI2SATo=z9A5Ls
z#qU$asqd+YoEj4UyGc9ZfE!yQV$(yGLO%8TOE(apqd0SpRHNv|G~+zaRIl2|C1S-D
zv`IP$zt_~1`_yQy*t$c(ZSR!8_?EO}(j=?Wb^KOLi?QV}LpSb|2gp9%^XUA%@o|*@
z-<{??x!!saQ?msH_eC*tXD+?F(m7fsXazlZ<gxH0o!B;*GauXSe_E(=Xyb&p&32lC
zWhdKB!5&-5;dqvM7?u`wq?w=f;X}aW**<g(mpkTw?#tQxw-~cgj(=<mk>&lQjZ}27
z_Ep}Hq=#(k8HYTwdVeY7fO+7P``d!*6IOMF`b4i)`*edv!&fM6Gf}>K$f`7VH@Aa>
zb7QFxdPWJ&?Y`iop3x+Ch3$9FRjKNxK8uG_8WTkIHWj?bXIMPEzxU|G|9%yoRqwor
z^{P(##LCv!ICqR1_nl8md`82H6-z4CXjlrAZA%~E*F6uJ<PzAWvz#|p-&`9qgMR<{
zOi`VW=(r^_qYc?ZvNAheWNc@}KYqH0l9&GzOThg+C#)fwx37lgf_cHx+8JYRUv)R$
ztz<r%J)+(HX=NrpnI9PbFml5>+X@<TYYwR;csNuBYKpzu7y6;9O328g?aZBHcy#f$
zL5NP>J+2%XWnAQ1Dk-WbhJl>h`UPI206jY;Q}FWPh5!{se}wDiVPKAKtU(e%-9eRa
z&K%46ag~;{cp`y&*U4_pCG!dqqH`N2)P?Q%;maDte<3rLXDX#yow~LU&&Q`c%vNx?
zDI|-A84-2_m=>4|zkbm@e%y~DA-WZj<k35(c0*c{Q**Gkc^4BCQ~r9Jb>tz|fzmG{
zq>T4;vgo2AiM4$&Jj`IZQMhQgpTE^$=ps8EbNz+FsI0hgb56?1yGXieJB-EYsjDw#
zQgWqy=t{XR=SZIkxt=+2B@9N-&w?Dvcbhj4C9$UCXdLNj*x2OoNZCGf3HvyCdFKa9
zxkP7xmN5|s-^Wd8q|oZzgN9%aZMNyi7r$(;-U&)gC2HQt7AC4voBaKtXXZoaHz%4?
zW$8A`B}=9FOA5aoxhEIBCd?K2wUS%jYf*O>ioAvgK}ITtU9Ds0<I_CaHGDEzco+V{
zz_1@#bFC@p5P*^4V`i3gax&oS@bUF+mkyW|<hp}DLVuu4>}fHrnT%aLVv2obb7EpK
z<I8VW5B0RKpGRI7PP-qfrwWIqk+(|7u?Fn)Z{38+#>f<uZAl>)IHsRvn1COr{rUV<
zZKiMmulpjOL)Y+aJhDQZvx23Se2(f@mjl=i*qa{JbYr34lCShu_l-@wy=~VszWwQI
z)e{=IM;GwbFi^Z_%vd=MWfN8Hk8l|$swLM&^UNY5++Qq@Jgv$xBNOuVQ`8ptdp->a
z5!Ub%ePJtpbL6m|wzGQ2twvYTXVJ;dEsV<O^|?t7-5(`j1OoL4TO-+@o9~PX*;g`E
zDDQ0FC1g?mL@%jsyhQQUCsI^hoJG!mXDN;28@j%VIif0;9y_kExq&qtV$Ff8!yFV{
zJ#Sk&sF4;t2Yi5fatfmVnb(5oBiQCrum-E$o>A-F8632kD`R4;9pCby2stZS5K%vG
z6VN06va|Y=rK(JGid6I;u{cdd7m1ZC0~lZN{pya617bJZSB|XLobgT$9^)OU;{=&f
z=H|w+OV4dSTBx4jijGOQ07vf!(m4|kCEN5#oYy=az(y8d)zg3Ep{D=bi2^g#3!Fq3
z6`OkH?8U{d{^4W%r4=`<_B$T0GJ|6|mML~khs!U^qT?(?Jf`Gn%$zr!kJz+(>_~b3
zlA!0X82kjS?6==|b6KMzye?cE%+eOzHJXxWDp=27FUr>=@f{5PJ6ukC!;>2mu38$t
z+(;zttIiyh{(DHIu%tJSXpfWQz{E`)xhWQsxM@Xdor|$BmC3PgSsNJ8^D#??5vtYK
zt-g74>!DX9c6WRzm5=a?KUV!kt-bq41l8@YUyJ?xRO7EG(EaFEbXWM;NXik>`K{9b
zG-hRhwsOK8DO0;-7|`EfjQLQ4<}Wt6y${!NQxS+0!VU4i?+?)+BK;o$SkI78)Bjb~
z`}$wGKCl03y8aI-R_uR5Ct_2Y@V`>5H2;<B`+xn!7xM6@bKMVSuoAeP`cNcU7cz|K
zI5`zTq`G))O`-LT4>mTooRSg^5*r`?W9PT2h7)CX_c|(C8k+C3RG-6I=cF|>5~*lG
zP7~j+Y4*kEFgOanARr*{8Vt-zE2XA}+liY3DPw4u)ncSo2KwBXmoL-(<ry-3_G7^s
z+IE9B%du&tZYzH7eMFZ2362~_XlUqYrA?;&`nZuDq1^R;w!`GhaE_P^T4Wgt?{=Q_
zj@oqD*ZP=@WkxI9rfCAcX`YKzTZtGg@FTM|j2YLdu#!m783(-rE_fQhjutB9M@%m!
zdYm~if{uC^n?ctdyY=yr*%yesauCCwaF}unOS^twcqzspvFOQl&rz*|%^6WKvG57=
zVP8D5yx@z;x1xDm$|qemc#wTK%r(hhy|KWcJ^5xH{Vh@gkm-NdzS<wX>M!NGgc!J-
zot@2Phl-%emd{ndhF(xlt|M|6=I57y3=E}WH~lFEx^wV9JxBj~`cxU>12}H-XZEFB
z_?{}$OTDRF*7NVEvXx8p{R0C_;aCrW|1mqL|M8*!U#1dW0}6rAiUgJVl_1VZ8iyYP
z*|LM>W@DtHmy*k$I`3qCWKz!o<#lo%XSpaf>-R@{Yq=|h^%^WJEDEg`r7l93WZ~dD
zPS15MYIhZa$T$mPZ(?LIiucRu=zOX>{6)4lS*!Z==~Msr@1^(GC)gb~wLl#GC+Z#I
z<Pyx>dcD|;ar1qrR>US>h7oGJn{OC&(F4?zVq<eaw!Njb)num82m8u&KH)S$*S>qo
z%Qz0RoTpbkM;oDQ^!ta2Qlg@wMJuwhvXO8EbNz|T{**`kWdaIQX#vF9TS;of$>c=>
z?&LW0sx}k}!Z;KY=awPpje~q%PihEnPHsV!=-01b=@=LWW}5uU9Jh3G?_bCQ;f^;)
z-5zJMzt%)TC<fDz;b7iXb}I_rZ3w01a7kWiQBi*ZLBZVtS;=g-<304*Vjdfm!r^eo
zE2sNY!;y*_*}_4j%H+iS<gKJNh?B(5Z(?^n&$--BzWnNpdx&%ff}e<(IP=4YYqxz<
z8XEY^40{DU4w@*RI1_tP(cZoLJj`}3h_ijhQi<#`e~@yOZ8nurdju_Nyb{{jT?B^?
zs)U6_x~(~ae)wT&@P_DhFKpf{MZfw*rO#S#(n?H>tef#Xe@ACz``{3Z-k7?4;pClB
ztTh*?S}WwML~>;)7HWj0t*)+;c3?Jm{PB$^?QE~|)hcDeY)^NGRpT%7n?=RPGi3H>
zXMt9LVQ(Yq3BvfDHVLXX)sM<(s?IGv7%3_mMrF79v*?>O#-^rg#Wl|<)jHS8x+4@l
z=qaMb=21N<-UZd`T<|4nUu`9F0)9MD6_k7a{0&vMbgJ-ZtxLr%2Bl=sg{Ng;5Ld6V
z?d?es?BG_x^7J_T-N8(ateA3tws9N5_t<Lo1?a^ucXCb*lyvjv{`{#lvStq|?QXk+
zilgObaViz&Yw9@|h@5b3Ykx<CJL-6&e%Pe2R^iuL0+(H%=GkxJq5g{}-_K5ucQ=~|
z=()I5Qbj_Bc9#3a$Q@SlD~v(%kDUykKOf}$*qwJk=c<*`BhPskr5t~S|J5^?uD^c7
zq;Aw57yR7S*4C~9b#t^q!%YPAVQfCU^A)GT0?-*M#N#B@d1oo*G9KyoN4Ri<)DIpQ
zAMGq3f#BZm(MsV8*jCA)G{(S(lM`2DAE?{2yB*mEk+29jj9d9NHZ?^iB^APDBp&Pk
z=xOx98B7-=wp$tMoNY$r6}z80BV|CWjK%?j4{h|C?uY5kyD-H-%~0Cjz8uWL;yEkg
z%-4(9bBa!db43kqmGqI8CyY)$7Nx7y%22+nu<+%FDi=;(^h$G3gg}tr?W^6NT&ZOQ
z31`l6OK)py>z%Jyl!jgLPn{PN9P<3J8+5j|w!*q7dI=iN88qwE;LfDP#RD~*7KqA>
z2SqV3Fu<-vU%u?&JKG}2SFMnEY|tg=eHkw=7$-S+j>-ff<qs;APB_!*)%MC@SFhf`
zd$*}U!C|VDt$P{&4h@paa@rGmYX{9=kg85gPp9SKQ3q-1n=}~iGyr_zav!L`BfeT2
zEtA#KO2%!HQc{XGuRGMAu<7P$@95}{e`>=Zfzv=K1#-c+{Utux{+>q$gd?(O7u|!3
zwVQm?gRkP?Y-=!a-6cbu7rP!>Bm2P2lr?W|3S!~k`9Q^PHYx||)#<^@AoiJ7vUoZ@
zAcDvt10NNk4Tc`pteAt~y8$|h8L#3*!z7Y;)lnpw{mQ1?i1S(dYP9D-_9Z{}W{|kp
zfGspN&4QcygGH#qi93<B?NUQMOfG&}R#w(ht2tq$Ott+Qd@uO{QGTYf-1Ny1YA}}?
z34^A<kH~zCIOzoU3sE==u9Ran_f4S{L+j0_ZJV@=jE{#{GwdcqFT+$y4W-Sm<09nW
zz+2XY=`;Y(U+Qu0zOcCHQrHdS0q$z&`~8rYVFfk6McIw|nscHN3_bN89-wJz(izJh
z%W08lZDT{7z@uU?m?Iy{Zt{_e=otnK&eq{#l2{lxbt9`A2!;2-!3rH88KH+f_tx9J
z^$GflF~;T9#rGC<hr%TW-5;AxFg;_nqZuw=Y#>PBbyJyZ4PkZ=hbP>b4PgFFWZFPL
z&g-fqn;}kHey6E&y=sMt_cOfH$B$*_z6FI-fxo1uqH0D*XX~RKBc7ceLA2zkjrjT5
zDZ5@fVK7L%tpS2N$PWy}McWk_r};pg+i^rxR3`Y>Y~9vi(kGV51z%>dCj&9u834_Q
zePrl`AfE|-5r$l5qS|42bEavq#>pmk4&!sfC!qi~lOcxPm0`&czOyfbRtP14CPpAU
zs#0bgDI+5@Z_7n=^Jc0<0@w04G7C8fIC;TW8YIE6{>I6kAv$>f7l0<+()3qvkVJ=n
zqjhf8gH|V#9~o6^>O)FP*>dO5x5n;;hQ$#kAn575)k2JpE&yogsHk2*Tvn)eKixq)
zJzic7w7~;UC6aHIVbuNymdQH?=jAb2&^~qgJx?dTcnr#`wZ{*hK0U0N>ppcmnA^S%
z=Rc!GV+BOAeS3R*-}lh_Wu_y=H!AdAzoxl=U(jSIulLzT&3$AnpGV!dz}vF@EGut6
zdHr0oSB+1zr=>FuU_#5e(7gY1k5W=n$%#BJpg5_nAo2(IM|l7FVDs#F!=+;a;vsBr
z%_U{gc^k`S^vTl`0eBF?(0!3$!{fa*wA!whw>P82A645kzrPQKx8=zEkeV8;_}OC4
zi7Pk*Vr>Kjin$8;wjGQW1Gaz25Kka<fY=LaY!o1WV#yyi03qPUg1O!kh$AOktyOE~
zV*{Yj>+|+)7WgXX&97LI(a}ol<*MbTTZxwRjEt|sN==5DUd6Bq0amMQY-}8;cCe^D
z-5YCfYkO+(@y8EwM)fK=u=w6VMa}%BB}4S7z<0$yvHEo5#tm2ViQ$~ddLttv=cDD!
z!HidA=+8lz)p<ER$`+yoPP=Yw27Hmtcz{MoL?qY!^oW$hG<tS!u66})%CZR$EP#~V
zIKrVCM&tX8WfmCv<O-MV(sZZd&h}2fWD=*K9wmCPrN*o-F~gXD(?8yr^6>WZ$}yd6
z2_QBBsa^2aEd`Pi5(?JV#b~69N6sY&9v=JZRobG{%7^2?QlMSJIta(==yI6;OvWSQ
zSQ;@%8XPS%2_SmANJm2>1@2w6)~q{`XSBp11RWVQYMqw>8!Ehb@j{5gIw+_8)Fd|J
zBZKm%yMn%iWF>CL_GVLc>VVRUIm|}WK%O`%<{{7=6%+Px@GY6uBcXs>M!lbeP(%4~
zLj~&eQ61DEegPw_Gv{ASxaG(VchyqxYznzd_qSw}X2W&J^O8y8Ix`B?t3|_x@|07(
zu?Q5LyPn$oG6bz&>h{(t@RgwjwR@vCgeEO&kJqcI+b3)ip4teb<(QqFoqc}a8Ue-U
z%eW*`D<SjwsMF&aKgy3Z!rco10<;=DZ={<-Qo`<fuz6!)y|UbN<Wq{EA00Du;Wz`4
z{75SdGvWXnoq0+%Lb5PjJ6>gH!gDwuJ_^hEPB_hd5X_8bn?WT9(7;j?0UsJ@0u496
z<roncmz$PGg?HzPR9LC`1Q(;y=j8FHnGy+f)e{9HB?c=1OyG&zySs-W6x4``eo<pW
zhdaG47TYDM)8nc0A-EMd%L<PG8$G=k8HZ^P`>@*mWz*}GIxYSI?&Iv~TzeajySY=}
zy}JUQ)@2{91q`yML}X<2NG<^PbzrtK6_OCQe=l@EEI>|G)@OQoT{t}e{hG0gLWMk~
z6hI!4KhOHam?Mp1!D5qYhEKuBSYMU=?1o$+cZ~%-1iVd`z|dL%QyYuYZS}(hvN<`}
z`Xxf?_BtayJ;5RI>dgnyVL!fqf4L0@=>^+VPyB*#DUQ@1D>oAxPmG9=V`xPSRwra8
zCpWAD$aNTpyo>qd`S}CMBwjah<gfgSNflJ|>G8hHh5;NzRFwo6!*S7ZDaDV;`55)H
z9&%E+0F7ND%nQd>o`Y%c^kCXii-k4fiZQcqEiF#1k>w!{>*LF{Zr~#Ic*{|a=%onH
z!s#pM*O$XN<tmFrB1zc{rHWSi`}^4)*44msO2FKeS5u3J&>))71fgPmyfQ3JQPZvL
zhn+=wiI^bw{<uw8R{@#o<omvL5cFl0JX5f3HViGSKdbuT)UI1H0#S_s83yiJA?0a~
zs@>WYxF{5iDC2H<bTrBCa=%DeDGc+CipBYP$B=gN$NGU>aB=-aNbn+KQi|!eNzj9=
z`6tP1w_S)-G(P;j0B%@DK)_4GZC4&<r#Qg8NtRs2LIqP((`;%(XXh%1+F>jTM}GJK
z=JnRuB}`1n<X__Z3Bo@sE2~Mf5K2A`{erEz)+A1cP)MPpp00yBt~1D8gS}gVf0wy?
z_wFRfjR-nS*hQt4td!j%B#dxLQphSeJsy)euK<+stM+_~%k|*%{_ihVbL|NaIXDzj
zQc@Ozv$NW{(drrvQJ0j+;rD<$xMS2*ef`n?Mj_wnGFXr-EiJ9{b|>d9cp#8e!oB?o
z!a2n3zUqm}g_w?ou#SbQ>0^dMsoK@zj`oQm{gP}LCA4byWUY%B`8FU7=Mcl%YakAW
z?>FbwI{X$jT7Uie^#yaXJ9ik6*3}c#=Fcjgn1f5H+$Ql&2_}DL45(uvIQQLu?#&YI
zAio&;YSX}dqmljDUPV2UOEtm`OuglsyrPpniE#Qi*Mk{9V)^wU1%>sg;^+)z-F>|l
zG{eMiCWy|v>GU9TnC##?*%T}q)6vnXiK;(de-@MB^p&rus7P;e5lp$G6Q+ErK{v1?
zfTzU96-vv^EiFIu9!y_co2u7TR#t}lMU%t0IK)gf$!=Pa@Gz0r$;sDhUF)X8sn&iL
zrK(uNH4$fj;=G)wwsy*XasM)dy=sYmXH66zI32p`ZZxPFAFHsKTCY_r9Wo9DKt>PF
zA&kvpl6%5!f5N!;lMkdAdL<+N<cG@SVpwwN=nM6xA#ZLhKp9fhGkXxJUr1bCQxh`5
z`xCQ+(eYKTU5vSQzIMFUWnZTP6AQ~^{(ERx8l>BrU~>gw$MFy@?AJ!+QD#0qJ`l9S
z_Uk=dit}xTvl&(V%Gmty$@|c}L0j5jZq=TJ-djJvTu83N(n5HTUh<uv*ksRfQ-rM7
zpPyMTa_wI-h0pS<KA?w~{EyX(E)C{JX}GMh=WEoeEH5tu=HD*JH;&CHE-sFUj?RJ*
zV+ihla!ojpL|UgnQNyXl<tY=BMEFBVbc_SSzIM2J!W<1z-@kt!3jqj|C>Nb<K1_0*
z<+(7hX1@Z-o*cNEh=>R(WYylNiCFy7N|+eamPQ>!jWXw5V_1oBrW|{J^C6qdo@trM
z@cJhS6@`!BwK`@jArMEn=#=>96r8HwMu)h4Ye>@AJkDG<HrnR441xRr$MOLjs`NTH
zBpQ<){TZ+NA@$3B_z0eIx(Q8F@?S;6Tip<zj4xlF0BC=Ko_-5BrW|qi?%f+GF_2_b
zE^e;wgj_9`zTnHJW;LL=dLXBpKHaF8bQVJ=M_^L8O&_#LfRm7V{W{HgZ#5g9P{xmd
zl8JL_zAaoHkQd-IS(uEIPr=FLfp<fOt=d<g9d>(s{5dpZ=@@c{$7=TN?CdNdDJja$
zw70hx{W`I6aM%GI`mfgMz@HGon6en$Avvtk+dOplmuo%C&s!;|mbi$8A3+5RD26}@
zOilr*rU<^e%z8nzrKKesO*l}+<<F1xeg6E&4mE%bD>WTq6%rC^dxX2e>j+aWch?GS
z;&q;XOl}momTX{FJ~|x(Z<kL*uRz1fS}eM5@q50_84~vPnp(gdKmo~Y|LPR0Jp~49
z^Oj%wReby*h;q9SA?x~-nH8jOyVfPy$nT!zGc-LawO#JJ(V-ik3$RJ0+%#tY`{uiM
z?_^pNT;4i{esY_(ug-n&_ZtDW6mG>hX?4{UCZ8af4W!TsKvbeRpXDDhj{%4o59UOo
zM-e8t2^3Hn_auLHTC4%%i3|-*iKJIVX`UZN0hY;yL(t9_O`&nwuVw)SVbb7<fdV*H
zX#)h12!+!N2yMe~*4-9j%_X4t;Imu+(y3s4l8~Ly+Z+Q|WHwSPgma6b2fkOP-E^7g
z)~y_ejY%|4ejpK_1z~<|qI&<+eu8|C^|^x2^t2YEYPmSQV!;3axm>uPoaAJF^mV|^
zv+MsL0qSDo>XweKF2r4UIiuAMg&C4b1q&TfCV=bYi*3;!TuPBdzgHVACsXzlxU1TS
zR?lpgJ^(z_x4I_>&sGfynG3RSwk5z0oH<aX@XbvghF<vPNva5mEX01t^J=)yAs;l>
zcP075+2@S=6VdPN*PpAy^e>5*$Q197Fp6#O>QaE92Ha>G1hto8z66v7Kt2WBM0=vT
z_v~mgUuvPaqy$aAEoWyY(mBKfa~pX3u`vx1d_`*ik4r}|T!$QPZLA{Ks6XTH6vOqF
z2s(vJZbt`8^gJWhbWH*F)d;`<m(8z0@Y@50ntVYNJSu{|xYANmAAq8PGbZUrXBerF
z|C9Y8EVq<C{mEF+PeS<rclAMxmun3U)JMK}E(gD1g#OPhaU7kS(;w;<)qQz|z7SGK
L;zfagme>CRWwD@k

diff --git a/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-42-1.png b/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-42-1.png
index 33a10818ca05af9c00406b90ce593809c25f56e1..eb93e89f44680e9c4039ec9365ec2cf69761463a 100644
GIT binary patch
delta 108268
zcmZ5|cOaJE`#%XqNXVu#BZTagot?eOi0qy1u4H>CviAz{*n4L^*?aFz_TKys?>^r@
zet+c;p69;Lxz2UH)~Timy|5X*V1y5nb=!u0eI)&<*L}P8zCUVjz3z*5$hczog~{{`
z?@(}?MOgYX@U-60r}*@Wh`4PF#GHHGTg|!ut&^I%`BJ;1wAuYg@O(O^!Z(3^<wqb-
z;hB>x6ZtMFf5I|KX5@vW`006Ke9m!V`y;PIGSWz>vyEz^t)q>>X{x1`LcmWJdbjwM
z>)PDydSudzCB4OW+Gw$M%1cP~_+0zm)2|P&sj>9HKO*;blU7FRaO@OE0{F`le07uh
z`4ko_H<3y{%#;^N4|Hr=;XdoJPHU<_)e$^ELC&E~V`RSD(4LLVj8$!5y&z#{hpZ_z
z*Q_@#si*cxc>SlH*NEu@*YkVQkR5J}6j#<n7R_*3$dIed^;?Z^i2l>?vFcMvd~xA4
zj+V63C%L*LsT`%xSg1ay?}Rr?q7@$@DmOIFu~=#@nWoVAm8LRgoTst0-<eq{G!9YW
zFQyw<=FLcN%%^%7GH(f6SV|iSjX~<*eX^v|e>2{^&eN-X-ecmXL^(*NvSNkNql!O3
zTJ=ga*SbT#vJDn)!OF>+CKn~F`u6>4WyMIES2*WapG1?M#uTxH8B}_gSEO&^{Fdie
zr?9U|oj&JPeA@|M=dL}~Wy5P4<*z2TuA_nd?XHznZ_@q2#4Zqnllfq&-3KpPgSXdy
z^##eX3pN#AP`?pgK3)yv#5@v`zZet~`iQ_QKr#usz9Q6xy&|FFJtAqTV-mtGf^gx!
zwK?jEj8`(xJ9|=%@)e_FBfN&eJABO&B)DE=xP2cnrh#7rNF<!=ugwQ}L~uN`ckc!^
z1`Oq=F*3Y6S;!yq-$ZH^(C+^e#9NOSp(U4zAY`2_P2e7etn`0XVkQrN$e!7?lc<9p
zZ*!*g5fI{f&M9=h$R*zUlXF~s!A3wxzy_T)=~UX@$#f$pI=teEJ%r$$%gjG=Rkp4U
zFwSbLP~q7G7LlEEJusokm0vkC(bu&w38Q8o-Tk&V6>4)Q@q-_}lzSmvT-H4Tk1p2r
z^uNM_KxJhrWxFhQ<@bIFzNfR#%lg#4oidj`{>q3GH<N3A^>2%)LXE1EkJ{@?TB|GG
zcex$S+q;|l(lM|pXdqUj-@TvrZwVRpCyBJ*{Z%Gpze){G>EeZ4k@ysZ(B!B5D&g*x
z^M6Z-B<j`6lZS~44+hlzvAH~i#?b@f)I}wrV92|%M)rt_L|D(nu+j+AeqijlOLXx)
zZU1%3O4dxe9D&;H3VXiszNsKkJIQuI-TYjhtX1>Rjywdue)GgsbZl{kJ^v%Ova^@T
zUA-wPE87)QIg9fm_h4e7q{x%Uf<G|zZGDr`tG`#4K8C6c90Ey~1nd^-MQlnCPsL>v
z!>^)?e&t7_2OS=RBut*PR<A7#HB($hT5CUWTnGdP1)UsDIRDCOkF~hypzH(t+;7NI
zc-VAAJ$<jnRMvdZwajzyd-~23l2fPoN;AxD(Tv;bpf}sWxP4-~j)p##muAa2S)cH7
zeiocwM(XV9OUr9@&Ui!kK7*^6cWvT2&WN!pK4R<gt{@>jL6Q;^R@pG!Ae%awJ--~Z
z82mMEetnIBv3F0v{XvA5Xj~uVT7;;DiCo8AF_K2Hqx}2G%qWxzFIpD5APIrQrtypv
zJ)T%B7Zi32!aGMc)I3kC!WeVMKw1CnnR&A!X!J)q<lTfMInNGs$^ERyDzoCralT!P
z_^FuwB<#9~Zj?WhX3Z|jY|8b?wpT3*ywHqe`t!}P-4esr#JhR9hx6COFV9^6a+<d(
zVCt>xRW6X}p7J9}&^{)Ln+S`F@z2Xbq8}Si8Tl3<>)}LkB)>GlJnJ=D?MfV&+61?w
z%q3f2OUM*PKnR{fbBx_SE_w!rhChEQ9_8dlzrJQdI&pMW+9s1|T^_VLei?Pdvb^}+
z>H*oM$oq)n*DpDW-;LLlUR3U)tFgEyfwO;4s4{OcHiIflW$``D=TxM323CkBh5Du)
zguW<zSjBYh=8-Ef|F<8zEU;MT#qjr%Jd{vL(kB#B9>S8O#S?marL@8#yM~rDL8~h4
zpW4@wS<3ZhPEp$Jl?}B>6{kn6^r3~7<RQ~mQZ=}LhA=;$l>JpBUQ&%trPrx<!~h2{
z#A$-~N5@n-%gqy?Whh6eOnHlEsIh<8pmxKG`O{Zxaer^7z4TjnIt?O)9s^r`3mUW<
z3_KF>9e`g`ahP=A-qg!v%qth4o7MhG?=78?D(+WTvf4}dDK+((y@NwRNy*vzPEjla
z4eMhh%xYe=6}IyIucV*FrYcmdu%;(B9$@S_6xvyRk$ykQ?6CYX5h<OvMNPsC=Y_I^
z%vp#_Y#mi1+UvfxPhn^U;5(qF6fgQ`32wJEHU3pB`qF|!oB9T08m=dwpL`#2yo`RJ
z(&@1SUnc!inC{KK|2|Ci5V`$>ET=r5yLKKkXx*rrRIFRHV=-nL6TFcgm-QV_st(+V
zobx4ybWJL0S|s<*owpy6?a&dMwc!&Iq$quEeG`->Lk>m&)Rop#A6l4QDZG!5zgTF}
zMi#O?^IB;`E-(<|Ej}vj`(s8@cP%HfhBTJW-z}oowWgGgA49B>j58&O<JUi8_qf%e
zt@KA#W?*R<DHjc-${hJ|B=@<Wdu`YVP~9x1KYGGNtlDkG3I;vEG}u~vD(jw#LJD2|
zuy{(>2ECDyfR{d@c^)bATk>RcLbHZ#I7-|h@z<_C4a=X7xrGtQor{7hve6^<Y$-eN
z@TnZdB38!IiU!DF7>||M3of{i?d^{Y93-wswXhS-4EVwrmvP<q8GL+BvtY8lv6{*{
zDsQyx`O%#(8q*a8I}`Z7*2L#Z*9@Rb?tiR>ko>U(-n~uyGU*uBsF0$RZ8n-KN~7|V
zGD2FswiQm_+4baf+05I1%8T}pq&zRU|90B^O33|h&niN}ay5(;Cp{5jL6<?d@i?!O
zPI%l@@bO4EV>p?SnVRD-KUa_Ca{uR+jhU)11Bj2T3qf74%%=;hI#_6!&v>B$+A=yu
zDuM`UId6HqEHH$0NSq`s4DSbu0HgqBL4~~wabe2&*hR-mJWpL*j~W@vvHA3uFZ?Eg
zQxWSaG+hECtpa9)mzw;sSPg^y_f;4I_lg7RKYlI1%c=0XAe5(^Xe$=|yR#O!;4)~X
z6cL)zO3b}19?9YoUeSTx2FV9BH|o{by4c)>3-=Ms%`4&<84XD16&dm#5dDtWN7msh
zdxXAS=C(x7+(~t@$cr@YM5z|DbpK{=RY*1egA;a1WHa@jd!|Wby!5c)zC9)fl2K-Y
zvvQU?q-jbPNN+7lxLwLq4zqBG&ZUS1brY$XctV`@SLHOaAOl81Mr}9qJIWMh!fFx>
zWm{cDr&pS8{O6d7UfwdX)kq8kKh@+1w@<jee{Q8+|JaPy7{#;OBvYoa`Yi9f?Wya;
zOK!tI9Jrp6$;w%rvdsv+boq-+;u!lWJ7Qm7e}s#zupw>ib(K)<7ez{)hVEk2F-@}?
z!8#|HKC13mFa(nWJC0H{FVcO%s+~_JemZv7<`n4@>vf!B%+gv$&(5`hrJ=-X%66M<
zRf>AZH_xCyQb8%Y#|GS$ZmRY^w+DAfl%(eNhvVN}1wkGH=l7A^1h4tvlHI~a#`Ysb
zMh@ClLZ$I)RMTlsgrY)ch@?#1_Acqmb*9~*R4P<g9`19;_oKl^qpq*&O@lLM9{ToR
zG<LSaDzer)lLU_vS@nAoUpqFxmxad@3o<^;SEJghM9+c8v51<IwUpYiVC?Tla~675
zZs(M_TsF<T&Y1t(N1%gqQfl}+uyHQwN>X0y)#FIc18B99A-nJy7;8kVS6GhXsAMP5
zmO&tfJp+D{?Im|6SC0sM_~qPEXik>&?b$t7_`eqm%>NmgJ4wDvamdH7cJ(?UaJXqE
zP4C=}nog|nDl@h(dc4jA_kQ7A^n_dQ@e=GqS8fNn#ga-=UB0D62IqY9?=CSvziB&5
zodYNA>q-((?=w+chMSPmmFP+dZR9~r2?eG*OBi0-Sjk?oJz^NG4>yGdJq%egQmAW)
zN+^Ocui^>e^?QYfUgiSHcOqira}OADOD<Xk&Y9)39G~*c<Ga_9$lhR>c|(z-qT}_H
z)j$lFGZ?DN`H8CCEg*P!9Qv=Z{~C?Jq~mt5B#eYYoXl9oLi4L~R2*FvWrBmdMhic(
z5@#=2n5$+C3Duy#p6H{Dund1IEPu$`TY+$(4{S{3QJAuuoX1HNR>UXy=#F8Ult++=
zXzu^~(P#y^z<NxQAO!jb*JMNgL|aRnLeCqL>*9Mk6GF>Go8Gw_n7UxHLunZ^5xI_B
zZ)gpz2q#1~{ppr{j~KVvGF@{{=ioFrTOrLJ$s({*zc^h0Smj9nOI}uT=PplnAKA6j
zqw0DRF)w+Bw8ZGvZ<2boJ*t?uXOW4=Snj6HrUsJFlv8~A>6tm{pzX`YTJDtTh1k{M
zxuBv8waf(dtZL{(fJbiUlN+qknG5ND=r#K??HY>{E3x`}yTWU&+Z|J%hw?4k$Jtl4
zE7>Dgxbk^P@*kzC=hVMsPh>?*S5n{y*l|_5u|8x(xH(c>V>wO0TA;PN&*W{M%^%Qi
z@L$^Qt_8-x`KlT{io%)#>MHzdI`1YgKCDjc$$uuq{{2s}HcuUdVw9Aq+2stWHV#kw
zghr9iRfp;~Kr1$6+NXDfEi&h)ON55I1wk3Lb@|irTu|RPKu)HQj?v!TDDfFObD+BE
z!-}f^=!{}ZH|iEjvV4aD+EVs_J-0gc^`EhZ+M)5R@mjO{Z;O()Hk;-S43;!-w2w{F
zTXYupPLW8?oz4)4kiW$W{fj?{@iBDLkyf{!G^On90Ymx}^m^0dRvP~6qjI+ux3`mC
zJ+5VBXbc3_a%V4>!!2~i%l#Q8Bfpb4B@SloU;8U&4_3U|2$@MX8#GgVWNN(`nBmQ3
zkdoOZtI*osn!;<HiLcC#sh4!lD2k|UJX@l~BP)pWA0U;1LB$S^X^3P5@#Ugp?q>Y2
z6s6mP_lxd7df?9xN#u-IDiKtwjN-jFT6KEo8yTf0@%uHVtz}gt%c(e$_>SSOsFLWQ
zulo0yhi0)b1(eP9M<?Ar+VvTrt++j;Zkf8<Fub{KYHDm5GBdk&f|V%mHu=8aP~?wX
zsJN9e^raHc2TJ>$;b}fvp0p1vjzzN#ml`2C6pglR1-sy7b@=^w&%UU_e(r8tK;NQ?
z={t+<FBasKe>p4zq?xMwZw5i?$8fa#8^9z+poH&}+o>Y^g6wq$8+}xv4>?lI+Vcp-
z*l9L)>18g?at`j1O?PQ|e`Y5p8rGlv%MdFoOAHu*L$f+KUvfLc&R<nF6&KrC>1?U?
zK?2pCvF97IT})<uDK<_~H!E{bk?DVX?xVs$07$cOI!)SYAl7*Q6PDsdTaGyMt6!&^
zr~5@=54xQI_n)ef(RHXgoY`ZO=}DvK4Wp6|Cwlx^Uh6?s>YYx@DX74~#@Av&??cJY
z_uXkdBL@p<P2A8P()~hQ<DZ%Hvm3rSk|<hEe?v1$J$?44aKIP7rm2<C47mQIsTU}Z
z7z+BVgLVHI4J(?i9WDSyE?kzv#lr$G`$<T3;%=7Y>$iVP6X9T&FX<+I$E+}_VE;p&
zGpHN~ByOmB1fJf0u%KE0oNQFge37T=(erlw<G}(~Kcy&AP79pLDjRX+h`{+qerEC$
z-=J#3Z{d%XO|+=+Fg<vMggfT84Xqc@Ubn+t`D;icieLHQp>q}jd0?{Qk<KzvDWEEO
z-mV_~I)8@>yeD)f*O8%_@;ZERNkE!Z%sE2xAibr)wvH~7^S?fp+J_uw&v)O4u(Fq@
z&SaSqoe#>6k=~NN#@GKaYL+3K7iST0`APobv1Bxx=9n&|k}dwM8x16xHC&^9gDRDI
zBfa8f{-F8To;0lPwTU57E3Jvkc?rGK6kJ|Zw)4dwdDt_6sZ`9d&g+RGIg@^@x7oJV
zy~M{RC<Tr?5^z>pIws$*7LxwhW1=@L>EniJ75^^E;W&)y@8m}$-(dH{M5p)ZG`lUG
zlGnUho#Pp}PFmPatr@tjCY*q{hI#D2`fgfow`bcXL`7HA)~vpg(l;0!I>N+vB$=Uv
z>-v$AR~Og3)XRy+7Zv&zhZ|~>V;!xnG2?IncGH8`>OtFmMTmTSdAfx1@sVB@hs0vK
z1fFwC(x}Y&=-IjY##Bvk+jf4ka#Ek$5Z%`k8*Zb0Bd4ea`b7BFBLL7t%v<T+-kxHu
zT)<&8st-^^j+oNT>Ysm*gEPcaQC(ro(Ty0G7|5+N4?z>NPPRu4-F>jZbPW+4FNNZ(
z{I!UcjgU~}593wm<cJc>nJ7TW;x#Tw+Hf<UGyM4t1fg*u{%h>6Ju|?z?5xivzi#OI
zUx(sV;#T<9<Whq+V!NB=JQDk8!Fq3Nzt%sr@-x5+n+VSk)BWG?w~};pe)KGf3fV4)
z9-lx9BP{+HIPY6m;(Ce97!SyPeSy`oGmRfFS<dR^k?RVlZJlw@c3ZKy!m6-b788w7
zF3WdL5A5(CFE0$4Q*X9<X)O!ZdwFMu$5BY`4CCArI3Z7}<M7R#)@blY=#S(`+^)nm
zFOSkK2}lhb%AWZ_;Y~ru!u~zER2gC<%V%U>Tx}<O>(~<PwFkF{r{|jY$ZOwwy%Tx-
zpLFuaVFJ7wNrYC{4<5#C?yaS<j<-1+&H4h;io)0cWEnL4Aa~ZiZ)Lvr#!s>?eP+=L
zEVMs=Io794eSO=J*ggB=(?!_#X2fJe&LO>T7e@1d%^z|c!~juDFI2K^>|2lLUn)Z-
zPd?TSl(!DM<70#oix*NuHjKSZJ|NfD<D|_=EO8<vEr;+@a;jFC%xmcdriq}0fq@`c
zRC$17rVbx+v6HIvDx5?LvRF0$84X^-oPM&?OG%KYcr%kl{)s2^b<rn(`N0NNY}9{s
z;9H5<fSzk(1MnFF_xKFp`0mlJ(WucGJQ8lHdlr)M*+v@!=rPr{++*2_OCfR(FlT$i
zB$0`Dw^S|3b+)*N+gs;`olxtA2R?9k9FddeT`#Y3MeDKTqD5c;k*h=ga+}wyE9|(^
zj)rJNLSXf>kpw`krC7vx##0piYdmEcN$6^FN5C{>lR;fa)nYQ09YsP&LC@rwJ0a`|
zSK7v<ZO0?$(<gqA+oK5S7R-TqVGEBHD{jXs!@5|3?-w;>TA^p3mQki(99t|-EkCUi
zx@1SfT;Ug2lQ6Hez`upWH2>PQ9VK_wpuw3bX|K5)cc}r|RY&3z7WO$PkCk{=ee81S
zbsmLoNNEBd`{Rqx*)c7izfPtF^*-LU(3b395E<hwC;z@uHdAT`2gTA~C%$n~Lg(YE
zc~Z8~&aPZvwNCcg@9GI?!a(#I54lQj<ZwH~N-;}PS9IhZzT!R(SqQ^R9R&cD1pZtg
z@f@r4yPeN+p<}Mh3KTGdywfpHOb_Z^y>#U#Q>tbhMY<STF_4#<zi!m7p8yrLdpou#
z3G+-@<y*(sD50MK{5!*TXAb#rN4ghO>Z#zq-M?&MP7?yMO?e966TY|qj7U=D*1P4r
z_45Bi&{6}#Q>V6TPr=SrQtU9U&saC?>gz+#{$9wlg<HS<7A_%iq1vbbU+L}NwTzwm
zG1`mb!ttToR4A#T<#OvBF>TY@`qSeW;WSx85*CNsGwstfDwt0Slnz>szhthOQj6(`
zFrtrNX005?4y={$sL4sF>%2zjrJdMjUkj{^7FRC5OBkkFPFAuw_*J}Ap!p$Kmb(J7
zq7nB&eILS8vd@Kqxo6gI@#yi#M7J*G@-Cu{^^*D4h<ADVycy~t?I!6Q%TqplYn-!+
zMas`k9fYBe-P_`zK}*4S6};5!+vXfcLhVdC&Mo7IzJ{;Qh@i^Go}#2(A0L4)2f7oI
zl|&cQar+cxA+YDE!oYD}{z9^E4LWGQ9qWMZ`{mw}%1Ox^aVSUCRC9j;ePenD1z<6%
zgIVYIAsp=GhJ$zmfV;p*O6`xY*~<$|cF*2=>0WSZt9!~d9vV^W2z^BCK34n4zrP%H
zKykDv;6S*zV$W<#sf8!Qul=B}T0)tTZqnoYp_3OKVb94x5hPl#WF$|)JeFCm=O=lX
zh#nE>wu$)uh*-9F4u0~=Dig|+Ha0ZxCZ)WT_e-R@Q7D>q1;r+pM4icIsyd#~*6rOt
zBq^;j-2!l$8I|(6YpdMZs&1N3En&Byu7ADT*PN%qfbT1A>)alPYmR@Dw4%pyBf<3w
zg17Kr9Pq{S*T9NDyGsRzQUfu7l#dRK>b{@r@5UX8gDW%SoTA(n%u8eGXrb0L7O#dR
z(z=!2BmFo2uqP~Bq@~h6*_AYUSh4KyIx!vB&R?24oz{sxCM1_I5T*ctHJO!-HA{jp
z$r@<)QYsp6Vz8$F6a;0#g6(oGG$7^S3<EBt)JBm9?aE^SQB9!>N<`HeRob3C6u#09
zz`%UU@%%BB5#C0^q}j$u47Pye?9gmC`sk&$Z1a222;xZ{``mpFt(XO;Zr&Dr^8cU$
z(m%OR1&>dX>rCi??2^(guSzBxAJp*!XoJI7y<gZCGLy1Hik_$<j_>FkCjay!eN<@#
z)L~ouo0i6D%!u{^^3Kj}HU+t4*8ct?(R#MH$+I_dc{L-eN5wL=hgG-K$A9npe;BJ7
z`kG1)AnLcouum}=9GqqHS?!gvWxPaYGT6L~%na@`lYzYiODM@4%(ag+jEuiZ!yyTE
z%M<<<ALqnO&MMEOh+PtbnI#Se9Pvyl0jBShxeqMKbY8NTi)5@MHr#F?&ZB>?HE0-^
z`jD)E^g5y5?rc<^tv~Fl3Y?LjnsJGaTkNsZO^MB;&AB9uOkLC2#^R=GN)^8=tyc=1
z9a)^@;KKk#i<sM!i9@J7M8t9HLJE6u*Ldk3Wzs3H#E4UpF`h_>!xo4s;FQqK&eDo>
zlU*U6x($DnFWKb%nkY)!o9d}=c61|erw)hxLzqki#E%<s_G#^QON5XZO3?tJiPqv#
zj*b!rAk0)@V{RirAXobc_uAb`a0-pc_KyaQ17w%)2d~8h(Be;rbmL$@sMEfu)yN+v
zQypSPNXu7c8?InCwiO_$he<2*aMz3W#LTkY+?AjhACBFTkhjrEKEusQ^<se{2Q5O#
zANo^eDn+n2Pa-0Meujq;p?n(dBym^QFZrpqXhQuh#w7u$<jW0ZNxK#!afH)XkD<^i
zxq}>qTjGlLKNi7+zgQoEy3H@IXZ@Hn4`#-5Kn^U9kiyDMeIv&fA29hv{!Hgr5?`R}
zcX|Zp5PG}Vx0_~R@xRPr%NjHvKPKu<>=Lp+{G)J!<tCHQ4_@v~2sf<QeTt9b$#BZb
z8z9Lq?xoPUak*eZ`^EP#bx>cQQ=D-HP*O{18I6A=Z~Q)$Q|V<%Bj;A1uAW%Icx6nj
z(=>s=F+HwwCfC-{Y2HJ%De&xihEy{V{L^fvjhkyDQ0wwo-)65^xq`UH$AM39`JxR)
z*1d>ISx8eZj@kS0*DVf6{rRghF7LL@-K*8%cKW6OWx{A#xPl^^Oj!SVW4>DD-=*x*
zMFNCRJAC@U^F2F?-b6rQC6l(7c0+!UshaXq2A2B;2Y~J?&fQd8RCSF&P@uBiR6&xE
zi*`fguQ{&^VWXQW08fF6)3IK!G;xMCv?lm2z;5`GQM_(TE)QYxJ7`G1Czh4Ig=@{Z
zK)QBMfgc&M7|VT_xS;OA5wU6U*uc=Br-AFJcO?U>f^+LMbeZ{6fR6xH^;M{!V1++m
zk>HH&?2=O57jQ3W<hb=%295uB?8SkTbFu+1#3v-|bG_gyA?7GoS;%=o(S^=>u$e{_
zr&(;aWO|ou)RFR6EmuBt*0svDM+_Ej?0S<Z5ObdaA2MH%ooA+OZmMPl9b+qGp``Vv
ztjm%S5;n7IaPFc%Mi(A{@TSfl!Na3J5OiJew;SHwnP%=)I3B^12i({m*Y~RC8~4x|
zF+S($(>>?+cY|pUIP9EMV+$3zV@;@e!U8JEfpXSVYgX?L$)^l|Fyk_zAg4GDn@<&4
zmLr!lakD@D54~i~{f}%0frIljX07ej#+iywJ+saF^pCBci#rWH`U>ZW&#FUfF<kYj
zdJDtUI(+SUJ?CL@6wW<FU~Q4>Ya34!k_=6(TPBih8sAuWj6^4T)Aa&#?N7&>cjr(|
z9<@)>3w2NR(2M3P8^ov7m#3FY6+^qv)pe})<~z5+QInDVL-7&f@xRD*=Xx6yw8v94
zPYlQ)8Y3w>u#;>@;KHC|p12ycagqxpU6X@jPw=Y{5Xx7puM!Ui?{dDlP8TV>ZC)Jz
zb2iKaa)a?dTAffpfJlyq^yPnE8Zy7TC3lWX=#*gOJ_3}@Ggej((_!_A53>`IU)K?%
zCwVL@TLte$0p{kqqh4(Galf9#uefeH#8hC$y&LEzE?4lD12VSH!09aWoD48x)PEwd
zWuILbm=|5q;Wv6HE(C-yG<@<Nh3lUZ<Vwy$G1?mJko%l`4aqAyPDde96OQG^Eg<?!
zGjNo>6biCHs1=z-vZ#kLBZhxRI1_F)wA4=PuI$W|2BufUxtw7ECX^gVn61=j-I)oR
zXra-jp?wiML!#(^-6TwO_(m{zY925k>8u1yfToZI*5Z$Se4#emQ33bIIs!zYUTA?r
zKc-$EOt;ey*+LzJHNUsIWOx8R&E)KEo-U?|#Gd|8Lc5sW@(tf>FW-tRwsfs6XPc$o
zbYbV3y`O~reYdoS$HeaWad9Pp^c`0+1@`8jU=~!tU6yTKvQIenHuSq>)O*ls4!R#*
zj2~x*xkGQcD_z-bPFcO;g<?=3U>bhdRFThf$8^sh))*ZThI2Bw!k)>eGBRkAPXn)0
zFQ)bZx@nL5G7f(QG181+K0p_h3Fvxo3EamOfwP~c9M|~N{-FyFpML_Uc|efq1@g`Z
z!O>)Xu&Y3XzsIwwPyAk$1W(yPFaA`Zi6XACYzXr{v^OInZT#~2_umkVm7_de^qBR4
z^6&*<`85kE=FVVSiH=6`-5yZ%9B~BQ8?FByH^Kti_+>t;d8zNRm3yC2@4^A`mC!4R
zap_6~dO}PkL4I9%{9I^#_t6k0kE;vmf8F=+m<ZEi{D7hbII${O&%Jhoh0LjpCS#!k
z&MXOq`rlts^XBKXMgMCeV(=&b<S(sfJWolGELPCamRNz1-`N^b^5a8b5Qc@$et&Ty
z3rqJ&2$&*~=_hZL!VRAT9dEjgoK8L8eXCSk4@foqM~!3@?grSz*#XT6iEoLqUyHTL
z<8^AmME~`})jX{V3)0@%s}@RKb);^Q3=1O?wX^Hf&sP+mQARzjG&cQdu%to>8y-v^
z&$YhvNT+Tg`XPB+TtEM(>M(szio^Qx>H}5=N8yXMm?-bc4A}3}Utgp%8rz02!J{xF
zb&Pl4cVOhhz`%IP&OTuAeh2z44R3AhBBkT9@rL#mRe!tlvCwtQ{s#~_KL2kE?`!*o
zc{)1`EEL#T8NST}tqnB7XiW`u39|@hUj83l^&Z_z4v4nSHvs4d`$RBNIK{nmnd~T8
z%SAlZNo_4EDwBi{e~Y6qwatsy=Gh?A!Rn^DmC*41(b3dMTj<#O!RvY0YaP;K5NQF*
zARwIC69wWh?QFMG4<ukbys=y?RT=^C#vfMG+SWAjvb<n9;CT|O?E&V{+l(wrhEfu)
zd|x2$hs8LeFe3IS<Bx3uxa5&$x2Cd++HcM3EVcg)2bysw()RHT%7<R|<X$M5pA(_3
zu^r`gnY~+6iP&<fv3wLbUG#xpfYEy+ET@ejy&$hqv<PmK3VM1qj@#%x9QNH7wk_oK
z9YtGxXRcqAh*-6|t|s-YWS#TJ>P~Oi@o{qbflMm>*QA=mt1Y$S>6!ODw*t+R_NOD<
z=v45>R#+4HI`^J7Lw;3N5az8&_b{RfQBIzC=plt6B(-=-4(u)vXPDnY%~D}-dqAS!
zM(}q#$C<9kY{>&(JmvHgm#d2&k#3TP$W)n;UG4<MIWN=h0jcbSpWu874olx*8)aYY
z7(tlm|FQ&Hc3?4CVCKM<&#JN!CdvZ7Jd)1Iz5|4@&vw|3YiEL&t+=vdJ$h%`mx-a6
z|5h|Hr~2P15CI<cRYoWh_f<#GqDiy*i_fXHmP`AKrkTAKt9XOiGbI?max1KI2<<<t
z<Yx^^93++`9xVtWMkkL#fJ{S0;sp5`Y{xe=)?owf-t{^H`MA(fkpQFzAgOZXP>>h+
z1GwlN@$9!AKp6!y>rO*+lu(}CMCeVmi>%80BeA7#0H0jFWq@g;`-czh`;9sXG6x87
z3(VQLz6{@lRl>uCa^|HwGM}*cCDq4byk6N2%AWWrTWMr9<*|1BR}DBUH~MO3==QH)
zY%QM9-T^6QfwTRJTSNZcytu$rMrQTN>uc9lfrI<wj^s)$5x4MSQs{pbBn~)Aey5&9
z_+woF>biz2H!E;Uj`DKImD9S5-G3u7Z|dKi1$M%V&#>Bk&)l=_y7Lop;AY!d^!v36
zB<9=JcqT-akn!eFkN6RULWCSa32CJ4btfq7m=?@idJS9oXx|?@?1n55h)ZhBGL(6Z
zCjR(S{K{g$TWb5_NsjYPGC(os&r^%riCP*Qkj>*zH_=f}DbSNbUT?qSC&|5!aCa$p
zy$5X5XiYxlZ*jta*{62f&@J~1zQyNXKV{nw60+Sl*O)5bfo{Y-q!N<d#A&})j!iCr
z9ffF~3x<^}x&Wbop;*i=>GYKq;vz}m25{X>t>t5_S)fcNdWja;AN?NfjGYX20oL^+
zJ2yO_q#&*G%|F+3Ip-^Br5*A>zkSAt|J%NAg#pCM%em>G`YwuG(11e$@suxwvmvx5
zcvzRSi62n#-G>*hdFCB-hs{&9wMioe+ldL2^4wN#y6=F^OR&iWo`)mB(CLD7u96Bm
zK<_8^sxSG5y_#AH_WkQB;#PyIJIK}tN(AuqPrp7UKbu}*xXn^*;rx#`DKSI@swJTH
zg4vcz1u;APUB#YH%?k&dt<r}Bn>q2MbX_hyQn&Ae`81-Dapew@?T$H28qP`|<?XC(
zeHT8re8>kS=<{m?%XJcQ*fBIw#)-Vy!DWu;eIqUMif4>?bi8_24dGX&>o=H1a2q{&
zsG73u8AkZuYIFl&Nihw}sSd^nfI(GgpR!Bm&XMCZqbxhV#gf_BGAkB;2rdCn0n@LZ
z9?}iGX%J$i{~G?+tnZc^@ksQY9k0064fN2EmLt^&LR9g6D*jkBBff@zj|>jaq039e
zDsXUQugLowmJlP+*~OSni%8*MYwd1J;63!{BQndWeswE*C_=*j3i~04(v$=f@*n?t
zbKAV&uVF(C@_ExsAwUr@-fuqzcv;L|L?!0%km!Kho6?II*h$_Nx#qmOtbWdL!)?lr
zdpr(A33(@!BV9~ZA10jc0p(i?Wu|OMCF8ciW&JM<<xi6B6<!8j5ZGYUj6BdC8NWR)
z(`#Wxf1t5_5^TD5-k#$nw}9e1d<l|37!l`_(bPwc>Q{}UoxU}lV!%Ja1@h;ddUw%M
zIZMo|eRtjivTS|NW8sq_2M-SSXSnJ;$nVV8{FS}CqbDj-7@422KaU6ak^5qW0T{o5
z6yAva`QYuB?D;2wrcb{@;OXojsejqB!8btVaD!>tF;V-2T-n;oEzH_IG2KxqiAD~v
z+dEc6n|Vf<dk&`_tyFhUZNx!5(q_jo_9*U4WmkQf(OZ0_v>v<~C+4}#;r+I_wCJ5Z
zM6j#5uO5e5#*=Hf$8NXL;)=<FN<Io)va4mP+$P%_S_cid@}mchX?b@GupcKY$U3Vr
zoo5&Ss>((DQy1Q<J0stgj>u$NdW_H~>6Q%7AgQ2Ajk%N53(GIVawzZqR77p_swv_>
z?sAO<-=x8J{?hQsY#bJ&UIo&HN(L)qU#7!{``r88K>s^EjEmVAD*@C=Ed$CSM2Qi)
zG}v<}HWRSNZZbie?X!HH4H(z(AjZrz%YN_VL!ehZBL4Utq$p{)5IX|jZqJk0&_B`R
zYgXZ3E;G5HBhOv}!UE^Md-m+}9ppJcC*4s{2z!k81$09^{mX;A1!o(r8)jQruy6e5
zpb40Xkrs(il@+4bW={>cF+v_^Y~Kx@F69+6EG?Prua&R@k$>Q7?zDtNkW}TDduxow
zC_4yER&cC!b*&R>9G(6+e+$2@uHLABphXOJG~K8>%2O2<L)1EqGZBtY;n))mI@Aq+
z*-<718OIajF~z~Ze9%ExN1Ndxbi9q+yWUb7JyxNXB~ec6`GI{81pvI$%?W1)d~MaT
ze7!gJBKQF9_IfnL5d^AQlpMI(G0{Fgy2;Z3Fsh;$al1E-Z~yNg#g6~+&sR%A^kra=
zJ^yZbnMHLaB>#l-0142-dUOVi>^?&7TC>ONrY8?cp(;vv;eCg?DVc;bxouv=+*FQ~
zOhBgfL?JiVYOZ=h*I9iCa7gK}t!V{$W%J1*#QWAp%l=vb0rSTD$cw1HHUYjv&{Oue
z_&#6&s|o-2IH_#^n-Hn3#br*5e(<>$(WgX+qkJ&wY(O>x3pV_~LE9O6rG;nTZ%F?~
zs+u6BaUv6Sn|boHcQHa>AN|w@Ia%@8^dL{s$Wr1mB-CN9r{^ZVR)F|rW8n?z20SPt
z!@${6CvG(Xk=YS`7Y+r1;-hDG%-Sl#(7GMnDmbTKX>Vx|#XWGgviFAz94F+ej!69?
zU?^3o{16*ne)wZN5n*`}T3VA=78Bt(5O;%mPETY`C8q76;X{T95|G1u@T(87Wnj%?
zwHswk?9b#$>Vs4cw@WE4Nb;ng{As7Z%^rxTu>DH}J*A+>?xv7UC|}MO%A`SrM@9J*
zYu8q96}!{5w1}r?HXE6221;0%LvO8s$A*+n-AeQW@wR-WUN`jVi;Y?^+U3lZUdUTd
zqpkdxcl^I0HvidVefCbpDtZ2@BfyE$MkkU+r4pDwUcqw0<<%d2s(L&Xep{6EasL&i
z!MF7&h24P_8{18#hd!sL^nutB(mq=7-_v8>Pa&N8#w@y@B>V~>QDz!)Ed^xXulhVP
z(uNPrVQB7j&Vb$Z)vesS7(bMN4->r#1zAHL%Ga|gLig?qih$qpL3JJFt(YH5Cf=;V
zO4BV*iH_AK@lYt01LA9D$kyWG99&F?Pr+P?jD(76gM7+tu!d^WJtSFkor1%fQXQg#
zv#N1o8)qUwUPY03hoW5yhSo6MR$J6Q5Cr#TMB*{?&8M#OQ~=&G+R85sU^zI<H_$T2
z3yRr$;NJ{8-iq1fHPH!c(bZlDYVc7*x<7ATUo7HZe>ecNrm54k8jsQw<?WT7oFYC5
z)V>LZ&93+ioNPkX&MtpTdv1cx1c{Q0_}JTL2{UyG;v$wV+3Nv?xH^K)@jwxp8%pL;
zkiGf=X6mYwLR1gAY0Jm}2*Z{PmIB|>T1T)6U>`%PmF5c5nfJY;QXd!$d=|as!C9%d
zLd~gzLKnYgDUIM6Hr^L-4rZ!{q(GM0#eUDU1X}GaCt;{%x1kie8Q_K*NI*6}?8b9m
zdWZxhZFMy>H$X>Fk?f*y=D1dwYWMM%O~2#YJpy1YUI%MtD}_zm&#%aS&hd^7`M2Zi
z3>w7EctRZ8Gz@xR+_=A40Rtefn&m5f+ckehUckg|4hF&F)p{RSWlBh-Z|!Y12arlW
zgHAZBQ7!=L8lPbC8z323`+OB1AMTGm=hEbbFg{;tyI#jlEhT1x43~|d2R0CXk!jE+
zI96XF19J`JY$O#OK`as%uP6#g7q<In_XXqGekkH)vh5DJF6njU&|_#r*I@+u3Zi{$
z4IK~A_(b`%AzKb)DuVRKjV-d=KV)0Ki`}RK;1J7R8e>Hce{>T-j6VUP;`|vX?9(MF
z^3DON2xtWh<5_$roYwR$;>H=1)U?JpKs^%EST&n=&!${Rj02g@T6H6}{jzP=c+5;}
z`^)BSpmh@HvT}WqK0+%aQEDcRqhu^@#(2I3seUc=d0rTG^d0!%F{NJG8e;L|fTetL
zL(#JBo?DU5Q-YrI$rrhv*$3`n6cp^IEB35pFL2>om{JDIz5rs)v*#DoxS5=`0<-GF
zFJhX6uajI3&Z7!<UwUxZ<Vi6H<}DIxFE20uJI61-N-N?oO*b`L1*Z1adAdW>ZtF~~
zFjDoYrB$o_?7w-*%b<}?PU-~1V61FmGp-j=1o-%GRaA=VlLG?olfVr0_aEbwU0ir1
zFV;S}TDxewtl`S03{0J!Ovao_()(O0)5I6T!rZygn3Uvj7CM$p|H`b-N5PAsR-{Zt
z#}1cLo><zp!Mbyke|5K0dJ>8~sC4+0`q0Y_MN*w#M%bFy)Au@5-Q@bKbVk*RPEJ%!
zPb`hY>l|OH>6gvKgU$ifky!hAzq@p5NXy<$7o}adF_8lG&2>A&O9w;nONNoIo8bAn
zZd@9WWjC6)$<CL=K^~KsZHpk$-J;Jk_oTaDY2MXN%o76>llanr9#WG(WQ<=JFN`3&
zT9e|}8eWKTVoxU9uKiVFv1NWoeJf3BJ8G5xDh>ldNVrtIHuck()|b_8DQ<9mGW~}1
zImCq)yN)XW3LzNx*n398$PzeUF#!f-npgNBKvw8jG*>BCTwq_wk>xwdZ?+RyzqGvk
zE`2DKqhHPWc`Ag_wd;D6H{xJ#|EH}DxuL!NmxQP9#le8W{Ra<>n|!fu0B~yHU@*;X
z8+LiJxNNd}C4&Ms37i+U593$WoX*Ivw3M~5<HUgzexo#jBd-6rr(#8ikx-tF`yE%A
zb=`v`0>@(jYNW*hVZO;QI*@W5+35ZbfjEw>DbUD9Rr)76h8jHdiI(keelHaDg(Y?u
z+`qyt1cZU->reRhDTR_cGiIqqN6cDwobPN`?x|)d4I~fT_3h#g0qQj0Ftf=_WYU&$
zsk15YR`OF6K>T*G;dp_b`49Z9hno&Pa((@s*V`i!s_qVi$>%bntYt504xol|5M`;n
z2}<tj(|(U3lJK_fGM9DWTK3@9o;gJYImK<MbpXw}b>pg$z-AGjv{L+ab_vdHb{wub
zz-sZae;0q#Z_WSlo7BnWnnB8hJR<b`;R6>F8j`fVeED5-jxX4uFOc31wD-_6(^o8v
z#u*TfRsCfQhPo_Q&;LH|v}%Vc!a)e(tfuCEsYD0MOeZxH%~i-P*tNS_3=@*IL6U>F
z3*ao1xa6~P$JtuWxP3usuRj#CvISsyiZtq-`H~p8;{pCr+0p<$xSN7jsC0(Osk(Ao
zw`@+MbO+1PG6536?}es)wm@%hCy07P+zBONqP>ld(hQ?pG>I6}z<rFZ7JRZOg@nyS
zuF$*&TM8Ul3%sYn5tAbsWi<DpYe$>sS3+vRy0<DmA%4I0M&_Uu-_OYuAmvpF9O8i-
z*+U2B_yG%Sb%T^!RCfFi7%DQOqE_nE+1~lojaOH@@~g7w<j55jED=P{OO82B4k7F;
zqP#w(4)htW@&WX$;zE&mMs7B_j|zbCSdIM^Ib!qDNV6^FUBX6kj;GJc@78;Xj65#p
zL&wF86xT@FhnMb1B^0N~kB^)f$hL>mxj@SJ2_IF2RZd;r#va3l(he?2$s_uIlWKG6
z8?tN~b)Z{SQ0$OGKyhII3Do9UIAy_Qc%E^l*!-9wJwrrnn$Sl2%j53)hKi(EY-`Eu
zH6swdF}N-_;y&9iHqML`ytiF;lH)hVdh5ZrcD8<zgj`T9Mq&#Q2c=Df0T|d#-Dx}k
zKkCi7z=nT-_eO2Fj<Z?~3}mbp`%m9_N*V&)T6j8Yp6W$@c@}9D+(X;x>Hwd>W|GUf
zD;8PhDQ8rdqrY6}e7Pt$G&cNflN5EBHv&ix&^bB_J=b0B+b`IT=LM=vt=j7O|K=U@
zAJ~3MvbH}od;*jnJNv_ap%Ij4j8qXoD89LWm37YUwz5ll!KogaH#<7*1o7`(xJRfb
zQ-o~kqQBl39@trX^}VSh2}C(1E};0=z2jU!*!oiXpQ7C%`-Ery6-cGiGyDNUb!wL=
zWxxF&*&~Q65~LaL^L-3x_0(2d7nmA89s4}$%2WKl6(f&KuXQMm(bjAf3j}3>)tkoH
zNyp9;A01RJ6<Q5F_>^NqPH|=O=%Z76JzMoJpAGK&ulGL>yp7mLYJnVftb^PKu0RrF
zZn-UIa8rf7Eb$iHz5MzgxKm^<k4pjUfaC|F$>t?PDXuGz=6k(*e2=eS&1xq2zkV;g
z@*~t7b)BD`@@@7XXr8S3;^({o>GX88&G(O`LGRK~XD}?gc5E1=)sSogsX)I1A)m{K
zk>j(C+1{&;Ex>3HhWA~+eF}UPz=P*aM#Vvx#Lo68tC@ju`RFmJ9+PA;NAb@d8B%|&
zDTSTk%4YrDmA`#U8HRxCnf038flJgJk4QU!*tS1+?e}#*mceK=me=9P@ZGIZ$RE_%
zGnXbs;CMaT+DlPdh$nUJJ&i#wAsV|M@zca}Z2RncP?m~0xWZ-rmg43!?vFJcHW6mr
zChz^Vg#VM~=%YaX0LfAymh=Z=v=dDRodw**dm|P6dwMM?#}v<y3aP+>%aQ_#FLaq&
zGM<)|GgZZ#i-(m263_ImilmX#5*=hR6Qi5{Ra++GkA*Isn4hx29~+Jf`7~Gfn_KKK
z1d52ra)W@~^{90tT#9ls_TVWCV*oIA{z$=6pU`0`TM*{P|6Ld_aL<NITMLCBtnO{j
zxC_OMY4mQxSr0t-pQiRqJ=xWIp&79DAsAL*U)ECWbV&l08XjK{5D<hDC;9p$(9uR>
zqb*fkl4P_{1DX4_g$(^m(9lOUHiij^%VAmOAw_F&e***))?q5m;Zn>dO=9HBX=mMo
z@-0TM>U(&ER}QkN?tMx>>m_=N#BmT6sJ`^TvO9MIeX!0{UE`Z1z~*Z5?`?(*%%L@n
zEtFdDypttG>RBUq8DJzm!&p0(Q^{2_&&z9q$&&4_>%YatTG=w+p4r&fx9cs&6d4}T
z-7;`R-75Yq;(H(<Q{7umX2VtJKbIp3+B1eolvSb?^djeA6&D*}WZ-h<k17*0UHef@
zFPd(n_sUhzjm;d^-<AeLsYTuk`%kIy!Z1@_qaSJQuzAx)CSQ$&&1mzq$AV14<%YG;
z%(a4|<IT2lobgw#*#UtU93zlCGKxW8p7Kh7Tjp$!nZe}xNaED~x_5V$84Bqiho>t0
z^H9pq4}B<l#NPXXL!)wZeyj?)EHPwAS1(2uE~D0_{@217GBx=B-^`I|ks9D01%Q$O
z5tq>&A&KiPn|OQUqv&C^A9J|NSn%09ut^2geUdCjb{kQPjuTeflS3lY{rjV8YhxMf
zwRtCM?Mej`$Jw*jG3ZG=+^)$=If~Ehg;R{<&``O*-L)`!@*;C1ol}cw7P=Sea#4H6
z^{m`DQM+{HV|QLTP?+oW(;6oqFmGIa0k=Neq(Yh4yFtp?<We7zIiZ|2MAO5}w0vZt
z;@Kmgp)|L#EC=pJ6e*_xbUMg4vUwa2^YeTkp>DR<@4#K<Ncej_^jku_<}%SrD}|%0
z$rCRs-r+0(mmvv~+;}UfO*;6g5FaWY=4F!2UpV;N321?k?WXdwF8?!s2cP+%o?e`s
z8h^CWx$_Bl$J1752J3wf@ErK|io$QwPX|7_cQ<DK1fawbU7f7+vz{uyimirTtWG+=
zxH0a#inTv*&SKSC1XmT{iK~Cj)*Q=Oi7nX(1ZphQN9e4q^G(}4bYQO?1Y^NHhLAkH
ziffB&T_kW}%tQMl&0?oP#gF^gtvhq76*_xEZj6~|E}WEK0n`Mr3n=hX6(4_`<Ve09
zZfOh=b4e3)EcRQV(SBU=cT_)Li}F?1#0_%=fhTm3Gc*)EUbtQm+hfBM&S<O@Ce(pQ
zoDhw}D8Q)(v2W03pE4ail2jM$X{ps80QNspIzsoS{N>Lfw+52eIr1p}Ar-R5JC9=z
zyrj1ILC6-VH(9&5odK?-Gfmn_X~ROtB7m;Kuv6@7qDI3w5cDIAV4ZH@O$Axka=F)h
zbC0{L%WgVJX==IfvB_N<Zt2A_$O_yv6ak<0Tl=b=?$(_m{m-4sq5(yAIy?0HYob74
zb_5;>dZs9ro4zF_jv~?4iI!abY;%8_Gm5F6=?UA$GxoYs5C=F2IYrc8Jk|C(4+EX)
z%>&GaY_Cj=kt}-N?(GZwPn^1FQ9P~9gYh4y{Ol~Zd4Y)z`GR}mYAzI?_NxU^k}c}}
z6tmj}#o;pben2$upnA$r*pF$qV=a1Km^3$>t{_FVp$~2;@)SE^t6xMlPBQi)a1Hmt
z4WcH<1T{aSl$?q0aHX?Qm+W-ydE*^*aQWk}vqwB<+Opx2N_}Sq&_VBOalt^idF)70
zGTyTa38Kjtbz1M=HYm&2Z1$PL{4jVq#s*n$(FidA@E?eKS$HONw8h8&%iLmi$3_d>
zfB$?}1nc3i16qHjW)a2T*r}u2mw9f5$6Km}Tm>i?clzB~o_<66{Q46<_-UX{2wMy{
ztTX-fr$fQe(C|*d^IV5z*=WsT6D#*|?A6gbiV$aap3QL4@*!$FPFk&j+e?=bJC9wO
zo%akee#)oz^Nf>MhemhBd~{1R)(6aFS+Fn45=3)mXlrIRvj$0WR{kiANuJGdaJyAK
z9Iw{J6nlQNg?788iQBrjHv{@QY4IpKE4|a~CUzgdOfDP>9f^2zx~U^1OeI$o6<$Kt
z-FlN3d_~ZTxuENNG?U`WZqL!t(SbS{N~DihCl6a=rE?bes|=V=v(FW?hxm8*LY+e5
zJgyypF*hGsFJiWJKj%2y-5e(%POxcs`a5anb#=Gh$g~UPmfqDvb&M9bah$rOLWS+-
z`UQVXT4!XkNfhFjmzWZrdS8ZzaXFzy<d;s%L23wB9zORPx{ufX(rNE|1@?*TuS1={
zuX{A+&z)UQ$d0UE$;rKE&BV!T{M{B&s7z6-VK9=d94ah!NGuE<VqU$*`tXQ0Eiple
z-+#=~aJG|mb;C;giS#<TCV39Ev(?a!rMnAT&DJ9G>mQgkGsLauN}4L&ED(-|n3URF
z3uXlct~VdvU@eo+njd$~7D`Sz7s;=&(*;qa2VSdagIeo&&9~za{+;^fa;iULLu`wF
zfoX?Qd~Fl)<q1iUyMNV**vqqkQtpbgVzZWGOa1EP>0K6&Dr|5Iz)Sjtc3mZ2cuMO$
z(ZGCE@%ab>nE-Fg<q^35HYN4*+YWPZ+!+x^L`&}Z2mXi_&BxoB)aTr!f_1pziandf
zdcV7x%DW3W%Z^;m?joBso=<guGE36dsU+_Ayd!FQb%%k#9Cq{tvsJh4jwqD{V;%bR
zVQGtgjCwZ5LDdoyFLn}Ew4Kv&Yu^|gm8x{IR>A8NJOSlAs;&VF&=u0CNf<LWUctc$
zb9JPe-n_HQ?kR>TVOF_!p>%c9Q{IlFAl~o<^L*G`b>}T9(UP2|_5F^-m=SwnEzykm
zD$~r=KC=ML6DYlzxy(fwJa3uJ_S54<>pNAxxB~BOG0|@CuiM#4eYv^IP|>NX8oA#=
zD9y3-jabM4S~s3*6vp2A{Geuq56SDyu^W}}JzvN4bij&`9x4Mt8!M*f;nqEA`>?+W
zGas^Ow|dkyulw$H(o%g_ysX&&UQQBhah7J&2j4Eme@|Vo5ixB{WBm(4DUSIuI-_}H
zPRS6+PW-XSu~i;k&vx)(@?+Seqxn?|vFkPcLCsR_C}=jT(2->K9&OrZRQ+1Zp|Ysz
zmi-wCeP!_zj)8mI&VEA^$1!_?*2oEeWMvGuhu<?jOgR*3tln9b*%!_*>`))@Kiz!(
zU})*kT3PQ~>o^X=-f~@iw3>%?X@6{W18%e3`>4$GjE-YT^<_69nnopk42DYy*2$Q?
z;krSXGdX1Sod{0wkf60fJ=2IYZggj@i%jF}nS`fm?Zl7e1j9r3H15iCy7(I1n1eR&
z503rW@;CCCI=0m9-TJ7?Xm|&Gd(X3U1k9y;Z6J&14xet2{^CQnfiVA>>Xhk1MNf&n
zgKC+#*+a&ZGp*YWyhWQ6)lxzF@~URoCdKz5PbXqjGrQ(|mxP&XQmKu3p6D)<szc&7
z(wU6{dRmx72mAKr@7Z<hw3?k)j58FivB#3(B}RfvsM*oMZo$6h(hdsWDo!0D>!*{j
zk|t&4#Bo7chmG|>%BI^>#of#jBUY2eAFRk;@M&K6T5XV(aAI5P2!_U*uTQAygvdg)
zR2ESl0n!<?ZjCaqSa4hYk!M;MzW)BiN*A$SxDNY;2`zN?kQ4hOna(L~y}&>Sii|!r
z3lEy-`}@&dDiKTHy<D*Ppy(Exo)W6dUIzY&6T6--nYK7bb%TCLpA>E0U8im)D>rRM
zsvaasX8M}7mz{f!^#}KH$PG4IRGdQI#2>7jHuTT#)L)hh*g<vDUVps~R5PNxXn#pN
zaC_RKPOMBJwt;arH}JnO0gRf_4zG^<ONHs&K*dR={$J5sO7uZ(udno``^HREi)VX1
z*mibbpi|7gx9!fS84x#C{r<T?9<lEBjv#R0T8fO9CxyOl1VhFlHrF8w6ZL%T5hTT9
zj@>TsZeKo1&p6QH3K?FIYP(aopQ5Lpht44sR=$CEUGp0mZ0!_sLqjj$>`Vx)k?Q%(
z^@_W^&6wI`A}MP~Yy2SASXBoyU0D==y`q@DrQi6wmh=U<Pr|&mqZf#2zhfapM3-M6
z(SwKYgu=@Qna&^QW&SE6<8wzWV;0?iCh2I@O4(C%l`z9y#2a}jqyO%@pYYRuz6~56
zf_$ulXwV!nS~rFQo(bx=PcKLxT2pfE^m2Hmex}&Dk_zLbY&L#b7F$=EV6zHSFn+Q=
zbCqC3I`6N`?>E}w@M)itDxsDMWt$xFi8xJL&c^sD1m>0hGWGqN|K(+#Ey{zR0;&E;
zZvFWhA<tcqM)rOTJKn=uF3zBnk{<0S-xiSS|4wwkp$fJK<PgkD(QuMpOU%3qkK6XD
z>a)vMVer*~&@UyY+piLGMyuZww;>jQRV)S{p50nfaLLp=Am&fZ5~O;Y!zA;XVA16N
z(R7tzQFUD#L_$(Ux<sS}1f)R_rMtVkyW>#Op@eixcXxMp58XL*$9M2~zaRW!t{LX+
zz1O;HFQiq0qzEHpDRaVvX}IZ<hu5-K45Z|{D+<3?c-)3=SWglrj4e)HwfLWRnDvT^
zeJJ#(*+ME&arV2h-tVdCmFQpDNqG1O<rK9sBvl~CVs~j+AwH1k{6%k%I|nDk5GzJ>
z<We5-mzu4fm66pAt<oujGd+4O^abR*s?bj+FQ*?cfZW$qiOaDs;l(_B{YO7U%Te;^
z3)=ez;3Hm>eDHT9x7Bwdge22D)c$Kebb@ai!{k*|L}X+@b1ic$KmUN<0jVflv!qA+
zv-;lWPICQ6HS#pg5}}fk28_Jg)8r=N{QN!W;Vk~U4TH4lol-kf)wQlAI=*-YwvPOD
zbW>m<#)wx_4jQ+`86o2#ggI4nw6~Uk&!atO$Hqm51tWCTO3|}vy#?9S&(gr25trLE
zr0)BUShL0Dq=MIX{qqN_OY@er!PSi34&lK!sGHE}-O`&l35r2O={$w)%BS3XOOFT_
z8%1y1jsm<}ST`E1)|*|9**i`)k7mQI`EDSwnYlO%8fvyrIB^g$wyeoq+1;b1u5QST
zd+r)4xSoYBP?_ex^pq$Lc-W<7Wcc*=Pft|m)z{x9U4vziSmc;BQZ$P{(}YM%#tgMe
zb)F&Bfi);oKC#cGE+Jk#+b*)vmPdVYxLL$(cb^X|>kbO=;loF0mmx(Vf|*_B08>g{
zlEr<Cgkfg7V{^gk?Tl$m>^BO%30??uB5IW#>m>va&6TEK%-Y_uq89l^cFZ`&Z+ULr
zGPq(}tDwq+Wkw@;M4$@4V$4r9OA(vYtZOoXd3)Tr7)T=@-YEQ{Dv|jlsqv{BAC3iU
zYPOC&O}0k85a2B+dPoZVlOyX5h!*`ZTScB3y?X+hBaG>uvb`Hi-mICvf3+l7UQy98
zS9hWCS-C&k;nNX>1}x1qw6tTURaI7(T~m44+1Zy-7Ri!MDnKAvLUDaPK1`zFqi<eG
zp6)<?>Rg4%t5=BCGKv}Ub0!X~FZ~6_?N-oGn>_K`EVxv-a{`oLy5Kng`F92@J8vL6
zv;8R1`QjK*FGJp=>E0gRmtLnnlN2Kkmk^g51U2^~5J$Sz(CcmAE`xm~1sqv3Xvhqn
zKdUF{j61EKRvEdSNj>SPbu-H)&;E9`A=@M57`tuE*W;$eTxy9ZS0*gDXp^{RrDav$
zruvy1En!po!;<#0rGP&LgR5^=G`8#N<i!OsU(xEZ*BkOuyS=3zVStZ*S>qOnap>=u
zwSV1w?exlG`aw}qVOsH|y4vmoeRF?5@)vwGP7BOijS$6>^s6T<EG&6>c^J{=*6Het
zdyCOs`^XKfHDt=UPtVom!(90u<rZ*Fa%v|D2c8A%Q{?Yvz&f9R3<b3aqKF7*@0yDh
z)4BgH<YH%`ww4Xzn644~HHp*FE0k>Cm~G`quqXC+DYjRjWD@~T&t~V)1O~&h_USl=
zj7(CHZvl|GEy`<mH1AM-03knE?pLznZO_v~gfWnza;t$cKUmDAA0Xp;_(+T6+1~#%
zbreJ_kS8dB@d|Z+bmuxjtzp;$rdL>b=C2Fbem0V3gM6cCVcQN9$iCPJ^2rly3Ys%Q
z-b`Kx1z-I7drstda%G&dlF)Y|aZ>1F0eh2E9;0O}zhVK0jxR3YdMah`8=wi%ST4?5
z&ovetiHGcke|yFa863j@P?)ifV5Ya#l1CDNh4K~QJ-g{AB7dGhAVA019=Q0y(D$`F
zFYYIy<{3Rf?8v(Hfnkb^6?OU>js4`7k(HCen+KF|tGhpr4^Y`=T2u0hxy}=!W$H7u
zWbKQ~STb;O2b6SOrB0h(WTAy@@3x!nrBZ@h<bhtIjqOR=_ZnEVa4qncT1{tE6j8vN
zl2T23;)|ExG`f>2M%4stDDv)}Rz~w{0vDS808-En%#z0+CExT2u=MOuEEn*Wy;${g
zj&j}~!VtGD60nT)@$l*~O}<_PKB#K!=PotLz7(u#8yom}+L2{l|JLV#fNgA^`*WfR
zTcKJ=HFs(7+~8KkpR$`wgZ>d2`p;tkMutX=!0K>+vA+d;>E|diGhLBIrz8<Ol8?1U
zomoNurjTX){NGdiW>^u=u$(lsx6Io!-4~V4>`9R}@e}8T?TLj{4Z)*Wj}Ate$CAfK
z=v^2Gv;XPvmD>rLx!U&EdM)OJ)=}Z{2rF_HOKQ@^BB@b|!#JtcU0<Mb9}>`8Jfqy?
zVK)oCof23!N_2aWrjxa?-b}0{v1f1`{yHt8&5C)pAQu;k;JU=&<s1RsBWi|j58dBP
zW8|`M-by8kZLHwoi1%{U-cIQ3z*UksV}RJ6?>>0O<*@G+<YaNZvM$>}{`9i8hM_?7
zYuD1xC&S700-3zRJ76!Tx!e!LR4>yW9xNF8)&X+4JcaY*uRdr{I{y;c)f?gcJA_!=
zl~{XeYP%}z#INeze<EVNlMx=Lun}A|uV>vYRt<sLHk_|w+3o`9ypW5pP7Kb-E#YBS
zM*oHcjLi#UVn1&^cnjrAx&q*dkju!b`@zX6xo@$h0<f{x+kr#b`FsxBFN*B^&_}?y
zW41?FGG-kA*qaDrez#=_V*kS6viEf18oXa$wkAh3J1Ee-U7wnc*b(u~%ll;5L0O~C
zH0hDTu+iIH_p#?wC%F2aSCdb(**(d+n<2cJ)5drgv!WjY2F*+7Rg0?9w<St!hKN#i
zS>u(@fn%d+!<m-gtFCdo@%NMXEO@W>1Jy3e6&4aC1X%$W0xSVM_FAqtlTAk$pMR`h
zG?~2|ZniAVeC;_>X{qOnr^$gpj!-Yrh>+Ub`sdZSXb)>QZJ&l4TI0+sgqh!v2QDra
z?s@EN`1F$e)fvRE+4Yqdg_aD@Bz;Ui5}aWJAd#vH8wgOJ@wy^==~Q5^9uQfcRD<?M
zvm9?{%_5&{Bzn1{$iq~ahrhV?!K6@LF_r#iqHiBhRZqr)KJj92zLURRZl3nuAkHo!
zPv;tL$-B&yI;@uPHW2Rj?_CMWDg2o^7Bpl=RBG-kP9^zPG-bUnR8i`XzfF5=$ZQ5^
z-eE<J38(f!*O$lYO?%P5lcaTwBIbUgH<}0#E1Iv)GJkN#qR4y|BRce$#ZU^Z$ze<k
zMYGH)`K0pz3Hsah1Y-B>jMAiuJ1I@Sz>=?=ZfX=m9<Rf~u)j)qr;fKljvb@%wTETc
zrG2FC_`12-s=D;5{Q9)@hs;`Nc<;gc#~%pkmX&{wht#$pcHzMv+`CGIbW;73F{uA{
z>`2XU01*)#dW$>pu964sf*K_=`}5>eSeTGz7p#gtdFXWmv0H$S9`tVpdlZ%xs`gAo
z1vP_k)5>gQ(#L8KCx(J$A`u4{ST=FbI5lXpo<pe83lN9OH9O-1sl9yVZ`sPO=OhVL
zp5EN%kDqlX3{fyeSZ<84A-(}m9Y_x#_id>Bu2ahs%q~tJOv)&&QFg=w(kt+>jLtii
z%F3KJTQe+*GJ>6E*hGB_?`n<4<{DnNoTB|{nh%&jRMKX`8z!*I&0CqBE>XExHyU62
zv$}@)?e5zZoZ%zj{umHK_%KCQszzXWvn+1!$a<`S&+RCz;yh&UEpC<d=8r+ggUxHT
zP$xfLxfW+!0-=Vp9?chEqofiZ_O_(|T>2&?E6)p&|676XHk;X)k|LX1=s&O+>vR5Z
zNRn!AaaaiK&X?2Qf+MSV^O(zyzP(?YW7Y$%Mw`um7S0~iFRH|zk81ORmn$I2u-0E!
z4>wN=5wd0jg(02&oo@;%#6Nctsr!XtTF2(Xo*(sZ%p(hX@FC*xIvuHiyioJ{_jMQ7
z+{GE$fyy<5d;8t`0hT|gFoCU=gVIq}Ap!0g>#{X_U_5H@6C^$LCOp8{6MiL#Q3WPi
zcTg%E&fhma*@qWC1%hN@hvQ;gLCuKbudJL*Sf&utVh@j$GPKY9#uWdC`cC<-g`m{o
zyp>WH2Hg@C<e>d1vVC8h(K4Hny+(HuajmywaW51;THYi<^metGyj4vxRQWf0k02Gt
z)Ywn^X13vfJLH5~6%dn?S|<?B0gB#y|BftCGlaB(+Gfb{_~C_8wwAuo2zNx>j*$oN
zYQ??zD!$XX8VY;e1JZ9Yf*QWGXSVml?886Kk^#SCmd}CT0>?$k&^Ui?GOqqjW+uwR
zofc7#bvkhcUOG=cH2IdwS5Rs*f@H~{r~apYPM6FtY^UK`upuC*16)BiH9X-_A#PD?
z4YITBv$pGMSEE<Mzrk~W*2C5n6;m5j*xAyZJ4wUuT9w&T1Q(PriV@RY?x){xX%M-R
zOOWtR&w6H`qx_Lf@{rS21hWwl+d^qAsGeA!23$WT%VFY3J?zNZO|f!JwkYhHgCq<4
zYRQ*Zjq$^oDXo-1BX9`i`R@8FwLW!&r{~o^j*Cphy6M^iX<R(O-P2~!FxO&DkarjP
z8TyIaD6}Yw=*{<f+o7gRq#b-m&|2%s(-8Vr;A32qEiu_cRE_<saX<H;K!#)^@ayIC
zP?E71U;yKDqB#4VU#2We5zSav4iU^v-7dD(`rc{>a4xi@Ki^^XE7>y=QYa7h0Bf7G
z_63G_sa~ZNUH0T>P{MWz=wd*h_QyMAPWatA+yXR^bFVJYBe^Nj-g>6UL@ERdnuTP#
zNk{40ohTIl6kE(lr=r8T-k*P&Wo4hZ6lF5u#sn&B?~KT0Rg!{KGdq^1NNT(!jVr6S
zhc<z=h4A~yhYdoh*q2r)awd~M7M;;#*fhT@*udZP%|&m$ORpMRJ=hOcs^!Jk*mrl+
z&K;Likw<T??$0ohc#~V1MAb)QuKjOMbZ(|QH2XZ~H0bTHHpyXj4%8F4W5b%5gbY-I
zrWW$$Y3_a6Pf*I;z1h;G%~-B~Agtj7Tku|rIbDKAdJjJgnWqe5W;tEG^UIcLm6DoL
z@&{-Vzevd?FSZOnw$IdrHPJ3%D;B5nWC#0{Z@rRK?i`--Znd<acfiFi$djITTbZrJ
zrrfrc-pZwpCH^-NtCs11B^1_@xEV(Nk3O4AasS(_?SjM{tElaIhys8e^SkoqwTh6G
zT!Mz5YG(Ib57tw$&JWZ*h<V62y!`W|c?ZAAxR1dV$e-o+Z7yxPYCJw4`a5*Z^#qp$
z7RZCbvvf*qM4vuik>&kBNyP(@%w6=f^8QJlRYVP4=2zr*7tdGKl9%9`VNNqduIwO9
zVpk`^Sf4t=Cv4w-bpynY+B79n#471Qev~xp7zgH4MeTw=<B9j5Q85Jurfb+(JC!@r
zyD16Dn__M8drZL5%5GU7+gslLcVDN->pwjfdu46wXHzChmx|DsQfzuf75DEGsT*oW
z7$pZR%WtCS1iZ@p?ZLQ%6ld@~#`dc9%iLb15upWu1fOa_{$HO%twP<Ad$b+e{o<n#
zT*%K=Dap<fxj=z4SHD~7KWgIN_Fx;^U0G;Yy&Cm5Rjn~j%FIEH5pw}^_Il|p6FP;K
zb?+8N+t|cUSmu(<tHX}b*)R8}qcz_@zoNbRomxIJKDr;u039U!##6kse<v!JqSXGU
zZXielD8<qXTlLWczZ{X!mHzBH+uB_ADn^MsKFd+t(!0MKe41#^EEx%pg?gJ_r;Yz+
zI@G@X|3#&#O4xavDf2jdA=&dJ0`7GJmWVn1j}A{HFh8zHCXz@sS#^QM{$LzWP?qtj
zz`k|ki)MCS@aaRXn|iJ0W1~b0fNI!aTUJBd^^u1nLrQctLlArgZx(mS%c^PrUq_fo
z2Tb@zC9KBwj`Ht}BJJF{sPAy1DmPz)i_5t2`vld~_!Eu8LlMpge|u7vEq3ZVPcvB`
zHu6g+gXVM%0z$Lg>5BU~XpmMTCX7;xFu$94Cvzbsh-r5UM5+2X1yzCjRDB9d5DcKq
zwrCSX@w(-<bt*v@0?JN&d4+i!5j+&AIUBka__tE#zYFMavZ7aGli&Q<cXF5GZsC0-
zpE6L;&haKqd3}idwK`3Px^K>xJkIIeuTm!eOfy|P9wX!?E0TF_l&C;bjX7q3$DG!S
z-~k^e4R2NxhOdDXjTeY`MjE8_h|gdQ#A3a_7@bPN=lSeyD37Q~=M_PT;w#VGM9(D0
z@_ODkK73bu-#=}!ekV42+m|Y;YpnVjoJ4ks!yb|8kv0p%o04rPg55HlFo4Fe_-!Qb
z`=@!SwOe-V;m3DTZNdv*<#rtg2-sbdw2~L9N90|lJb~EGq0|cKmC@||sw}dcUEN7h
z!@h*1LvuP?;#cMV-(hyU+_xeJNrhHIzf#|d?RjUhJF`zz13j)i`1$AVz>M49L$UU@
zsnoA>OO%B(McWM5IPl3rJm0>l5;g!X(S@Bu?hX@Xkel_WYe%GFJ5MepM2HVJwdlb5
zal+NkSK!g5AOdlvwHyZ)i3E}*vk)@<jAlN07=Dc#`aG=UGF-nqOu{~nV)(NS8XxD$
zKGcr`WU%w3{(h><TDobF+$s3zjHBO*OHyjlscXh4@Jrd>h(PPULEQt8dyahC7?`fT
ztpp#fU#=KP<4l&TNI-`rQ0*jPY9V$r2Y6kv^8{17PP$*y0_kFy%lqg=RCO#UM6P0=
zzHy{dtm^z3HN^r$m4#OGK1hx${7Y?WQm4Hqa4#l~K<1J27(sgV<lF|n62FKpS&+MZ
zHpdB+_y2a@>;}7Dt%71)gb3W2QcSttBPdwN`Pw`rr)T>do*ltyAAi>yo0|d7F6=Af
zv!no_`R0$@-!t*9%uy{|rM~jk-LZiL6Zc_*6Le1`6ht^$2Jhp7ea2n&@So;2U^+|O
zeea`{3&IiKoAkI#lBfr%7U;C^JE=4^8EeR166)O~LBTt-QbMgA7W7FFo&ia;8u)g#
z<W`Y^@DjxeV71!y>J9FcAdojWQsB)XTZmEK2}6?nLVFeylc2Ei-&L7f_5a7e;Ef%X
z5%=(ao(*?T8J(7DqlL>i{Qa+lVhSeonN|rRJx6eApU0<jGSZ^481DbdcRjMbs8=b=
zoqdnt`~|gQtdm5_BOtv#MIe6A@DC3&hKB*mVh^uW4)7-8%QeeM1;n?W>GZb&DAdoM
zxA20>X0FqHX^&MW-Dn7Cq|7}G=*3A0hFoTr(CbYYMfFK7N(q-Bxk?hLL$5RU(XEYd
zhzS>cYcGHE+E}BM8uOY~{qdCJ{RnQzcSi3@FNXv!!ZG%7M6O$hL2&l$aS7lB(9L7;
zsR9tD>X8N#uk?uUp6)ys>>kWme^9jFm0Q+fO3-%1MXku-Zwg`No*|y3QEa+rHmK{m
zKqmpu^`&Ka=_dkDQ6*J+<%~|45lVhV79zomw22+#0lj}^(s=XZzl=_0TB)N9_vI-@
z?i7CYNNU@q%axo$g4`8?A#>jyfL#W9M(?soGR&Z>kSiz#gJx?{XZ47Ztehsio<~}2
z;ss{trFXMsv_5tz$hjTRx$2K#c27WuAvoi%cDryXOvu`)KEe&XqHi$!>M7T$(6<2-
z6m`RrL*Rs_fzAYsZt0{LWhg5MgparP;e<o9!rcPdYCP}X<7+Gu0o8*P%mhmssI4p$
zATgin|C&XE1(ywY7x{9Xh&i&xRW?8D3xRe<b_P|lLCGb#%>YiEIF6<~3ZIrrd-HON
z;bITwXFRhVXALQYuhvZtjKPXVPwf`(vNvg(9IcFhFl-*te{RF^FbI>a(!!|WlX_-n
zMMdb%V!le@-z0%Wln=UsCIiREr0nX!S~;&R-S0F*czfSGpV(SLVT@;CKI7*F0qbv^
zt>V?9CrE!;k(Q0=2!r2cV|*I@%<9G>c2O7{=vi(gK}2i^KOjr>z<t)e=NWY0!>p{Q
zD8vTvm<S{~8Rp<ku^_~tfR&Bdar9vg8~PWjLB;^EZydbP=jrCvvO@8_wI@9JjHuKk
z0G7;rQJPXan_Rm>&_%eSclFmY^b{9tSH<R?i3csamreu$i2$NGAtaAxC~L)@kh(0S
zPZaW!&p7oGa(nU#)Xew7$k(NrTqv1NkoN_wX9(~exLMIHpQWZUv6Rh~@kLZXS9cNc
zoiZg>JwrarX_7D%f6KKsva!-%V?5#Z*7Fc?eFfXAFE{5CZ%n8kOVWJ9uf0=u>AW4c
zZ6r`q8_&fR*nfL<hrZc;PQeZS`F;j-gzD9aV|LkT(~}SGsD~3N4W5s@p2;k+<3;eL
z6|WIo`AD3QN)_ho-|ej`$Ctlf7-IrpIMo+?36|?{2oM*#d;Zz?W61xrN|EX>iHC(k
z-YTD86cpYbGDkWNl9}{0qJ&#aG;>WDwn$!?31-$K7Ys6?XHh+9WvaJkrVyu~1zEq;
zQY(KeOFcy2K6eEX)y`aM4Jry4n`eJLX&|2Ilv!o-o@x(4^58xt3Rrl%WA9Po+T&Jz
zV(@JI7(KdHdo)K*!U4ZrSK6mq=sk~6J?yz<Fz;p(k2&|RyM1B&E*i5$u!41(EJ+>v
zDhQXIOCMh{q)q47?QcK5ZqTOPUs4wS&ytq}B`5~=CA!%Vg5}15UrUGhM5)F+G{~YN
z5mE>Y1Uowr;5ta%;q@lJf;po}nTX$|r55zqt5|1H_2dIrv+B?hbUJ-G8YA(Pfql9n
zjG=?L<+tR_MRfo3BxZi~-~SzO^^YL9eLlPKOlJbRPK6(Jo)nSom~)HsTZaZBK$_)>
zhA*bvV5wxFdsTQJrP3dUexX#rDhzLL?*kl|o<>!Sh#mhi!~r#$Hp)gye1bcQ!~Pr#
z-scB&zhNHIy?<bV``HfF14`Hj(w%1_NFEO_A0xlAvKp%P3kPIGuL+5CWS-w$e-q+T
zbGVSZ&OR{3_Ljg+Y14VoFeFlUEDGTbkI8GwYGPgJ6+s_YyZ}frz_1$t_G6#=s`O=^
z%;UpRKN_jAy$#B3su^KK%1w@$cSK3=tK)p+=FljdVGq;^aUVN>APf_B1X;ZL2$lT%
z1>BM4G=w}F$4D3Hi=)La?a`v(r~XMN0#X0Sm(6dEx?1Mz`UPC~MT=O@XvMrwEkEkU
zqW2anxT7zzH|qfMZ78MGP4bk7DZ|PEVHSsgqD4;C#F(Y=yF-6L77m2#efC6maT=`4
zV$Wv=rJ?w<sABWg@$okeB$CN2lRKQ)uv6v{GDsry#`Ut!gqlEzi!(etJM&Z0^R=hI
zg6RL);8_0bZ0c@=ih=2^evBZDefahjgu8*{JK*3F<&;?>>*_wwKFeZy*Iud4gpnov
zM=sPXjnZ4j!ur&v%9U8dj#$nCFIOHk)S<8zOt{p;RB4}TaH|x3D<WvY9eR3kt=O5%
zmgEvDIy!4?eQDFtH8Zr=BfM?Bb!*(cZDt1jt)`(UqG?Nk1#bYlFe?ibFkCbBZdNG3
z^G2%t@}HZYp^ztGZYa;<Q7&$PZbVXZc4T|0a48pQm(kZiJ;dc=IDB>`7rS=u4tCIp
z%2~{v0@LA!<>9#$5MAQkoqjP-`Xolkbw_>2(8u8o+KM51H<aHUIKbM-?#iA%ZTe~H
zXfmAa;LZVsvYIfX{s7W50?hS<U3N5#Z4a+H3%&fp$vXDZkhzDAWUaN`;-dvmXmj~(
zQiH}lPaxq~ShG`jC)r?62DMqLT0GsId($N9l<OtKZ2u~jn@9+c_niLMZO<qAbo=^Z
z$|g&a;3ByR)#hi_pI=$*?LRtbtibOoNTuM%D4?Bv;=w~3Y7OMRQ!{dNCxaVlSXdf*
z1r3C;jgpj(^gzW~_ILA~CVyBFh2f`DXE*O_X<1z%Y51Ht@-fKWf^?RckVRV8D+)^<
zOUV}O{%;k#K5wj^pt>IWBW$A`PKD2ERsZrd#saaF>G_F1p-RA<eq}BO4WEr5EEpc?
z={tPY&tPx518ktCaDGl8ZSWtD5;(XsU@k+c9$laI4t<1f%m?d@or0KH<#y8P?*UKt
zeY+xBa&n9!6#}!qlALY1R}-)K$F1$5^JWh!vL`5ZW%Wgk<}vzJPupTpupbi@vzHv%
zvSyrjlW0iPbT!O!S=dRT^P`X`6H2sJ?QU8aT&-~opeI@M@k|;yK0tQ%WXe>KhbPFA
z?V_df+w<Miwfz4?9{x%ibmC#Z^U(Jo#Q`bAH{!5FV+C~F!Ab&r1tToejjF{y&<pPE
zmfGRk^c6g$U~plm%eh>^U-8!JS#j-2FhzC$x@uq>#T4CnQMFW!`y%xJvjl*32(KSU
zQerP_JZU<rr({oMO<)+5ft2Uy9S9;cp!aXSH)KqkvSU?vY1TA)Cs!CK#68*Tcc{aB
z5OTD#KG=n$yt(j>cy`MGwK!~4I7r(g$#$bKMu8pUQ^L1{61BcSDZ_po4B8@sA0eVk
z$-O>^yy^LWU-%s8xV~0j2Z}$sJkhrxC~oR%#<iu~5Ja0DRm`!w)=hHMh-PL#99t4z
zTX5aS!$lFOBsVp@wl#|e|Lwzc4W$Y70<TuztrHF{O5cS;KpyPq*bQX60FPY(7KJ$%
z@~emB>E=)9cWQMuqu@_R|7t-9|MM=i2|9Fo*pm%-PoGw=dIRj07Zu6lY9hlKhReKd
zT>i_?Fmo356ODSbLuxlomQlOg!2x}r*@C^HMWbg-BK0i?NEaW8isr-m^tDGyq$ax|
z(of@7*^MA^&~P|*ciUNWU-dSPv0W}8(*91|=oSeoipsZ>HKD-z5$AS|lf&QbNz#J{
zPq;fzz#rBLP<^^0r?bSvdu3(=f~<*)radCHUr8NOR}T^SYsk3ka1|Y3wM4O*pzdoK
zTRI+REJx2eIeQ?ny(*Eax=ksFtw_?_y=Ph{Go{Jb$sid-f|nqg$;&~55a)6WG7yu2
z3)#Yb%;@xZ4Rxh;xm)H<sO{qV{FDGo*4R4G-wX^*QC!ntdBI|YmRT-fzdM_tm~J_1
zh(vl)dWm|{c+7`2It}}X2|IJsI?ywF_kduYd;oTV&JeuvXWm*Y>(Kb(KeAvykG9JX
z0_<O!@K(vv&0AJ^rkIWPsG_LnwGQvWw8|dg-o|17b9{qovh*?Y1kBE;G?M@o@jNYG
zzg?bKTvuXzbRAD~4X|={V#Gs?ieUv4EX7=PcTycqD>1^-1?<NQ=;6hSV1sVM2<uuJ
z3er<++1kmLqT|LDklA`u?aqSc^;>GuM^MR$&a2N@x9vbGdgDmq@mP7GKB@uO&i1Y0
z{1%NeCnQyI+Lc-uETcTEtOsbf`KD6!EiS<6C@L?4j?h`Tw-e{LznHr=Im&li8Mgre
zcN2Ddud_7TP=coEY7Ln>xG3LOMV2=Iwl7UHz1QvKjO@;j^;8_5DInUDJbIL4!*Q-*
zW;dR_+(izz;+!q_{x`tl{XeyqzfzC}3v;;vqvkyWr~o|i<OAo~7Q&J&+%~vxMPKdU
z_s1FKu-5jq#_JjLehLQNgFh|vvF#IRl=@)A%lyWb6NL1B$suG>ydJqrUA?F*2#%gj
zsyC}oAKv$t3rv_^SpRH@{&A5y@+b|J7Lqcue7J_4#0LXP&c7bSQabcauOV3^njOJx
z2-0#Y;!k&iz=Vywebf57=Ypi9ZvD8~)#K}ZS8_82Fs*1k^`^#|a!zQXa8_BFW-CYH
zSq-YB6)$P==yfrlI_7t?9;G__xIM|`J~<JEw6}yE9@C*0@V=ma>BKAYwI}Nv*6sB*
zXxq?J;+!_teOLKo(eh_J32QB+A5)7NYd-vkt3lNOOre|E8^cVjS60uBB2e%@(Uw$|
z9MOp~d&uoYi8T;N=ih)%Wr6b;&&~vw7t^oQ2o3UG-+U<)z<e<vG&Vog!uO7f=ZDp;
zHo|YR2075{<<;D43%_CRlL}I}wS1H?u)%(&LSRH|y|b9a`qIH@zOly!<D}S$1W0_k
zLKJM0$vfJg=}oC%yYa0-+q@x9nsv=xO4soqt8+N|WPe#rsQkzTaFcX_X+`kCxev2P
z_3<gL8BUKR_kw{mISozW<px2fH7zz4+~MJp87ekIKV?O%Td88ztkk@@!C|Xu)t08*
z)Vceb9;!;45G>dE)*{%rFbYRtL`?}Q4Q89DwYUpE4LX&-I12M+kbR#e>k|H`;mC3y
z2|Iz&y7Ym&!#$oA`ihW;uF%v<ha+eg)GBfB&2@aV)+z8dUL_~>@uYl{g&R@&_Y|{D
z%ZM8VGk;4YvX;5u*y`%(^$iZ*n`P~|67qUBX7RrTk8Y<yH&16G;D*C9G|z@z4y>{c
zE@!nlnXH%<d`uNL>@AYP4tVi}RrPj>H&`b^f=^HHW`K|Y@!4JbmAqW2vk&1Fft<nz
zsWrC_&k<0LNPpqvq6ucA9^#Q{kmwm33w3#94Z-%~Eaz0LW9-hc31WpIU<p0<Pn`gT
zFt}CU`bjQs+aCbik;eHK+LWi?lek$IKB}2rb6hXna7avP#Hfb2f||=;!Wd~e6+es&
z+2hUBv`IfF3!Iq5^5hRD{r*Cztge}2)(O@b@g#CtWZg#9nJ~veJ=_mnQWTi&#$Sg!
z8y8k5>51FUo>tS@UpeX|IwLkZ6B?Mk7*a(IVbYPI0GxMP7Zh*azi$x~IetM?6>Q0d
zp*_hQRcdF#A3F5O<^guIo60I{zL8O$YMVYC?4LGZ)15Lu`8UMxUM}RLa|97r_h|Ze
z@Clm-?f#RTu*w|DjSgrhcAZA3K>f4549%v0rmNy9f8!zu^`q?%Cs#?l(2y3P%*#wb
zwqOTXHg<(Dd3m&$ug>1GOZFsZ<bSDg3=K);It*~iyc>c;3Pi4OK3WTP-xQRp5$sYd
zh^4(xl2qi+7wGVqN3@-hdZIAd*FmRWlo-$UeP=|J#lBAkV_M{2{lGP@BmZU5idb8w
zi@A08fii|bf`^!r&fYrX)S4Q;Z%!1bsy4>X(Tj>E;30T~$<QR`G~Vay>w6(uj!fY0
zX=1PH&&&9GsRl!Dc_fjenl!wS%%s0Pz^FFBdBW3~ErSxHo$EuI$w`&h#NCMduL;5t
zr4{>wAOwWxq8ApGkRbZw$4eMSM#dikeTczI1%^_`*f=<tL_`uUE-r^q*Gu3)WF&>h
zfZX<Z-+oC@QsaGwKr&KN?b2B}Uuux5RKucb3KXX5(aN0cbXVYJn2){p$CPHIfBg<9
zoA(N7;gT#4D^O&?57o|5oPYb}XT+&$etlv8sf9ev2LvZlU2B}riFq_df4N)UsZnwf
z@+8TyWSCZe$q}wwok1-*1i<)!HFXw8SLBL?U>1kbyv>&xJ^ON>afogRFQ3h&gWirF
zChqO*%TQN11CFk^Pzo>#J2toQg(n+ZKEY$__R?al?yMBHXs={D1gu{aT(%oXA=q@f
zhx1j0oc4kJ{h|;E1PKF!l9Mw577k&wM2nZ7pI^EvviXRv&zu#=oY^NJrYd$GN?A$f
za<a6rt+80BHeYTg;%yPUEPg(yezh<~$>>)rw|7gdh|iRxM7AquisjTGs_*al`ze<1
zcd!Ho=yG~E7P5JuWMO<dvl_-?gl#dw(m{~Jyxk>DXC;{smA*B2W8S@Ad}&h?DVWp6
zGFq`~qyEt!sOT}7=-gL>-<ibCCvaK<SK&UZvNBpEilWSPD*x<gpuhhEEp5o*Le1Hj
z5DDz@gcL8MgK0;|A+k-D@#RZsNz0?Eo7-51KqTZ@!0w!qC=x7^(W7DUTCxXkib-QM
zAnO2(8<Tu>u<G_JRwzx2(NBb5fG`GWSF{CoO<>+iVCs^H-q`=|LA;*rrHL-BH@0ko
zTxZ#Z4QTy5hZR932DEFscxDp=Nkt7R!@f_DL?oQjL{<}iFgYVdp0LfMqUsp#BE|tQ
z=Fj)}?%phZ+PiXBJe<8ZbPgRpuXQ{%O^%SUvpaPphyKX6JxWSAcFY?`iV;lu&X5Xh
zRD=9WkcEZie0L%sQz$}nPg*s3bW}QIy5X%-Gq_)gxf}~G9cPL)65U+?71x1qP+G(K
zY4A%LVNpFz8H4w4N9vsR^>)aqC_7SNpF(6q)Qc5Epj4qFT^2?ij^3iZmv&8@_Gd?y
zL}IVDKbb*p6+Sn=M*^`jGtB@%X9(sTV!9q4Wf+qIrjY2xAVTFzG?0~pY8sD6Hky!`
znF_{w3$y(0+{L15HBc73H{OpD%HFohdzIn2Dg+)u!`_Z{>vLNZ>!tan>Hx9TX^P6`
zT-LS|Td;A}B77-sx!COHYERwJ5Q?$GF<rh68s^q&BKvz02%b2g%P7@DF%dGgYv1^A
zxd3rq4}1CYW!jV}W8x5)W&4$!EaC3XJrK*#Gdz6sR*0^%YwF<DQd-*k$jE2`qA&Oo
zaGC!GD-^ry%-Fxz5kNH`lzUsn@R4S!#L={N81c6Ldms<H3204F_Uj0dnhys+$D^NG
zUgzDD>~N+plLI^{peJ{rPLvx=D-+()aob6eCQl!WheR;?ilAruOz@Mz2qS1;$cyxI
z8Lz1F?(8FcyIjeYd3)Y7pyNiWpoXG+we{hx;(@_wPKBA8t8xZ=rk57UwX%wzc$N=o
z%I8+N4}_rGz9}eN;HU93ZhG?#2yw#;mwNt65r1Ik1R9s>7S8v*hL;c?nm<snc-fI@
zT|#O6OflbNa<Z9~>`xbh9OU3Zhs?a+cR%sCts{Qh>~3{2ftm?Cn4C>lOKw#)SKzQ3
zO~&-+nSXt+C@Cr1J38XoZPvQbWdg;D@e*v4_y4F_<~3=k&Ufac|Jy^ft)oKSv=8h`
zz-BvUK+NAAYlgTPIAW8`j~EXYF(1``B}eYKalUBY%a*@S$@dCCZ%>!~!im;oN#H6H
zAygzY@sJIx{iOpVhkz{zkK6_Ayw*Vb41VSZWew%=!?{z;#t|+_+%g<7vahCPzmLJN
zD)|`KAj(#yFGEIJCMF)D-1Y6BnNOP11;ED&uM#U436Ud58~>7wm6k*8+#IHs3dK1s
zD>mm^v_UBPjH_IVR{uTMk|K@xembbl7^rJyy|{`!?@mU|#A{X}UOuWpH--?#rk8sD
zu7!7P_kJ5lYXbRb5>rZ#0oGj3&~VEon4hoLXmmg03X6!qrYV(ZO5A2&KR(=*0cB-v
z-QE7@_M6My;ORT<#FToCjqRVAN%nAivEG5fTd1Md{}v2Z-e0drK-|xUuJ7*jtgNI0
z(21On+Ysj(oJzXR<P{amtylT8C6h&!m2s=9tHCY&*>Di|>*MwMeB2W~3rp2y+}irO
z+84<t&5^aPka$i98nDcEO-}>ZN=izu=R0koc+ASx=J5*K5?@NdQ~&jP#kZ;k{6iNZ
z6aU6QsXI2dKfKx_MyIG4|Fd$WeJ%B}mQDHwUmvl=$xmiUI{VY)q}{1@Wl>tB?MmT3
zhydi~#ScDJdiva$ST_<r4U^i}H!GbX&LV-+x<9#YFdMvSK<ronPO4YNTkFuP0-QM8
zw57ndtMYjtGxYWrXNpOF|Hffjz27z@=Esv-^7~qEMVMmFYbACFNwAES{_4R9mFP%`
zQxqvigxtK~{hus(F}3{xdEuguf0bSy`m6I4&3xgc>qudjd$_oKMO9uNVN>-=Q|d)j
zJS=m!Hgs?H{u?qW(5b**^hj%AV75@1a%EV7*Q6N%rn&W*PBvq;#>@1CmVhS@Sy|cK
z++6Leqebr9olNlIT3;WpB5bQSyK^ipEkSI0i0&Wnp!x;|G71U;y1E}yJ#VQ!ZqIe@
zuTSI^HFMD>wEoa*1jokWL{Uhr#cFwyl9T^TNWf=eW>x{JOtCO{?>~S0_HCP7`pHwA
z4$sKYkb;zy6k<Oiaeq3%ki_W_48B-YRFvI%1xCyBhJ3vz!svWw3@wFuj46SqjewAF
zFp^vhCXP`Lp{}lOsOjH&1rCIVq&C>DyOPAc&6Ph_<uCB*ndgrlcbZE0S*;ognk4I*
zO>%iarw7j~N42;O7PuC})`(p&0mHOf<R+a*$Ad>_24=)$6=Ywr-E@R0etK6Z*YWi8
zJ4g*ra%Qj2{<sD}K^hNUlS_}Ph^(Aab-t&G_N-s`$OE%$i-{q7L`Q4e!xs!o#qOmb
zZ6AngG}L}Hx#qzHYkPN#XU#bO8qK%^piaUWwhPZObv!83TTo`+!34*hiX^h>cOJZw
za3@%0o(3uwKjjLh?4e6UUSJmnG1dI1GtbjL%E36s%;7sRQPG{TOvL#3c!%(ttwbwa
zHcLYAxLh2W&6OL3;xfL7>a%aV+a69H@Oy=cpOjOUfS)sGZn)9=8(gm{VZc4Z<qKz9
zVtl;HHUO^K>XedJR&H+KuV1gh)1wzo$iXitDA=?xb=CQwC8NGcVeqTel%&>FM=*D@
zZ?U41#<>tJSHb+r7Hi7jLdjeGQ*tr1D-c+GsUYY~|K_Vmamzeom8^)TM{@AAVh_j$
zy`}D2L$IwpnAW01!0(T^#47r;s>*!w*q8vOdQc-g{kDDko0dvP^@t+POsk&7C#j>y
zMUV8AVlo{ARFIGE5zRkr!GcmBLRKntLTGj1T^6O4hNNOEv(QSw1bFSP_97F!BT-ch
ze@u@Y6a30*NdPnpg=qp1G1~J9vn;xP#(5XPi_*r->NW!H4|Nyoh+{kS9#u`Cq<^I8
zx|=n>@aD}M?u$ve&hBnjYSP<o0;{w!<Lle~5^5Tnc6}xp4Gjr#ad8U*B(R(;IxTy&
z9ZpP4)PcPK*4oZaZ>iBmUtfPx5m5=OOIFKY{m}>zT8ERlRBEkPUCy5O^IV0|%p<qW
z+K)A0Z7sL3uv4E&=QD2a$cPa*aYa0yscUcDJe<9vyj)mV7$(hZrns}SQ(9FuxKBSn
znT&%Y9@_GhlAB9ap!aSlbF74&|ND1>Pq2u&`T1+B0VLJVniX~7I1<m$@pWvT9~PrF
zYe(2)e}+cb8+{C4?%buM5zvl_*=twSZUqJzYY|+_58_>RV)fgC33pcW<q6&PqWzR5
z(GF;RI<mA8KwL#Jlq|qEQz}N}S=2#0e4f%Ur=@~a!uWwOy@UX(Aw+m7Mi{(wvW>=3
z$K&0<W^2G$^*NyXutyRN^I<hn9=#;6t>*i~t6iOl<v_Pm95Al;s-s<8IRTAl5r|%D
zN#Pt#ByGUg4PFt(gE=By5`Wwvb^o9y^x{!=bZ&vkVkxT7c~$6=0fGQcmcJFevJ0pV
z??+_KLfG^T(fIjG0gWmoU&4J`(>$5LxB>jm?H|^=!`91sh}V5FzVemKoPo#U6#RLy
z0)>uL&1eoh{~WJ&yh1`M1)P^2?ygcNCT74cl)~foyk4#sdjcTnDNrh*prZ?0S+RR6
zN@g}fhdDYrO5%2T%T{BDqf%8-K?Wjvhy7_ru+dBxs&t%<@Q!TsMUCKHP0C44EqOwl
zuAJ{rdPeHBGRIa{SKCT;I!vQVr;WqepO&z7`y$CJEfy{t0S<8SUZJ6FhI1X2ate-d
zCo~E8J}YZ}>`LQ*zW0Yc9!Vq`d-cTM;2*tmVsyY;2nd*I5rlI2r%&ueJECc2w-Shm
zVobQ0Sm(YIlEM+ZzcVCOeXuPMJ!)SjA&UTGaZd+&_PO2<UrtnUbInsLZ`$u}a*k?H
z^)Ks{f>Ho(q#{u`F65PvcUeu{T$IsWli|^;2kYHCPm+ey!8j#*&)Eh7x^>M4P($+w
zW}UJ3;`coz?e580nOZ+yrB;rFSD6AY2-_?uMKEnv1bnQc^?N;Z@Nl(kwA}0witk{#
zpEx)uzBQ6MQfV@Q1P&ECUTOPCPR_!<44xbfVB5!nGdUb&5%rhbEN0V5b}gDEMvwP5
zFwbywn_S{avPkg^VmAW&D=yAivb?gsJ^>yD8)mf0&579UpbTNXTF|ZwkB~4_F!LSN
z!CXa-3V0}T#;_U}YOJh=*+D)7s+7tCIcI!DrtV!^Di%bK2_-Q$r;Y}LXEip5fLu*v
zBiC;*s~Nq8ojGO-`aRE4*cX}el?k2=+n*C->`Y5wTe9g^#z*z!@9vH%=o_(&F*vIf
zW_&<`uBR10>|mE|YdZ4=RM)8os0H;v1*0b{jo9J@N;uC+k}39yoioYgEcGvUiC@$L
zX0OC>Mn)Pc*aI5*-Hj(&0L)@oiF<24FwxnRk(sH|<jP`TU~sTd6Vc>$Rqn8_I9;Np
zmG}!%??4MOO1;hg7<H#b2Zv!6uvQ~nIVhT&n_DEpNk{e@-AxWBu!IN^yKE#iR~cEQ
zKKR#<85cG&Ffd>U32n0Qfo1#<N&?*5$PYoF&2?*)RaKjJP@rYIdcjg}Z!)j(?x^9p
zm7nj=L<9yRar5vbvRIC>he;k^9xXO-tpD4TwM_9t^6&n5S4`h7H9T$P!=+`URqonU
zfsTH7V4FrDuDhq3&vd`E1(*()INzZ6QhSJy5lA0@0UJwkX{Z2OteSsH1B!sTEPnt-
zDYKkRU~!ac9xdjp%e3veZTh~Sr@W$qE-STFAQ550bz2ztp#|(59E|$Cul-T++FKqk
z7{Gf!A|fJ<Pmg!*x4Stop+sDNz;R)aMTYyZ_+P(%xn89!76b(bzNe-R!laZ;vRhuB
zi5TC%z^-E8DahU`5#Xw*s4%l~+8J$I<M=cMsN1{{Pv(pPzv9`nniV7P=&q})YYII(
zKi@d%A=z%eJL;I3soC`o5&T{J#2=_QpwAQ;{rbCPH??T+skN<R&j5p;WAa^Eg^54g
z>e{IeYm^<g@4ehy^ma^IMNQkBQXW#JLp^s3f|`bE-)?PG^ya(Fl0xu?2z-e(Z(|4W
zkmyk5nz_X1s?J0ne<P0_qX#|m15I66jgCT5B#p4(@zX%9>Jac9LATEu<@7@%Ak>a%
z|Dp<OTQaMCc+R{36xo0@JDa=Fpnpx-k$##o8ccMk2#r&T52gPndZ^z+UL8zhCpSr8
z`qbOgAP@8gN@ah^L5gY%%miXlz8=DzAXQn)d@d@&24nMj@nx%RXdD$@@yYu?Niyeq
ziXljv5)h?$SkPK^p$EZLZ9aI>alXo|hceZjUt9aV-PT~6@t1P#wikxONh;f{e+J!1
zff9R*;FgkaQ;?~KQFoh^DG*9FKU}Ic-|o?lkB@_WO?xU|!Tk&vVjGNS4&bocItI&u
z&33XA>%Qkt0}$5TwWBahYW`qoxwj3mAoHuW-89H4+ew_6Ewd!e>blupwAx<zk=nLi
zcd^9rvTE1))rWUDVh$$cTrDbMkTs%rIjI00$h7CxC%j>qdtOhooj%PqKy#^zFTK~7
z4ZO3Qe$jl$4Fr&=F8zafOecRxgi<gnV9I@R(c5EKl$Ud7>An`s3~Cb8J7{u}dGlHT
zez&RO;;M(r^e<fl(%wT-2M3%R2&>BXZwHx2iUQv3F|10XB5qqytd7E0;vZ_1a5-~e
z-YP3`hC%hDk2z^c!#wOBDID*p`;o<<u8ZwU!G9h)c7c<AvRSvFp8S53s)OW;pbnI3
znFQc!a4=xoO2CpLN2G_(AGqRHl&ogin_ij-ibN6kC5xPbT}857^_DN;Z((+D1<LHx
z0B_QQwO@PaeS>C8yv>5eS{i~9P7OaN5#xfC-F1z{$COZpk&MP<F*&FC&^^h!vZl`p
zaSF>h;(&is0Op+11gRRlB#Uwm(^L}KT_Vt;_@WsBlGZQHpND-1BZJNN`sKIRmadm`
zMgx@C3oQVU1-k^IlGW{PB@<rNqCX)x3@;!od2xvgMFf3rmj~%W5yVya`-gMIWo2c8
zfb5d;`aN5nx+IAIjmQ7K@<l*f3n1~--WrSt8ZV~RxbF@sKtP|_uSYwHmWk-xa=jLE
zS_w@+Ff*$Ne~*MgoGjUM+qGP-e}tGDZO|*e;r#NleT?fR3IR@ZSKUVTkEpj5Ax(*~
zrB?e){@gbdQZ0{Iwsp?$SHEFaT)ZP#x=W6P5Wqhss(epFV~_uyhalTOnliXP-h8GG
z%sZvDHe2A{ZdO$Fn>XT!o(3yFrNVa}a+5u@b-!yTYWoh0%%M-a%1>H%qH=Jv%Mebt
z)b|T-Wc&k@e97U?7gfh$^?|l03xS+5pFGrg4q@+)YI40w-}xD{WNwmF%OY~C5O*0(
zkQOOWU3c~^4Lfs6kNZk2HS8R-N2iG)09~s(1^nYMPMQeS*7)O@`-}@sQx1C0z8GA=
z;X;Yo>pYydtJdhD$GH#hF}7h=mg#Z@hop)iBY#Hzq<^B1VhZpzPp_+z7fcg|n-P;v
z9i%i|pw_KU{%}xMjC#96El!7uWIdhkQ>m`l!VzKCll5f1s!frcDTutQIwISC1H52j
z$SFS@h#|c3KqND4=393h$_GDKOpLnK?IR+>UM#G80Ks;y$xQHw%tSFAN7Jp>WN`y*
z+Z@5d$^HjRdt{&YNbM4)RrT(5wbw_97i<qV+o?Z({E)wO1@)sF+eB82Bj=%og$8$*
z-SMm{^dwi8#e~809ui<9FDHcwFr_Fd>sBGc#Kg=Nk8fAAYJzXAci4~PcCk%17t2io
zJCV=md*dp6&)Xf9OY>W}x;&q0xaDkzryA*5NU};U4edw6ZSxG)p_LL|Biqe2yrmmU
z9_&8|lFDRXa1pmG413rLva~*EteB8czrM9HB{~@P`5Dw{&Lz;@rQb~tBtce~WbLAu
zZr10ZSt~MR@S#`UNJhVuj^XL+-eHr=G%`nRVj4|c9ro<rySFqfDDY|I>vk2`1unax
z7jAs+NAzzH!xVZ@OUGBtN>fR0FNV(i#*XR?jAzi+RsFUQ(Avdgr1E$QWEGk+pzIPl
z*hNwZ-qo0qAG>|1_n12Xt__F6^saPc9uU!xwny*jO6jru>&4oHJq)OIj3{iGIfh&n
zx%cfjlEdk=3RimD568cvn!MC&Y_0Q=MHQtp@AK*sNz(N2I<Nhha(gK8iOBq%M2|=`
zL}X)R+w<>dGO2o<6Lr0v4P~3Ykb$n!_rLGS=04%y(S0}K7VAA81{5xrZTZ+K{~UR<
zP>(KGt=VZMM>EZ)hQ()PjK6p1i<pj6jZ0rjh97bERihI{&vO`O_x=MnuQ;LnToZ_i
ze?S!z{y$8R3Z{8)Y@MB*vkpBUz<Vs>@v=IHYE`O3C=89#k8UlVPM$b5rKQ@FxiW>y
zWj=Lw+e`q6p6i{U#aGH$$na!B;rbxL0X?cDkmvt=g-NlqC}j}JG8{!I6`7hU1#YV>
z*qsI6-MqZKD&h)hH?$9}Zp>~ogM~VVAxPa${hpp480ksqMC4oKakPVsSSo23h6_ps
z?3E@c)8R`$Bvk_{tTvx;6RFLOd5$d-Dhc|4OO@|Iob|p^yb&uWshX_0j(M_KKZ+uc
zu4&#OZ=41&k18w=KRWzCKB5s{;e}N&s*UeYi$Jlr^};*RClb?}^Lt|Qys;lF!WT?7
zG?1Yd(@G$RcjDq~=Yo|zu3}2T{>%fz$<<yHJA1h)dH^9Zxe#WFFyFN|cZNDq*X~;Y
zuTm>vpfsYevvKVeUq?Ie9vS}Z0O5M;rafL5DTRySOFg!4nU%llRKJ59uA)xdw`Bw;
zWQWnM`fiod!8NxG)}aa6oLc-CVe?n44_3-0d0bSpF<Ev|b|mU{RhqBn=hQw@pqd!>
zbYC_V`%ZXkk@yZ$#S+tvZ>Z#=XNO4*z(A&6eGMnSIZ#k+zC_W(X`Iu3@hA{~^U1^J
z5BQP$$X3@P>&4=7wAYpJ=EC2(wvqRjRRW6|$UKv2Q(OO<Z1cId!wGZ0!7!6sW#nNl
za%fGWO`9Hb;cpQYD^#9U8#~z*Mju>2<<<I$k+`oF>!N_H&X6UN!{=`HEBO-=0G5xd
zif|$>Cff~RPR9dbK0cW6@NlYk$ofopATelf=LhlFH7New+(BCgW(Zi#m17?K{r#T>
zKYw(h;HoMX{4oEW9pe<YBM!+Vc33`r{g2Jf%}zyk7c*KFx?RXH&+=ZPRdXLGfIN2}
z<e%qHufqhtfBOcD!vJXNgH1dxA%V$ZZ<6aa9@LO%m5N_#uXP4JgYc_mAl1tYW^iy2
zyv+UX-8<{c0~OD^LpIObUCP2Dvy@r=P@Apo?TG><22k|9*~yI9nasl+9v<%1XX@|o
z4-5<2%!<{lU#~^J0{>HY@2{vRxn@XV{Iw4a5=Q0#Vg?8B1wh&QbBP*Po!P%~bZD-T
z$aGw!G0T4Uw&~_2Wh}H{{Wr1$9Ro)q8uXDKevj8&^Q<S<LVR@{`?ao+s`%T57FC5B
zxT%gm42r*O7*$+D?=!a)5S^@WR#BlWaCr9@!~DqOwgu+IRY#wVo;Y97Zyj4mPqi&B
zuvGb!__huM(U~EROEjb>uT{QWs#ThXy^{w@RGwdO;SgiV594NJ^^umeXm4vmNHhxr
zo4hTh!Aw+16!D(4NN=In`=|spzuaZ-=Z|Wj33*U_C8Raqx1q+4$(KRRClyNxy{Ggk
z7BKzV%~`#=xZEc-<gH0B3)7qrepCN1J~I!aCO~DOrya4uPx6?Wn7*B-CF2c4+v|%2
z`TX>3`_}sc>yL;+BubpT%QgL58d}6uJSNuf39u<d_h%-(&6Tau>guQ;mWT`{MeyLi
zy?&Wa%ISJeWz}?n&EtCBIx&GMCME{zG{5@v<Kp79jf{jq)%XbM=<5~4?j^i}=D6wU
z2GB<pBibE`=NlG=#uSe#CnpEGbL^h??C+-%G)lF<{oKF1y#@DL`{^=py^8n(4z8@U
zl#e1R=4yLqM;Ls+xUU|ZeSLjS7we$707^khk9VLLiG+;Y4uLefKOGz%?oJmGv^-oU
zHU|tFufGI+ftTO?QQM}bs_w7VlmX?}`2R=KSI1S=JzvuuA|M?Gozh4NNF$w>mhKLz
zLy9P!A|(RS-JODTcb9aR^t*Y!zt<1`bAijb`<y+qX3d(lNqeJTO2L&@RTc4+vW{a1
zA~YT!AL}-H97m+jnQwm$4{zE%PZj}4YHz9KA!6qIKJ|hXtvygNXQJ%RdRzH+aLu=?
zSlPL0vZmCPW;Ht-Zts4ZO^rr2Gg7(M-R+^4(lrS~((|ns>YWBpeA~@HtL$LthF4E>
zj6QE5G%o%(n?RvtmJRvGlBI^D%%^-8ryxAaT-mt2q;t*e0WMkY245|@yAwCRd;E7F
z=?%Ydee8^|pOUWfqIzmRMMcZTPewkFWt?MQSeV0!<%F0T!5s?~mFMX=vwiZpOF8S?
zBr<Y*nw!r*DM9+}Bl<HsBP*jWd}lg!oi|3^yih@QXm{GB|1-}7?b0|ok`^ZtfT4f!
zHN0&(-))^RcI@2$64RS3Z<+tVPr8NDq)R`@x;*?)tv)ke;fjv}U2s{r9vP7rN+<L_
zSEDh&HqZ@<XZWV0bJWQeo1aQO(mwcw?r-y#q-=iXCLDuqIaAgApOZ4SBB6*BSgFU%
z2M~)-zP{9dsZ4AvN6{Q`ck0zif*(0VZjZ#XOXg;0omM;1y}cP9YA%w@(~NE|_W9nt
zne;4#Q~f1x1Q3nug<3<#+TLQ5fb-_y@80~KIt9@1<#XLptJ)8$HyS?O0@U`#1|g%0
zGD~cJ<MmF#CoHKaK{raH+kIJ*22e*29ahUlru78Slbe&)kaSh4?tWhI{j&U=B3w{*
zMW3FDer4~v!;mF}PSAQPG>}B_z~4^!DP6*0_TgM4nR}Gyn=+S30t>Qj_=~DrI`=T5
z+t|S*?vSm^LHhop!@>UWIYJs_hB<+^7T6Ws4@DLDNV8FRvHkIxj_Sg__Yn1rfLCVa
zQI_UFMZtq@3a?EcHO)0?@2X0F5#vxA4(W76w|S4&<>Z4Bn?gVl346aD#$PX(bemi_
zPBB4;Ke<*q2C|in?mjX14WVGbq3^pRH~uM(ERH937<rTSwjMgIZ+YX$#y-Q`zbOo<
zr-s9(BJRrMV8pvD@pgIP$B=L--PbiaH~iO@>>M9A?t&t=0_16pJG@#llOSxKY`M+c
z%$K!HS<YFLGOoUt<{bukN`-BM9ifkR)w<%lk1sdI*-Dbg;)kow8G^k(K`-mrE2GtJ
zW`<<zpJySe$Y!HYOy-EmlKjHp{N#JZ+!iP>exQZ+2wHu9m6k?9;I29DO~1yWc+(k-
zNd)RrY7r4CXJ_ZzyE}Arbfe)PqF9uI#Hb-V_!1x^**6}e0B*QHnKu$pf6-}a3ZUVn
zD17ybt?BG(b+vk8CqiD3xRn*7fZLuSAmGUbJ)(kwkO(l4i;IhqP*L&m@I-;$0UZ;w
zpfa5<%_9H;giVlofkv6#!W$e)L1_gA9FYEkf`eZ>EWecuBhJwLh+-VB%9?lzS|O*P
z(k%78^Gskf7zBhxPhVdq=+nFGjE9b7N^yc;Q1+~%<o+Cx>F@8~pDqG2m)|wz7m#45
z2fEu3-$aVj7Tk#=!>4x3z%Npmzw^5kCWKlpQ`8WvV!y^6!*rsHOX)*vDROp&TFGio
zyY?EQS*s2^CkbW+e$2%F&b}L|+=DKIWH(F+v1TH$w$?CS2Y2rC-6<h?R9O|Nk!gQA
z+k$NQlf}ur_`dgzNPx3~Gj^V0h$6N-o?jZyB{F4y?`hGE_g&}MMv?Nro_u-3K>}sG
ztRME;aj;p)e;AUzs(P}b+YmZ$_r=9O+_Lk4Pp<@CKU{VoyJ<7aXIr@=hK0~k^DUFk
zm!ZG0F0}CJ)%WT=xFm}tHP7^IZk%Y)Lq259);kxgc|@2v;m-<=1nS%4bj;RWKcFoR
za%Q2k6?>(pa!xs#4Zyp|6?eq!j&ehm){LeO)^?^(c=t))Ep5NFB*-<#HV4T>v^G*(
zM`sN1766v~$&!f;n{@{_8;~7FgDLz~2OabCDgKD49AMc4yFOT3_x(FwBN6fS4xc(c
zKbOcU&@Q4A@3u5DY6FEWiBP>;sJ7Sw2Ysv8_1Sg<YZmCQRKm6$J0MQ;LA5tB(ekt%
zz)r^8e3Nv65D#gkEza|v6z;&k32UeOWoga0!5Md8_p9;96K5hbdo3wS42vvkrsvq|
zp(>t8^^|dh2~n}%o|~H1lklzRka<nVN?4s$VIVe23b^rFRsGVJ<|<6Tc?XmK8D~Uc
zsCR?4o7|oScUWailmVf{QZ?|Ln)E*`E_<pWyw*{0x_ZlXmB$~qPH4Qp;}=-N36EEV
zPD&h6VKBhysM=c+3^Q|D8XrX=pgq4&`kovoqiw{1G$*_95)01^)%;W@yjjE6kEm5B
z_w|}@y4Y%7U#E?bOnV#Vdd^=Zm{LGhGGc^qG1u<Q_aFZI4mQvpY}aU}c9zqni_^vH
zBlXAwIi2sJwo|dGN}Qfwa1*xMK?K+SXiV$9?J<ruY<K^0>w+&_)<8L~+hko<Hr%66
zK|#c$P2e^E04&Q@dTXhU^635|a&~ns)u=S{FeCKyAGh;$q|d2$Kj2uJ5}2)%5O}-}
z8oAuO5PpD~iYj#(Duzc|rGDClRk5$Fcu<*kE`593{u^yGH=%owsS~&4iO7p3&;Fy=
zaKXHEYn*5GoYDTVB&H16=bo8a-cF*Li1smW_XnfpKPl=wocNW(G9|Ttb&mEvO<Mfz
zSMYbdGu#dzb;S0=1j+LME<WJMUXpFZ;e1Ek7;<MoYe$qfE1YR@eX}i93oSGfggKJ-
zQQ7nUz$_a4(X(ypS}GDB(QR?~a?QoLh5q68^$xfB{Ezg-oe$|)+pB+Xa;v+uy`iwz
z#2pQ_i7qe-j@bH^G%R7M|NSkoCQeVksg910X+4a9Be)93GP>|rU&`^zN{f7cWtT&*
zh^oqo%}XRD$x|KJX6hH|*`l<~Jj|1N$$T^nD$ahDw?F;)z%M!<F_x9mLTh@}V=`Sk
zy@IzV_w|ob07`X%Lrle&>A1s4mBW>#jjq=Who=TM3wsQMtbXK#d?Zo7lQRmP1aL`-
z9>|uoIl>$UOy?W_?p#jK)GoL!T#>Ltfn^$~GmHe)37)2h&o&91?hlOt`63RWt;7uS
zw?FSo9jm$p3K>{lh@<aKJyVq!C|1?sHExcyOUWc1=P1S{IqpdTI@Zun1tj2+HL|ny
z2ZWZd9s>C31X_ykWaOV&diyu$m%3|fPqDU%Vi|gSrR>ZYQ!{z)kuK0PA&DjSbu;V_
z`8bbT9zaNxPF`NoCp+;{)&{KmB8sy83MoE^#Vjz@9wImCyQ+N8PaiW@PSBbi4-_`U
zp_*CuI9-fw(^@Xq?(v`3)G-PLYtvK$GdD}&1e*klo^1-(bQ~hwA-Ch?A8OLgZg{Of
zf8+Y*rg)yuQ=wpF92yIiBy)vo!brF{mVq$k>g~o<jTDo?=!ckP7s@jp!}3zqkLq@t
zqGR;Ks7aw%dSb8F^I8}Emj799Pjeqkw@YUhPz)Iw{G6nW=^9(~du!(MoW9DVvS_C2
zhjJeOzujk`^a?A_RlCl_D&-Z=oQ!rC`r5dw4OWc!B2D28@GmKo;sP3Eg)V1~5=PcC
z+VaoiEU*eXyw<MW>?dk3e8&9&bp*iumkP3o#a=a@cv%zruW=@wbg_8p{UF3!&KHy1
zEs^=`o~q#Ps1V-x-X{YJGlQ`XkG;h)V%&<NP9|xc*3`+a1CQKCH}rPM7zrDVcQN9T
zHUeL+kR*}^vS_a6fLBQQ!{+-=D%iUMU)87{d8-M<TmPl(`xoNj{DLc#Cx?p|_^SM~
z9sHwM=hXn{Sm~fp>*-7|FfkkLk3ZnmoQ4r>ZEYa<=u&s$5bzfN=3q(b39%(1IP6(X
zS4@a;p#NmFB&cc2V}H7sL9SnlU2Wd3s$Vq+r3-SC_ej{;$&T5%Pc^sHrH&QgwoZo4
zqI$TtUXXPkJpV|Q^zHldd=_3(qN<S;ZCLJxs&tYpg9m-=$bBB#8jgzM<F#LnnU*qP
zrX9nd&iedX^rTW3Nf%SrfM=o`SeZTc$a6b<%e(!?#*_(}5&jFe0|zcoPOLjua@81g
z-yQ$5?2)Nk{~O;j^<I7h=^c`@@7n9>RGM==vYu>*<|o$XI*nqtR#Q?G(t;mD2}=eu
z=s6AQ+t~Xl>fZ#E#GfT)iPtydY_`s-6;rhQvClMP|5uTNPBD_*HOC;`XL{B75aWHp
z`0aaZl;r59-j9JIXSLjf!I57sP6|M#wo_H-;!)2F%Bb+r#)PTp$NR!X4b(9bbiLQ;
zxb%RJ?G7gkffY?!QzXfMYHAX?zrU}vTlfx&_SNNKcB*uIq_)~d@jGCe9<*LzUg!N!
z@m3&mtRb@U>y32S?B~>mm;OGp<8Ac8UZO_EGFzWn0WFG~pc|5C9W4QUFA4y}@wZ-d
zDpl7X?=w{@_Oc(G&b7!rV68U;)UEXa{z-39QMb*r=)9jv^;fGhC$lG)VJ>!dR-S~l
zAJM(!E4M(ckMBA9OvX3<Li}v#75&7pnY<!EuY(A7lLg#3`h)9f{V@n!w${qDh6c|%
z2}kIM$(Ak#&J7-fCqVx}O8fkpZIpXzh|tFBcw+1Fozm_R#&O4dz8Q_+&X>R%%mL3&
z4vy{wg^GdERcJ5+CVv(ZdK+dmip9F#U%fw7o{8L1Y&%Y6Uk)`4v^E478`6Z>DU1$n
zd;eQ~k)v@SrLEg#ywsv)N^2kD7_H)W8TpEV-ctFi;hKHukFgQRSNG6MWvZH1Wl+!b
zm~$zERb57rMgA;4Vr<_;I<^Db5-D)r@IB-At0}FA3j{w><d1TvU#{-*?6Y#vVY96Q
z1FcN~Lr=+6w^o?%$OnQ;R{85K=z2fUh_`sa#<vI`W16T%YZ&FS9#HEr#`iA!Qi2}f
zBY&Yhrul>ta^WC}wioJJ&1qR1PA^&ABZ~+$|E>LnkfBG;^s<o-o<T`9R_Y}9Q^Dr0
zIDgFwT0S>}$}?N@?(ZdqqwRE#N<T9e#-jPI)%3<bJ|etjJHQmt=9-ao3>2-j6dA#|
z9!Y?^-H9_VceU(+)H~7<H#K(!<^{xby&yK7MW{ESAoOkqu%6CL$U?N(#g$3({c($`
zAr)r5-v8d{lV-{d?EZ3UBJh=lqjs7##-@dHV?)BW2?-psa>C~{k`(lw73&o%%CjXY
zr9n)R@A^!oU)bQk^W13`ZS^u$yoMSNHuE;hc0iMLQu^>^QG^{niqW>As_ALVFSfXV
z1}MYr$kK2ipv1_Hy@9*&W5g*$)$wvVk(KSNq_R#?2ST{1ehw)0f3%c~_Qgv`U&uxc
z?ZKh+ZZ4!tnvsON3Nre5eRP8f0S$^?9jc2Hhp~Xp-y_p$efV8=d?!)MeVJdc8>rCz
zouNgxZ6N~Z$jF(!ppAcL8LlG0=;WgUR5BZBZf!6dzMVJ3`hP+3GHN8fW@omFNw2{z
z8~k^_T>w~<$98u3*LyN%=7avdM?En!GZtrsj(-v;9#(dD0nZUu7DA6EhPXDx9|iD6
z1|y5%se81EuCQ+A`N`Pozs=+{f=yjI4{53PGY?&Li*Gh${6|}m=2~`lj7SSw1BZ3>
z2Y=GV$lxsm!2qAo6rP{Gf5CIjU+<9^S9T$TdDkmCy`a?%8;oA6k&C?4%vcftV0jB$
z`Rd51@Qm2y{a!7e1JWm-Q|~JLbY$Fvx6{UY+01_}M;xlM%;rLw0&0F{D81j@yuo{d
zr(1DONj=1zM*AM3cV(`*r4<dgkqt>ogx}&e)vDqpJRk&!{tR;G%B?!mXC@yin)#9>
zyA7+=2!LudT#8?8VMKv|zBNGC`W~K&Yl^L6azfXWDyet+bj)Bh>ytY1?F=7zV~UXh
zPk$|ayh2_Cj#ezfLsLL#UP?FgUCZ*>ZBsSKHUN@B(lh1U=m(3%nJ*iMiNj?(5{43&
z^5=B?rzy-IDgH+0tp0`XNbdG0m3|Mgw61ek3ozN!FXzNr|K+a$2R0I+)&}b-R=2%r
z|F2)w5KBQ{?46X9Dh$@!6HCYc@#NL--@k#5Xtcsy5ssXkoX=(J3Fz>rh(OJPe1$4d
z-qu6`k!&`enZh%q4AI^5zwEPNdy+ExjLFSAZPBLoDlwXl%ktVOecRaSM^khw>)BmT
zpWyWmv!&rhMha5{2$uQal_{DXi;$A87e`4_@>v%1A*(}9*?|#v$}BDH{z|={xnNL#
z;qr7zMad*AksxI>u>nESD`RYDns*aAWK909&FHb3IOSlu!-_ns|0RHBdvuz~sR~Ir
zJg{Un+R_E<_DBZdX5z=$XZfy|>0#LtCb??%E~DCcE^kF?;5T#$Z#@ZV(?jHGim3MA
zrPojv_zK`sd$E~!^+S4$HID44`Vlb*dpoiS=80dH1GWCGa9Fqm2~7rrt+)WqiE3tc
z)atwVe55AG%cef;?;ACN2pJ~YcL*$7*q*H{bftIudgRU47kE81h{8MubG%i=n*Ql@
zEK358iu~SIC;oj=ip|H9si`t&_c*I#h0g&^+CTD^$+!Kc#XL3<`kTJYj|Z(R2`a9+
z!Cyy~;Q*NE=yI%(Cwzira0dl6@M628Qpr80PQZ!8^Dv1>v8gNFXV3=NWTZ1?U;QS(
zGd5>IVf<g=F^|cS8qFkc%|-ggb*J-qt$TaEjuVhV;;beAFtarZXgD}x78=~~pFV8|
z@&k5mZi#G_k$P8aFoFZGRbhtR5rWJ6m)-h(&e%w^kJtSvl>WeOrM3~d!Q-;4Fn)NU
zI5SkT1V1ks5)rfdw4d<rx8C24f#<S0Yg?<Sd;iR^^Hpg6nACFbhs7$QfiM(DxEYn~
z<X>4UWWjAEzwT7uIGwI<FZ`t+4ZoFMU3Q$2f%umaJ>%Wp(}VC@J!Kn@OLxnY=YF&r
zqd{*K%v9boLfLBh8y{cJl~26fCjq;LfG0uyqJUNP`29^!ZY87EOtIU!py=#@aksJ-
zjD7!)JOg4ShAA@5<2x~|;kByZ*zUD1h>?m{W3Up$nJ>=P&c1f(@@V3;a)nWNnb{gX
z%)hsoAl)uIFw{A!isOhsorqGx+Qbyi=}<G09O8wjvinZ&D6f-WfhM#T`?d9#aywn;
zNhb?q+;9RFc2*tqO-$-M0yz9Sbe~_5Sz{GGJC5V(Ut#W_2hv%yxgNUS$CyoyMg}IS
zMXX$hUq;Jfa*(Vjo*A~@D5~)vlStzi_<)Wo-}m;>qgiAIQH(c{0>Q?{n)-A1AFHMe
zmCYbw>zu*OlQzszDjWIiOXZirajD<4gPf5^-{Z87bjPnO2}Qb-J-~2|6s*EYO_QF=
zG0B-!&2Lxi;#RoLKKy85lIJ$mB^z7LL7I)+9caZ(Z`|FO-_^bycdm~MZJ;i^*xa&0
zZ}dR9Ob~87n+mC-Iz;@HRf&|Uacv(LAFmDxMM~z{)6o3>_C3kSj8QgN=#MatM@H||
z*RNk0SQFCR04*$$VL_}I4(EM7p-C>}88=a&Id!RoB%1qDGSaL_zv(;B!y+E_v9+j?
zP=iSuql5W6Boq{&nf_gF+D9zl#%yM0_NcM=^G6yE=x}EZ5gG%i$Db#K;V4Qyg}6#7
zP<^jLPLm^IeYdH%6LA(t`hhl~RhVL%$MK0mb5)A6o&~V#!C0xHFmO1#q)5^BiYg4^
zs&(3Y*$zmx+n0gc*C(x{s17kh`l9M8v=f<ywhlyDaJIP=rgSIL{O(n=`%E@0AH64$
zo+@@a8!no$Qhk>6Vc#rvV|~8>0h}5PE<__aR2MHPG%U>?i)OEkbuYak6jMh*#fg?`
zaD8cHaQ1fupbNGxPTty(g@(R|Pw^&ueSYK)wm;bobQQ9?hmePJ((;Y5|LAb>i&P6~
zWUyWb&R!7KuKcNf26;+LFeqQ5++Sm|K3f>+nwJ%=xPK1VhA1@!C<~>SGBmz?S>Vm2
zumhTk;pEVllprjp*otfpEcH^iGUL7Wp3Xyop_K>QjoxUj%9{gWsRI^pOl`h^t(1Go
z#oxXr4a$^*T&eKmFMv5ue=5!QL~o@JowsJJr`36U`sqAjmcDcrU(OSz?)+U2%>UIr
zmuVv{$M>L94sG-SvTmXp7p?ZZb{;fPd#1~b;kUp%+11kJ;gWqP9!9|MFd#_A9`5M2
zE)Oh5Kg0reoUB(lPcXf#_XNq0EjD!aGdtP?+rY=F2@Tu=cWp*<w5L8s)<{zLZvuu5
z>C76*{JTSH3*X*w!zlNOq-O3_V|2QcEZBE$J%~ookIpwKC_gJ~-C2`CzE4~o_NnSC
z#h~wjWP|0(@_zgBP@36rx3P>66y7*)BkTON)TOkoZ7xaDUl@17#OiwMOxNIz&{Sgk
znc$$EpxEDdwQ(9oa&O(EL-KcP7Xv$9pFPV;YkppiuYhthEbJcv=x5URGQKr3_P2d0
z1qlZF6bww?)d+%`zLAAG<J)aqiWOob+#eI9Gc?*CQtTkujNZ_{#0{4X4=)fZhEJRJ
zdP5>y{6^*e<BIWene%H@%vf0w$j)a{TkW4`cu!Q<cbKHFRRr46|KJqYTS?rbgT<yB
zjm=*Z?vU4~!~I>Dj!q(5PIfklDP$8sJ#Sz_%Q3HWy}US@(;V`G3r^wvyo1#TkE4oP
zWmmG7xA-2Jk+br`CT4l9H9pQA-AdRX8l^pde}rYJvN`8DJY738wvr^>+nY68^~>zn
zx8&_~^#wx^rg=^~tCDis^y2?6)&8{&Q|kTO^(rL$N@9YmF}Z#6Y0KaQ9YnzOYNLB_
z7Oi`*i@?;31Qvc#OqQOF?0Te~$VLmp{~vz;4DdTQXM?&sJs_vOZt~6VSQ+cjP46#U
z?ajfzls0tmSm6pW<&e#oe`M?s=!?Gw2W)XSvWeyzOK2nO&U!`cS(6Z$Ze*60bq+2>
z)270fC;L@PNc*8l-ja_mn<EQT&u^M@#w=t6U6ZoL66vM;Bnl>_Ng4+PEoZsWoC`jw
z-W<72D%4O25Fy$TZ~5sGKD*E$6rNB?vrM+`bY6cYy!+p!J|TSp3P!KXgMr?7CU~tH
zJBlH}Q~CIPUb}gZ?t0$%xcTp|n@sx>Bq_52fI2FLoX#BMpRJw`J!|aC{)y3pBKQBr
z1Y|z-#EXg4<I(H6OncY96p1+R<XoMZmI0l1fKT;&eMEu#d*$fL-<z!%f0ERC9iQ=B
zLBt=-tZP6cL9Sh!x|-6CG52TSJv!yQe_9Fd@8(Z;i%dAYB5bULfi12R;RP{T2v!tO
z#3CVd&&R_v%0IfZ?DU$WP9~7m=lOrh{)in=XYTDm=o|2hL<Zi`Zu_$%`RWB2b6vlG
zAMaHRB|O~sGL^)!8Pqvtd-K0|(LUWl02YB#1;mF^x9aRI?~J(kze;^1B^7Z`R6iFb
zBiapE^Sq9yo|l7#C)jdoIILen6r)rr{aNTW-{<@S83-(wJsHS*Xl_*A>@MlI9@8|C
zaNopR(Jfk!mLWTFs&aKZWyoWGflW%Z=ptE~dQ@74Rmq)`G)`qc)jIAa2c6!wcPK_~
zPY$TxPtqoeRzDpi{K22lm&A?AUvo?!cyEtoE9qLY@Nehs+epKRhdMGG$V6OZYS!K5
zZGIWqlac!N#AG*YF|!xq?ypeC->?^%4(xA<{ZSpM6?g70!Y<ro`X34!M}NG@=R3ds
z;1Q=AKXL+pu;pb#aExm=h+dZSwXZpn!a6(OB`|BZ2OwjDWEK{drI;?f(eeO#@*91R
zAn1Y}bZmn{pg4Iuu>z@CXO|HLtpqH5Z|kX%a+6_E@Y2_w-XIhyzzBG^+pJ!u#qQ)p
zT>AUl5u#qDo%`o{fAfmpHGlfg^`E)m>81|f4R?Me1Zqe?6oxK(BizddM^_c@Hmtva
zD?)M4{q<aT-X{T2C!(2Z;zUv&Z8fXtos_O2Vi0$sv}DDLVBj|Wf{r0}QI)jovgemx
zK&ew@#y<xfUDi*(G_KuMe)@Cgfg@JOo7-tMB6$9}Thr(v0+b!qJ7gXYn8;<)l$Vc%
zHaqRWWXrA>nrG~#S5|WfwO|(5xbZzEg?bGM&&5BL`rNp|J!8}F2O~&KdO2X1*g(^z
z@c6MhGgP*-l}5k65W-bjPYq6YcyFbaryDGZgZIqk1=6wA@!q>!@R=wTo@ANp^*AG!
zzQzV07PQ2*cmoNL5N+;Vk_e_xtAmDMe&*asiO=1R#F_rxa|8=;XkVN_AAKhTw5HW5
zy~8|=>4~b--d%3GGH_0JtF%C(lz%??%uIs>bDyWL&NEQ^t$&BtH#(q!jptLhP)7a@
zBhRlRU1N(~xb)64P&4se)#F`gY2I~dp~|!LZ{7F4w7K%;_l8KCF%g`FIDWXcvS~h7
zlqhK1ceGwnw;W5e7i82!#l+8OLag1@PV|tXB_I0THWyslz56hY9svDUMiCZloVuu2
z0BO+fWRcBWb@<WIk?dS>aPaYv$U~|55G6Cq;jN<R!~Gozztbxq6KZ^U^xvcSsob%O
z4$uOl#(h-Xq5+iLjBJE5yB5HmS@>kqL^WXet?n$8hA(&N*F@5fRw!KS6<deixPiM*
z@kdtLbkp4$DzG^qU6z?jL=h`P5&j%JUj7}IgORGa9;x+5WLkTF`AJU~;MBox^eeB$
zN;`=u54Qf!MP-Ic#f6AZk}P~38*n!N9aI%nWMlZCQSg~1sAN?%4-~9#47XjU<09f?
zS$rN&J>mz1#J>-(Y%ePP_=Nj`T2=d0lyWFIDG`cJ`s1-4TlQxlERE>J|8^I#Osa%}
zDF?~pI1Xqkt#57ubzH|{lNXqFV<TeKrDkJ`RLxO%ZP*HD;@LmT{P<g4!$!Jnvk?b4
z{EQxJf;I|?wD|d9lBmg}D+8xdIX@Rvh0r0D@y|r0#HkS!ONG%1bllNaV>=ztJCuiG
zR3?5PHKEyrNyZO&HC@rOTnC9dG04BcyeZG7HZfI$X+>k%8~CYUH__t0EHy|5MV-;E
z48VK-aMm8PQZ0Az0Iqi|C@1Dm3s@VU?;pYgT}Gy_FS3BU)b_Xc;R(ImYBX}_oUHBb
z@8k*1!oJ^OxKU0hdQKjuLdu}K^RzfYdUMM`VPZ2^9us=i=2iA%bo4M)`Ja1<hcNd4
z5%Wg1bTzBCwzifdKap8rm9<34kI6A;r^iI_q=(ODfqa_N8nM3-i%FGwRHFCV`E_Cd
zb&@P!Dy{$rC?NdwV?#!KR)1Vmv{0*2bJK>7Z&@#!BMU((psC(Og5@j)LWS}!F8=Xs
zlx%YEI&D}AxBoyoODrwU4VMB&=Y>;3uP|S-T++$5sly9J&vlU|EPw(Kg(<}wUJDga
z3~sC6WC!OoS*O+A*eiOrec}SDv2XBHg8uW3_Z7;v3*;WY=MBv1e90i4vA@Q6BOs<p
zcT|%A(fTo~h=0f<h+Bw9H-&-L3CZ0lFoOI(ZXrPX%PH;l+?B_-^c$x&PJ2k%aJ%Fi
zsUn+f=d4jtSW;Y7!%O_~xIY8ySXZ|sVxQcL$JYKwN>CFQ-?uD0!XGgq@O*vt@S*qB
z3z$A0V0@~mA7iDi@zb#Z4z#q$)2pHTONY1sC1)ac2^mbh!7AKOi@H#c>#AE?9!D#t
z3Y_Dxl;TLECcQwj-h0rK=_FF0ZPp%|JWG2MAnCt2fg`T|^VNzWmR3(jYT#mn^57uB
z>~yAJOh<fX8Hkd^Bze_#IAI3YJl%<VW`}#>ape8}f&7Hk@yR`wBF$_Iv=B~v?}%i{
zK1kFb&-%<QN9e**mx^dOc_bKdl|x&7ZMTF>`B?8MJ#GIJX@GD@^`iB0>kFQd)?gam
zme<IH%e=7LBvBbMcq#H8qkNj%sT1&bf1AWD#@or>X+O_^@2maB%(5lmxZJ{Bh9JkE
zl&JDHr(ndD>Iq&L1YK&pN^cU|1MRJD{Ty~q+`2q+GEJ|>_U}O8OJFW?Ns9ZD?loe#
zEOy(4q1CciNjwXA&+}6it3dRl00Qs*e>PNL*NL`gU_K_8`A+#S<(Q?V73~0I{nCRy
z9j>Ta^j{DFVD`dR{Oe-ItJG%s`9V$-UHM<^oK#?!K-I+;9yRAV5+=kE;d<dUf&Z%y
z#8UC$^&cO6xTqcn`|Ut4xx6lO?IbRvII#?KY*o=*so<om(SX<_BZ{6ZVYMka<!6aT
zc6t?|vjECAirpPyoO9aT@DkfXcDtGEFNcnrcl>|`26QS$8{I8Z?WAaV+vo)0(vKj&
zHw!>w{rFcM4#&0qG&4s^y)Nx|?B@Rf`U(tMh~^YnNuu`r{z|lKCH?qg!#{H6)Co-1
ze=;h6So;u?9f-|=zh;hBAQzuyAWwVg`7C0QwU+0yy%K($x`;b9Vl?0>2Ioms(l^6h
zGlMm_^}s<j$F)Yrs!#52nUU8}p{BK!om;mSmQyCKYA{~F@Bb}8z_8zV5AQ{7EuZJb
zo(vfN)N^5byvoa>n0c`Xwi)rgi!pHHoVu++K1MpTc|ZM`m1~>8V%RP2clPb+$Wzt>
zVL8Xag({w+)P1x)GRX(rv_|`YrtPvR{xWYiM}eUzccI;gDJbP`-V%>Q_TM)&av#dY
z?mpWEyYamDVBZTAuw*|gSJ&d7rRFVY+3l!~ZmNuef!3&{3e`rpG`Kg9UzATIAG*>K
z6Q$;=Z5bHrx$7i7Jor{tR#snYg72?nWRLCe-Og!T3@%D*8|9rz(QzJ?!Udz76wo(o
zyc?ethu#$ly&EeQ6+GFVh9NV#Kl~%Q*#wAaQW`3a35=wf{`1@bTP7Pocjpf?`U)k&
z(ua8+K#45!@}S5u{x4z0RfXxhA`#|^K;)xpI3nj^1QJsdnk**=U&at3yPP{!q0p2I
z;$P2j4SxMn^ne2miIBCr;eoshiqt{W@Zv>?YR&z~sjX`wt^J?iA9epg_{}pdRyg66
z4>9ll7kF3NKAgcFg2}4Bzy{zav(f60Slck=&GBDc_80~vTw(Hs2&;_{IxPqBI*Ry0
z+mLL&tEXxPDb9Lpx{R*ylY!fnbWW#>ci&xXlq;|;tC%St5YbWDq+3GfIv{#Q3tRV1
zBy~oXl->~2sgV?>N2Q3&8=s`;3R+^@^l%!c=keKW121l5-%Ka9TPSYrBMkm^=aFX!
zWZAZu%5qx2q<EE9-w70DMr&3EB8Vk)pkYJJ8J#hgjLpV{k__{mxDmkidlv1ggIoQE
z1D}NDMF(G*4eo6$j-uxTRGf(!>`+JC-BnzB?M3V!^?#3_(vW=E!pg};UtgkT5sW6F
z0leNW{vJr~4vNUy79=sgjgc<eV7z}IoJVFLo?0UG#SAn$0MugHS#Y}aSDC_*>V%f-
z3W9m{Z>?SfhLUm?#*n!KQENZa8@^WeB10=x_IO95Q*?-@@utDiZC3gV$Eo&h|D}pU
zO(r5UR$z3RJE}XpAi+@~?(H)Sf1X!TdPY5`B0m{lw`bm6;(?Z+n3PMr-7{u|d_6kz
zlNBW#Ae2DDJ+PREpA^g4yrzy#Zap#iqq4K)FNu0uj4*uu8=jieH`V&Tn@U=|7l820
zW~M6_QZ#?UWArpFvwQS*h6o&eU@NR!`z@AeZP)8_fFV18`D?4Iy@@(CcD8#5j~dIB
zW~YBOMY_{YFKY#uv@rHGy46rhxcnU#b7kY1)7i<?WhVTK@PVMK;g0KD&vHQmQ1sln
znY@K~|7C%GK3|oU2oRY}n$+y1J$6N)mYDPymi@kN7ORBtoewxmD59k-VCG@wY$T?h
z${KO(G(2N8%h_7hNb{F3Q%-!F`DjoL7?7}kV#@lsbuUej2&|Vg`RK=t+qo%@cDeZs
z{DAVXUDB<f!tp->V7N3cj^S!Ol7@?dH;u^r9-RGv%9rR6+}|k3nv(^UKg@q@)~Zjz
ziCr{l6`*?J%Ae63!i-jBb_Q2cQ{#SBuFJ~ncKP^JI4sc*GMIoQrlS2RQfJFR#HEne
z#f2I~JSn36+{RRmzUmT9nLe->)H-+7V}qi-z7uUr50@dnKq&^u*H`3UqODLBPZJ}I
z-X|ruK5(OYaSlR-{L(<HQ%`1rXtH_m=kuXK9EW;KBa*dB)?0H|R)Ti2O@OBV;o}7f
z&gKnv!&H58B0HJJaX(-h?R@-GC`wzRg2~C&@tuHquZWm?lTdzL@?qycXe@>p_bpl`
zlEMDCXO$){fq{f*KBg)X`?=*PCV_%i{9X?}zpyVBlrG_t5!ysCr<nG0%sN)Pvgpg<
zJnoNTy@n5Vb_y^U%m-J3J!+(7<IziDqYoJwx$8jxG4LIbfti75S~ap|SVjK=6bsLG
z6A&5}*kV`5Z8_UQe&@{F`;)`s12W=*74LNSeA|0bWew55KdVCQ_7#;!jge|jVg?sx
zM^Kzv$$4u@ayVc4jrghlbF3RHK_^xGZ(f9pe5g`RqIV_IB{;RZSW9{VyhGyEF;Oe2
zBu6Xw*!gk(dF^3zz8|p<OMgAlL>fx<myn6l7e$YFfemINHiudhTnhE}h&dmyQ@pLO
z7XJ&^E6p(pv2u|;V|;{A=%7Aj%12*_xdmsZ<Lv`&t(NwO7qLdBX0oLPoMUf3;h(o_
z34h%Jj#Mjk_AQ%<Ef0BzdbQ@`R@{v7`OP8EiTy|Dw7oXsHU9x&9{hB<DJCdOxGrcP
z)51#glZLy>O7?L-&$GpZCuUMTJ%YulpGzFtt2hgAWicabw-dKL1vv2;o<L3Xp~hgh
zOt;wXcia5<S%~l-MB8_B%lDDG(=%FBUv*EH$W(dB%XvMOhQeavVV-hJ`q)~`>edax
z{m?4Qx`5`u9{7P<mn|5Jfou{8ULT?EeO?+iBdB(&Dp(XBUWZdG6|YQFH4NWJBSEiZ
zv=;%l{xOC*o1heG#`}Y_h1b74)Md13)av@`G0C?3h!>3<{i92d{TAJG^!J{be3VLp
zTQM=YZ$o~USmARr=1UFx_3#;I%Z<5jc5OumBe}?J4w6I5=f}rJ3-zvWjc22BU<97n
z3;%KB3@_)&se@#VpkTJF`E_q$&eZ{KdG`1#VRK3n6vfC=>CSE=z&ok<wRk4W&Xwl)
zo#2*>$~Kz#YFL63T3KdcY24ZLt|ZygT_+F%v6VbT;c*)L&>ngs&rSP84-U0r#YNbJ
z{uu2ddurXNvDq1Azx%=}s%;?|lp1G{Jy(g{U7W(&%_`6<hSE3)65(vT)x~=@nNaCl
ziLVHO#d-bUC646&UvPU~ZZU_0BVj;Y7|qmzeV}RUaOrk;>wT4>h-4N1Y07*^qM$$~
zoJrJzI4rF<CMHG<D8T>So(_Ek)7}<<_gL~i#g2}So}Qa~S@2f`o05{Ui!uf*0rk8y
zGD+QNKtr&Rs4Sd&@VGh5E|{%AP8}NMA!WdtiwHr6>y5#n^lAyg7`5Na3Gu%{tclR`
zWTpH^y13*cP0RZnG;RYzT;64>0a2SAkd|POVS=JZX&N?p`Q-fX)|zaKyVS^GYBYf<
zk;=E|uabFfT?3RlnqDS9CyO7iK+uH?#*%*j{C(+-@lu-B_Z_MePhn>0+tQGv^PUsJ
zE)KcLyPD!IO0z0(s9k|Z(TBDG<i~bHc{vBr!jiBX!ht?R4xsMhi@gr!`N8a7$KQcu
zfDqf3mePTM{bbg<WM$oD_t#~aarf42mG#GuC}8+lUf#|Jf;G6YL<*irUmb#xj+Te}
zWH9>UbhIqWMG9f3zHwdsO=Muw`5Di>(E3j^IM~_kaB%<-ZPQiOL{lXO-M@bagWkoW
z#~L=6mA7`4r$PaY?)CNcqH;?nu&SEU_bTez-R-1!`ndJre=yd)Mvn`|LwJwDaQapq
z*3+r-hT_}z@n%8sE4QpND{ffF^=qXoU($_f8J{>PJu)O^ex3l;o`}3rZ^w(i>v^ov
zzRlOFiZp!8|MJICEB`YxsSlF0dW}NcTiOjzEjbKjyB*SWYeLn%E|*T>N+dI%zjzUV
zh>A0Rd|6>W<OF6!ny>znm6w+vpiKr4$bk_!QXuneTszzxO5eP_I`uyOO+?P;;0Fjr
zD89wbYfm9_1Q<*lY#q{OtH=PBV+IU=Q~1~~)W;X;)QlGE30&;W1jbLKDzZ$P833D`
z-a_q)0}E(>mirmARx6;`9=%V1e7$A27Ts0p=YKgK2`J|hpOuF_(dtjQE_Ey`IWh$T
zs@zJYtme~T4i->|J<zs>KxOP#g0}|D{dEZ4PNG*OZ|x$^Xol2`eoY;@kg`lgbXTm+
zaTlhXabl+5ey3r({qxBk+5~!ZfONJe(!(oKE2B(<(5cF^g+a8%I+Xh_&Wc);R;NYL
zVo;JVn=6U<By)8Eo#x@8{%o}^87dArH9h@jFt8IRO9Lx{5=KW=z~DP@<gp4?>>2_i
z>`9#FPYDP*GsHva85l$hCSnp3yBj~ADlx```7b3cEws0w-#W7GvH`!ezHZ|3M(=89
zxh=5o#uxO*z{2qYn+9@WZ{m2_U|7*#;Oh17>dNu<pCcIZFpl#%K0P(QI@!qj_3Nif
zUhkz51j3b;j?U?zz~K`jBJ+On`&M>Ol*5*nr;m~Ah`rpYw5xaI(`x4ndbPlC_=k1<
zsOHagT()Y*)#ODC(nGPcRqjy;4c01#5*!4UjJhGNd4~2FfB*nmt5?<|$Z*r+?8Nz?
z2x!?RZfG`;8m#(aeUJ{TqpC|vIim9#h3?{|tl})USD{SZvETsG@XQl<r@1mt!=D2G
zyNelZ4G9T2Z<&b91Fy?o^JbQ%vK;f*5yZtAT7Nw3)us?|!}0^WX;HcPZHCftxv3nO
zvYu<0WMF1CIa%-F6A*9)pPkvBtRyv+g01=A3otAN8q?(_Jx?E<PEMdH`vi^L!cH+5
zjP!JKh>a@w5%gGRWMqITJaS%MUf@5V2+T0x;o(vC`+&d2;dL^2r%2&zphP{z1#>or
zU}S=@_@)$a<uyJx7hMM`|G@(h4ib1F9sH_`C**(p?~AK;rLyB`I}s3Gz%-|`ET^ak
z3x$xpDq5j2vEI=pjTiAYTYr&KDcYtmY}Nrq7WBbVUdkDau1%+*Q`JOGq9WW><+*7A
zxz(gXs^VS*=O`6yM$Dbg!U!?KOzEfpItO1&pqb($5?h|{%6w#)!sn3HW@@BL*(O=t
z2vXVn<Y{WZ?K-yF;cKF8n|dJKEcWI;LlT7^FP2hz26;5xhfWSN@W!dU(vSR4?@&t)
z{9^K{iiCv32-u9IMHN8fQ3+$Q2a`M$U^+_ZV%k_fg)h40{(3h8&fXQsKj2&O6$Kro
z94={r1(-f`=i2;m?*&vlox3G1cEinT+2SEMa5jsL+DINu)UsTUiB^mPLMI!Mfc@h{
zAFnEW{oww_qyX(RS}~XO%v#lR>e#kw`EM>77zfo=Ay?0VbZBcAW1uJ-xag4)FQcSa
z`=yIh|9Q6h)$juWdB3RHi_E^Si239(<+WkK#?64jO?!5vF88mQFzhG_p(*XzvONp3
zcg|JEq3w@!rr$Hj%%qnr(6QH$`b<mzJISrmK^5B-kr*rYc{tz1Nfr^@#~lE`0AgRy
zB#Pf&j7s25{Y*&3LKDj=_lHh2FcbHU%kn93Pca9R{I6(fe@<u!OLYy%&uow77~9$D
zuM7<sTD!PZ>aT@<`GR1$-0B}+4Zej72V5E0Jq+w39A_&np`cHnfHp$48FzAW5)Rnh
zY*pdnV&W}73fF?(jAp4)>l?j)UWN5nZ=xlzfaYk98_rXaLfMiNjD>m6fwQaV;!TK0
z(S&i44z0?uzr)&iSu$!c*X7Q=bk1x8<1^oH>~6UfLc6>wnI$r9^V)60zrOEj(Qz^i
z-^k)uT46&4OaFokgT@&(K^Gei$dHc|-vF2W`iCzz@_}o0d>;H`ZZ5cYQ$&xr4xnD9
z<^vV}3TF~+lylV|YEIm2R8nNgUf-KARU~c1#1Me-j4`0F!^rt#KsO|DD*e5pVvBX^
zxZ1W?l|}*3f~fy<c1Eb0qf;peEo9!-s)uUTghdYrWJZA*7WQIdq{#g@jUw1F@It^N
zB$QBAj&8d-Dy^l1UyO-)`V=3(&ijfZk<Fl)DEYC|6m*?4mcgnlJ9b<w7yMgBMQ{3h
ztqRuR@3(m_Aq`wDwY`6f2O6`2BO=&)aM#@xo0Xme$oY@-g(=;mF0K43*fs~MEj@q5
zrP3<J(y9?@rPwiVltpJ63>cK-QO0E0n`!ocv5uq1-|=wiPOO?!Q<K<lFle3dDetcN
z*Z)eENYgw%AwVz6em5&?+QrjF_RM|uy90K-hFpuAZTM^ECSdbfW-;>O(Y?!chX7<p
zT6Xp*;OX#GQE{kY)hYsth=2pLz^f6Vqm0*ep6V@GvB_<7$^MI9*8l!RSl;t?-#7NT
zpkZZ=03inr6BBc?SWnK=)AQ<`^_x_j>`}B>O_A0e=GKC-Z7pCW*4EahrL8?u>u3xx
z%kkx9OXdPt(mx;zB!M|SFr^-|_;@2%?(r-3xxj%B%a9Q~20laUjS<$X1hwWESxhj-
z8h#q3kHV5|GpolrrBN>{3S$MdNo@7&x;z4<$&fs(UtGv_VI3{DnOr3?hUE&~#Ywwa
z8#2D5uWYYL4}s?%&^P?u8Ob+qwdJlA5?b;m_`-c;7Zx|5tM%JG*oN@dfuuPMl?ChG
zhP;^q`o1e7`M*fN(mW<yiX6V6%kLr-94OOMwIg0IcD#U>H{QA?R8QjU;RzZf)XdCb
zn))AuANhalNoD}m(tR#wDE7hhDg1P;W9CTraD5dQ>Yt1KIVRf*=S>B;*~&a`%E<3~
zePCh;{vYsZBjK?jE`AtV^a+ttH-m72HJyBK7H`ZGgrKP5r&t-X%2A|v@jvNzzhLkn
z?cTo~u1eHH`&@|eQ%jTpPxOhBr3%uv&|6ZPpeR|-)Qg)n+URe>oHEoYL#OL2W`CPD
z(*!tQC2<}KmDKR^@bKiS^(H5SiCQhsHv_Oh0*|j>z7TU+JONOWj*&6Hs|*4a<^LIg
zH~<TXPtsDr`pDub9kiHZBRKo)eo8&*yq*n{QUIkwp(DSq^ag{xKVZP}sKuH67|q*T
zUps3L9`a~msvLDXb`aD~%)ej1J+lvC-@;-am9h*mjL?7)?@7q`;+4%sYVyCjI+GY(
zV-qjMKXc?g<l|wig$$BtynpT;=e4z3x^vJvaggd@7fJ`9M$#kChv$`wU|yKD4>25>
zVF`L8@%;L5xY~*+mqv5D`0FlH_t9TVjEk~L_O~_u1A41-vU^m?*n~h=j+BA}AAkJO
zkBhK_rJ-2{*rVl<AII3a)l11RI72Wr*xB7Z#q_J8;e;3R(bw0Hzr>|@Qz_^Mc1iXP
zu&2>3HH3%Dlnl>5bu4(3SWa$ZTi`2ly6AN{!G9l~D~i#GPb+T-97+97vi7IT5Uc8U
z#N(T^mP@RFgBhA*6DP@gFg2+Hqds%fyk78~#HOuLpCu(9)VMxrCduqqNF~E1Q=wMP
z348bl{ExH3Pm=4?oo3YMBk<KJne3#tf11{Kz!F@x@5|<P9a63bpEkoFTqRLm?(VCG
z(y}`b&oKcPvMDN8bXm+#$mT2YoJTM&Zx0+dgEq<RmA@e&C;bWT=#=9V0Z9K@gLa-<
ztd>5XALWk2x7?P8wS>O#S-7dP3lkYSvu%J%fw9`rc8*d3^Ht!K;9>I(D=-_Sd;PlX
zmeD<PL>k0YMZsevIM6aKDUEDs5C9<r<p-dn3U%vZz@X*H4S1sXe6G2`>t<lGj_?8-
z7Z-IIq}=mBcijK4C6LsV5S<?yibSRIe&Q$WbL{|}2~QHxM@B}Vv;8@uR0SzuQ%oVC
z(dbm%yg8bqNCjus|E&!S)%6h>_x<@}2>iYJQw3@UBiF+J?7ZK892Nux4m>fR1GP@I
zjcAcZ=7>eUF2mstQ7hPZAk@@XsXkiwlIpd^>&l*djcYHiJ`-2);+D5SUb~}aBvSY*
zCK)<vKGgv|o8_bDG#0wQ$a6(C{XN|uRLL0WO7py`ov#9U$|(gikFYCgVBuNgeviyx
zr_#?7)ks}lh)r@RsdK+Xf!bG-fWiD}&-($%Y>k!GeaV7Sx5`7RfbhwFlKw{%u_vqb
zIm!jvS37cN(1k(aSE0V6Lygm`TqBpWvs}0I4}p;X)eY|HWO36szu!Y5;w@3E;fZJJ
z0FZwK)JA@j2p?Z<Z~e?ywq|Jn5J8bk2`1v9svG;o#=7>)N8?6N?LInnJ^r`gA_eq_
z^<*JKdUOK-m>{pu)Ywx2gk3<u$;A~1l*-!PtcWOBk62)DzOJIz4Hv>BVl@FcV4|yG
zMB4}D5a=wFn}wkJTPyQ}1V^`f>+Azy{s0r80H5Wj%({7;uhoIRM4NJ^$}?`SR;o1g
zAQI6dEuGIt#k%uNjFS4E3g_MBs)2!B<ymr(vFXa^ju^$`4hrYO2`b|QEipQMtzY(1
zDbPq}zP7Q2tsFc+!u@)O!kxB5k-yfW>{R~qFb;b1uePD$^zP%?&a^*bV(j+)l@qP)
zbZ7l^h3;UVls#9Bo^E?|9GBU{N55jZgyX-(P17^8<t~3+D9K1hOzI%uOI2vlk_Kio
z@ra2(OJ=q$ETkHBd@6I<R{8n!r-M_z;EY(d%4_>Y!4KdJsK?rkcfkdpG#rJ9hzOX^
z{RRPXrUeoKW8!(G2GY`Kz)@%Apy9BqqXRLp%}*hx!{pl9)S^hE4;a>I0mm>9R$77c
zEO^1TW}yp^*8FaJTEVkvb(^5bnFiKjV3A*OaS;GB?Uoi1P>7>-M^bI}(4|+dLN6C?
zwR}+QwVh3s^8cBtO~~8UlwC0eS61emV?BDDhtq7H5PpC}6lQ1oeyFTaOvm5^b~v0D
z7~CLw5uuvCY&l2_RxjZCG&xb$+fc=dL$LXb5wmgzb~IzV?YO6xFC3BTJj1VEF{}+k
z*JTV~rr^tM>%&2yh!1e%?jB8mgcPQzeaa&Ga`68&#zi#0i!3M+@THhr9&ew>?*_i^
zS-)wbo%v&b&LaTtZb$nCgrVxiVWh<W493ND4`Clamftn;Gix;^)G<NJ4Y?N0U%wQ)
z-BS2?o-RpIJ+IVzyGEmp43;KWIdf-X=)#KPjSDpp>Hhii=gr|*31rZ6!gO4asZl@z
zl#<5X5k=0r&V#>!8DO4j&Z8#;<6%qv5i?`Cr~f6W+&AVfE-ubzYiet!duJZ^m#ZOH
zx$C_mxqew}``&HuWlJ+TIJm%5*nvzut%1y{?hSJ5%p767WH>_}S{SX8j#6%e8k(PV
zUp^a)v+cI18#GFoT{~O6vud*tp$ZN-g_C9uII!z=E5n<l=}2(l(^UFMXHJ$Ka^zD)
z-MUiZGb&r^KKWrYi(6y~CC1~78S=d#a}>xNCe*l}>l9&*Iyy$&@;zk9we3&mPW533
zT#)nK=>OHRo~@q7y+1OUT<m_4v9yk7vB2!1{}`R2=XP&l0s?05iHG6@4h`>`eQf64
z{ifw)vP{fYSC>DhC(<Ey)N}lTMA~Ojd0gwov^sZ?_1)dwV`5|B*@>|>I9&>?CzOCg
zY%|4^;mY(luqd_1ugE`d%yo+>U2g=G!_B@gUbfBgPS-%_cBb#|-HenV<12Mxdbqvi
zhh0Kw#jlIICO=gR1=KlEqYSQ#e*StTbI^zILvKkzeQ}}K-kg;(LbFR-WqBbH{jFik
zp!y=kbP1<d!Y#%P=L)OAbUO0s+KTLU09H}u@Qk+Qijs*EMwfD3DL-~~R$FQ_I}KdK
z-?yRBt)lP8<_h2Zt_!}s?Nw2*70jJCEcv7l9So1^Tlz;UBPiGDzVVHa>mEAZG*2ex
zJXD5-)Ras)%1h!45EXAmM)DdClf-G)KWCN2H8aOv-+C^b-d|z42dnf)T(U9l?7PJG
z9r8orvjz`^*yq0k3p1gpAvh~-8RAw{%$=^iJIIT^ch_ZR1Eg;)>z09W*4n=U^IIH<
z*zk^*6}Y!PI_rLur%g#s)dn`i6OrI;R<B(wek<@a-(%Be$}of}SY<LhU0fz0pv0=&
z-k!YpO@wiELKQuKe$WxLq;q;$G>9h@hj6%Ka*EkF^6^RhS8F+3n5(VkxufKn;nYv;
zQeot-Dz1|E;{Ncy?-Ap_D0H+JkVB&fe-$kBC>UnbvD4{Vdtb4`*LbuC7R9@P)<ABG
zy<a%Pk>1nW6og}B!%VCl!N!wS^!GrYhSEoMN40&N7(e+GU#DaEnQZi>745PbtG7=y
zla%V!wZF8al)|swRG%Fea<ILr(jVfsSfu2L7EU$OD)O&M=?L4ty!zrR2N^FEmPm&;
zcYaRnlAA+$Ts&Zfh^%2?Vmn{!xG@JZeUa-eFq%_o7BXr_D){?>Bk6Nk*t?P`V|HS8
z;=*B6W$^PA4!-tWWClzu2aEfdU+q+2=xP&e8kqt2X-AI4VJE*$xQ<U-T4-04bi=#1
z9G6&RTP9OUdJ8A6N1<fHk&ke8sXQ3!+_sdrVslRgbD6RTIxc0-MQ~b2-~E|%9MkhT
zve?zzcoLq9eMc1u7c0XbQADbL09WD~bfHe*4xhrs-c0CCib9bRx4LyqE;#Ekc3xah
zC_x4TxN}M9I74NAe;Y-veHkac_qU$X;OlYZzd@~Wa9Ex%MWFW&DW^C2hua#1CLivY
z`r=mYm{`1B@a64P?|!-#RUVaGus3k>&6QsE4eaTkBui{E{<lOy7QTi0%Ygl1h82w}
z9?qDj(+CLY@GB9rkOHo-1B5DgprJAZLSl^pQvem;Nk}{zl@@dXlYU1Z&Yf`Y%T83;
z*n8+T%gT1^iFP)3xpG^DzeA$R`kmVe0y}8|h25^<&{5JQ<5D9&Vra0d;wX~01CGMx
z2uBvn!eLW(*Gh+ZM7N|hX5Z75NG4aXTK0LR*BlqDI9T_Z1DJF!l#6=WeCo)Jvv_u=
z`E##f(`}7WC&G>Q3kWPb2G;Lenr?*hUPFmLl9?Kd2vn`J5xTD0oc)QhB{hU@Ld%Es
zup^4DV%J3O^7^0n9+q_f)L2+BR3Iu<t#DnmdK#HlH0S#t!V&vS7BGR?sGy(#>PQ5o
z*_Xg(;S|UWGsm`jyFuGJ#60ceFEP1KgiIRaUdq+q*EcrGfK?$d<?R9<uu#!6O>k<#
zx&0WV`ICv+-;mCuJnjE6^&RkBcklb36v{|tCL^h=jAYLu$`08wQg+B5@isyc%FNy?
zJA1E?y-D`oBYXSbpL)K(*Z+CFJfXaubKd9N=N{L6T_@v8HHxcom9%#-8%%wU^BO00
z^$xq|<7p(d`wbFXjz$LPH3)K^)aRsw{ItbcQ?D_zm!q9w!@^KdaNWjmUJIUvD04>J
zlFUZGnEm=aP!(|^5pXL<+A)WPNzM|88rMxQxyuJ#SX8W#u9_RsHR5911Bo-Kk@qu(
zS^(h9b2x35oOUGHR7QN5W9v<TXY>gC$lZ6!CoAf3n8wG?Bo5|gb&`K-P{W=(ZMaLQ
zL5n=sm;XzAeUat@{<8^8GC@achh@+0ZEIiz{epuz2I3PNpx_XKRD){<%o4i&6p%bd
zT(h#6!NY#f-uo0xH@(evlW?%#lVRo2z<4e$#;+qJuC(LpnKt%d;Ck1)uYOMTR!P(y
zF#6?;iIZbY`3j?hgdTIyJD_GvTHyRU`<dbEC54-rit6!+jt_P#o4S4(2UC^GQF!~M
zBh2o~cPlRpB-hM<)Zm>Jiq%r{$>)eWwr$pZmq+?kbOWnOKJkRd%fX6wPSs25v0vz3
zdZK)ytA)q(yV^9eCq++$$>M>Pk-Nm3hlx`)+|_!0Dx;T|(3jh7gSvLAmRl}!WQ*ss
z3N?AIh;D{I-g=3?pJ9tcP-d`xSDyPw7o!8L*A>7i4m9%IO8(b@DccH}??`q&I?NJ!
z5eXqiK62)w<+NJ3X^SSizVJ}J?GAYFA))64N7@MbhA9P189MIWn%2b8a;xgUtYp-O
z7t0wlFzWG(-GDgUyLZB(i2cT|dga9(M=|V%#?3#1cky2S+|J`E;q7P2id?I@zOEze
z;_QTZ30qJ1(c<he3m4L&;;+TP5r$L78T<hmm-CH_AQi*>0DGIkYiufgdw^5rJwc`v
zjsEg|vv5oAgSYoyvtP|${kkcDms7TG5xxIglv~B;vO$mqTd1uwzY6)a%AK#@uj<ym
z<zOUR!q?Wmt-ma<Zfm6Ha_>Len%}L1spSfJ$z0PRb|K{K*b3@cCDx1AusqIbYimP}
z-q%Bv(y6mwR9^lnuwBFhOm%yyzcrH#XpTVAB(3F0Iyvm|(vG8#uWx%te)=N*;cx8G
zn=LVliQONl95l^dHD)-bllzHF|Gd-HF5hRKEIeLbu2e~{`Xb%cf_q;uTCtikX?*%>
z;aZO_qDr=k?P)PrObav-jSU`eXAEX^*dubS3Ys!Zq}<KRKX#QnPG6rm|8;MxOe|Gd
zpkxKJW8HVM&D@)}_=sajLWOOPBt9>{HY=L-QL$Io>sNB>dlw;az7`Nd;>*nN$)kt7
ze#k3{a2_fK{VY_2U7J1Mn`N36NIq<(>Br|IWlNJ--uV&&F8%|R9!#Ur)>9SVyGQGq
zvdfv?;fMLhhK<k4#I5YT*UAZF`eA0RQYi7|%ojs)T<WJuyjsTU3D6biSRSeCu#vTl
z@8Qryq+&5|gy!YvCqm;UV0p7!j-sQYS~PfqgM(GAGXX}W`26{gYT&WW*Nft{@<Il+
zk;=yVkY<K_SA_GY6lDd|q3IEdMVr9^R^9G`Y%?a>h_Fvv3@u)Ef=+SD$sFR12Yrqw
znyYLTHz>XA&es^Sz3f`d>sM}TA{S28upvlN+Iu*ABj^$wpQ6jQ7D~F`gqoTq7*?rG
z7xhvVUXpbZe~ym$nCPAV`cGT11HvFWlsjcY!+#(-+tl|;0jJrYyj5T5{676M@l1+-
z^ZMr9B&ILpSEpDv5qT8ZHWprg!Q$0+QRlHUQj!5Z-BIzl-bF@f{i`Az!hQ-9S$?vq
z`I7f%A%a1}c#V2fCg=1g@vg((YY%LGcG#4~sqYusNFQJ%d(n@)un`BrZ2X<}{1=~k
z(O08`8|RsXY{??cxIU(z>u7r3UT1qZs2QRF^^FL^f3}{3b?~v<Ah|UkB|1|8MF|qt
zUnFE?Kl{=ZGb%bkwE#GK^$ZNW6B84~nv5iiib%A4et-JP7C*SyZOD<<kN<rwu0)pX
zlh<a#$3}4vbsL1Tic0z}Tqm@)%lEHcZVZeJ^vsSUOC`y+FWRCpe5S&OzwDoA(A)O4
zg2cW^NuBz$zUOt7>ymf@sMxYIP0sKVd-|BKSx-vgkCDE6+n#rL9Lgt&7o;}gvaKno
zwARPU{c8K0M2DUhZ{$5-xc{7-Gka_;jY9@H&qq6@h7ef%+Wx`HmO=OY;u}L@?7>2<
zyChNOi+bx*iaT!~*RzxRk>Ozh%kwj}zG*Sd^=-!&4q;da)%36i(2?f%yDo#{2WM8%
z-kFM=W`1$;lf+V6Edt2AZEC4}(9a$svWJR=#`paEkDopjl+dXIza9kC*_%QP8(hM(
zn26R5e69%nEp3snwMh>`NbTQna1HiRzqa{eJtYuna8~1aa-VoTS4q*D#q*OD!;e!J
z_be-yr1hl95%$-&i!O(@FH}(3P>lEq3yoc*vkVqVEUjjjuF0c7@!>Bjr^aTG5;WyI
z>=N9j`gpc1POOudvS7-7-5}`WkmIMUFTy>@*`MLXy@iXWD>>1zgD&xJLj0zAEizIM
z+^*r!Xn324as4uiT;p7rocB!het>DPMWd2)PsAX*XHX=g%T}q11Z-O(k$u2CJ?^%e
zJw^5BqRsM(OAGjpcv$B_2t;9`vKH(UxE-*47k2m{Us}YqsczU|>~yC6?AbFdWCEDn
z>&H%iAMg)1Hi|no;dZxiMj8*G8QH!?IcV5Q7-?xyL0pJA0hH!BMRd6!5@3Eg=T~zh
zcPJC)(tvtNFX>^4s1&o|^JhA)3j)8w-OLop%Sqnc)E~Zo_nr$mH<FVgTP<I>Hb!*=
ziV_lH7Iz8CzVb%$GLkCz<{>t@^{h#6+Vy1a?z$9uw&x=?uQ(pLp8vF|gyUSb+hP@*
z<K63!OY1Skym)+iDu=d`&eU=Z4NdOuam46_$A5ReoP8p%Xs+||Xhimp8$InlL;bO!
zomoG@=ASioe)z!zi}n`A>&1y@Tt%063zmn|TWBb~t!W)w<!48bcIJ%;*YQ@Wa<>02
zOieltT5(Pr91%_K2pPBHPqoji^p1Y7k&1j+Za+=dIM`Vwmi%HNiTY%gZmQibab9pF
zI0xAmeI>v^ckN*6bV#6TjyS5)JhN&x`ib@YOy|+0?}ZO|>1X$}2wSQi{`{#`c==54
z536ZpH5N!dAkzF~i;AVx8Toz+`^&4B36FJUDZuPbXI%OsP;~g3VkELE%gZ;1KibQd
z)^Yj1Y9!uX;Wy+n6Q8Qp^+G?djX%Yz!p)9L2LX>H-S&p1*hX#?=o4jyX}e#eog~=J
zv2C>Z`7_4!qo&y8`eMuUz}Tk@S6lsunnKLw1+~>0N66tiE80tVOX(oEmY*UUKQ@Nv
zl9{e}g*#2Z_x)?YKxAqZX9MdIE-A*u%*=C_$hI+J5f1G6y_m^jyJ{62@3Td`>0?Gx
zyR(i#syS=7M_Mx}uktLH5^k8)rj(y;v5{-|S(YC8NbL}8$lQu*6Vo%OPb;Py;Nh@6
zdYfE=nB;glya*02FJvz2SI1rVe5g&p&YF_9G$a=sUOnhK9CIU!=tp%~S{qL`LFJ{f
zk8Uq%qrE(*UGUqe%qhoWI%vz71g*z?Rebh#@{VvaWg0uWW88YzGHt(B=ASQ?rA+>T
z(2;Cz@9@hiTs>Z=K#3ZRc(jn?7X8()yGVmXc<L)oWJ##yTwmm+eQzi$jYV5su@kMZ
z@03kxozrwr^9Hdrfx8YJn8?|MZ=?93^MyafO_~;dHcQ5Dw)*DYN;mTqBy5rHeEl)b
zJ#Ssm%kRYwX+4>>ql?29|7_&qqbpMid$(8nHWb({@K~&MK8X5lpSyO%73b?M<s0XS
z5LQ^GRM44bZQ^e@azwRZ-kdhw<Wwx6+8KB}qV7=VkN1_7;Ev64jh+EkoBw$o>zP+k
z>Lmj_$8+9Z*93BzZj;nW87x()jt*gbbs*o|-j3#oiqjBXE{a9Feh@=@S%?zZdtOQ4
z<=a`N0pB_}Ve(?^A$G1xeeWvj*8$Sq`PD2}t>VS{N<FijpMp<?_U#&}R2r7*L&}$%
zx^oP#=f2@!2|AO~6z^sJA!`~umQV(rOme-?*YlMqIX0g2ql%E+qIZvCQ(khA4V79{
zZnl3WX{melK(0Pv>vMXBJ{_T{ls7}u<ik*Q{?ba?^!s5v`Lj<GXSE3FSV>QjKa2U2
z8G%sI*p~PetK{4*oJ=;?$n@Z?uiq#8MWM3C+i5#IjD8Vs*M()5H+CiER#IhaIvFJM
zDi|@bR3k2SM!ye!DYCZEkQ=@ifh7F9B|qq=5A944LuAyDt}SU`JvOg$M)m!*h-O>{
zE-&4KQtM-ita5BM=IZ5I<TGjG#AwwW69Z?@KNeilg7NoUN!GBKSAsFL9q&`LyS*1d
zaTJj^wmAoP$yb{AAxndzEnd8p#;w~H9y)aFQwmT0B<fOwst@w+S^u%1a4;I`YA4pm
zWIAKvBM`n*Uq)2YjLFCA;hm)rRiSwS`#Vi}!TxaPdHoBulcx4`p-68vMMCR`-?)i>
zwW&eE0XYtyN}bMi4We&bR%S?SyILvLe|uz+X9v^!E{n~(U7{F?DskGSwI(OgnxBTe
zg~LMR;_>Ey9fevcu8;tBEFFE5Q6eVkjh{XBL+^M3&-2k5`$f2Q2Fwklic6UMFTZey
zgxFsD-5;@xdx&~UuhN6%CQ0>Tvvwe!Cf_C#x!9d1eYayxD48hQu}p&`BbF|zZFXf%
zkz3P^bRwnOd69!7s*u%LtZhJ#3=o>0b&}H@MO@b8R2`#d&Sd;g-ekS?!=jzv6Zc5O
zsxz*#u2wz~xXBWmUoOXK<hHj$oqF1W9o#Tww)^am3^A8d`OSG;`PxH0naG`b<1EAP
zskvRQ{MEB3&!_KbW3yDPNgu9>$(r-k#oVd3T>Qnm&sn(1Lv2^LPWkYx=~k}Y+qAv6
zrYmnsmq)!9cy({`3#-c=XVc^HKVbCBx9aJc&8EV8cPI-<iAWpS|F*|mX9C7LpoYjR
zco0zQNccWgam#1RC(h5K$GG;|IY~E0{zV^R_Hv;iUX$x*lSHT~JF%<9(-*=&E~4y<
z{-AFbj3c{rmA2z;3Oqi+3w#WM0nFV+E4<${FPKWt=6tdBM@AQna21DZgHb#|*kL|5
zrBVyR40W;NSz0ptGI$MZnqumb-dTIzL4NW5;{46qCz+fe^_T#*`%PBn$oHaeU)@dC
zqwk)cTK3mY<tkX|!OH9!tnQat{kqZI8$3h$#e&{Hsx_Qwc4m5E(cNj81f0_Mjmj)V
z=&B3Y^FHY@P6+F5(%8xiR941u4d&?m<i9G3o(<i2+LaZ5VU%3*?c4jP-Cbl?zkvY}
zh|b!hAm;dE^?aH)T}iSvpp50V%5GU}M|BZSh05Dq+r-|Djm<%Kzbt&)qJhr-4vCC=
zRcK9V^e@xTU3g)jt{zdi317<-TwarH5%(v~PN4UjoR~Cq$Xt$UHoUbs;}9r!=W|H#
zIY#0&mzaIe3loE!aat#c9Hl@@3~lA?$B;$~)2aD|9INkj%h?62If>;JYybU*x@)js
zfdeKp#2a#pj=6R{2{u+q^rcsD@LTSz&+)0CT+6G&LcW)u1``qGL#BvA^OWDgh*!xU
zJ@n2WG$Y49pP5n`$M!GJ%=-9MG9Ui9)pb3@rFpIl5s#q_7+?<ev+`_7(3xxL6;;0^
ztHtum@J&`w)W7+OjKtkD9id0pU(%k#%HcQ?QvCdWZy-kcp^=OQZJEVQuK=&gwp~n<
zW~ojk$3~$Ef=*0EZGwH#54aACLvYcf{~mp$3mcV8`U=*^F@5kX3YHSyMt^lh1SQvp
z<L*{{sy^o2O!Yj8*fVge?VFjW{cq~+ZfNU0#YgP;#Qe3MSfIiMj;1^co*?fhH_VSk
z&S!=W?@Y}cp4^Mvk6}02Q<mHRexrgKC!oMhzN%%A_KnUX`gc^9i|qp=BN_$?I;Q&?
zJqjBMN)yAr{~5ir6`Sacgj>X4iIqJqxVX*v;DgP+H|Eqp3!&^BqxKwI#Kh5Ul!*B5
z_m{T+cGbnVdVyjCaH;V18yWI|vsy|Kf-(uxm2N%6r01x4GLI@*EiGT}*Dm6Pu)sf7
z|D9X+D|Z8o$!v9>)U3h{<0q<NpWb$34z%r4KoRe_98(!7ml`ZL(;rOmdle<^C6j(-
zQnC8>)^_eb2FfFr`uw*PXMYG8JYp&%ZSt!WwD@t$+>2sjiF0JVtT17)J%5TK$#2V{
zxBNw#mIYtyAL~M5BfEsJqU30V1+n>`IBBYv5q6dT^%=AVe55}t2e+YbhUH+%td}!4
z&XC7%66JkdotEAXz013LN{CB}1I&5hP$n=cd7Sge>zZK3+4=(zKb5|eR#~DUj`_*#
zl3an(%Zl6!9@&RteSPGdb`=M5FvjC`0g9Gvr$A3EdnT+vp9LPCPqL!r{m$c9eQ%_N
zF&ypQ=VP+*D?2dv_h&QZ-aZG}R9bBWiEG9k_`A9Z2khjZC^W00&+ua!_7sJtx_l_3
z#+|$hN^Y}I3O9wzJlS6zT|dy-kjgwAG<bC<h1;e@<ZD(lZiQoIRb3byj4~5&i1^3D
z(=>Na_3`dWWmZnuW}{AtF^|4O0=K=AS|FY1ruj|m>QUL<WW0sdY$DTT#3YJ(+k$rM
zCAoL?xYBf2jZ5@L`vaG*8<XP(!J*`3>M@BW@9cMPQ?%EXi!wEKAS)3JUf(#38xan%
z@Y!5C!jxEdTu;vR=cey-Sn7uF>#)BaEGOt@%zTkJ1Ev*bW+vdN=J0Oo=aK0<J>G*w
zUW<I}wu|naWwim76On5O8~26ABezIqKs+=}$(vuvUsY919m@o$BM??M&9;V_FHn$+
z1{E7xSX)0%?F~W9*~5pEp22gaC-?Abluep`t7#`_hhD5ss#_K60x7<Ni8wE5%BW@!
z;q00YJhn%XF|{Lm-HJRV{M<VX9A%pcD@j}}nIne^A!|-NNM!x^dcvNEhY7_PBshQZ
z4&_hwzb31C>Wa_MV%66w<KVv{6`QD7n!6==V7gLs`l4jdgD%g`mW{UA3v6%QOh~>e
zug%quPde)en{?w+dzwmu9@zh4eDFEQM;Ez!9EhbNncM&SzW!LE^~Z(ozSNt4S0y4n
zAqu0a0zob>8U)2h(}IubFH4MF`pY#y(AvrqTce))WAL#>II$v$W#@6zrOJG}-rM7C
z3)W>goYglty&Jn`FtuRYhvHK9-x%-68sanO?AGwF<Nny9LX372X4Cl>wz{rx6mgH`
zxV8BjK^^hY4eW~mtM^{Aw)Rq1SLo2Z&A*8FeBW2auUo(Fs%?nUm_K<s;vclJBQ~3X
z16B|(%9c)$x&(iRa}G1{`~ScMeC5A~hW)(5I(5c-vX1lYX~p>2b)JY9PM3dLsQtXC
z_XmeVud(2t?X-bhmg-5}wd9iFdVH<*i<cuJ>@2%#rq5lOxnMP=%D;?A37mKSgZpNf
zM*>$DQ|7dz3VN)1A|73&$&3^GOudH)_`v~4D5jXLJ0-JJ%%^W=sv8={@3gJ5rl1gq
zB*QOWqe#BdyFuhZNv0z+FV$`mclBci`aJcO`BqqiJ270!N$_r$0zd*E#~g#-pU;D&
zPCV1ST6jjzLv6mx$fEd04crYY%&j?B4OdJT&Qvc5(&61UJCb-=L_IG#R7c#QuIVgz
zaUt=ZX7{b1IqAJ<nO{ifX!bRp2d2(TS(L$;fhJ6AT5q1>K4pG%CWcX6kr&m>Wbop_
zo1SbH=!-L1A~GYB(VFlJ2ryqFf_p^m&t)@DI}0YrXd+J(Q!Qm$q*@iYDpu|hD3d1&
z=u;lpn)#W}4}?;mk?XFnJY0yboK@iZYva1DrSBH8s%Dr9k6w;ik{@BVilsJ{qWiO;
z_R=mYuKta#vP|CHgM$Ompa2cqQp;H!E-t08l_JxjI!e2d+8-YY4*;#tG>UA_NRTkA
z|HwhAPg~%SUBbb6378N8U|6<yR2F1NmI`A5kWNr&sJOB+$<|!E9tiUR#u0iws@bd;
zo<Vj2Ao5f2)hmE&6M>{PAeL8)$2)*0c%dg*D!Bw$qS2z5nk5$JgsWQ!g`n`iKNBSP
ztH4O{;42y~4_{m(QAQ5(xWlDQf91MCY7j6iRPba#(au61%K!>sY5hCZhuu+MZ5%4d
zHG)XtDirz9_<Sr>L|e4ej_`A`qW)spkn(tHF-N1{ucf~)M&yG@^K9CMY|W=i1Me3m
z^|h`P;@mB8AA0-v4bQJMOf3MXiU&~r#n<N>5Fh|s&$Ju?XAlGq1pfRWBsPQbNXH6A
zA8gI5ln631`hd{j=~nC+x-AsI2}%qpS~*Wrd(pb`DKe4DlR^MVsMXN{qHG{RFRyb3
zbms)DGvH;yK7INmi8-@$I1b6+7|;e0I^L+qDURj0^B40aLBj$AgODI#OJ=UlNlJPT
zQZGg3<6640Xw`b7-n(cd1aM4(-OIGDSo(S+zo;pLK{L|t1PsUrtf{H#Y||tnl?vL*
z$BP;Qs~1M{2-buz>`^Kd#R{EkWql+r(cr8QLe?>+e|nDLug!U%>r)a<7vgtH6^gku
zeVM#p`wuzN4T9GAHFci285P2Gbl&(kxF)8iMApZMb7GD*>IuSy5g`yGJqD^CA`+69
z9OTU%9dqy_$uQtY9#~((!c<oOq{>7#(gq0B!A81I?gv#|D6iB)a|9ZV4nK@md<6LB
zCvjRx^P;~vJE}etRSK{r`SJz%@@{|9``6GA8XN*yI{10W6kmaUe}MQiwns%n?*J&-
zq5CdYM#}-I<gJx64~){l@Puxlpt7HE<D1v<``Rz4CG#kS0;2ko_<bAvhsnyie1TbQ
zef`&U)ox`<cwI|m|Ji_5A*I2S6!K8IYO`?a+NT|lkEckSeC-!gxW!%K&&k*d{o6qo
ze@aTLaQ)WE{mkx~kH7z0@pG_mc5aFMU?Ul{fq>pq2$RYt^lVJbJpZE!0C9pEt4=|c
zcvJ)+^+D>z6_j4F38;S5*2W(TfxIr7a1yq)oZt@8xXaO-<e}=<=;|99>XVMBI!)A^
zLz7}?#uIx0x#){BYh6H;+XywSte65s{SIn(eO#{CY_w%PzpxOEp+4TLKFi9Lla#~(
z=%bngOP0o2|5`Q_ih*qX3N<w~8X>BwNdZvM!a3dlRz`mV1nEoSc^r9qQ=YKBTGBBg
zZ{xun4`V?pxW))bylcN{wN$H^2W*|EmHcAYXap!1UdP|NYaa%(*22#Hmn+H4xtVXR
z|1FYujy<&TXSKqK>7bX6o?bm9pU>?>#gS=L*ltY|Mdh+TiBk;vbn9e(;%G&*_!6f=
zr?paSSD=hnaysT7vrlY!`C9F41W=I3;N(5yqSvh?C~-e>rT7w!An5qOVAHL8kK_4^
zJtigwt&N2;9VyA{eyO}?&xx6@y^|^%UH9r+JQ5!c{V-WnieQdQ)W@6aC3whiaXT3D
zZTWfYGbn3PedOkq9y&ewdJUzgQ5SjNf{}D;^W4JT!bYQv*54@UqWni>pwPNUEM{y|
zr&N+!345?H_U9i^F<}QCyt$5O4Xt3zr;?J4=9C+ILrJe6m>RT4G*q9RtTRrN`?d7@
zz(~IB{sKfz+UD-t>Bjq@dJyAsXf%+e(KH<(+|kZG{;9eesU+p%f49%}(9<eIMorfN
zig5k@Xb>u>m<l1#KxJI;cl*|^V3>p54)F_Q?!O|Y2ljr-INIBDdS+4NG(_{Fm8TuQ
zhco3V+n+-3J<-YJUJG_o19NUZyt4F3LXv+Qx9?|rj<s9Pdy$1%GncrRv`IvI^GOY3
zY9`S$OYvaRp%o+#1U?lM6+vO++yp4fAlPUXvXET~Sy`;$TV`rB#eAI182SbVDzmGw
z`cYu3!n7Pe!tq0!BouT203(S2YsH3$LrkHBl+^jOTvzos?Ck6)!|nvJLcJCednq|N
z^M5mXq|SkUH<P!i_*04O1m8^d``a+Bh32W$n^4x)hgKAr1|qUrh9!^x;v@k~bd4vj
zfO}<3x)p=611tJV+&LP6I*n&MX9PInxf*DRLcL~xrRL=IO~1;~o*uU~OSAM^Q1kQY
zK~LQNY9u{<)D`euFwbOAHh^}-T+qnjvje`+p?stMZ^Q<jF<HF_fXF_F5IZg>WOk`5
z2P0XM%F4=~`+irq)Qvh(Bt-U5I{6q7P<#j;^~LV4)BTbH?;g=NZOh%WX1D~5&Jm}o
zn#ELd|6Mg+=;PEemTkd-{c@`v_a%nQvw`^qODBTmJ1TO3QbhsO>Y6KbCRrV9zBa4e
z8DKPs0(=7DzRA@hucUtCO9%mh0f<jh@LIkHG~pGG+8X}#3dPhDr|oA{gM}u80|#HZ
zN0T%;)ds4I8Nru09|z?~8IpX4@X-VsHny$^2lP(%6H!r>KBktRNOQY7HizxA;dSQ{
z_$D8#o_hNY3^+=_;Ya&pOJZ5_R?CWs`?Fc=I+G4xNin#PohaQ>W=h$)o&;wp$fJ$k
zbQtiAjinZJ+MNLsHX!cY1V!X?Y{13B*c1SA%C6uVpj+Lp5)t*OrnB2`=%7HLXo<sR
z-{LiX5TG9|4!4cUz?d%!fvv&}-U9BU%)T)ie*Q?bZfVQZ8=LMc_ag%f3}HlXMUhhL
z6$*&5tnD(+sNbpND~gsd9p0jppzG9N7XCNt_YRKIWKUZ^7quk(#O-j;qgob-Zu8vh
zbeC*c%{e(=m|`TQ|N0F%+&4{djk#6(BqH-`YaQ^?&n<xT41lk1b8{CTpC%;Sl8qH=
zhwp>@uw12;!``T!8Un1qA3uIv*lD?CE1viTv?^lkr#_NfWiSpFk2&JH0_)5c9A$-P
z&rtLyZmgOmHN)oBFXBBtV2ucb_HUy8BdGyXCw-R?0JZ-=p4K)~kqZARyemNu09m{*
zp_;uNwmW0|+7pvdQg7z2ff|(*^3M?IUWL>O+S%=FtQ0*>;rQ8CdxptiV*b3z&>djd
z3m6N2c##gQtJv>3w0ipKBRK(L8H>RN3d%sK`i=5z%9k}TG_=uay|M|)G}@~r?UCGr
z{9({#rUgJ&<8^||m$Pkc-zuG+tRq36Ki3$<IyYACjcLpxU%%JA4ZkB5dgaEA)+*;C
zSTT*{KhXX(E9krYjpij6ad<7naZ_Fj3ePf4c~mD=bc{Yt9p7rz`|nb?H#g7{B*CQL
z4f}`ZGbRDGTLWqE^)f!VZ!aW~)O!)xFQIWWA>C-aZroWSZ3M>|I|xJ(`@{l<KMQ$V
zA({LVE<9*`Vbg2GGb>xX3q>jx78a<vqg&y=10(^YPZicRg-*5=;j--n7ZYT3AA_ne
zyk$@oRxPt;ke8PSnbdyX81*cDkQPSEjsW_!8;D2YVZh}*i&jGh=7tdUyM~48XoS#G
zf5!FuFE4~io<DaLv9Ptx9(3S*+?LW~_`u!<TIx8~BKo$C(5FV1fsR$>HLaq7Lv9b5
zEVJfvVPp^XJ`pKF!<V4Ik5#!B=*+p*3(O_FhwC<QwzC93pide&`I0fK@9eGnTUYis
z3@*RR$FE;%%{}~XnVOGJbp7F($`RIC?nCj4chqkju@G+gMx~=kZ|bE=Ik*o1-*rzr
zE{g^5!zLBl8U5_WeQAz+3CtQvnSAJT325XrD*=#|1F|tdZD#gcveD_k2~bfq%*DuQ
zel-OUWdm3QaH``))&qX;qHD+GJ_fD<->>S;wUnv&D!;TZ+)D4=NLax*-Jc=dMOcS@
zs?wG)wo=YAlWg{GddX5hc^GR_Js9+FE`zrK>WUt#GabQvA1n69Q5ScndBvYnt+gFP
zS4p+)Fb=H~wsI{nj%=_(Fqf?^cK;wxs)>w@Y=zhJs~SN>-;=@^+UACvsRu);W*!@e
zNZ#N79Z9-idH=`!^cYg+X?dW+3_wSN*#fbd)Opt|axBM3=G=zqm|5iP))se3ZC}v*
zc`BfNX2nuTcbR5te2@D8gW&H%i)iL(YB+Y;^;jIrejC1}Bcd6QdFtQ&iX9>ld99IE
zpcjT4K`iTZ`Pn#fCip<V?O|d!Y6HTcd;lW?vR!W#u|~VpKF=BJ8n+dT!nE0iJGeih
z-uta-ET^924a?<j*Op4YwR?)Y&qtxPh`K5SDSkd)QG@{CK|<?T58P`Syw0~4{W%Ht
z1o+n07SX?fUF^<OtE?%9CY00TEjbhx9$5{TlW?B5FCFfaPJps5E#PqX_w5T1b!Jcy
zO@U#4y8fW?_ab`#glr@CFtRdek3(^e@|UP5I;DRFdU=!+o*1jWrx8457Zy5Wz=}SX
za-pV8$52=0<gzg_-kw*0nfrkH%VKJSt(C*QvHz06khaTpbhCrLi`iia<uacAIw=QH
zH=+HB^FBU4z@?2qOw{z>uQeQ{D4lLc7&GQi=}7$hr}sL)n0}?izDu?<;PI=xFPA5+
z`NyG3sC=OI1*la<QTG)VhQ$h2^j?p8^J}l4goTn{Wsl;&`&?v%4M2LMg(fNowuXj=
zA$r+bPqR!z%D@s=9dAmRq@uH&&(oKYXlJth?H5o#R!a@B#TVl&5P@{!_s8~&ctqqS
zs|VfhKdmTD-=DG-`(nWc4FfgxS<A7nn3x3yNL=5nXJwzA{<q*X5rTkqvi&pFu=xhG
zWH#UQE1g`DxvO}v2VEw!#b?{SasnWJHXthjOPtp8rgPK};MO;g@ne$dK~mi}>?2c`
zCbN-3`KL^7iM$(6uwE_84;z+jjbdM_-V0Tqram*u`Tj=^>H1pAX#Sr2p4`kxtiexc
zOejqDc6j^uj)nhmNGY>!@bU7>g8=BFt+22#i)IPxA;@4k{MlQoTZcW673~ivOFof`
z5bK(S)f2=e_KYyqMexG1VsK8iV*16>L{d>LGG7D!SoP_MPj{zmqfj@9w+rLwGvsSg
za{CP7C6siFZ_l#kn*R5SyH`{zYi>|bNQ;YOp@4B4tQ@MYuC4{QLM2<1fpvp~gd}u5
zLs_Fz4?RkGrktA4k)uf>00Y5cM%-KFNnsinYT7#qtGkM<OPKKh0b=8?_BG^*yzIFP
zeC~QXCkx9-L7V(rEeDJ^o&Q^JGeejv1Sqi(v!DC?<%=8mU#Q}77k=Th;|D0T2owkj
zqM#kQ<lQ**ov8lnW9W#g99<APB`FK2_MDn}i9$dnuH7dDgaR!hy|d7z0Q!LcLe?uH
z!QT&qhoBic)SOztKKI=bv;i@X$FRS~fj+}&Ko7lc`+9%2SQzLz24tamYAGox?rYzI
zRMgoP-be50X6dK(uQ>Xv60h{<S}A1f#CPd9<YI^#Nk8yyHyYC9v-O9N7?I#}i8=nH
z*=qH_<$NoXdx(dI3<jGc2eV<Hs07JT5L#(D?<kEy2#14%(_+p_2+DPE+nTR}`AAT|
zoA^=j27%>v(`eMN6INo?OPY?M;Shh(`bBOB9xXTKNP4Nam6m^O&xhay`qW<dTG~jC
z3JV$A&91UR)@L<>6>F+;iY*TZMUxsezgZ*T@4O%D_8lYi)|ofdnpFMgC(u0Z&3AMg
z938D}5gR~_z1xxjyjts4r?8xhi_5_(AY10@34|Xf=H)(#hboYKOZek~j^n@?#Y-*t
z9b8@A_!TR0zb-b5%{0te*f+@fA8%k+p58}3Oy)Q0>6v@Bt~TI5(lZc+N-F?Xynnr0
zSa{`<d!mNAu)V07I{8F-?WNTWBT#$=-%vLK#<NYn&hGkB40L7Z9iv;Nw%OdS%IPD>
zSG9M_{cpfQd5&(R-_}*q6w>Wi@|r-|mL23vA;_OtUN*5sL}}i$Ay2jk^fFy06`>#|
z*}WRi)m4vgddk(nu7Q0+@qU&x8toVacpO3@2P`_)uoMYRO*=jU0u`JJC_adPc6ii5
zQuBub>f$|o70!O5GJi|>l#u7nZlm(w@cNS#msLNGg)xt#({N|A*kEHCH#YFy+%3^F
z^UUw)IynONR%WGrqA7w8lP3EHJ{rfRo|W=is&qu{V)y>FfFJow;Q{v-@HU>A>>DgE
zu}0hFdm{Vw7tQTyy`-8t0PY|{rOrj^_af+cpJ*^Vh7c?7C`g-OBoNE)P=s{;5R)xJ
z=Z^zhEwh~e<K*;Go718_M8T)rtY||WyQH}gS#o0pv~9)6IWMRB^1PMP8B&UoXh=mq
zKC%3Fmn57!woXfbkXC(PFp}^5+r;mOw@lDlY&lhjjPPOXu?D7eM=4PIM~hK%!zjbK
z4ITW3(F}*7+KMX|ZKtLt?Tf04%WAEN+dV3M&wFT-CANKaet@OmyIhrjX&>P@=RY?=
z{3|f#jO_XJ1{jy2lV;Y5G<mI@eqy_`>|*}5MGRZxqQ|E{8|weqbOPl3udRC=1e<YB
z`0RjSy<+P!e8~PA$IkqGBsd3N8iKl++9$R}D)o`kHClk48HU7)`PbWG`Mt|-kKX&O
zVyj3_3<X_5LNtzK#l*kfz|LR=+*i$exs?82XG5>pzB_afv~aT)c%!_Ho0QXo9taSl
z0PK59#jW{y^~wKk8IJ22=*-}h`m4ShzlyDy44-cp4wto$oba3TmoyP1j@7uCyA(V6
zCxk1cmq3~LVzmKEv%3~oe!~p370%U<BRJ`ccL2K$J&r|5(^J!pKXT4IRdm|l_yPIM
zEoLqWqjUGWS}O$1_uPsuYYGrGB@@oK`~F)4IQrc$2xV^D@KK_QioWwzUP;gerlUmq
z@`q~DSs??*4~vg#io-sPSq4u&SP#KWe;jRS`ljTlngBw_E5ma+G{_SKYvm^j8-M>7
z!-F3=3&9xX>7Tz;ez2c)(R?>z;zykih}O5|0S?q!uA7rh7tQzYxj`?%9|ByC2cQ2r
zXpj``o537|mh{b4XwKu%y$dZED?Q!$E#jTOeP44L8xfqULq;^2jh`BO&ygkKj+rJS
z@#uPaNEiQ+#R}3i`q-RTvM!!yn-r(j+rHp$iMz~i)mwYemCw7%I&*txw7&g2g;Qz0
z!6oDbL8F7wJK$}5ep<~9Qc~|lbLT@mEFc)Mw&pOl>3?w(l6^)-MvukCL&v$GSzfB=
zs>lc9*%W<-8&$TU*jsttQeu#amvU3-i;PZ01i`6`-}^zM!!G|-(KQ=<YF1K?qPY|G
z4w-gk{WV98{lne7Bc_=D&34&c40G)B=g*tFI7%QuU{yH<In(yQzjhew#iF+ym$IRt
zS>iV`#fw(y&jEfsIWhN2o=o(kmQkLLQS0ZPGit2pvF|+pO#`7JZ|BVWsSZlG2N3LC
zynNYg={cnMgY)eclb5gG;UviP?IB|kBGJ0c`aXo`sKjmT0xt=P{%j%h-*fQ(${?k$
z{sxtxV}XLNS8gsF%pTR`Pbr$%{k$|sGP$n1`zlbuXliQSlMuZ~V?5j2tUbgNj%#-i
zK}|15FpuEA-L`0KZtdr5zNAYN>wlWmaJROZf^5RbB`$~;WsiNN!SLXJ+YP^ciUc1?
z?KXtFT~}8(6zy7PTq3{X-Z+3&UnTvM+f}ijTBzT6o|r9SfsF?*<ab@`q-SoRzeoLt
z-xesB&fLtJBPXwG=7!u%um9EX5!l4>YL#O3Xaz_~$y*Ez&A;P4uhf~^%BqX3ui#^1
z?bo)&$~B0q#$0^3_orC>K0nR9UL#c=f2xk$2j{0Y+i@-LVP7_wURxJCI`<+oGVXsb
z3QMmWZ|obSoz(b~u862Myy^72YkcB@H%|A+j(b5lfCTsM1Y&^DWW9&s84gQ1dik1o
zM+b$v$&EMXnCG~jHP&LkSmu@zl)|=}K8JP%4NbF7=iUB2U#9LA$2BWUYHwTf+YB^b
znV*OX{5$Lq_6dJG-xS39^kb;sJ!9g12^L2Se+dXF_w@Dx2pUyjK3;YIG6c26B}Y#8
zHw&><k+UUs`v+KBt8qk*CVFhOQ@FWzC?U5dUilzzGKe@K#_`084dQ?+)l4|2&R9^#
zqADs}|Ei9cBPMEmpN7`yDp#3#J>9>0=i~Da!ljQ(ZHQY*D@7DUq(k($vQCL^Xz<@!
z3$JqysB+Knwxcy*cyFUxDJFV7sQJe?B_sp*f|x+}vf#~dXk6S)cXxNjPe<ctr%jLq
z|H^4X;#ldKf#A|SW`<12PglApjK0a0)>uNNC|=Ivm5(i+h*<6KS5Wo)n!9fVHsc1X
zF3r+Y$n|s{$pOp)B|BT!M+gmJpHjZrbw%np<4E&gGQhpKF#4t2is~Fh8lnI4hl9ll
zb&&R?=p0eo`x1P`!|o_@QP72jR(qtj#ID0)F9W-DwR!#0+8J9K1ea6iV@g?rd!L*c
zF7>sOT)pZByePoUU`Enp*0NPf#rYF&3><xQDFE2ij43sqJ*xr4H>zZB%xMuc97L0D
zzrQjIp{BPA!cWarb-;W`&062d&^1adqFK{R;v@4}0q|0j@6MsBqBX3>6VmMqzQ|cO
z6iO}6{iGeQ`Uz2p<;zi_<kh<QOHfq4z~kzR|7v+s^V~}3^CCC6gWn#R5GHYY{M{=h
zYsAEJTT4M6ANg%i^CYVa!f)@N($F%>yxPE7KqPe5j}~k?O)sDF_?8H2Z!HDWRBVsQ
zZBtl0A$)Bv6fLXbGW*Vo$5t_>b0_W5ShEZusw+zPA^C-7Wx^t$`qXWh$zcoKMzh0E
ztq8-&xso9Ne%uN7(fRW4TZubSs^wvU&T6@=89h8ucEl2R%de|<lsLfgON>B7)ep(?
zpwt_8`UK&?_i2w<-fC8I7l|wm*<^V_j$Ec<LGB1;-Bog7`F|!Lvkh?G-5NpYnXK<}
zKUfKm?JnyRkZ7vSx>GEB>O*o)aOh|>zhZAw`0LS*(DAU1SyHBFZWJ^LL2}{c>hD4n
z9tF>5W{sja<qSZDE0G}f;Hj%tRRTSCid?MOYB1OXNfx2CPmarv%#ha@q?dnuz83jZ
z){>-1U1Q|9!f-H^7VXjU$<VZ218@Za^O5l8154hs`Rm_V4W4}s2>`6JS(e5XXjHsx
zCMy%2^6$JVTj>mTu8dV&qw6ffbJBm21ulxs^+P-aQ%!>6>@`=cxKMkfZxM%1%3pd0
zZnfQYYT@1jiM^Mhm3xfOf*Io6BWcz!FCR0yAJQ`}CGxzDOS{-9CuR}<o#F#-F#fH?
zl}fw#WT!iC0>`P7iA{0e1O<-Gp732J5Vag1+>wY#X&+*WZQUYa(Ns;J+nqbPe=C6X
zNhBX)V_Z9n)~*&3nvb~!E#DKHq0AZrFD|2^!iJ_Y#78yKdnqgXt@>=rtnCpSvYu!Q
z(h!knAy*2kSj0!EN+iPqs^nj+ujeSM*xVYyM4=ugPg$03@662Kkq8?*bvqXQ7UuYA
z!Ht!vz>)Bc`bjER&WoAFYn=?Vw4x-wVo=G5pC3tF0;Z174KKT$Aul<wGry!Z3~WYf
z_eh}Bwg3cA<h+(i<ebL0eH#)WxXPd4DId*-4CF%~PM3f7>@(!KGcOU|{i&aC)Dq0_
zUG-kXeLeB`KfaMiZ4LpH{3-H@;_1+U36YCJtdR3IWHE`Igr#~s>>!Ln;JLx-1%~mE
z72q`PdkX#3Ahi{VLm3$qln7bdg<5Rt=a4M{0r;fDo#kaT69dq@_(s>$3skx9wY{%|
z$_{3jOoQp~spTWNOHf>hVZnBF<F&N3d37Rmhgu?;-e&)Q_<?8QFC#GO+=-N#6m0mp
zLbYA{5i7z{O`l;G)wlcvOgpc>zsqv*NmEkV;p38j*8i-t{X7c(qbK#=HS4Ywj3ky%
zjb19Oj+fTk|9aB57I9zwbjVr#aM`+Wb)>0YiG-D&{wXxscR`}J&~jFC2Qqy>r(wu~
zvokZmwJ>S$0?CdkEG;c<X>RTYW)#ks$d3>CW>!2Dfl1IVX0tY;#&5TFujCCh?gJ_^
z8l89CSs&Nt<>eJ~;j1nw-I|YnVBqgXK&{0V7QDxB3YF_+D1Qw?BfF$P%5C0gSu{IR
z1Sp-W4X%$@YeQ5v0L)Ruq@<~^&_pA<KM3WB_)a0c9>Mh{0u~xcf#T63$ZIoL3ej%Z
z*FZ_o&AnCGWD$cbvVCsS7)2)E6K~HP%^^381cV6EmUpdIDGPI!X8Jh_T2>ktWopbh
zd{=z1Gnl$nsM~>)98EuhsN4-&66aPZ#?DsQY8@suZUHg~6e^8uAZuSN@Okz0&Em20
z{#6SP>ASJ+$RCy@&|!WI_4fhDuKM8FgAV)-M3qdpZ$Ab}on$Ilp{wLj=7P75o<me2
z*<U)2j%fb=1Toz9@CQ#I;T+aF+XIVjQ;*NR00chFTPJ1ErLx^6N?^3|EAN|fR)i^6
zBoqYGp&0n2Q|Ks@%5j0l)7v`?v9Ew+yA&avY%uKqU<O4Fpmz6s|DK+kTLt*ix}d;_
zYeQrR`Anz-#kaDQODlbQ{$<Ei8Bm9SVo2g&D-qRe{L~Gnq|^A0$!#hEdk}nu^`>K3
z&p{n9S^Jvw>l6>;vH+Xjc<(HV?Mhg0^6Q{^qebtn8Ec+9h%j}!R7oHGGn}2E9;;+g
zwU!rQ36xudCQZjXOYfY_caT$(@Ds^B%c89S{D`AAgBHo=&(eYFSW?WQNsKOm?F<b=
zNj4086y+)ps3MZE!Z-#eKsmBRUG-qZ0ln-G!y71vm?XZbR|jcK?_;YdS%e%)O8#Ok
zXO<!{aiv3An7#lzP3!o$Ht^*2!E}j<tZg+49k%8q0;!z0ySIuSGq(_{XvRZRKFMM5
z2`~(D0Q(I9J)o|ogoArD?`#ew;)0hAIE9AtyE5};<*QsWF#@F%;~-X9cC=Q$1GT<o
zU<DcQA!f2(EgTFr%!Cl!owzzfB}W-K+Fivp+uGVv>`js3G#e=qnHPk7if1-1K7POJ
zx%1obgWptgv<~;D12|od_j|@FacZXp?aOUf)h$YIGm~~`{u!=xvXcutUvo36^m4ny
zHZ8LYk{wkQxb}^<ed%`PZ;`U%7>?VL<mxJp@oHl`1l*rRoC|lC8xA+u?0gZ;MUANH
zl;iB{1TrnSb~%}Mov$)ij*A~wANI%;Ch;#nmE#JnTBEP*)BHvncOF^q3$D#!aOuq3
zbi9aOUexR0KzoJ2mOkJkJQmG}<(%`JT3*iZrgl*c^jVQR+8s8R3S29lk74%{2BNZC
z8fh%*!Gq@Zu&E6^%MT(DZgUsqF%@;#lcrw=3CP$U1{OfE@LIr56SYI+_zUQWgnzdN
zz!u-61Zt~9IHGX(Pq?=la2d07JfU`eZgDW6VZLL{i^6+(S`z+ycbG5TOX)C9MyuwA
zA<LnTZvaPb3P)*H-|eT~18W0&a?)Q~7x~r&G!U&sO!E#TG*bvVu_Y33%(tr%1tn!)
z?MW(9i2Ww%C)HD2FZ^8D(s5{=9zD^TX$+6eUAg9FSsSO!sLYD{815{`G4~E4Q)j(P
z4*)Ha0sxLQ;My?mNs<U}Gy`UlFM~i<q84a9^g4{#X-!Gxg;EOdopd1C64Ro-dX+|^
zrL7IQ_eV7^FE23ZhIvI^pAB@b$$+MN((hvP9FR4ED@g5LsC!5cI0^JSUx_`N&W{1~
zqZxK<Act`uJ={+-GS7eX`~g%1$@a#F4<EW_!?f2HEm34OYYmZwhrXT)A9A*pnbLK2
znq@Q1LdFak#12g6V{a6$Mc$bg5D-L!#H^a$g_4d44L)DGuJ+vlmg9j;lG#;BK8hS|
zt;4cc^XJgBXZi4rb8FW?f;X=<>n%koS8^;LeFhTeE+RQbI-satWSS0GofAmO@i}b9
z690fy|L5C+C)>TRsfJ(X5ji-&e3_7hU<@Wa&(Ycq>DIUE_AT~Nmw6|I9E@Ugj6BS`
zjmBFPnvJSMD#IT-M3Vr(<L);CEJgfQggl9V><PxeJ9j=>Ehgu&h=aD#z9b1kKWgFA
z%;<IHY|V7@ssnLqmwnw@8#K-%o%{89op&s$Qiy+S&|W|jxrBW3JIaeYAu(@m`Hdvp
zA)Nm<8khb30*N5f82C|`fv%MW{~N;#^<#qUTj93_`yZ2+J<^!Zk}p0wWcr5hTz?Kl
z&&s;uyJ>r=p{1oGba|U)%FXT0aXklPR$C_87BqJbPMhb*e+hn5dz&SWEBeTAw{Otq
zRd1*9kip~drjNs=k4tx@Z5A9}gkwCl3zxp@M<u9!cF>6+98SBYf&fmkEY-wha%}x>
zxzm>C8%Vu!myR^=VBhGDQN1J0^K=C<g~g?zLX)sYe@Z3<L7btDcnL61w{G2f#TEIQ
zJv=xlNYQlv4H1=gEz|5y*H80Gh|%L9cN;^z0i32MnHGCNazgUeE+;LCL@45s$zPRx
zrCVuz$i1ga@^!?8X9EJNOV`ssDH{F0FI@Cf)bjhv@BNdOivE(~L55)gH~q!k-+@Us
z<1tQ!Z0P+kQD;OsDpec*g|BSZ?hf^^^4!}ZmajJ@$_hGf%7rK!A23UgJ1#Rz^Z+T5
z4bVXDtPCjv3D3^{J17Z;+|E+XOZcUSZUUe9<ItJ_O>Tin)W~t(kNJCA6{>+8z_)-;
z!Sf6(Y*0>)`J;*GYWJa^Eq5kDk|yiDm*M&@1KLbqL49Di`dx09vj)Vjn~dilViqrV
z;yaMC<b)PW7G=U8++%ezVq+C&Ijb%09bd}cOX-?rW-q)DV~>t$vN?^ZMj<d4*UkB5
zgYGz|n84QAzVUG(#<2Wt>SEyn-E4oe1DXCr^0YxEyI|3fJA}l<$+FRWi*VWfu5>CE
zI$C)k(F4n>#<I1uHmVW04Zs^F!OD#ov@ByUNZ_@+Uh65_Y+<@10rjs+G{h{yYI8j(
zHdgyfaW%{!h5_mDZDEFKLI1}RK>C8)^?OfGjtT+c#;o@JF&z76`uZsV69Gyt@(_bd
zD8`bKRqrj90=s;Nlf|v+B5J-pfz^>Hx)Hf=)M?xMM6aVHt<3MC-;isQrc>+F{%?cQ
zO+_d1ocZEm#`3Z~Uu4wz?wEg>%556|65CMz26_A>fQngyT-c@Ju1v)3#aN5pUI0ZT
zLjL`ySuxyDT40J|HCe)<WK*Fi+MoC&z%sn}9&o<x03cB+5C8}q1|3oEGQTM<dtas!
zNCSqr*-VnlfTj8py!X}YSMH;Z-Z!DJYRPCagf1^zaqCAB7r?M_8g@OzCr3C+uJoQ`
z`@H?W!jDX;q1}c1kv>CB=5`Q+SSKNJc=(NKAw#kK7Uj?9>hFBqziI8iQ2J|+Jk?Il
z@2tEpbzF*(DzK@&@H#o>*&sh%>$NM&#}X`Imw&jV&MJB{Hic%_ipF*)J7a5?B>29C
z+gCCW=*0CTdS2L#CcC=2B9@Wsu<p+xJS%WHafBOqKXW*O+dLaIlMJDNviXN$urT4y
z=7?42BHC+Yf9-=ViO<}JmVMTe5x9B5$(4fgp^~+=b-W&(OhnRE+Lp26(pwt~Dzku=
zQ?Ibk12|I#+EWG6hZqW~hj*TL&5VxPVzWrdh9S~IVtWqsUz--kI0A2bFR$GHJ3V5N
z)x-9kmNShs#{OkPF!v_AUmJh#Z<$Db9CjFZ%R6Mmmuu$lLOHJDd&>Qox5#?bdVH&F
zYjUvS`N5F8hesz4D_Fxx%V<_`BLTb83t~R~6%JP6?fQjphLv-wPs_MWf5oMxAr<X%
zTim=`ccH}MyxB|#bjewOkV^rH(VwPV8CsQ&WdJ8DgmP^^uoz9)2Q?=_2&5QvS4IJ%
zm)Ys*L|`K`f#@5qyCAA}8;@AWZgg(^#2JIjZ~-pne(+OqM+eg6Y0)MNvo-$Qgmi6x
zo%PG6s|;f?g8%}|?6JHPsaRWB=NtVbkjR|rh~i*sH8AP(()>QtUsF@VL|(X%_D4Ib
ze}3I>AW*wnVP&tAY4>P;mh{Ru*RF_`c3DN`rAwun4ot4b{U#~#0e&jdIm?xr2SfVp
z5lT~fpj)K6@}}Hwou!*hIYS9eIe|u+h!oG7e@B#av?{N4I#)U5luY52`lzkgWFH0P
z06roDmcSt(=r#}&pzU)5)@(z+E|FQz#EV=VH()9A#j{JU5N`|9wEPGwjcu2T@$6E+
z@$k*{y2Atwrr=+TTPB4j&$a~)26N<W%wsN;Tk}Zi7G}`1-TGCThd!XF*;@kN5ics|
z1Kdh(y?iqYE^+*bqM+vR#6H6E;1KfE+@DKJ-RmTRe-611bp{4-NY|8a!z>08#C%`n
zsp1rxKWvy7z!}TTyoXQDkv1_7YL1RGU!L?qSOA#@DfBy1S0B6kxOl|U7kvOHbTF=>
z&J0}~u$d!|^9ycjkzc!m5cj^EteBhUd}cmY?EHnBd!a`&^d>SicxT?T;Pq-l`Azc}
zedl4R0z>qOU2(Fs5(mnxek`Lu;L2)gkq+t?ex|>LITYH$HZ?uX3Ah-1O1|fCD!_8*
z0Iamo{?E%%mm_$k5NS1=aygYM5H(~GS6Iz=D2a<BUP%>{duZ>8d;FauAf0<l-cDO@
zFUWJg`TacbRq?P?VF5)=L%j0v;WxrjRlcuES9(H!dp$Rj`CezNBEQANN96DU5(OQ_
zFYeBkC3Uv^q88wN65$|ympbU>+KOV)ee}3dTDt7vL}WygB^nn}d(**2iTT~&4vmeC
zMWpWrVPIfb1A#LW1B1k%rj;Lf{qp8vA77RbZ*+t|NS)X$FzO{Dv9%R$fh#rNQMWBd
za5#Dq^<4aJP`3S2V?}Gw<)=sNQg$xxFC1xea^%LFuv<(SPO)wF&dSx#={|CJWNB1C
zqZzinWX4b*zYyv4%{Vb4nV%XxE>{HS8(8+=K%Xh@PmvA*X-iYn69}=o;~sgSGX+L~
z&&eFisu{ttAI#EFu_y&z2QexT<x;e8fS0@`jU5;GZLl=mQP|Prm6fC%bghTw%B{Ic
zFA8fU|DY(O9+7!tUNRHqjr|_eXv2x%=Zs$9*eKnJwFsjG^-h5%vUot*lk&XZQ2xnD
zY-Q=IoxkQs=L$T#7bpwkfJ&ly(t@6b=FzzGUi9JKIx8VF4iiT)#u$M7h%_(b;oaLL
z%jjpMj;%0~jS*PvutqWF#SmzR4es8&^Mt(eg(FgJU8lTV%fQ>yC))FqSs_Y>Is4Oq
zP0ol?LQIJt;#4r;%I!5<$s#ktDp>yFces{ymx+~RbyP6buiFhNY3f1rqlev34=`0n
z%k$AohC)$IDfeCtTaJA^qbX4}WE$GLXl=lV3W;iNO3Ki#c)=Ml*zL}3(7d~KX>p!V
zA?RkhZXD4XQ!;Oos9Mjib@OZ>?3b8;7dbHmFCvj*#^U$C@t3}Bl=W}dFUW;dusl)^
zyVaj7nI)2RQMh@n4L6KCdCs#UP+gleaapTWYGA_kf3Xir(l5IQ#Te{gysO}QO1U!_
zJYKjn&nY=(CHXt(?ZEv<N}~PWwV&7^8^)gfKKFVBcKEovKOE%H?+z#b-WZjn9JfSr
zyL@!3Q5g~2CRt|nISQ-0_^t0^|IM=n_Z+J$4jtPqRdkXoOqf?S^>Wm+%`b{A($YbQ
z>gT$ITS{3*SqfRdnEi)3RqosED7C$C4zXoddt2^L&&eX;a!TSk_$@&_Yq@oBY2MB0
z!r$jU6A~cVKx=AXR@$B+%ye|AkDDZV0Q)$d9*81~ZcbvvH~A)6A2BPW^mNBUjTR#$
zGKh&G2jSs=kAZ%Gt51jCxplz1&3~D`X#aP=piptIe6%HT|BV}!)-|GDdxM&KJ)Q5r
z@1T-rlxuE~kj(wxcO-ghWXG~m2-{_Kj{p9V>!Yh1e}4n~j{pDD(tZz8l+L?<4mfdO
zyMEAZwJc>;6%2Dr9tzA(Ol@Ipwc~qQ*%!xxmo0vp<ulx5dw4gz0&<9u_xea&@!#Wh
zR4@GBr_cWPDT2~@wmpIAirYM^AP(Kt(?PSAWkj0H|Btw@4vYHv{soJVBBCgwq)NB6
zAY~!l9gB450>Tn+L=lnh22nbeT$Tpu?xnlCJMW<1@AuyO&wcJczvs&XJgdBSXXebD
zb6)3l&df-@QhIH$oftv+=go8iS72DhSknijy~9gBsGUwfae;Wpev99j*O)>XXZy-E
z_IIfCjpSQ@q~U%Kj6;|=-$;MuV8GD0HkQzqSIbGIA3rcdV~2%yFwE_V(uxsF?{QRX
zA9OtYNQ-0N;rbUG=bS$_7VE<!&^cLYdH?txo?6=FS>!MI+#EYNbGkMceb0tA&b+9P
zw3s#praxg|aWLg-=~_ro#_)@lm!5yPI;`3cAzWB=I7Nj_QdI{P<<@A|G|ueLc4pZ%
zQ%EY$BeAPy;tBSFrSIsLtjjBjA?{a?Z>AGfTK^g62fE9q@3<k>aj92*LioHSX4J)X
zs1K{({!l~YmGq@%$Oq7xbj&1(vZx#p{hjXHKJG+V{HxZhFpCrWY(>m7fswb{Tw$mB
zAxRgU!OJ%mt24-kQN{`-n45pv@D2F$d=7opf>v=RC_>&C>&ZVBJ-~pd;B=M;U-`zL
z#~4U}$TxqCg2S5Bil|wWD#}fVaF&Vb-yJW3rVMCOoRGqQ9+XYLyLrVOr`|#+X>>KV
z+z-QKSzZ@SWq&K8b*U!^?}T6ZkEO**(`?xs(h&_Fvs3|)8r@1)5F1Ns1x3i}6o;a{
z!zc%yMugyHvlxO?WF~Eqlm}T*gyMFdf)mlt(=pbx6s6!$TmC%FDkCQWy6BzV8z1=v
ziFg(slaTR|NzZ*-sI6zZSa>Zh?Jvpj4zgF|Lvp}K4tcS~Z23rAbl{`n?GEt=7e<v2
z2iF{UN&UB3((7#D@J_@SbVc^)?q;~0r6m&h$_g!!sJZ!VpT#QPxI4UYgbQBJ%UhSM
zo6>uPD9qJTwTYO>9cLiK^;~9-Owbj$ExgrYwZ**==O`8+bbg8_^gIz0M4R|-(!wke
zq|&LR2CAf3GfljmCl?%770bq0Lei&CpLR`I1CfQ1Vp~1F4&AuHz|P(?Vi2k(tos8k
zk@j5EmY$JA!tJ4tOS_ANX4@*S!`%FAg;;@V7JE#4GCAc>WmGwCD*xRknql0A+yBs2
zZ46bryW5hC0K6DO)ohAFln^+E)K1YkvK$^q8*CLgmOOJ*K-NAjYfo{<f|NI5v-$Z2
zxHSZB{?9lVAqh}Dv%ND{@z%gl5Ofpd(LH32Zs6dkq)j}{R?c*BS?s#0Xz1c~)g4F4
zLXl&mriin7Oeok&qLZY+Mv4dLufYf-#Ad~Elr?Vu2V<@z^SwZ<tfy75%O<9fLn~1(
zD5JaFf8o|&>lb$G@<~-MlClM@!)Z}y2<K?pdR+B$nHXUjMD%AxrKtk5R$a?CHY_oP
z$`myVPEL>xb#tdqO%l5Q;4XO*@?dYr<<p->fAl<=ypDy>q2a=1ed@Z+b|MPE0PeJe
z+jh%alT%o;evdnkS2J-<1M}a<9slUpvb|--igp#PRlpvEKA86`y173sBibB9LlM}p
z9&9eXM*<It-m`!{Wq^bR_!600S-Ubx_AdkYRf5I9z_3{t)iPkd@4&nH%-LV7gtsJ^
z3b^5dvQDo%Jf^=h{p?PPdBQk_VCs=i>~FW8<V16AL@I#8xH1a4)$(}>7p*Xbv=_ff
zAZLe&{SqgyKkW!7zpuu{np<p<?jGh{yI&yyFFHTyWfj+sW6G9hvD0+N{M6yu?S-Q!
zGKlX@!O|Z5dDo+RK%$`qO@Vj)%B&(A#$8jxEr);FmmMq~rYqQ%YJSfkNZ<D(&am>j
zb1PT9T#Mn)yPiL~G^xT+mYZ8?zCs~#^}*Vj{NG+s|E4kf^9l<X=YJdO|Dl&4{{J6u
z<NwtJX#daGf2$Lh{_lTS#Q(41|KVSN0RC&8fc|SO|84m1{Qvhl0sU8JAM5Y@|9=d>
za-5|)1RVT}vY8P9+tj2%s$WH3J{Z*5_7z!ZT;>QtQhnFK`(Nrb$reU4DJ3O^0m(3}
z#bZ$E&8%503(lfej9}5?1gDV#B7{{DWHA29KVHe?3Jg#iAQO}cE(<pWd)@%U1=tNr
zkm&_mflR>$UG)kC3S4lbF=o?9HqP?bQ($3@Dw`lSzkd0$2sVW%gU?n>7AFLs4H#Yk
zQSCRzwZNXOaw-$R-~p@(Lq7ah|3+0zu8i*hMhW^%)jVfQ&&=H3iuu>y+)r4d-3ONJ
z9%>e;XT!4ox-Bd${$D-2lL86g+*Q%wONI`RS8#GO^hj8IJZ#p2B_eb3!su&*$V_nS
z2({uP=>;Ke1@Z`=m=#~vDhZba&DC454JUHVzK+x7GEII87Kfu8Txmd%3lg=n-;Bl2
z!40I^g8l4yIdK9FJhq}?P={F?T3|7x`aGLNjerUWH!zC6G`R%<x4D)u1Yn9foj}3!
zS8y?ifP4avDn~)N)@#oYJv3wE`Wl0*+@J-wRfPMkXFdZGvGYysO3Ic<bgS_753L>h
z4mZ<;FQF000n154(*EH8dGo`j&mbGzhVpvDf;^&ovdv1zC>-fjg-fCeko|UNdjpwZ
z<L)@iUknMP@l!r$(!NCVvSXLLBp~9(po=qVK2|0s$;-28_xA?}Rl!UEa^5MU54Xds
zlQ6!1$*yH~hB5?=%jT}>^(M1uqYvs~*05um3Y9O&R}|pa^!sI+0j2tzK2Yi%p$?Dg
zonu(+l_l@=Dr{pq%X9mzd$6jcDIW~DP*V|+G_BgN%LDZ4_kRB-kH5EW9q-0+80lzG
zJV@wL*P}muWE}(-v*nFDCHE|FGnzcY8kf){?Kn7}n6iFw+VO!qhOy7=a{v?&AbEP{
zs-kHa_vLU9pa9{L{Iy@|c?YOgmMvU$aE;Anu%_GS=j|;gEbI=JJiprZ^|{T11BaW0
z{{jP@oide%1?blF-263%vyD{k-F{!S?=oWa+mYP~l1dQ|+xLcZ%P^^LPH;67PyS*$
z*VEzY1&Fr5D<1#-ed#64(!^rH5pHOR#_s!HZGwWO$sbw$Do&)|c+G}+ui3@i;hrxq
zzCvky^zUK<fMY!tJaTf>9t!zHFnaLJ@sZ|_nCBmB7(8n!Ni#4oBHAt7T{u|wff*`?
zySTc}Eohibz}^HO@o!vkaBzf0MbV4hf^Of=lC659&J=!&@Ws{5a`B`4bV8$A`TE|M
zRTXS{4V$_*iggBpei8HU6Ry2*c}pYax;%iv_>rWEy)Ayt?8X-qmAq#|yH|^>P_NKS
zMY<gG;;lf*f}}^eSvO!y>o+HuwG(x$WMoAgie58492?BYJMPQLEKPjMF2)xg4<%B2
z`7fr9o$T(_)JIp?mC*PRC+eK#>^dy{QO9a52CZsm!(tLLvh0O;=MKHhO?RK}V*zLT
zIfYIJUdlHWD+Aq!l@)y{>S3=N#|F#2Ve1&B0*BRp1>6Xe;SJovgF~CM?cvsKv}P>h
zI8A3<s$9r<0+osW!X4sXE)ggQ+f(Iew7b+(o)v9YGm(cd!>9{vuAlG)2mZJiw;Azd
zdHkm%k0g`T`<Fp!3%D-K7;G^zov5sUIZj=#(W=Iy4G!j7tXBz*4N44(d%3s}gY)q5
z`uq-Cl$}JYp}Nxnv#Cl&#w9588o$V$r1T{UIRba356%YYb3rj&dR<(IGONyKwPP-c
z4}D;(PU-Vw{K8Mc%VXVALnaskoAGZ*CZY(8)~WU7ZWhy#qHK_gk|@S4UTO|VSrmQk
zVQDfXX|=aT55IY$I&kM^KmHq}O8ev!*Q3VQXQhS~yejH>`ord>7h?yrb|t%2na8HW
zy$#fmAX?_E^!W7noLvlY_!&PeGF;is^Dn5c&E#Yz_9qv}NS4@9F0&ZQe+RB5px+<7
zEWH6$FBP)M*k20`F0fq-jPo+CT{G6b5KJByPf@nc9WR?^A3LsPMOZE`l=7wQyeojO
zT(}*pHA+)2e@*eT1E)463tmJw;Bi8QT4qusg-TYPvDr2Uc#K{Qe?oX+16dR0&3Is%
z0sHmfWTO7i#^y(1<uN>nYH5q5`xdhv{up*Uai-Q@#&G$CJNMeiM)Nj9A?w3ada+}x
z0Erg<2KZQ66V+0t(8K0tv%hAaP9$0Fw??WoS$o{efe{0&um)BL$_b*vvm8%q(~G3S
zGaY{$BTZvGK9Py3w4uaTQn4Z9caqn@CspiivPNYqquOTrZr)UmQ0Lk^$a`TiS|-yi
zB|bj?mx9scXPCPM!&<i~)f~3pO?!6r6hp_~mYv6vuA~rLc+$is6RJG5CTOb|{NTP)
z?^@nAMVVG=BlOq39P$7(?n#xL%s4P5+x|ls-I!s6@GVZFLN~H|+!t#Oo=t3I5e>HE
zXX{@3OI{1ly{*Rm5h$Dq0^^{Nm3`I6`=ny>fmN8bJl`i>LKk_hHqKeTq-9j}FP>EY
z-|xZ7d<SeMUwnLGmL_!Ydfc@+Jvd2SDZ}w+;jNtgrIQj?-ukt?eIkB`Aj^Qi5fB3{
z=-<2Lp(4N4J6_?WeAE3esy1Fe5nIL}{ctie4u^RxlP0ks=9eyQ*-GViA|!SWD1wI?
z-)m{0c0YLjQJ^2}PbaU-CSi}Ugi)PU4YnJQQ?dxWfrM6v88Z_Sj_sEh(|Z^=!AqpS
zx156|vU0GzqHU&aC{H6bKfkZQg5=#b9OvBEe?Ch#3Ixz}J<Ju(>99T1V~Yvw_6aCg
zQjfoCm(^fyemQX^M`>^gua@P8%ozLI;N9aEjCfNU!`@N4Lm5o7NIZDAEt|)wkozzG
zkS)3VwBV}xpFHbTD#3Q0;NGr}y0+;Z7uu^dt|)`D3{TIcC=Dx?00PFWMH)Z*gRlrj
zK@JJ@>84LUoQztKr*WyTVrSkAvi*)+BKXC~ClaBT{MURff>8Fs%6u!rwC@K=XsAW6
zK`*%J{&($y0PVn1pN|A|(K7=v1tsLD6=*w3E7iVc6-E)Z*1ac}KWXqQoNi*%I%+2+
zladhH;Jq28R_fLt5REpfithW~V<dC7a7FRfpJsf0-HEDsEB2~giEV1_#oJo?F&u7P
z*+Qp3=ez<K2A23&iuGfpdeQ5$yM0RtI7ZzQsB4E$XQne0qYTR!E%P`%+uJc?O_%VQ
zSBIZ|8FtxDJ}<LNcA%yq<Z}SSf@znCWD6d>_Q5BYTe-2fw^FP^_?TeYv*j=07IcUx
z!@?qr$r;Kzcz>FptIpXQN#DB*e8REL0L)X3+4Szil)IVlwxXhjV?yQc9`Bovh=jcN
zlWm0%_7$2HA2Gh)x@ypkOP-yTeO^N6NobYq-jU4ybYM%Czuai0T2e)Py_1?E77QFC
zvD+F3vGd89kl^m**!15HJiVO3pqH#0nXksmmI{ub#l|qmWN4mw*pa<*a-SjbX=#dt
zujR(bdlkFYCn3$V^`?w>AfI<u5MGrG<qa5M!%kC5ptJ03oIA^^$3n~`(HLa*3d`Eh
z6Xa6Py#2Ccp86k2f8YesnmAL_gf^vRvGnE7om7P_dV2BpCPlw*)}A+R5G=>PIa|i;
zwRDwT{us12cW>3b<&9%`_fCYN+H}j|PgxXJ^$)NzLcJlNBwUow6gi&Nj3htu$8^fz
znf7fu{Pj6M2mrg`j$1uvYjPG}alDjMg>pEsB$JU0?p;z;)KB8$v`o6}ql9`<Ab#c#
zT(S1HzLSZwG~U+mmBn<0mbRki0dr+z*&n*mT)j=Nc+zA}9L8A+UW<#0VLG9leu_L2
z$Zm61j5;-$k)87`Xj!(ZgUsGOqT~BNclI5k{*0Ihh*p?aBhMfvKAv3(U^Y6tkrdaH
zy^~>k1iXO>Wi=`~vrd-b7WWd7*z&zGJ|+`~*5yYI5li^Y?hUsib6(z-rjGv6DrJ*(
z?My|qI+yG5tlsiSVnswowrVic)4SKe!?8CNAk-WlqOiC>r%|+=DMZ8oBpfjDLRMzf
zj*{f$a+3d0(!#tp0Hp1$Snc%?^-_y5XS`kB?{CMXx@lo>@q)y9x_56g3)sD#Fds?B
zb3Ry#UVwG!ea2b#JMjXN*k32x{P~$ydO0{Kfc`19QgC064&*vE_|seWr)h?v><@NZ
z+d?5;rK$~~AFMnZT~mjSd(gbo*+Rps%k_4-PA_f#=Z4CkZ)j)cHdZnd*VLCvY5e*!
zOp|KPvM=>6{7@mf%sIi(Vj@{370OhT2wypgIL|b+H+3#^xs<0c{U37L$!`07H2=qV
z=*pSP-uhg`60(Td;9Ldw>+-8l-;r`szjS|cIX7DR#a4-KxE_4=r=?lWG!oGcB^dzI
zuefq@Av5Gq7X40SF~khnMU66zx0lp+uUh{4HBF^rn79|^Gp>$Ow#rt|{AdWJChgj}
zo*vS%Ds5=xl^vvYhIG<VzSOuH!_{8~sqD&UL#Pkmb`MlSzFFg$_p7q*7QdxjVD0fZ
z^Xfb6rY&5p*fNzv`$uHpT=qZIGXRZ?<=(OwqV3i@Te71_{Y2Pjqgd?C2#VpsdVDk0
zQp!Uq)G0DXiIz&0<00$i5nZL$*H!L<XhKVbKgGOEYiy;jr>D#kJbtbL<<`rv9xP|%
z;Ak=zG$)x<wGZ@ZLsi0rDx|a_AzY5DnU||aA;O`b=V}i9<vTv-%AP%+y@3~CnF+>)
zhuAVm*4Nkk@%@h>h!+L~8hLnNa7L=&xuRi+7}(m7?SRP)KOCjGHx&(^vA?yshhfxq
z&AtjMF7=WLp-vxd$pN6CT>s!=vR(}^jG8iUm3)rj>5g%AZlPZ4zHrSNh4$96R~y{e
zwB8dC*uh&35$P%_VRmciA^n~&j4TQlB73NlMK(a!Ih6GO36M6KF3I!Uk48{xVHVTa
zNF@r<$LuQ8yN}6S>W(xXg2VTQcdc3zYraFxOLnF;u^ZH4GprlQO;{K50bACM4$G<q
zrTOOEeo^f0z0}I8IMT6Qsa5Al+01wMk&8;x);tNIZLhQACSC58vkf<<Mc3PxlaqFM
z$(()$<stw#GKe9TdpZor7xGxT^CR1jEx|eOv-60I@>^>Q7_(3dbf!Iq0V*vLt7Z3_
z(qkv&U+0Km+S7L;9RAp<(lTYYe7<i1X7`7rULPJ__gZ!gW2@yrd6kND@gFBF>;(Kd
z7XGl!;kP^S{OOCt7)2h=q_&wa8rjQA=puP`_CiD>3%;_djTE>Km1+s#KJkhxd*F1T
zEZWSRODG8Av>=Auv<sx=$MqKcv86~Pg55L<`q!c_hk`|P<B}-u`czyVU$T~=0(Z_j
z9j`c%sVfl(zc+h8VyW>xzW&|K?Hz@XP3sa(mbI4-qxX}Q9FIC*exxY#SkN4;$DI5z
zqnFMRi=XZ(SI$QhOm)vez!^H@lM@9hK&JA=F_f61uP+*$5uR6H9R(oS*!zIX2B?x!
zS|$dvo(Kt{8=9d}>28n#17Ejm2S%=AZN`Q)H-8_ipxyYjbjR!GGfGT~&Q-1~wc1}h
zb+oC*$2ISU?PtFQ_w*>Gs&mzITBZCJC~F;PbcA>S2N8XIa^<WP;diHzmaa=z4hrEz
z?G8@9G9Af%mEHi%Za5XS(TGpE0Ki5oGIfkBly9UXB)C3%^dS+LXCu5+MkCt|w{Nm2
z;fWBs@C`~84ULjBUgTvid<yFmxl8^a>GR{qEgkkv69by_Hro{FO`lobKYAO6jm`SO
znk7O49-iIhJaP7;>f>r?>m0;jCz*M?$9X!O$pwB4wwO=QXA3(QUGQGP0q;60!uWUW
zCx>*UVtTGyyCJ5^sSZ&KC9?Zu0lk>cbCnZ+GFCfC*?a`qxo|<RU7+?6Hm*V;u>Y-X
zeh0_MK6B2s&b2YU&SM3nM-0tJlAH5LQ1g!66s}evAo}{z7~qd#RPe-7pvyvfQb9x3
zs;NqLh6N73Nn`q3G@^N!?9Q5$bT}1Pjz*)xoRYD9rb_Sc5Ly%mo7~#iKd7AqG*K)3
zGRug?aP<nTy83b*j=-E1GQrTOFt_i>PaNATv2wnx6r>@O=<Ld`eUDn3d@8_wMCrPD
z77#5sPhSslS}A8ZNhhadtfXBSPi&RLa$f;Py!@3Gm-jy^-!MNz9!Gumdzp@DoGx@2
zutZf}T-a?Wo$Yq(Q3|*v2p9~G+~37eKo75#fi;Q&hB(8}Ixrs&w|~N`1d;P+)n%Ex
zazmdgVa1|8<4y_@g~o0jHY1$oKTLIWj#n0KAZkMcMz)t4A3M&ZNTll|f=#`yWrBz-
zgNT?ZM3OH!9GUDDVA|V!O@JK=5>zdFr9EaHc4_xJKYPtu<)xWrhR>aRP`r!YUr)Zi
zqPy2Re)X@`8yyAdiK7?NOjXk~pswe^mECJ7msaJo`~!Cl?2S(6eTiU}fb~fzz2{PX
zSaYk>jtjVougyNL_3w2CD7lD#A6ClAP+?t3VI=Mg320Va92_hTLFg=LAJ0@j-E*5Z
zaXs(5Pu!FF-g|z?<}kO^*HJQxJ0=6?`f}A6YxkIiMvHo#Y5y;iCB#e)*RkD<vg`56
z#&9`Q6)y9LHQ`o&QR2?G_b5oG6DHOZ_)u!^b$~=46Z_9D8xTEAtd0I6ia1YpZKhw|
z8VrfvPH41~scQ8h8(E$C^-7?0X(j=}Tn6~#@n_;iLw&p%UtvoaCBOw>s;XYIh-C)U
zFC9@cqE<KM_TmL4&?wVHJ7@nJ_8Td}nV8Q9mD8m&-zyQD72fKQ{cf=`nE+D4?0uL?
z?9EI)YB80@Pys6LXcA|;fmUs|?0|l&{<IFVmOdXIFi#0q<17>*Vmx*gg`^pOVz*GQ
z?_s4_bV0RSH;FfEBmn8teMUBHe5A#bsa<>g?rT|aU!I};ncWUn#Zao3tu5UCAZ7+-
zdg8y+5fUE<Wn2D=mrZF;L85*YS`Mxue!Pi!R;xvCWxrhK=DOP}b@hQE%5=|Pt8@;<
zGg_Jx&pkUgPzZ#k)@CSGR;2ygbm(WK6wO%Q&va${3aP8n+~Hs=!r4Y;YWoR(EZjfU
z^6@jEm%ZpU#r*aZxwx0cTV!wgQomWN1WN{k&=gYGu_DPLa)xEJ#cE286)FYqo|2(U
zKYHQS4K%(-AFMHpAx;CRfk_EP_-=e$!vIt;Ys76f-Q*>p@!@)Z=_^;f^=>VjrN;>S
z`epfqjo{tYX+rV5A<oqdHzKJG-OI(5b{n{*&j7g*@9&rK1%+k-++?C#C72)j1LDaN
z>-ojjjmWf;0dCQLCDV$oRe4wB!;z(<Tb4Wr2c{Ci)n4Z0o^ky~Xj(%nNgrl3#xgKr
z==}t%b~=U1xzf*_mXU2{jlG%g<?4b}l~tBzuImoCM;^>>$B0SH@<duCaEAy%!Q<G*
zfgG3?rnYw0wuCbuj#Gg_8+k+A2^?aVL=fwKxgxNRI<H=tDpkR}O|zJ20*DhqOq(rq
zFK4RiJdrfDpl$)l>-PHKj9)u=ceee_gVhafn1Qlr&IFHQ;-^p1xqQlCc_?SVg;Wb*
z2^PaCK*teV$d@gt!k1DIE&*u%_WS{ArO<JPhRg5vz=_H}#A5qyvBY<=4>rJUl16#+
zW;EwPs&2ptAT`1*h8ToERUpUavOF|<5)4=e`A%eTl9y1VVhclSV_<}K2-wBcZNZhM
zHX|Qr#vQdgWPe8Zcrs@<i>ZQRrA#*<uE5;WKdxnD)l_q1JsG(Zg=8zuf{_8hwAec_
zA93tL3GJX&3yn`>fRQ(z-`6U&oDblO4tMK-1B+cbny;neBJe(nx1==7RilD^7m2HA
zQ{SDW99l`tp)N=hLVmg@BT%KEoNo*7bdoL=0F0?>Rw@y5#AQ4Sb5urUyoLQA$m~(u
zY>ALWFl^PvBrltwqn>IB1GSdMk(<S9vZw?#3hXQumWN*>8%~pRRE;uqE!J~>cO7Tx
zt-P(b`s>hCK!bq5C9EnZCw`?rSyJ!qp$0DZ1wyZ(O*bIsw3lD3N1}p-xw~ZyIyo3v
z)V)8Jly~lE7JJI<rf$3VP2Kg>^L(dKKKq5OB=>vR$c>03y``mSsK_J};litcf4c2u
ze#h&$(XPjX2exw=DVUYkEqX<0%ZduSH(XRmTL>P>Z=8~4#-#?&zBsE`w=OB9J`Zg;
zLy0g8o@-KMPh^28^VniEbGbVO6|wOLj{Ybd)W>@sC>bM#%VVwxpzZ*Vh2p%8y0W)o
zV>E`4<fA+kb3h?mCW41cR$#m0Al9>%tr6-(4rv#BEi}&XHse{W*l_DYcM2%X=Z$h5
z&&W`Voe}pd?ER=_4}fcA&La%FZ9(+bI%VB-niD9%#w^8HWUhdz#?l(fb&Fc?_RdI5
z37Q&q*bD?NTj5t)v;leRH$;&|c7?WN?~goIGpOBPE1I-SE<}$<yajid18$330~5Pd
zsJ^$|RR(~`r*Vdonqm+;pPz3E2yhrg&!3~jJDJrI7s9gwhxFwmtCX6DxOqDdYdZiN
zv$S{eHN$Ohd7P2>a>dIUUMqM1w94=D76dEDTv^&Fm@=bftsH;d8EI1x(RyBiL7*J4
zuuby1#%W)AigTjO2Ev}xX6=T>FnS`}W)x-W4Er!v=0=8YNWb?WO<9Q&#r#wbvY3!F
zuAd1%!<n<@vsW7%W+y6G(h}Q4s=P0|AA-`^824cTCEXxJu6}f0re0-~13N){1^jDP
zU)ybJgAsl9DE&Qv8`b4XQIL$4U!{R)l(K}1a{F>$|3|%q0l+TrjcpS~S6NovN#IYf
zE3{JrCvKcYbM;jzydN7ZC{yR5JE3@XIYO*)kg<Q%l&6qNC5C$AF-Fb^`ObKYd0M_i
zAmY`3|K7IYNNP)<r9T^WRczI;MSP{|6mS?JAlN(DkMq1TKI08}lgH6owbf|<@)5HO
zX^1|nDHvy|4g49ysSN&!FjyQLXZb_H$x5rPU(*zpj>sI0M^o$YkrkCH?z3&r1;{G8
zt>|rU9rCZY*z=8%Ozxi9%}lXcBiYW6h@C#fpR$cTs$H@xy%-F$W0B!6rVd=OYjOIt
zu3!L9tQ7fMe~;@$ST4+aX_%uV!cE6BFqaEPnu@x4dLcEf=^w4({A9~T2k#qpD-1qX
z&6R||p6xiGg}_oU9Xw^*Zsd24|0jxyOrkZgch1&~(CVE3-tT0M?qeWB#6}qf@K%Bb
zLu;2CEGSy7)N+vPRi#zr<?GP|0Yv1`gQU7XPO_~@y^mt8^PZNOkHNkizToKjJg1)-
z56+1RFKM!s-YSeQwFK(IF&6mGx!_qncED;1M--VR8d+7n6HqU!ic(%(7l=mLMa}1E
z<h-GkW}p|)n)>}JL#3h=x%w#kX$68V#fbyGpoVhwg2kL?D_f2r3$rA7p*WT(sv}V6
z{>qOSfEllI=LHMubHUQc(+XFwc_xj->v=2oi~8jtoOy7MRI}^nR0>R>FHua-D&kM1
zXMJ_!>nGbl3-pSgmC^Fp&;(<zVMr+YR)OgUcUw~FwSbq2XEwCE?LX^5PD>$Y;-%sn
z#NLaI;`WvPAV4^VLGS$;Uk*HlCh`xav|KiGsnYfCP5B1|1iZd-H7q(R-QI@w!Hmeo
zz`N9ir7l3?)ntzrX`!pmQ~c{_A7~n}ozSsAk)<Ev1_HiZ=pvG_Sr=~m$$|*;q#JvS
z4&yduMW*SljWdOZlR5%D>8?ZTzDdQqw=IOi=)h(xW?aZhc%Vq`YLu%>;q0ayuO^cH
zV9I_ep^>HZ3-;(2-iZ~!$SH=8!K}5KE{ryiwK3)6H~gWHzA?`tMdR)(*VRYIw0J&L
z-PA+2H3pWj6|s5ZO?)jYBg*djY5#G6bNf6foX9x3$Rx;brv!QSW}Py?KxHF#K~O{_
z@-cCsk~}2-O6=wdtETRSr^KPNbDqJuX1P}EiU2w-RB`<L48JJhX62rh2>WA}hx4=H
z;rGbzOE^k0O<0b$<#v2>Ia}o@+)ablX6o7`8`@7Ia~C5uGNY0SZg$KYhAPt0g|?PN
z#kMU#IWuKaF;ORndgUDU@)eCotfj$$)f!c2zbnBd)K+IP-k;`)=R(5?hzX^P6*;w$
zBh`cCV~3{wy#e}hMz8K(`}6(f;!4mY;TMtc;JDf^*AuMIkMV9N%g3*Eg*P*wEl$sW
z*O{Y{J({z<(?Gpxq<YP3njq?UiKzSm+jlENq0gWE+k4x5{T>gpxySup7<Ax>rqABh
zAcG!9g-MAbXkZ=ET$Z_gy;WRgomwC4>$ocJmnYx49)o;`y*Ydcu1~ngH6VezX?zvW
zZfG9$jd*$e`EaM#{C+fHz3G#uRqk5dc}yjDMI>Q@AwoyKDh+;9BSQst`)8P}R5j<t
ze{iM=L>U<x>T5J}^wde5$gngdaP`%ZVcC%9^@`zRI44tVMYoOb<{t8)0?~qe&$o{D
zRKEyXSVw6{E<@ONISS?v6HZd>tX-I>1-?RqZvjO9hACKd7}M9i97DDO-?dph@82VN
zzmjTHu=NZ7Q8#s3$IkG_sb>MM`_xDE)SP*_lXKPkzS}#AQQ7U;W)ET_Iy=%--OBGm
zE%9)o7dE^en2qxKG)t{z{*K=giM~h1*mH!0k;IQSH5(m#t)?X26Qyd1?X335@7LGu
zm^B%6BqweC-VP)E=cdlZLc$Z*FFcwfh~?WdE_=-N1_qAm8e8(I`@Xw-4Xq8w{aoK$
zn)OcFABAh)HdsmLAjQL5NWh;k_|SL<A3_{^Jo)GE_iF)xQV1KjMj%i_vCbVmK6koD
za;>D)+NuAx+oQgScydd0a2ZyAgNUT@R4cMsZ;0E5(}9MDX{hM)!ov$yGrw`7_W@Tv
zOf`RKdC-FBl*4ne-YrbYVf$r;@vRsk>g4x)`)sN+SJZNT7UN9nPEAxJ$_TMEU^@Ie
znC)qQK_ewGJdoCR@$C5$|Fz=dRmt(?Vef}w&01Yq9{Ytw)1LloH(ify=J%FwnI5|B
zaF3r_q<5UhSy))-=cQ;S&dw+-?DECO_OMq9I%B(+^ipK9t=F?rVn=?QzD<be`>vFZ
zaj}+!`@x7gGWrjnZnPUr96>&hTJ@Po1OxAL&~S`ZVX|jg+$EnShWOl1*j?&C4aqz&
zk0K(xb$w=@c~)OvKW^Thjw-YpP86uES+f%NlHh3dHf&?`-bBkRi!m}Nz>h2E+70WY
zsK@1FT!BA(*c(_6zl<$(44B03@K`>d{v1G;llCF-qXM$s&7<>V|49>sIr1m$-cL#4
z?@ys_Op(#8Z+Z%*-5T7pQwbkW?|pIcZlv=X$I?tE9Sh^k+4`$5LL6}sSFdg&CD%RX
ziI0YcHVq^E-*!Ydw|>H&B_!h5u00~f!-WESiT-LW_`UmtpxmQN6kJfxQF-_^rkVcs
z*4Auw;g+MC1_m0o{1!k<Yb1TCRfH$_+bbPod^0!>w?7DbVf}>fIN3~0A*|+6hojW{
z4RupXvvoVaOp!Vx!v*xN!{~R3@@y6(mfvRy_lOttb=h&HO<_|PM;Ide2AIF!%v^%o
zJ3N<hG>#9>{_oK)StAn4DBpw9>v@P>SuD4~az8R7sAJi@z~r|kRW*gqy{t0Cua3qm
z(ZfZ(=nd+eEK&tZN*MXe(uTpjF#9k{_5!DsO>czX)y<`Jc&wzP2u;}bY~U+CJtE|}
z4`rgxrQA02wj>X%lvYyAfr6Sp5(=grLA^>==GYx8c@P@9XdV(6)Z8aiDV#39RrX_5
z1Ni?ZSYmHqDb7r~U3N4D)mP5N7MPd&k;jdK`z|<WbGNgKhL20hD2BZx_LEgbfPC_F
z9yjw4yUqh(WBEoz-P5yU;D=k@s*fJq@@+G0uxKC)+M#YgaEZ9v5uPb3dXDnOj$9}5
z(TXD_R&`sKd(O#$Zg=%KU8pz>bjk~fsj2ae{7V6W8sZcEv9b9ahosJ(+CQSQxiUrY
zvQ=C)DM-w?eMvqkKGRv>R+@c?DZdx__L<I+hSj90QR8rvS=%@~(AHp$(Fs$n@|(!<
zVvyb!4^NHfKtR&h*RE)&@&h^d!-sG64LQum9`=0yzGBDktT+7#4Vlp13j0p)>dcXm
zxvsK9adFD1$Uw+(s<EZM;BQa-tEDd7&1LWxuIY#Vj}t4teiBk;VRs?K0Jbu$Qa+XK
z)~qd!)lmIZOLVbegUExhaLl8Nfgrsm!dp}XH>+T`7EqrXmkWbcl9dZ4-#$->7xYqm
z_6o=zBSd2A%Sz9<0PR5iS99$Ss}W&6fnrpr&x={d#$se?B6fDsKK2crT8SmB2^)j{
zYS`!Y<TxJ;N0iIQODL~XQ@d;~h{xp>5M7Acs6Bp#>^NmF6uPlThwA!06huhrg_Y~L
z)2|-Vf58Q2L|V?|v6cN=T_HDy2di%m*pVp0Nf~LKr#mVTvgAF-ga4??`t?8J-`*Oo
z!an?IC6nc6Pb-aD>Z$XLvd{=p(QFIQK(F{(=vp=U2P{nKg3Ih%RhCMpDI7+9U!^CP
ze<pqr6WkC8aqo0*aYQ~M-WoXBxH~wS)6W|f_2~X-?Z?QNNNYi^IO*04-RAn|;tUTU
z^_iJ%Tt|n_R}B`UuM#z-WiGh0TsJJ@^uD}(2d5Uq#-)5yxj68vACo#<jzQgz?^o8J
zBg$Q_ewxh&7MD!FF;Y&#S>pi%VZi~ZZcn%tm|`;vN)?79#%<zH(0Tp(;tMK*cKa3<
zCf%j=3a*ZYFzPC-&9@`UgHh3&E6I>GB1^roGSX3zOY_%#@8(^NvY^h+*v6WsO=W_;
zf|hGYZmZLx&|97!Zt>)j=4vN<<uQqhEkCg<um>HEZyTU3IvmOXRypT$dEl&|@Njos
zdBX-vaJBof5uAgYT&Owh%e8Dau*4vL(yJf9x<?@eF@{MwJa1gfW$D7oAx$ST^6L?L
zGlOn4ku(n<mQPd^z+9PD$H++T>nW4Hc&mv(P<G1x3xZsGRjg7+W9HHI&ol>nu^q=P
zpcNv=M~h>fX0OOp@40z0)5`1w`4i`0VOO^Q1KTF+6}pHHxfdikoR157ckEO;qhsKm
zs3>l$bB~rFM2wCOy7r?ahQw?Fin&wSzRXXn&7(3sfJ|C2v$ADo^XB#P-}#-Vuxqyb
zibKOJ%h$Jwj($#Fo$u*V6qYD80qhZ%$A-2Gvj3dOV)#n+?F;3>nCO0j?6^G1h8?nz
zt@WUg(@HaZ4jjC+hj1>opcS#^Lb&bZ=tas_Pi5`ty@$>IPz>8@@#8nR&KjlMHud$R
zzVe<Pho`RrJ1(ypwaXda{DMu7eNLRZLOI8%HNCf7P{$l)BR*%l?c)<9M&;6v#@TSd
z)os|Xh?S8tYHL_~5f#&&lm2pD`JrfFpr&Ho)))O?T&~Qbkr#sn!0JX1sTa^p6W81F
zZY@1_<a?<Ek(e17h<#W2d}Wu<3Jbf#e|KkhQq<)cv^v!2qi%X``O@_Uu!0I}8hT2t
z85_}&B*g5DKbR9@`C9uGjEu($O;C{-oI8n%>5KF883zq+Kb-u5qsT+s_LjW=9Uc-L
z-QMZ2nS-(ukusO^J<=ZOXA-2S+3z)SGSF-{5WNQd2VHu3iEiF{P5FW#V<^&Cm4mfH
z`G>NJj+GbMj|(i1rnb(;?qPMK^0LoY-&Y=a01T;G?!-t-`E1B>%@9=#X3x=~?)f>7
z<CCulD#)Cv&0BS!aU~#baxEDOj16lak1RH|xxxNeOTFj=<|Vw4h6=???)DgRpK32(
z5y%<wGoEiRdoN8uuHg#9@?!BEXX^-ldUo{Vz0ud~k(RyHQH_hSc2Sy{sQq&$;xx6T
z4EanOPL8>@u*?S$VLFz&w(n2zW6;=L?G-lxCK`;2w03FG=hhCGPW92bMw@DyeL*t2
z+*Uv~ysa7&j^-vHnH>ww=J^R7*{unko;NUltOIA=_rT$7T+e@4DFh&5)0e;U>`a5H
zAd!jfi=Gr9B~X4mkZZ;X@W`d~6D1PMg)GETodaY7<8(8-o5mi=LHCBPUkA3=cURK!
zJcjgE#m;&Ez(3kn4-@dGWJS|}eE*I0P9;tcE|v+uoIp?56bfuno^ckRkAtw<*UZ1Q
zx6<-*WcFWUH8$P$U+(vm8!HV|P!%;+yiLzLL>I~sJvh|e(0ZQ-=j29vX3MN#MdWBI
zk<+{Std|I1lIG^w2M>Dsn!T$Zy7?}Wekmm9Qo!E!;j(wt801+UJsMP}3=^dbi7-3k
zr&lPj>05&anlLpUU8$*8b3y#R-_USBLdCr#=xHT4cR9fW_ki}e0bz!qNh2c(L-||u
zL%fG2TqQF*fs2c#GotP{nO&l~Y9atCu3^s?Wo8gbR-Q4{@vr(U0HP(#{AhW80O%1$
z(5JKMEgrzcn9T74Zl#%|IYk-kH#1}4(km)xAlF+(yeLIc<juzQH|Y<6L8)P9FNu$Q
zt6`OMWADH!Ff@@jwVh;%*xRrFKW^e&;H`(sUE)`6bMt!rG!1y`tws6xS|r(+x{jXU
zbFyI`$2+LJyyr&yO;dh75#a2W1v87$TsUorXf8MRGqj|$2bv*BvzIHn2P(a;pgv%s
zmyuznh*?<6$GwG>%hnBRsUGhOJ&(87urX7xXMONc`6P=$CtIdsp-hQPMvA0Si7Zv=
zY9l)jdz~;t_SQHH1M~Ijlurj&PA^I%nwngw2+E!6>mwgE5y!6P@L3(%kIKsZ$Amg=
zg3ddprD05wG&GPcK$=NyOu(Ml6y`NFV#0oZBD2Y}7+fxNu1~a0won!`L3Q@{NE}YR
z{wJb`y_t#rw=6?FrY2UWz1-=imYMIm`c|}D)>3VA!~D36CO(LkEK)xofqZZ<I|n@j
zb8Yu<bT~PAIltutW`SiDmg>5$f1Er$vVwa|n_FgUpl2Xlyc?k+Y-mCc^u|$Dt*5h>
zOnc;IXMkop6*`Ml4Uh2~SyndI+5laZW=CT2HKf5pFig`o{S!9w;k@0)kI5lE9P-U^
zo;(>yIi&=&nQo36b6mPX>~Y3&ty-(=#>)IHVP%ur#^yBAM)g((-LBDzX0;*xi0~<M
z%ajA?g_FS?oni*@h6M>OE&)4xv-uaj7AAV^AD(5-6~kdFiq=f4>`%M9sSNa=1&Lb2
z?SA~qVXYWB(la+6oGcl`_3u}HoBTk5DyO<e^nzD;3_iSW@o$@UZE6b4M9mujKnvCv
z*#xyt@COJ&Onp}-5(OxP*gmsltAnf0hA<tewr|pMecLoH9Qf!xjT|VbXjrS%T|ih|
z_G`v*uUhWd*(uj{_U@{blx#Xiv!m@bP>HEo$BB7uB3n-nBIwYY3kP2zQBnEiRQ1rn
zZ3IPvW5!tN?8e=DyqB&}b*En$fZw~}L1G*lQa?`}U$gJX+<^1ZCpS1i7Gz|Hxkg4J
zCtPo?Xm<3Kk6X#ridc-ZptXM~W4&FNFPPMSu7P|nmnd&yv7oEdvGR&6ax9UI<_7z$
zIQ2k2{5dyg9i{`R?vmqho+V)SWz+X-)ik+siHR1@#4^c2$?(wAFwt+FtW-JW<*zbf
znd>7dT%&M77M3sX8bObs<_}gf6^%D@NE=)0)m9e9!5<py2;$?LzdWo&1OT}{Vc?<g
ze(MpCxu3Tl9;RJQbdhB=6koxG)Gy-XGU(q-1e^x@zH&ICa|5>NVE>eXI&9Ndvb2mR
zz2X&K#(GJ{!n^;y^bftS<!n|bFU+$Yg>=3kw7CvpLGPV`p-m}?!&7c(&<_hG2#)^^
zV)Hv*Uu<4b`7XTo`_K(9*d2bp3VYP`6f6C04M0?2=$-4lb)}30a#tUUe7(K1^|Q?R
zz9v&!5Mx-~0+ID|mQNwx^rXI0)jO#|s@yJj<NL$)!$W&|tMHT!k8?uJ%@?djKTK+D
zC_tJiVlLNTyku`{#MORD!sG`a2FsNE99K&d>j!X8(9YR>HtEIsiaJj*@i^}{>Jf+B
zGegeW7^j`-%GxZy@*Bbmj1nTp$6fu+<N)Dpx3-o<ht+~EvKSe%8MIp7aN;q*q$QeH
zC8oQMB!}&@<G%6W!_J*Cysf>zQQt`f)jD1*(gXO&-OzB`Z2nyY4NI|Vq(X-5-3aaK
zPlMMwsFF830E{6=MB&Yd1S_hFCTey7>PcM%F34;Aq!T)zOk9crW1dW35N{gIx0R-%
zp~N1C%%l4su^Jia_Vf(iCu}U_wr6Cu;o*vgU)k&n{H?C{^_smdfHUT0(7oOt*EaTB
zY`)&OZhCeaHTC8aR*qOTJg@r5Q^CVDN<jh)L0xTko-|kD$2ft;HrA@ARmIozqI`kl
zUhM57<B`A>E(ks~eoX)vC$&NoTxfl)Gx}gWW9tR^m#3fNURbkzvEb1o4xJoxP}m{G
z%XWeN$BuvTf+lK%$nKM?<(Vw)ChXktj#4{(IlN<k0o(0WPbkEjz+0^a5dUcN(8O&h
z&u2wUZ1vFb{LNNNVMQ4AXFhxV2~VFwt)j1or&n$#$<Ny@)d5n#G&|(jVL%uOPN5&~
z4q5g0pJ|Reu?!=N>_H!#34S47<8zMR+7S{-m+x+d>b*fgn8Eg`(D6sGQ0=p$#k$8V
zqs86_yB9X_dWy)UG^x}1q^;9JS6UL}_^i0{ovFuaV1{;_638Vj$4H}+5YZdGl>&n#
zWQ`;5rLCWEe_GG%4^lc3KaUl=U+6!xJ}HcU(I5N0xMp#0<OONMH-Jq~1j@F>Ke+b~
z3{OBS58u8u<lZ3V{he6+7LlVCkQzp#qtVM1+^o-OlS|9X7{mIRb#o{1oQu4S#iv6n
z>2-2V!zk$ram%wdQhxWN%>M1YAzZHb%;8lZZ%R&q`;>34HXN)XxeV>lEam$<Ia%Jt
zEzV<~t#c$@`roU|v9;Dxvavt^V18f;mVq6#{4G7`o=O3H(2XOS&8>Y|z#s5GtsF#H
zk?LMbvsG)uwGu+-zUtZY>H8LkY43@$NyaOP`#y#KeD*N*MQcl|$|He~*j44n1Ty#}
z@*BM4xHHcClVb+0N0dw+yjpxy#wo7UWh`4q*H6iu$T;-$L>$&^@eS$!9dScXMsfE>
z)?s&v<#bHw^%j_^*f@LuuHY4PwZ&`J!n>LtnmUZ0YHOn}bjY2t7JbBex_RFebu8FX
zYKWT<dB-@IikqfkHHtm_sG@Xv^jfB81#=9IukNc2%d}5vqx`WJ=8mRm`(dF68s_SJ
z_lexT%O~rWLX3BW1mzZ%g(;yZ-vz#+nLc0o!{On;_?Y(r1LJ+XU%h@&x?UYlRLs2D
z+4K{TQNKA-pJz+?5Z@0c_yDl6bbD-#ECs9X&Nh9h%H&90*f}<r75l-~eGcr)DgG5j
zk3&Jyc{29duAgb|W1&%F5?lEQ^S5VC*H!ZKO_o>d`YO3-@zG3?KG3zxu<6$pR|IK#
zbA93MN+F8x*ADeMR=64}ge-17Oi9sfZkbw{#N8vOP0jM<xkHXKImh<POml5nU$2%?
zuTd7MxD8AZ-O5p0dY>sw&^k2u*7eQ+;!iQ3EBargE4FBa&|Z1I&~-C=FW7A7?C36l
zS8N<?Q!}(i%f14R5dD`3==AZM^iZW=UoV%LdPtd4w5sanNG>)L1>2*jw>7vzr*%PH
zxroj!nDO##Q`XcQ`e2pKmUGI!0}bGowrVPC71I<!WHkCl#pCsVY<*f3z`OWu9!h53
zzCNirsV^Bt@P4=DTH>|{WM>buncN2a4Jd7c0zoLTz9qSm@;d$Pm8K;5{s6L9toUd?
zR0!6F9iF~)gml^`k6?n^2Kmms7(SAtDSn0rw=SXifzH+E$eHrfRdeuZoJBMZnt<{u
zlhUwjC>wc+6>V1L+G89iRy3ma`0>`kpw2brel=9Kfq{vdCutYePac^^Vu5C3(6X;M
zHNpJBa?9!+3%RRKp6?zlz94EJ8+-NalD4uRv1Ca$=rbX8Hyh%W94^0KB7c+M=BM^E
z)A$HQ@t0OsRMn^Ve`om@t1sAWx3os(8ysX4ch4Y<wUspH;?hHm*Vh|Ba1VQF{ooxQ
zm(NCuVkY3wZUVaE_egydK6wI^!Es*Om^!__D6ISH>3!}Es#Hm50Zog3t%~0`jnb6Z
z;pMyhH~3mOK$NpJJKzKG$0>gjMTiDh&d!V8;lfF;->z~Qiu^}=A49`iS4Q_?(L6IY
z(R{r5b0VA@WW3%P{@#wB;`Oa30k(83LCR7BksN9#437VXNYKA9($ye9%yo`@)HO7&
z{(kQ>Th^PnwviGw^;~^xAuK`F%KQ#*U*{!o%-S2g@`n$EH&W~+*2gV@mKECMrYf>5
z{0cK8hCKEPFQ%!Hv@|*f=5_f~p9A}M_8=<~vxj(HS+;Rirpy_{1Rn1U3{G8V>98a~
ziAUvENUFzF*YrbmOVFF=f#N0Bl;26jnz<S}&ez{#jjxa^F3hg4RjyCqf={_6E`I1x
zJ5M}g=vwv&rS2CbDFRdDJ?poeMF+NWS<c>wRzgDWV7~^I@sb`8>OBejshk;gG{?hL
zbkoOwYB!8gm6!xzp-9xp6Dv}FX_!+<$IJ^TBvyO_L-HahQZl4P|9Dh=o!1NwF9MpG
z?Rj0|4dtT(o0r`s7@oW3TAFUyGdydPAf8l>iOIA7{eDnw8TObVOq{oB))}N$cw>S^
zI+a#<gH(68uG0m!Ivq5L#SiA;hpD%(%27u08oPIJOBAg=;i8SmYmZ?Ys|xE-wSeNm
zO@2X@<5UI)TRZ-7dzKLE3sq3;e6GnPC8g#f_(e>^AxKs>t)RU<mS4njuwi>sbTyl4
zNOK7GA5A+vxO)>e_37}WNpAMm^@5j$nLDQKWq!Lmmo&P{^!;M%GZD(pTsPZWKE+X?
zm&hGnx`^HimVBmT3CT!}wGy7(;hf*y-Ci1oupFs#uR5fF8!*|D?cwT!Bl4Qc4M_L`
zs^PTQ;N`94U_e75s2i{Wkpbv>YCwSn44h|k(Fh8B201WuNnfWD1$WrYqh@`VE`fjC
zwH?+o*enOs+Vj>S^D7{{rB`8a0)YC;xoRzKYKJW!MDO{<<23>9Eg-!HK%D-ge$FGa
z#wVPp=^vAPeIcRGJj;QBJ-=c0@cS7K_Tdj_qatZiPgLGLA?;hgFjkCCGm|#^jSct1
zVlI6ffI>{Tejn<NL29ckh=jM~l4(B;3Fz)?2?hjpozWu9WVB{1u2+cZXR_Bz09YWC
zi$-EFO96NTmINRbhOoJ*l5aa`C-9y2<67!k4+ePE#?I`Usg>GqNpeapy5^1b+V==0
zZ1PalG^`1)AS=TrA&o<US$4olB|Rhr9ccuxZJv}*x^+T1|Ip<h0`D*Q<S{M55%B{<
zRqEM>N8@K$Z7Ud*G`uGzV-0F?#o1f`m4~vLuT9^ms&B-2YRTmtvTC&}l-<caLOcMj
zS(KIKDLZ3~y^7Uyy%X4_=(-Q7)SbEK+VCf)7E&&5Sp1t-_^0OcQ*O2U<li!`4c3Ko
zRk^`PZaz;;rXcENAAIZr(mWcRJ*&$bsTMC~uYRr&$iw78iK-S~j~#UkjYScbEPzHB
zT#P!kyiH^!g8e)ldq9*S=y{$e1t5+aSzkJ^iMKXLt@xXmR!dtZU%ZIkZ&w4(>rKRK
zlv}0ajxv@A=T5&G9L@dfG+KvFK+*+q!`AZ(<>!}GBCO-SKa<KyS<M(ri}~)Kx7j_6
z<!>GMP4o@wXkQ3XSU8*w-Oqm}@BIbv%q~J{tXOfHvBi<$zx#WQ9!(UyM;&Szah*w;
zFL{AQL6m@q^`!l*cHhP?fz@+shk9*KXN-yYFwKZ4<plkjCg^#Q6Ng*z-4^5|y!<&L
zsbl5<ZuWXJz()-bRukrfO{JMUsvnh3@R{a8{9VqW!(=5%8W0w>?>B8aZEEy>RGZPs
zpt<w3<N&LYAoxYNEL=d3f{5tc<-{IQWTId;!4snJO<b&gqB~xB-(fF$U~QO>ud8M}
zobl$(x}=n;UyT`?&WR`^g^r6|feFjOw&`|_OEwtTre0z*1H*>c=KGHJqmSC8c2B?!
z6#r3U$J|R^;0dyU_5)Q&_wS*+S7&x`IA%8@aPU!Y&xj{U^feU~^jB&}$=mvNb~e)>
z-}8F_$;O$hpeLC#4zdOYw!FX#a-z`|vf5gVIdY1&-}^g5*bCOb^*YDWMDO!@4r`0y
z`}kv@U&MDu03EU*U4w2)b<WzFlg6F8Y#f%G_GW{FNE8WCkpl&|6#J_+GDNSBxTs=7
zYo5)Dw`XVqf(Fpa@x}wuAIm<@@3&NHjcJyEY4UCj!Vr{$_Sw=+0EIBEtmZs8*5fhZ
zg{KeeL|!8Bx4P`~@9G~k&%KC=nKCo0-83BE-8S~<hAg<gl$Dlj*Lun#srxP7yueks
zFX-aJkmF0zUWrS@3xbEk-P`%MZsnr$lhS`tUSfXH52-iKM0K8oZSOF$ZnYN_92zvT
zobw`lld_n;i?=@K4AikOlt%Y6J6W8*;6rCQ<Ber(_<S9Q|0vl2K>;9^w-UuRJyMby
zYA&GmNk8wIA?NceI4qulfk9#*D{)6z*|FpWN3p{D8#oRRs2~3)TsfZuGgQ3M!pP9X
z%urL?l0R;*wK1=YR3nTGj~RKS(5s;?P3}HEM`O95?j*?*5=^1$!NV;;hD8L26FYcT
z*T07IAJx~Z3$bQr5v`cRstDtuj~pDF)h$imu9WQK<*@JUFfR0Nn$7B3-qE)ec$%l~
z!E;%N{epcmtQre;Gt1fOpY|7ls!`Xfv=FdyGBz~4pzwpTyHLTE(-lsCvpLOP+vNjN
zwWuhz^pC7Gd+SO$4toZMx@|Eb4Lcx=@$1pX9C#MNRkV-e%TDSWYGR@Ev*YI|?F1YF
z!hb`8Yg6R+fIq3^-~`7gg~_ML_C30!0e}VS`>E?<VPR}(<r5U{H`o#?js=oiyU{-K
z$<HW-0+I@ElbTyt!>!ZyTU@mw6)U&3k+REN0V7IeA)fJ>%V26m+nZk{Dgq_*LX|uY
zM$q(FKn(Y_mEb1d@>}4#s<lpGO0a>i@h`cekNWMh8oquF+etsF%*<p@PQjO87BhkC
z&4`esmfMml%q$<Ey%W31*zn}IR6=fqB8w?Nes$=7HT4!?QFYz>@E~G<fJ(Oyog$qB
z3JQYK-Q6YK90Uag0U1gf=|+U1Q<U!RkQnLi`u6zz-sk^uS(kI>oPBnzb>C~Py>G1d
znWz<GbK`)5<RbmP2ASphwDoDC_T@}sV!B96#GyLhr~PY`eB<_`|9e{qEI&(TkaXG-
z$$K6Un#{qXH9^?HW<Mg?^O$2X>oQpAVd`^zty3yHdQ0_AIlGyF_w}_#it&V>`Z!@}
zGVnFa{9)^gB+(vm>F`njenPrB?R^=eWLH<`f_Ce@c0hjg3)xsvN^xQ!9Of9aZ}lPH
z#ClPy5;qS9QYa^OP@BEm&D0|G;vX#oZy)Fh?{|Ou=jI82feLh_uKto9(W!i+Q$!zC
zQj)(|T2?-MR{cJYg<*La5dLXx+`Nc`bmJx2<)Y_XGrF4_cf!K0GHo(hFc-F$)i!y=
z-Fu`Q!?@&imj%Thj~HCKgZSc8M!J6*@5gA*q>b#TwGI88h4i<@wz69OF)9>(^8sYa
zHQ1UT45tfa)Vi7-wHxuxnz$DT_g2-3lnfkh=*GT(@r{QycD(LFL~Osb6rDyYyD|CO
z!oqod?(h`v;eQL|U8}0EOy;BM+cP?qQn-{M3b1)3W}9V=DH#Z5MQNyq7k#`7a)$SI
z_Mo)-C_1TI+{z&#S}t1;<>#(jCp&QikJ>CJ-5rK%<QG#K{Q`SA8#<#E#Lx*CW+I|;
zcPCT-xZB|Snq*{7>=)c6R|jk$Bl4>zqiucILCtp<vvl`Rl)zn<gOyQ~4kun!*3fxQ
zd$Eg+)x&+lNgIC3ZcsGWKJxB)g0>bk*iwJjH=$-|x;OS`H@n|Vbz&2Vfmtib+-JSC
zda2DB%EBUM4OXGZ+<G^Az0DF)ruqwH9L}AcLyY;Swi1|a1LY-ocx&y&$rNTufNugy
z$OR&DQ@^SwQHx$Kk~b({y#%{&v}jn7XQP+AMbP2cI;al<`%*zbE(s?OTJ?&+yoSPR
zDy>G+{K$1wWmyB~F8!Og{hAUagISm#XNLvCFX8Y%gFmsDoOyZ;8k?ty3U<7ITM;|V
zhG-WRBj-S<pS0Z+bn^2&brt5geZa?DPMkT!_3tq7bJBo-x|e6wKvxSc_?PkC-syq_
zGX_++uzV@ajL=YozA8Hkiansi#1!Up7-xL@ACd@2<19tHBPsXoA3=*uCve_?K7~{!
z5Mc0k>`B~}RlF@M;P}jKD&!~No98Brr@ucj9>we37L1AhQ9;4=lK!ET%iVm}Ze(!y
z&W5|VqQ25IX%eQ{RpRzXoC@9duCKWi^G9wnK{=pcLa}1z)5>mN<L}>Li9THGEXi#2
zZO0qoZ(p|pMhy5@|G+PqlKnjk4|-UUXd>;1GUB%znWyJ#%^entV%=ZEP5zbd{0xi!
zcQ8TiE$%WqHeji?KD=UMHNQkA`VjfNZH%B~I9XGx@D+_1l}lXPfZpVU7s7%5q}kUv
zviaq&W4j<G{T(1KScw8@4&TbD65Tko@Q{tuV4<k^?HAHv2>Fd%lbjr*3FKUmt)fCM
zPdGVG--Dd3ZddQutQ5A#*LQ|=Gw}EdgDNV&|Cs2FEl<+Ds&-1gMixv(;JZ&zQAq=I
z8EbA{e$elCCGEc5+412e6dNSfU;WXK!Y7N6s8mE450qToa>@1fo%h~+X-OI-ew8$x
zPU~Y^=-KFfpYGD<TeY29J^7tieQ45CSm*OzsF7oFV`Tk82<2L2#*sg+t=V6jEFLWR
z1C2?$B<MbLceYNBe-Nl~`CU57vGc9QneUCnvwv)8i%UUGNh}m3bw4So^?DNdPowT*
zepZ72)-W~m-|+{rq_=TpJi*2UcjWqhw*LH*9M!gSJiY*+Nj+UPP6BRuE58F!P*6E3
zk1$7Wq#8wnZsaa9pqZwkbP2Cq?AE$EKBNVd9c0m?qCk;PYsmyvhaN2cxdi?(V<AK#
zkhB>kWnyRBoi3;IrO_#@_bN-7#l+hMb3U1lNomxGXki9{6RPQQ^bi)Bl9<I)QLfwj
zry57ld6|_;FX;|sE{x;tUJyHtC3o4caRakzZ{xMp>FNE%*xr73ae4+3X%y+4==nYF
zyl4)AbfK8cIo8?tOnu7qC249Z_&(5#F0%F0+9Z}Vg$)8b6l7UG=r7t<c?It0S$!cT
z+v^>c(Kh*X?;h!R9wsX?;W#`c{Yv8wD9kxsH70#He@>EGz-7O+I0QR68P?kCLOPsf
zzsi$n4sMgj-1+?{lCKrcXZoaqm>7hLrCu&a&>!+0kCSy0VyDpX-Ebx#sE5uhZU0dk
z_>@xX55`#^@7|+>)wcMpxwUXm&?H-VQ&42DtN$t~K?uwJ6%Ie8c=OPlpT@fEtPlYc
zZ09JF;JLZkP)?pmWSRM%{8*E9hn|17dTiYs{6o8i#b8%USXaDOo%g{_n1ftGzu@i1
z{|mc4y5seiGd_X%c{6>N3r=IDq$oXoXff+uN|>>FY*roIR@+nQZeb?e>^sqiyHtGa
zYd36-138HTEitDjC0MSxD)ocSvniBbwl~0ceU6TW-IC!sJbn|2B}TTE?H_-du275U
zmE1ITe1R2Jv5{Vt9^h{-c7ShL!gJL#7wfdpDhCY60yJNKSDQfDy^NdNcxHUzwarjK
z^af?N-`V?K`Y#z_d=dD6Za0}s9eq~SRE?kuXrwLK)oC}&e0aBDHa>&r2o-1W{06(=
zgKiK!O)hm0ZSP(=tq;cVfF@Ow6PGxAFV#RDjCcCt$;H8`S6Gz6V~gJ_$?Y++AV4zE
zQL^%o94eE>YwKxFS^7b1ZaWt$k(IwM2*w$8O#NHi4R3U1h26*4E3A~Cc%?h}iWGd(
z0r<F*u5yXV8CYL7bR`RhX1_#Kn8pe=>y=ftTTf1DR9NG9N;EV;)S^J1FHJ;H#L21E
zbee-h6WX{bZa?#t$4@Z_PjEsL<cqk5p|()4pymzL&slY)c~sows9PR^C=s(-)%e6<
zNi0h3KYDO~ivhqC>JNzTPHw3q^y62QTp!dAef0da6N&4F^WfI=^=4hJt-U&>L7?zW
zA7`y8TD@nTFT61W-iS|gr)jG3`mKbPvz!_Z0ReySIqc)}AL-6O|HJe8!IJRr0(xRX
z?)efqUD>Xt!S3jmIiIgQV%(2YvTBcwZH&9mN?QtwBt~DbM(%Wd#wd4nnRi(t;!hPV
z0s~G_@_Q=t?&>dw1O57xP%AQI_}5$LjCcGDAncZ{CVa6xx!lDT?&f&ZAC!yCr?aUp
z)U-^h`EM^iB;A3=7(>`t|BO`NZj1B3u($xOMzLU|o&F(l%Dt6MaQtPoPgqA+&fFxf
zZcpzs+Q`rGsR~~9DZTt{B)(yhZ!v)9AwwF82xXZGFJfRXJ7wF=w1{ii8$OD0I+)_L
z;TYN78&`V0F%$JOK*FnNqG`Dsl-nqQLPWPS(Gp6*5C}M#porPF20ZbH)A)0Cvl7PU
z#2FVp9+f(qSsM1}j9btyU8F2MyKa5#bb105ElBJa|EbPLw5Ujv&+ASQG6^+B)?r#m
z4JQOEbkNC3h{!QZfB*lLG`(v@=O8h;+?QE_6+H8*;Ti0cE)f&kn7{%hHjq7I&V6En
z3_eB{woqhngrSA?5APX~b-v#5m7YJV3$2Hlrl1CSluU%3`7;A2JVKS#NXzo(=Cx*5
z!iNIF_JYRjR87V90D|W7YZF`$|Ikc`cU%d6p&1{)N@+5X%>mNPvuorZLf?DVN3{KX
z+G^p;SCG<pVo~3#34+7bP%81_>c(qXC9Zmn5)@cFAG<q?D<rCQ^Gw}cC_&W=Y6e*0
zf)P=1ubVXLv23|LG!6}|f9&-@rcD&N4r>{+Th^3NM&2w0X#)vs;RuJRmiEJ`c23Tm
zs8r6|4QIL**I?B+j5<i%xDd>w0coy_H~&P3gUTpC!<i9~hN=w#sKz~)Rv6grLxN!U
z*VLTb3Y89$GM+->4s&_Vq6!vFN6;Y)U;~sN6*7AU)!2pIW!71zK%uE$Jz@Cu;D~K4
zS=d+fIEtkeRchGf;DWiG>a1^}8?gm|a=+cOu)c`Lr}c^IkzyraoGTIsTy(0#K@5S5
z2jX8z#h=a6(AlJzpK$=m++T?mVDjv}m`fRM=xwOM>kaSTad@Hc$o0y7yNGM6;E)32
z%Ul_ox)k5odeO(QQm%b3SFHIiSBir#dH*Zq4S2+hSK5)V^VAmZdduGeb8;=_Cx%>r
znQXzz1+i27Hou+OFC7Qf&<Qdh2oF^El#g<J8!cT$JbpMo0fXcj8OtsYv|5jc^&T37
z3WoB-UGt5)_di#b_q(?ONw?syMVP>D^==SK0|it|UT#}O{K*#=f(Jb)&0%FswLIlg
zo)h+4$}9>MHW_2PuzLsgA#q3|;AQB|%@-ClopDXrczEoKH36n>gbYDJGCO{Rsiy}$
zwR$XX8L}0S(N75F#}=kax?g1w3FKtELu{Pf&i(#gJvRFn_@{fRxZMWzu`RnFDZ>?h
zj{bb7j^`$rkbuz9Zv`A-kS!_`_*yNxpwgu*l+EV(1#&%ec?Au#6_7TP4)_8obKYM7
zY7#9qp=srY$TpKG-|!1$sD6DU*7!NMQ>~_zn+()c{cfR7{@B}N0VexY@6EEO(x*55
zQ|u=1BiHcH`aI)66>S}KKuALge?cMb4XkVA%STz?z5sWZ%i3jDB_jOoI#j`GYfBxq
zZ*FeeT~oci+Y=<`b2KJ?;l+|<!1;dB%<erX7luIkIFl&ueq{U$nfEvR<$EqQ&(7G&
z_jil;1&=3pce<8H;+_n*g_>84dL8fE23Mdl*Je(jr}y4Z>r|5H=r{Dd_jCwM7%TQb
zkcFivVZXI`krMlER#oA~*E}Qk{~WHU_TT6Zg4KYI=8vBDjGo>tZ|^-FsNGC{Y$wEJ
zq+kdH<S6*VBeE|$KC}z}a(9>s`GgDYZwB>T)R~anB_kW@q#*4yEk8W^J%vii(wM4Y
zuqk0;jMP(n>3+;Oo<KS_n|A9VZ`*-4xaEuI+(-kO<Xz>wXk6zxCSksqiBM<>;6z~N
z5t@n~(y8+N{I9o$LLrBg#P`qO`ksw5z>Dqv9#^V6SnS&n_70!^K`TFp>`2VYLsuFv
zSB~j7Yjw3cO(x~mxv7l}j*SLg4z)Tz;!jbFEBVgJn6jU@>RxVTZ_)MAz(Idwm>T4T
zdm1OE20`fc!?+-*Zz<&;s}Gq3)#e4zRd2<R#g|%RzA=a(>O*@sBQ4Ll`X^P|W)NxJ
zmP=IL&OF)lt%G~gUQ>vEPOaQFU0*e%HD3JC#aTsU0*3R;)>~1rG#Ws#q?LoUhmVJf
zP$M};b2wB<RHXuZ%Nxk`cA^k#*Z4kz#%J#(WUkF;4_SNO2!QHmzmzT8MzWUo6auS5
z#dd)!5XXNbbmt5Wi5gULf%j#l?0$Xd#dqd5%b%d;SN&_!#*Wd#LakMc&yteD!~~Sm
z$#nEe6{(L2aUFrUN}Z6ED2RS7a5?JAy=<EPc@uI_NT{tI(Ka}=@6bxfOsjNl2}d1?
zDLXl>O=rg!u-$IFxaSW7TR$KZlaB75mcE(fqn<h&4o)QvVn)VFbyE}}@dI)yx!2a@
zTj6-+$Gfc~#eRH=7|LaQZL22x*KCpumWXRGWAYU9z{U#3qAF4qOi5rAz6lgHtj6B#
z@r_edd?~vuW$DV5?JdyivK=IwMawPr_Cu9Nv@C;Nov79NVv)PV1!NB<LU*cA@+nVF
zl#c3iYd5_23=D!(6N5iz^-+v2_}WfxBbP<jYkK|kwe@y|?eZK9aDkFZ>_eTtNV%5Q
z>iUT2jSX#Dtp681n)UFw9wfOclkMb9l#Az+NYAql<a#f0Zf?$CwjARagiGBu-1Hfg
zDiF#t1VM*MwI3rNB<t|!ICq`d?YUJV_VO34J~`(IF{z<)8$^?ncU;`8cSNmCEUcbg
z`yT-WIhW>VpIl2`ocvWaBshkF+I8HvaP1;4cI<juCQEtJ%X=iR!CrDgThq|S$g!`@
ziapE|easLFvA32ur50u~N!q70BIqS14ahF^e}uPRk+zv_+Fjk;957c<ZlxGF%jfY{
z2dT-?onKOZ^G#_P)+1ytsr#!-0zKf*J|P}b78b50PusnI9J739u-tZ7&0zJ{uLLo+
zz5ytCzsOl*dcGllOC(1)Hu(Np>#vG6PrFa&wM*e*$gn8&VGtl*)ACVG$$%2TWD6+K
zHR(IupMEL(4;?2yj*7}O<;LpFXsS^%W@%aBWgEeNo%{9buNCQ*d(F(s!N3lWf>}t{
zpWl_2$7&jqIhP#-m0lb7#M-K(c>{El52;bg+P6UM?+W#n*YN<U&&>4am-I=mWF)iH
z>Fl23j?6FdLNM)N6P^;N415!oBqy`DdUWM=48jNmk?~>Rh`!OKuW60MF@0m;mf01f
z5YpStibbi~Q*dh1l7nLJdNEVuys)7++p-}rp;ml+JLvWcP;7Pe$8n`dSJK5gfAov^
z(MC&qlm7%=@cwLhhv38q8ma#{Cb<f<Na#!W&MTi)4J13FXMJ<u%EfG2+}kMOGxtC}
zs^XA0P}`>j+j*W{@<xM_4|Lq5zUm!M$7B?XYowNPC*T>j9PEEaYt-m^cNX{rztI8b
zqo?E;%}VHXT*k(VSFPO9pC%6iBw>P*Mw+OnFc53VmdF~cO$F7TN_o)8r$2*QHuW2^
z)LS04f&`#OOVvJQb`fy;1yJnKGc&#8Y?JOTey#pKE9<rH4$DyENy49D)tc(+J+*6n
zN<Z5h56aB%U;I=Z4JT^_709G7=i7>=tgbggEBS~1ADbpeSJ&eIyd5Ejh0Y*?Vx+Z^
zwyQ555ZFj$WNO$HC3)K>Y;`u9s7F^%C^`BE;4}5DGMtYW#y-Efhim$Ja|~cc@*#r2
z+c&;}iZE;VIiNJ2tg5WkH2vF=GRl)$+T8g%B;Q}XI$c(-()VRnWfc?g#q*C3yaiHw
zH=Aaf-(pbRcnPY|JvB4+eWTf8d>mJ9DDFdih?n%;g+Ae8VxH=a9XD{;rg}YYZodZ<
z;;jA@5A1zev@_}1VlffgJ9(~u_g)&PvjUNoJ8#@fC2$O4@o3sdEK&_YK?Mk1lb!T>
zZ`8ioeMa+N6zK;PgnsYmscv?=$46)qs8x%ft~$o47O$lHBlH{(0yO!H66uH{$=FlI
zMNpJ*W9MR4(sYC;zECmG!~7Tn)aZ)}I6V)`dbE|pbPKl$c!)Lv#ObNAX8jb+r&sf!
zf52++aF~qte$o;=_T#$eDQ)D=Zb#WWGt<%9%jrdns>rBUPkCs9Zo1Xn<{=dGUAqet
z6ZJ~~aPZJx976Rwg6t#NsL<z>D?95?-gXz7lto2a?uWN1if0Os1tEj#-j#Kws`1Ia
z&Z>b!`)h}6^#aem^%w7?+WLEXVpMn-hvtSPEhs=@1l=8#m|iMh6}dNuIIB~lT~f{e
zLjXH-jA$NEX4foDN=EukfH8c+8&3%;RvpdLvsU6V;kReX`*sIkTDrk!MJ9PS%Inxv
zqJ)~BkdWJa@krRAP_<Z<+otas-cQE<!4_byX-ZUTy?gqb{Ffc--vTD&Kqc0^_MyPl
z-dpsa__oSTunzYZU-ivRdTb~~8mNL2o@~{xhSROd?<`D5@{-9<SWj-uB}29EZZflQ
zj(H!Jr|B3vi_fNrMg}+uWe9*2hMRoyB&aD_NxfrkWqA{1Mc6%0v66RVv$NTK7tzGJ
zZ%eGbH+NjP&vB~9GG)s74#3^t1q^;o|IF;QyT++V5o?K6$=NNhFkU=)Fdjd;dBV^X
znmoBLg$f7tkUMJ%jWf{H$N!3=MQPE-%@COU@jAoDU>);4&>%Y*V+2cNgtSDt;1tz~
zTT^rq{Q2NDd(Z3Yo&DpNSv3wO|APITS6#%que;Kw@?8rXy)a|K)mki62LCj=(LvJs
zZNEliIdH^&Oit@=gHRK&BBrYpB{!bf97E5{w}qbPhyQeatK0h{s`;KlW6Z^#(`+LD
zX(T(>&Oj$0%*EB2AuIPogN?urD4*#h<h!oApPbki%~5;u=Y$L-<4z_foak8Maogml
zORU75#QdBXK@k*yv;MO-{9upAn<ZgCSWRql%XpZ`$?tAY_SbUbg4=<@MHMs?5R%8B
z$14HsI*>_p-#c0$Sc(RFcvAJWI}?M$c|ZYSXN)W{Tp;x8v$tMUi;Xd$5G#YI07Fys
z8$Q^ycDQBYS5oXTrRlt*e5b92j@&!szoi#J`B7W1&wY``(HfZmwDV?xE*_&o%0*gE
zhQRWb9=;cLWm2sh6!_MmQ5tY9m|jBFlj$)EaUBYf#qVH#%)ko|q*CGnC8(J5i)%ls
z8Rkq(l#VlPtg4ex4cCz%t_zz7i9BX;nn@9%-=_Ampcq5I$!UU$%AIY1a(p^MZeQ$Y
zPjf?LM2#b59pR61g+}_y+&cuAn76zIvKn)41G+)l31wyy5<81>#lr_mmttc~EFUel
z#p%z-8%a_h3~$$hSWto(KLNm!=pB0wz51pLcjc%|KqOr1^b9ONEDL96St9J;`#9sE
z;41Qj9#m4=#}eC;3Ivq@e%R_T*$g7{6E*W0w|zCQwPa&LBGe0409k{*K|Zl4PsLA$
z607EI?WI7Fs?-w;<oZ#X5)WCC<TM&2{C(&5A%_yLFyz9?-IObHE|7Kso6JC=XlO&q
zR+W}^nz|&|#v)CC7$=~T1~&lNWOiuX6$DHNPHta)#t92y-1(((_l?_9H-SoC!+X7%
z3o_ykoO3)Ln-fJp&%-RKWD^reg*7Hx<LbJjsUJosnQ>c`ZsD0Sb4+_z7Z)GTi}YiH
zM;9c~u`yqZr|%DUcW+?yNqO|k%D&>_<^B0B=)2<&DA0E2PI78obhM(f@(us2EQZ$B
z)?*WG*F7Kv>9jR11hf<n(sF3;{D9LwtUCmuh-0aE-QN8<&V_BdHpM@XW}bzSb!Qz}
z#O~QUN3RI<_DcM?lHXwKDPOCHu+VPvqDs7wWs_v%T`jj~{}>caHGx8a8Ck@f9AgZ3
zmV?7eUs8BPL6>jJLW|L_t3c8HO=EL+%dX?Vvm1a^_0$nqj8hTmm~>zP*sQ-f{TA#T
zHZ(GQS*<J+MH8EjfL^xGes5?!>yuan`-wvEH44!?vH4oZb+ouG2A_kiqSz`Xmk0+?
zGVb&J$U%Ke{Dbb-esWO#-F|%wf2n14=nb-L$R-J@?Yq|R1F4ZOV+Cy6t~kKIcO#S&
z9%}_924XteXz)&6{M}n-&-AcPf(QiC(Eo_I6$#A~-CQtYyuzm!B?QS#&8txgcHe!z
zD<Plg#t3RU2DZfXLFEm0j$KFBg=!_bKWNg>M&AFAKk%j><mJE)*3h4sS&<1bVaYK>
za>+@g{a@i!K8gQCUWi=1DXaXx)TO1s1+BP*zkBa^HHixd0N8cKTUyyN*A)tiW7i<$
z<Y^<bFawr<_~YyAwFVHaOKgpxFUyRs2V|nJmjQrI9!X44+Pu0j`i`1f?nzzIJy1g`
zWpi7kHzzwPA!=XIw(i(qbf7P8CiO8rA5$<K_4o~sQ2g+QE-*B2kd+zkI#3PLeq^4;
z7~sm&<&n6PwTT=d50-bI69*uAGi(0e`){1m108s7xo!Brpjz*_(olm-11?1|905uh
zx71_l&CQhbmEM+qH8Odv@V|)ULge4#`d?33&CZ&xUek<Ur$mf&kHnBjMLZ8n7(^(4
z`V>ZUHd0gbg`9$`&iQ_e$a-V`RwFx)g34@*QWq-)h<BPpCzbim24Jsf<r(4K+4B}a
z4|9U>8c>Ux2?^ZA((bK?fBbAlR$9u>en0E#=7g1tOCK(qn#+P>x%ku5DiJ~6D)IlJ
z<9E_A0BpFbYd>b9eH;!<$5IG3NQ+u7EQDtu4+B#Y9u0TqbN{2J_}Jl%0oGoK$Qj`~
zNczPH++du!FETqK*b4L2fA@muz$CGD<)%fvps0K^|C>Q}+5Jk1GxFu1!S7go_h>o~
zFfOk~*bLua^9xA#U^gcURpYKn<33+r!RY^awJ`0YaDR))UAhMj(J{pVg8@Kwwj!Y3
z?_%P~fCmlZga3UL1hDWsX+-(a{RO2M70q_AvA_Sg#~tj(BmJoT>pqQwIo8R(G1yA(
zun<~SP1dY&T~dFJ7m@#ZT-+~JhCA=3gfTs#<6Ff{9Oavs>>(k%;Fwyalhb^gKz9xs
zJy+LeKbBk^&O+MXQqdz>cd#G6u)ZTlc;lJ2gai|Pd~`=Qbb9$Vo*^4%#T0ewSIjoX
z*DH#O_m*!pTIr2PZ471#%&L!_QxA_u*Q>YPl09sDoXj9&a9z`=yl)G4dQCPVv^G!?
z)>qPoJ;#&gJaPe*gPoDfM8wul<K)#o|2-VmheCQMU6mzxh;L~X740ObC)cvRS}g|W
z`>tn$bqbM)ON`kiXh(Tz_Lb<DAaCQZCUitt%^FaT?G$)xMN=~C1$QMDn7#hQpOa%F
zw`O!`HCZLGUsq-oN^kW%bykfGe+T<g*j+WSUy<Yb&>x|Yq@39P0lR0vH@gUt`M|fh
zxYGlDP|AFiU;qh2QfKezH;e21=$6%*yb+r6Qz%7*x51?L;c_wk#68I|UYqn7_vw6+
z-(GBeT4^Hs@!4Uu2#%L{(LGP@!Cl*!Wd=>=iq!SLuBW3k6TLLA&xEuh7Y<$KJ~yjT
zJ;*)0fqCNq6<OI~mvA;0Bi?S7Kmu=Tbl)Dy#|$3kW|Yb}Z&Q<$MYm>$gRqWyq1k;7
z8rmA>of+SA@^#yn!>*j_s}1{S^`rg_pIsAQr@OH0d1whU6KBa;%MFr+UYqOOg30pH
zl}{Np&}nBVYmOg0IZw&Jk3(f}ZmS`bps?8j?z8{o7r#{@*h;>(`i`&WONe_{q{$id
zsxgMX{z^QGaK4Y^Bf14?-<ds>UJpWB@DPQQXK_J6e_H+->*{;**!n!U#(kiFoLE-G
zQDWP#IyXW22GOEL+3G$jMg4V4(AfdXbmroemdbr8#l03hCx5#wb@fZjR~Z5Q67bsq
z)9URc3;h#Coa?i~ACeLWZoi3V>OHEt_i;LBTBFMi$Efv0u5rvr1P{B%2O#pvT8o~W
z=mW<6g~J0#ei6oehgN-K)^4X<x6jP+&|h+FS(72meULA=uzh?<a^>aHX_0%##saya
z>@)K@3}o_YWXiQsas>@sf8*uz;4;Nfl_RSY)b+sUzJ|e$Y0*oo@%#1X^OxL@_wZ?+
z>J4puH`m@>W3whF%7QMxLpq!0ewlge50U3fdB%&qgoJ-Wmk4o_uHUy_TEsgg$-~9{
z3R7qVM$?zjy{34)GxO4eutrzAhD+&+Ke4skL1)2p^0`#5MHFd~!YhrzEqG17+&b&)
zyQ=!@*>dGa$pQx(!8A52h!Q@#OGY`JMwx#wu4;IBRjc2>7mi(ocGwsCMTNCXUMHm~
z6_Q|k=#ovzuIDsnHD1h#^EoWDUb-H7p?{B*O%YL%7Z%>g311czg#@kKwRq~?wx7?u
zQFDLgjF$d-swdfu6LBdWzGez`<M7(6e$8<2<NM~x1KKslh^;@Yd?+b-*qgdFlu`Mq
z|6==LxyR)#h=E3Sq+uwY`<`Una7jr6)uhe%2;9|snh*Qloo~Y2itD@~ANs{<64TqH
zX0sJE7%V%2o@ZrJjn1mXe<uqfmwv%oY$A=@SQ968u*v~7BAC|yo=T^Rhs7y6sF3uT
zUXNJ6WYC!p7e9*L>|$%Z=BeEfZFXb|nTU4ZC#oXZ8y0VSO_$rL=S%uO-#=B7!XQ}+
zS&3WCy1RHd_->b5PDRM`6dSL+tFKj7@$U%CFncU4IMp(&rYjVxlZ&?;O(hk}_F)ZY
z#yxJrpoC-*K+1s)uAzN2Aw<q+tGQZTH$fIG89Hj$|ND^QaQ`i)u+v)f#uY}$N88-a
z6cvL1RSnE;PFjzT)bpmjvX~IG`agXSR{sESMv&$3OQYszvYmpgV0!(7K?HeSym=yr
z7_R<Blh2x>!QxWQds*<}>^x^z7;<Fna`yG_$(E?pS!GhN-cKRyIq5TYwV#kExoY(0
z{boagL-4OqTN8!S&X1O@?UMCc_aTyku|EYf-hA1<iP$|+2{A<@!Gpu!^x=mM>3T#t
zg_rW?W>(!@L*-6g>SPFe2a1BJDWrO8HkUOiC41ad(Qu({(YU*!a|Lu=*y9b294d%q
zptkBH`=*70GCcpD1qOT5EO>r5q}rj~P-ZaHhM##(t@O~@;KH#Ud&XX_&hc|!G8bG#
zKw(VmDq7usBA+dt+`8o**O|vSE;;d(L}YU_H~s09ZB3nD6{G8F8sDPyYO3Td7T!ls
zIeQaJ+AMl2&i?uLJabr6sNzog7Yse&^%{FU<hDyTwQhs#$_xAJ5(77zEd>#x2gKQJ
z(lh!TWHx0o4h=<{jaw?I*<mqnP@8LnRBBrul7nB^CYcv{rwGBRTa_e!d7f;I1@Yg|
z;}2}77ut4SXN%@bJ$P{D(IyYS=wTI_*}|E$zv_c(UAD*AOi6=Q#C9!SynC<aJUD5#
z99wjD5ll0}TbCkM>-p`+<`K?p#Ug{!ndwOwBD3v^OGm4^dV~l}1@w>W9V|{+%ZEgH
zCHg+{N(&1KNY~)sVpp#V@F-dLEoTiBEY^x&bh-8+9Mt}}*_=t@|7oS3vEg;L+xgM9
z4>8Da9@lDv-Xf{Lq)8MQ%5^JpW#>`Z=G0am+5PtcinkpRe9`$j>zYNWSFuu;cKHam
z?SZgegM81+H+tTuW;KO=aICXa7PyGCjWWKC%Gm>%L59Qi^^CS_jldcME61~hIJb*;
zqgNN93Gc^-HtkH<MEkCfHnL9upFcds7hl_rtW-1l_pEx@@C(piL(6gc`L5e_d3_K4
z`YBR5hhwAb6YA}xKRU`~r1LuN9C{tZKFHU2OQ*3Ouw7vv1w;uvt8Ign7CeQer1P{q
z;C8c7=>!(v3CZ+J@@a4UeUUW0^!cJkaa{~KLsqVzuzB=~ai}40Z(V~K>M@PHws0Gh
zJ$thx(LA3V6|TrgYhbWk&svtb{n+^QNVpvAl*$j@t-2MYui~2~!9z|%a@%Fixag6g
zwz-lP-(7avAMoA5%^UWq63jkt#EGnuQ`!50;>i!UO7Ru*SX?L151+c&UJqF1_Y^x3
z+^^h!Sfy*Pb^-NO=i+ob>r;5>`P0W|C!!>%P-5V4SZC-OS;YCKcIKCUC2SZ@SVt6{
zm{`!={kp(JRnyS#>V<=|j0b$iMXmDT7zrVv>I?cVwH2CqllUA48Qg#yY{?l$I}<|O
z@+qo`jA3kW8rD`V>6EeDx(n<&{d9Y9wwRAcFnyl?7=)D+2r)g9dEYBRl;6Kbnk#im
zvU>_NUsO#Hp&E*ckB+fYM5+AULC>Xq{(K*ZU^lIi60P82e;1b6G)m~~>(c-lFSdSu
zL<zLQv=RsxZh?|P$8eF}-UkFkSx@~G`11Q#526Jy+-hz4dnH%9LS(OCwKKozzageJ
zk3>nJi;IgTn)`;1w3;24QxCL&Ca^L?tZBe;S6^S8p9p#Qu}wSpS7T%9#l?w-UK~~`
zY9G?qPubh+^^3*1dpEM4xIVaiYICn(-a-#H`xbi3+HPLfF*oP!o*MeQv*p4r@>Yv^
zlmXUp7Y#Wdf__ubEghkinHrJ$*@)EI&kAyRzn^BT%P;w2-DUh<;N|urHpJ%OL48u?
z+|X+|?xDBi`oh2gc{*YG?PZ_>&%Ks?pLtgdgefV}H9N-kyn9ODHHw<@HwvXuaCQSv
z7h_HAit~1UQ&X`XSktjvGfWx`qx`b6;^_w)lzJoCNZKDi@}B^^gZ^2g#ny+?uyx4?
zo?Z&Os(QJFi%pak{MfIn5R*ahexKu0Bf|anN>@@WPQ@4cq)-n1>IBUwD_jJXUTKi@
zUBO^;qM$Vr*L7P?K;2dJ-;0Nz5!O8=S{+)zp5gnWP1WJ}r{aAm=h{0P3VaL8AHSCo
znUyprH1BWTQrCF>e|l2G|DeB;u-lIb1S*OU+~I3$)w|kpvHDvlz%N{n??n}#yL|nQ
zljP;LK{_*7seqgH_mDl9e`NtY$Mxo#|D#$x10ZK4=W&SZi*&2M<+E3P>+);0R7L2>
zh#UnQBfIJ;P&qd@U!4N`2L==xEsr}?fIrrUu2W^amR1(j#s3xdfO$N?21j6y;NXo0
zpi)Ms9W>V7SMMM6G++Yo^p?BLV5l@OTdqb2K3S319R2SL=i)GXytOvuxhF^JQY3X8
zF%;6HgS5Q7b?G#p7M>)W)#MN3vmRHhpF)m&EAY<Kv1KC*GIBNj`(?0!<R9=ra;%9>
z^l^4U2Wv*I|H=fy(ETXZ>Ci8yE}hFaZC|;kkP=<54?o4arA>94*vZU-O`KB!L8#i;
zaB<@XyZlWyq7)%I8SK0Gm4bH*m)wM7Qh{I)cR;`r39&ht-g!^N4NkSRu<gbb`6*DW
zM`pb$IZIAd)$}rl>FMn9m<2>hg8y6tOo4S<hYAZepW$JF9fg7MD$ZdD11ri9vn9@c
z0YflK5tA5Nc5sC`94<zPuPt>%skc2PdLpM%U0scR?_M;Jh#qk?{2CB&gNT?|77-8-
zU|O~-@WcRUM{~-63(Ehttu`huj!9<k_&9FHzEbc(Avoa`e6wVJ-Voh0l$y+(oC@}p
zItapt4~-qkH8}(XG%zqQ_Ks1DNKR*ADA`Fk8KJr^rn5C(+Ol>BmQQwlL+$DF=kMIM
zrt^|T+_kB&)PC~HiRC^6H_5h=RW9<9yk$cOC-Jb`vCFpdlBH~Ua?*nD(Fc>Mvv}%Q
zXW_4U;o59oHd2F<EHm#Z42PwWlbH{&-G`l2k?4Co3nk`2(D%giNGA)seiwZ399CdA
zTc>qA>s>-`_4C66ua$TU(En-tq+&i*gM))hC{zwE1-~ZM?WY`1ypL?_fg(I2lf=MB
zDeyLah2R62!4D(mIK?3hdeE(uR?$WY*;cN1V7a);Dk`pP!z?n&U9lWQ=vSwOsdsgB
z3;<OxRvGK@Z(kF5t<e98eCFf?O8l|G5*K#)W9Ui=LpBX$NV_hzKaf!#%9bN~@W2RM
z*6=Rif3j<Jgs*sEaf%BVEKqdg2nMW)wLqg}G|{0!YwJs%N>099B0pdBou_urKuMfK
zuSWIb$B)8Z2f2@{mmVIj4oQ#cxhDe&_oDFIX}1KE<-;h2_5xaGow_*41?+YCm~JTQ
ze96xH`SZE&`DW#0y?3RPlM^8jfU=*dvD%q$4*B@8RG1b<^Q;-C$pVVcW{K-^o#lf0
zL|7%9{SM{ib>Esc9?QwjW)&0D-`(BSu6pzJ;?k21>m*>m-hg?q#&z9V{zh8G8{uTH
z-yfNx5eVgtv*BUYtnBQyf|7<o-{YB_=D^!J(f9p17gX-TQr)jFkJno%9Lj}n`D>dt
z2Vf-#IHX)$ULO9*h**R23yPH$6+dI+;JBXb7;)&=jjw%xS-%8c<Z3_lI@<Fb^dz6v
zAU#>o*J8ceQA2K+`q{ydLJR_t4jd5ta9G89tT-9m&UK}m418S^H2G{k5Z`4vMk8M-
zLrT33OR=r=7AGtfoUNyPXTC9<M!eW_e?<l{>%9&C1^xQPA}FXCWW@aF(bJwpfw9!9
z1E%m<tVc4kFJFG9pLx@Q`+aJ2Gzde{B=)f$t4sp7S%|RfTBg5047BL%aNTBOyc8{F
zMbE;;HPHD5=5w<Mu)!uFVs_0(s~g%0l#ZBbqS;o~3qRf?epbGP(HJLWQ{n+<Yt|K;
z6}}2hl;zN~h=|m-I$`YS5uz2I8@c3E*fffx>Gj-aT#0iq6sH)Nn7|RW%-r0{poNvf
zc>d(Zq(L?A+mB@?D;*5-On}53y$qGGOKzzy%!g!-<L3N+ET=&+n5Wu#d3o0TX>Y-G
zB{n7X>fj>a=+xUV6$80>c+?aXp+_>-!?}@e+p|SjIK-Ay6?WLTxN4~0WImTa{oE0w
z81p&}XKQRSgLaiOt=_kq+FmCKD#^&mc&!&z9&D8LvGVh)0|{JkGXL4xdI{2QiXUk`
z(pjd82~%ZiSt8tp`3&QWU!;v0K=s{IB5oVGs8lHB@Bl{h9lG{|U+h3FAt51ye+Dvu
zuE~LN+YFe*2~Uh9dV(+&3j`_g!l>ceh=UHM)Li{~FJdsWhvbumvxJ;i{+w?&a=Q6i
zTIOz5E#l!)i)w;a5NBb|zXEX*|32)ycLxIT#d6V#h|dCV0eXlRED2Uof|zfO-`Oft
z{cbD8S(m|;F%Ub3@zINm8z5ZvmSp5o#q{W;KJ{-_&J}<A_GB4_S_jV=`$Wc5{$@eK
zb!%D+2+AV`xnKo_295fNVWf>lr(TmZdg_BW<Z$!_3d+jL%F)TmMPO>^e7@&j@^v0-
zzAR+4M_2>26$sFms26FkzP(M|57tq1WF*x5F&0Ln<Pse=wW;2h0(b3$AA8v2U}Fy+
zjv15yiD%!Fh2R7pi||?BW4(g(8{ag2eCjgGqz^E6;9YT?=%%5UwuXVP8c>joZftA>
zlk4HbhX)bRh2;8Z(NK3h&q&%c%!0)mxr>4w<gn;;tD)?a8y;iDdg!?YsQ{%&cw8_p
zh5FhT49r_E5qOk>k8*~z=&E7EkFBh%WEB*ES3upA^~dk*>KX#iqn80TO$E#pqWkwB
zwLBptC(lo`uaR+|^R3+5+vBz9r3@Xqv4aKXvSLtr6mznu=iqO8n005hL5A)4H={fg
z2}wzKT&R5et)@zL$e=+N!RzuzadhmaPybuY+w?MW3A{?+ti(K%v!hM6)eTQ^!NUjs
zYf{ZPFthPWhgACA7}loXo40P|%S16EGKX>$;)Gj)${7WIQ)QLzW3&lC@LKji3LS!d
zlNb>2xYxq${CdF7zOq(|8)L@~-K>yLpYGXBS1BQeD<C@^aAB^@i*+kvBEiN=uS07Y
z<pTAhrR~1edoY7CFrk6xL_g}U*~Ek45GJS6@$n(M#qWSax}^}qb)Nf9p@y(rY;b2E
zWuVOs#4<B3&yU%}#pj%T0X7{i(#esg_I!@`YmyunTh)KvG<@rKdBy^?9wD32!U1p#
z0O!HG4hVxs4;E^b(dUid9fzHSb2gkvAb6dYUWP`&zR9YpMr)T_{v(67yXEfN`aris
z#zI`d2blQ;xOWA;J{xTKUW+LgDNO9o!RinYm-MMNnXIrYh|~A_SYn7N_Gf3I1z0fI
z(q+<KS5&h5-dhNF=UIGwJa8OI^en*OH#awfWo8%yx^?|)Fg17|^LP_5mvdNDvftd~
z^tYBqdkr&FQ_Gne*U2i!jHAt|oVq$uKI`F^z_|4LV_>@cBGT9~zqexqFo-r#LTtFc
z@&+HheDL7G+GtT?5I#+@c~62C6}~?fF^8mfrdFAGp2J*&*TrsIQ8;FkF;Kced&7LS
z0^nO{gIm7-`D5w2F|H&q{U%Vm(%#PXIZQ0z|L00=qLeV5A;o96@u{O{?7?Im18ph_
zU)-;-o%ns~o0*lRMTOSz$bEn|uW7m>?pwpJIO{X3ss#CL#}&aFgP~QLYw$hXX~Aa$
zeQ+C(^0QTfx^$6awpAz4++1~0_Xsa6WLIDj^4wcueT}*VO0Mv8Mj@foy2ZXr-;>|_
z570ua4IOi9;IA~d&d$!k_DDLU?Zo$om7uSf>F6ZU<6OJlz$TL*4mGrXB2yXST7Q=w
zFuv@hm6Ykm!fG4(7#&w({)m3FXhlWEG-|k)C0M2E(tqP0(Sd(I*lE5^<Iy<-+)w?{
zqzyTj@jEbd>=5_?jvVwAopiY6`O&6!okuAEd+hRm8AC(xzu#R+bntr2#K4ee(iZ;9
zUf`ajk&zK_yea_SxbF7|jJZ4qEr9u>%>?Wu`Nt0lD#15T5ySP~#}Ak7&44-f1;VK{
zhod?~1OyoX9RhQ&jxW074^&j4t{FFQn-l{3tgI{<gxlt17ML-A_PRLHcF*f@Ez5n@
zQ(8z!2-Sm#kI&0`6=&TU%~Er=RzO5Vl!d;R{fssNlf0^gM8H_lx%SAp00)Phg7{@Q
z81f?UIokMgHeP&}PZjfh=J|REeW6qsc~)*N^GJz-0n`$Ns{%};Mg78hNu$17tzL#y
z*q*IBz+e-bZb*l@S!X@aFYwb4zZBpA_RVuQOwccDEfNl7cwoQAZLcrS9SjW(V|r>m
zc8$Zx`DF2eGsX`)Iyz+K<wt=RDETTGjD|#DzA#jXAHUN9U>2a$!TRG_9oudw2$!PV
zsO1*?e4|Xe*3GUoU`H1V3k&$(XL*&I6J^rgM;il4Le9fw77D(gud(s*OWGyDkyF)>
z+|V$B;T5lG`)6a53WvGk&8f=u7!Cb_Y`Mf?0C~9g0f1c})o~aBSB2;`S`N^1XqUeP
zj%Z{OX|wDi41TF;LYkVkx=OLMpi#imjk}^mSoF+j&@*Zt7kES%jL@WPVmRv+iOgUE
zpN$KEfwFx<z*OT>370=VJ@E#Zda&s&<|{(aMgXL@W0>9h(16Nyb5d<}b(N@7#}st%
z+FT>hb3jz)A8$+3_^PIp!sA*N6tAx?UqwUTU%Va~2Il2ebS-M~yIE)u9#wu_U7b^z
zs_Or>gM--v1NQVTFlp&1Ct&9emZCs_ais^vasO3RY;5*N0w!4mIj?0}gxJ~3V0@a`
zq19;M6v2*OfNyvWn_x0%>{)Ku9N>gX5_Zi8p#ok2=aa>Jwe2e*9iWsuS#A|&A#)d;
zi4Rs<ZMP7*4bb)Z96;rtot&NF(jUyh0UF>%YE}|#=#x2UH3~Hy2O?^-Rokw70c#+n
zrY;6r%Y9d_!pW00t_47umqA84f_en_HjTBTyLazuH~3U%mNBi093QKKC7SAYCIfKG
zKnM}URu16x!zlx=)hrP86~DeHL)D~0S6bOt#OLwCuGUC6I!VxE^9B5w<#8||Ehod^
zXKZZTy}Q_!NCRDz#>d9G&3ZDv{$eQ0O-D=Ht&$@jBh<U>1WKTAxf$Z^Cd*kIy#cr=
zr3B*9oUSQ=)0Kg1QXGNZtJ{HYU)G9(h04zmUpejcD$nF-dv3cr)o!Xn?T=q?lF(xO
zA~5GQe>MwDt|vF$c<AUdhCRb6g?^JAH(}yO3ujw4##9;rsSh#|5)_%nq}Xh=a}MeV
zu(#OL@%%+D&}!Ih03CV+^Jcx)IT#oitias<Rji<(Fg9K74Ds1d3j&<}+|kidRz`-!
z6%UL;wKGZ<q3^wk<Q}V!>3xj5JMG*jBV225>&YZrd%jiMC3>ne@F3$w!|6&QH;N^q
z$9zZ<A5Jf$U16*BekIp_1PSHHC+7oWT|PVo2+v5ZyMxRi@Tae)Q=+04g;?FYzi<ID
zX_a_S2)GnA^PtZB?!U{r?k*N5Fs4Ib9px=|#j&xmy#k}>wTT?#wH<%*J}AhmwNSra
zyR@`4D>e;G@RURS?Z#_A@Byo!;P^fPDQRv~(^FtMz^`AwetA~F{VqJ>t6L*zpxNmc
z{0_6aD^<XVt^$nN4<`LcI`jeu>-KH$@8g?pGoGkqL~rq!HZbB_zS>`Fj<&a7UCymf
zA`4Jg&^EX{rE9S@jLkg-NIbuP_byr_JDqi(6y3uPzX6p8X`1u*@89PcH2STLm84!>
zpv1d)<+3s}{}ffu4x(|d3^}*yhj1#9><=IA^tqv8E#z|`75SX~fbFt6iNo(*kJ;RH
z5__@%(qv&^c;S7##o^iwI&n!T{ZIn%#R(8VrZynRx48}q4vwp}1M~~6;vG}9IsLl|
zP^0BulfO(hEsR1SR-~Bre~tU`0Qk@ObijK+!-3iJKAW1mVqCRVI-0P6Qi&)u^p@j|
zNMp$N2c&8ARI7DBSQ_dwPB0Ifi6h!a$6S%P=uydyzDvO9@%Z{=G2)kfXn6RM?$y=#
zHju`^y8hYP8ta(~CN#T16kuH`aBq4UVC<M3bp|GDo;`bZ20Szp5C=0eGqkqtrkgi6
zyKdD^z|mVs4ZM!J6hKI*y;DC0L^&88>khQ%Ji!LCvM@y!(4{F|Xq#>a*!7|A6(ErX
z8)D+8OHob$efZX*oCdTDUZB|_aP=~S0F$`@o}CB!1Gvbj9&<LgB@0(}D#e(;4*~-S
zk3;pP8f=Y&RTt|GeXn#TD&j-S$fXJX`#`j}LOE61&xFhn6B1@d(n-rA#7^chi}dSj
zWVf703$=j#*2t>__|keXQzix2(T<8<y_^b5s?kA80GJ%=cBL4i)JxVT8)F_26K4Z5
z^9ljPv`sF7Tg?i0uAQmbZaD8uUY{oT22sJi%GV^&y^Kx(fVuzGqA$e@2#+HHq|M67
zVdLcNE33%2N&w(Q+1NNe=pI=*Kwh3;v#as*kJb9VhiZMx020Zih-v|+2O_!0y4z0G
zE@uZ;42+D{z_gO67J`6|7O;7Fs5^>rU0G=B2VP!XSs6hOS<N@SWspB`?O=%Wcs%Di
zriT`sM;qOU&7U6$(C`!hQEw$C4z^ddZkwi+Q)ZP@UGP$^7i!=DmQNb`p1+){ZN`ex
z=m>SZ<W`IHKDNgt=Y7ehUdZg8(%$~EZy6AmRS@ACAwaK{9pk@f?j4w;Yo%?}kczHu
z5`aR504ntDVeu!*Ei5do4P*qzvTOaGSV{5T8X(fpRWdGK=}BB~!52TcIG%E7EUK`Z
zVr6E28D!M{<lgD|IU6ml1l;SOKT1I?uO%2iF419csRLZlPvt)S6%fKcN)U1y1df?r
z2GGqjl=!ulu}<yGmXH&40NyerG*so~%MYm17cX8|fgxpOXCDR)X8V2~0<eDc&}v=U
z_mdg-xx?sY)H#bX)p_Oj;`$?PT+$~`5omW(YSQNArvUD3J(L}$UaXr3#^Z2fLIq)3
z7UQEs4u}?5^s<Op4xMb!A!I=d6So0EazTpWp#!%~6pA}HVE(e+0HpCbuXJ-eG_Dud
zmK*=Ve)0UdYR4sdWE~nW-Qp|93N#b|hqne(Ft4%^7~hoN-+00DLi@a{lQwa6;q_mv
ztbu*U&c*q&jWK2*bp2RQ+t~AyGXPTh0Osvel2KDr>#tQgN8IOBF2~n>3~7w~&4Sl`
zLC3RTPdB3g2L?j+%v@YbXd2p^ECRkOmMr8f1+W>Qs;a=U^Lp$U;*xPc2jeWborHNC
zPgP1P2>ssB<+=etQAKES0qn4zQwPIZ|JOh?54pNLFL)LAc)1s}irRDO!QuLt0*^&6
zG#qE(mj^WG^TCJYJm#Oa>W}qD0FUV23=9lJnspI#XnuPNE*u6xO;uA99Zx_93ljt#
z(*XlUx&zglS$Flk&PyF40G!KhC%B7rtChFwk5kYM0?rowAYfERzHpfUK!y_fN8i@g
zHgEp`j1RD|&<G%0NS{+Dfc0`@V|&32g{@(?unXIQC7DJ%Rp(iTqPGvmqe2G+Ipp7U
zg0&w9ely9Sp}Kv|Dq|=C+FH#9GoZD96c>Tf;WPSi6P%E^Hquq|1pJ>rNd_9KK8?{S
zHGY3}b=<fPUOBR50dN&=IY?z<VgiJ$6x+h7kN}6u3<5txL`G)iXbisEB!Z5B!C?8i
zZf+#32?!Vrg9?3YeMO4b8Vfpxwj92{zmF#O*33lc#~%Z(2mWS%0AdROj}`#_`!5fv
zBbEk}43^djBC4+`1nmEIyIO%duN8<H2|KTRLc3`YltKq?fZVx~p<&vH68gnT%)ew?
vX;YyM6#D1Qu4Xz$oZ|up9|&Qa5}#euzT>j@$d}zje<<@p{&}&4VZi?fhYv|i

delta 108175
zcmZUbRX~+pw}n9z6cALpQKY52RZ6<M6qN4nS5&$Lq)SS=q`SL21UB8>%~_z|e{-(j
z0@y3(9P^oDQP_-B-Go#y#seMB@Q4|Dz$Kl`@c4Yj8m_~VgGccpP5%5qSneq;sgI|2
zDouR<+lSBZKYT>~GWBF8tYXLeb8QS<N9wZ^3%b!~+Y=X?E(Y`X>;}Em7&gvVCb-cP
zb1b^-W?nt7t^|e8E)odWj}tmEtk|OHvISnwo|^C56m@Aj*cwUJg@idl@AnBHwwfCg
z^E#Ztg||CQ`!nWsiR<TAbl;x5^?^bW^PhkZ|90AUu*UrTbag}M!OkzI{B*ls46JH`
z&!t+3N-9R&6*Y8khtITZ-5jYuh}N}POxqTd3&TbI;b<vzvXFcgPfbnzl85U{c^6Vg
zpY(F~FE!261q_i|JHoV7C@N=>z|py+sj)MU-eQf!5BWf?@FO=;X}*vgChNIKc`06y
zr=HDWV0TBK<V&s>pTA|0vgc_wCj{!SEah6P@*Rd$($|bSv0C^?dp`D6308=T4f60c
zr!AE5L-DC8CUHX`Q$KCLt}0oJTk5Q&e*?GRn3pF3=Eie3H|TD18%&n>F3NIe2d^0n
zM-QkKv;0VO?KGs<rOfY}HrMw@1@U>0y$;7(Ssjd`pyPQ)jNg9!DaEID7wPBhq%$rN
ziRQO>QG?U5(Fc>EFJ9bZe_DP;^gFb1g1?TfcP1@xofx5YtU>Ds11*{7i9x`8XUh<7
z%*3GiFt7R{<VMUVIP_aDEDxjD=INMp?ov@j@G~N-Xi*)<{A2$-LHt1C>GR-)*&vcf
zT!&a&qID?>3ADAeR5b0sG;XYL@gm(uZ#O7|C;VoS2JG3x=RycWwHLB6W!vKFwv0*M
zb(yaBgy#;=$NI5xn3wldHd^fA&;?W`+x=?`aKa&Vsgt%k-PLTD!h=ouUlk}nM^}eZ
z|19wC&N%UM2Z?IPOLn&iFtL}sUnCncuDY}i$HXc8cr2({;fR0Hs)cu=&6w+5t!{3E
zBd$Vo;-vrhYx>q}SbN-KfmgGAxv(+o5j%^<o0+#M0}Jyc677=J%4?@n#kgPW>F-~=
z>pXf6eS4C8{r!KZOd(T|_o5ZYkk$Fr!ZC3C`xO?tGUh!fQRd^U@@CwW({2ZcNnXFW
z;o$ED9i)IG*Lj|5q*{4%y%X5s;~rc_JB+}|K={`w1Ml_a1kgq%f1i`l-zRLH?D77<
zg08E(B%f3+VkXMsIEgll&!_BqiOd8&Re}W{yU%aV#1)2Vr&9^|f7J;j7og2-EcA53
zBfLC+vtJa7yFYutYBW)ZNQ-s5*7T_jw0A3gzdzXwJt47%&`gZ!qK;6kz2OtpWN1|@
z=JF>?)1#p{*b%>%2t~KnInS@<$}KDPY5qixN5XQ3gGWr&=ZfY#@503)sk#;ShVe$<
z&2HA9&Fc!Uteo<IVWHBi>iZ=ki5?1NKP?QEGab{huZ#qiwZz5!&X!`!OfovpO$NJ2
z`oTUAF3=YpT+zRr-mNo`R6l%R@75o<=)>Z8X1j8x)@7U|{~QceF@n|6^X?*-e;Plf
zmW;pPqnSOd#50AUd?FQBmG7~{7bm2`&*>yr!>-`qNa4hU-Y71=cvFR4=p?O>7k~N4
zgvS8e-NC0-dut=-_$I%AF|q%EQZAFtmhi?yn(Hu-&X7Z&7>+>sRYvc=t+nHr%;Jjv
z!pMo!E5g`~A1O!uV@#!%%{%?d^3U|LDjLtFQU>&nLvs&tAx@9YAMb7l-2~&4RNOTe
zM?Sd5SCQ9z-JRxkn&%w4g)$lPltjdvxXE#)i*J<xcVvzT?7@?H>BaACw{v^={5mmd
zL8pEwn@E7(EwEr#Ax`z;9-On+FWe~Aki_tBrR7hkrYDnb+9@4%Cv{CXX-l<Un{gJn
zVfiJuR6s4SySMN+;?j-E+PG@W5g##eP|?#1$@Kli&(0x)U_@}N8)rXyxeie!cQ-Ge
zM55DC&YseH1t!XCEluMD1V}}fIFdo;R|!V~)uKZ66yWxUYqJy~g!^4?ew4+8lcdXW
zT8%ElJ2=hG7>TKTrFA=Yhv$#A{|y~#$(G{aSGW6uVt_X1;7HqGfEE(^M9N$pbvdGx
zvyweYFaPNw@&3cWnBuGr>#m2#0&t&;PU>lvm>3INds8sPI4&!k2S|2~YqC?S#T76;
zHzNCzEj|Xt8?4YQS?>`yF9~*4xSyG_DLFuLnIou)oZ;7GiL=cffpL4&9VOo(xIu+^
zIm`zdE;zh3lS64cxX$QKsWYZrFTm3J@Pz%lnNOBvGZl-sxP3~skl<xCH6^vR>Gk!O
zAtG&MoivIT`1i)s8J!i9In%RV0s|cMV`^VH7%0OUzwo|_S5KB#GK=KGFyVk33|+`d
zDZYpZ@x}Yv%HQ;}un!LUX;|{w2toCBSx#__c9HMaMeYrL&(;23#UbtS$be66ykNFr
zcASKYpZwK{4Np21d7u{fvH5GeI@h;%H`35D&ogdrxPk6|BG)_lODXH$gM|gZX|K+q
zzPUrf%O~{P+ICv{p5&B5dw<Gyo-CeFMz2x-z{E+Bhde}NNO;;r4skgXK*&W$APb}E
zQBaNQ^CH2vT<^yYT`-r|)bvTFGCVwFDl1NMD@)?#{sbSCSGmLIQdd-JgGfqASJ_!%
zzG7*zskNk8$;?zF1grOnJB%8<`jFkfTH_|bUX~DYdrq6Lq_3luZ)chuA?)d0V0ivc
zAL5$`&lPa%F8L0}JftAI5gQsq*tYa@;qttPS37#OM+GGz??i{3p}TLXX{>d>(NG3$
z=iIk=(NWybzn_C$G}l1LO78jIz}Q5o{iCpRjlH4BBHM(Gl_i^~;l9vZ6Ko8J0~%rB
z2N_4G101D4Q_Fo0f7}n+Jd4g;h$#dEfPfmv-jm)TG2=c#BBejH4BrRQ>pP(pcDT6I
z!KhYt67zJ1A`O0-p^Th11vLjk_K%HY7Ne<TLAxImx3^Dj<>P}pjUYtFZD)w0(+h=u
zxzy}7<=Z7@4FthfZ>}FTT>Wcls!#DM$`o%|e3y5mjo}LZJ(=r<xV}<oVdR&FR-}%I
zYqy;~-+(6wgNih4QdKg)#LC*N4%ffdCgD*bpG|Z$du^_C7AiG+66;>{C0cI}^Yg3O
zWHn|xp`aW@Uq~|Kc}K9H#Hk-PW%t>XmlFGhMDt}RV&W`*x7^~KOCPhk8t%KFFx6#4
zMzFURqfqW#n$yuHyMqeF?!L&5tbOA;{?uANg|TG7jA4erfc>YTzm=_7fQN@1n~EBj
z)<w`#=L3PfcuoIYQObDhADBxRPYWJ>?sXGUF>ty~2~X@ql>eUH;ZOzd(%p_&4C?m2
zw5zDIfE&^Pc0pMEVCDmw5=6KX5fY8yAY0tB@4h6SLpdvuLTY;RLGKUvZ{EsgOS*Pl
z?<799l9_e%N;Y+-v7uKxB-Hj@1;qCG3@+1u7WWLP;FGH4>)nk;!eiFs@cP9rq>ZBA
znqoF!yi=2(t(?7d)HZ3u$NB_8PXe)KpNCAJ_?OX12W~Xa<&m8mpP+)K26{>#iwSPo
z?u$`#vyKjFXr)AqX#I+2bzT4cdm*sl0}+^3+b3t-OH}mj&kft&!ckmWNZUoEhk58-
zdwV1?C5hIN(VlrpbG=1#7dCf(6@L-XI@3`TLyLL44jBZ4vv)T#;{it#!`{**CH(41
zqlK;Gdw6`lrm9vDk_T4SMv6%;=sh1ext^Xw2p>PC4vSP^?JtK-Q+4;XB&jio^l|x?
zn8doYmY)TD?5<|P)!@b;af>G}qMTt_?m7$e^+j!JpYmWjxU{kmI8JaWQ0H=CG46c?
zc0WCXG*L~)HgngF4Ku&yhBiY{ldsTg!@W+CUR93sc;Tj9D!5^|JM=j^zYZCjvR4Td
z&!R_J6I5FMy$QxIWc9&3Z5|^(@rC6D@38UXFrDiHudKflbNk=KOlqUlDy<zht6100
zRa+30Xt)EMB3reHDae^(dIUy;TZgwF>xAIqc=(3Xe$}4wt1T4v)a%)bWn2y4${2Qz
z<NqBW<<rj-6c}95TL=k^rO0RE9quc?BBXU+jRVIWA>F)nxHbMqDsE}+Rcd45<2RBY
z4c9zA*6)q<SstARWBJiicQ-kVgbBq>qh`ELqu}64L8Q$s_fB;2jxcixedZK#r4v93
zG|XG8!skewyGvBXEuOpw=kmNUXTz3hcjvW46u<iR8|Vq!CeWjCJZDJy_6fyAnN2Fw
zN;H-H`&`b5t(z0&aoxp!Y$e8n=R2?mKZj1y#mmGpUIO7~G;rdCDLj|`Kn7+tweQ)N
zOoiOzA`mE8q#d5n*54h`i-ChsU^Idh9#ofTcKQJxj_yK6Sgx2*=irrhEuM_C>AQ*r
zEa9#m=qGHVFau-k%Dtg^GF)U!BrGhv6^Q#W{WXVFi}&rDGG&9~*?fB{u4h*@_TSE~
zKbMvlOzl+|F~yyKqPeKWm40K6zmO2u(k}XI#CqgzZfp{0Y8ZE?@k2VAIpWWepLi!z
z5p&X#IhuxpX3tcDa&OBZ#Go@BHf2baiz!5<9bC8Z#*kmrlt2me?5ceuK8x6h+^H_x
z4M%ZL)AQ~bbV>j_{QUiaDToq2<<GvQtGLA5Of!VeTUZ~33M|_Pt}(5d@bbtsJG6X!
zfl&4oul%tI2MJ;9QR8^$12Z}J3F@cKnv*i_$}eP{eYEOuM&-VT3k{#eLvh^2p1ytE
zxeHY^pFS<LduIn7Y<Vl}PG8Vcj!DmTw?>y2<A3oKWC0SZy-Sjm{Ke;HO95HT0I6;#
z^U>ejR~1k1W=j3v@%+$mjRbEvY;GIA&uYx!McbxqJ#+Iqo;~Z>>;j%@TIS?<&a#rR
z%p@RO!emg70eWiR5RA<~ePhM9ug)m!`9PJEwP`TZv-coF(HOskNO(15XT0nPN!xLQ
z;C901bt_Yofr#PAD-j6My8~?WB$s})Z7LzyZXO`M-}uO74s){aeC^v+V>Q;mH_)B?
z7M+VUm*fF%`FnFC4V*FsuaH9Oe7bW_IrO^5<q?SF*<x96dwuSwR<%D|Lk_140gy2}
z_UN&lZ02pw6zcgWL4+Pl_|^HUrv!Vt4bdzWS|1~ll0VPl?i^IR?8$~xttig2rl&Qq
z08@HuMV=w~m4o+Cv!qAQX=9?L?VAeM*cB<)H3gh<FC@Ujhk)TK(%ov%Y|1<jHN6Xd
z{G5Rdtt8<&g);p($F^V6L>+!P4OT?&%*S-0(9kDZ7Wma=PG`0T+k-?v)!Ho08_9?2
zws&<jceXxF7wIvaswSc-*%YU|RZLz|fA7^3ws&%vzl#PLSxX*-XNgy!O6J11cfQLF
z8nlfFSL3ep@Kp-8nXXNTel$lhkI9GWW}6WUh7CIXW*N$g+i?FLujHKQ<=|)^Ax%e9
zO>>@a5UB(u#slGGR)v<e!F;EC*T4N!^~T@D(;6E)gSrnfG0nunOIh3dX!taHeh1YV
zySrRpZZ6WSRJzHa^%8%|n}D4cUFQ5gpz@3O(f&HL5xN2qQCLIMr!bslls?zWcz`6Z
zR<7b;b)W6(Ta+cJo+y~;LvsRHO^M>Htj)wPwgnlB3NmeLC*ObggrqoR@+1B*I1Ve#
zutc+Fv2Wx%ODaW6G97j6WG`OuvYF9z5`*(OqEepJZH|2V?yJh}*W^~NN>&OowvfZ}
zE=aYibbw0+RH|@j^a&%1(dJVz#b28Kyz@uSQ-2|RwviB1wx+&p4KmYbjgwrepYn3@
zKY$j}u`z<lTMy}VNo5-tiWd>f44QZIM5%Hke0Me>Rzc54j`(GHzRP}#3Ib{0ECtDB
zv)OL4eQ8UlPqyOYDeoCbW7<7&+YpE3ic^JkUZ*MRtt@IMMO{`C+UM)E=Gp0lCd9~Y
z<LY79AO^cc_{#K!*I3<EP+9N(+u$6x;18e-j)`3<Rmp51a5CqCBXUN~&l|L5c5PH-
z&p-dX>hD5L-N{g-UYb0pOCK7J83Yn2anEE##r9MNmsjhr#Z*W*%$ns%xO!7kep^_$
zD}o{oVa#!T+L0_!P_ibdU~|a#CarYe9%DE!YJft_H$lc-W}z1yg9zZ45OkR;1Fh(_
z4p;V*fjTsrBFn>MxvBwGZs!m0O~u@mp8>t*%U$RHSud@QIO-BnMp7d{L_vUGyKIB~
zez>^sZTvATX~`I=90|wV7S-0jH?!rUr)zB0x2M8zcCeT8DD|btjh$hvM0P{eWO>Ty
zYDg@$;3byMJUz+B6$>59_ZKw3>dL&f5!yBsbp#!Htv_i+lG2d&k>c>Vxg8!IZd77|
z{l1YqTENJ-yUhrrF70HdnuXuLf9*14b3{CcM3%#QL!vD`eLhR!#;W%x8BDV~nh|Lh
zmfBFzaRv30!3Lb$v$>bJ>Q)5d)a_Pn+L*Q~Q5#_nNa<sB;^H4kTRbZOBa)8vj+jbP
z<M#(u*B1xqb-Tx>(>&KN1lI)ZebTLslf<H9tq!@=)YlX{)%}Uz#tBrpT!xnBWUL?)
zmgK3cFaV1l@|^RnxI>XJ=qzB>M|Rn@^05@3;lAo5XMbrIY3~Ke?4&Q7<B^keI&VXj
zgigwgAW6lFDyR@$xi3EW`5{|c+`z?Cx@&Ip*YT6-hebY>>F@QtO(emvxcZ@dhEJOd
z`D=<PcZC2HPe^@-5aggHd^RIt*j#W}Vt8|OAPX%~J{|RZ^CUIZ<g-?RBYjV0`O)GP
zeL4hE!#0<X6x!LguAoRZKg}0SQ)ba?nB)m^ak{1fTIG?7jI3sqgf|j4qit(KRo*vt
z=Ax7hvP2fgb9z9}e354!)U)K%)tE?~IdQojo+G?$o~f^Cf;dI`4%&JQ7KzZ?hF0+g
zjr(m|tG9wZQ*aD|cl29A*RaYjCilkuQQ$eN?C7~N<-&u4AC3!wF#sHc$GpA$c;1J0
zMUDG}2`s_t{%$&|v(q0fbSsup+vK)I3*6H|<wMi}wVFe_%iwG6HT=I6;=j1nf&mR-
z>wPejiP}NGzoVsDnQr>PP{NP1!XA0=gXpf|a?@&?tp=UOv@=@FJB1zTTd?we0@r#l
znU*jXbGvPTg!sYx;ijR2-*(jtsx}gTG{&%8z50@Y?shOOzmARVdw4FRhE9V;Qxao8
z?LB38=|wWL&>Bt}SAEX>z7z5AG1NoD8VmrVSIse{h`#-0#S%|%{AJ2pE4EzWKg6g+
zH}}re#0XD~UpYlD?RaY>oEDG~UhGjT(qJj*LxHso@nv)U#uBHx*BF?KD}x`V-cF*r
zr8MS;h_)fE9dygUz0%laorpHV%9Fc`fIt2LRUx_p8`I`ZnNsLyzD4^4oz1S+(aSS6
zUx&aB$@0$HK5wg|X5&{mThT>_4gfNkWX`)>SdLh|%Gqr9LI1Rp+zU8tt;vGAhJ5qV
zSDnI@1$WFPgN5lloW9=6`8=CblB@XPRMnhm*HzZ*6!EduWtLE7g!ClC5>w5J)u09W
zXDJgkbF_f3LWgeGD#Ol*;W2I9y<x;Mt|;|=T`NBo3p{j&5)zNFa(Rx)*=T&?cO0Ix
zFV<_dMPlMu&IOZSR@gpl@pO`OKNI9W!E+nKg$cNm?N2_~jI~X<%5VzL8z?#uWZgMp
z<%%{k)wmE#Te#c!>y3ZtOilva<)#bTAV~MqMzdX=&iJLWg7nq?vjLpr&~VqUjZ?PF
z7<tdE>{&!rBz6B>DucSOwZ@2m|7?vlh~5DvP+E$Ii@UeAGiGIPAq@f-QbeHs$^48a
ze{o##>ihkou#O%dGU`j#TZFY0uY8&t8&sP5-TbO;?;08A0(7^_+H(b>R(j}l+UTmH
zLDj%P|0<sK$1!z3f8;JA8{3_HOf2mF-#3bqxzn~WZAA<BP4P=sz{$ze;m=kGTK|=9
zQP0zLJm|8TvY0Bew=(#AQ)@60gsWR6^kwtMVAEP&;^I`$$vAgDIeaWk=)o;?TevwH
z<y*OBPjEzSbXmKd(-5?=Tkit27TT6-VDiZN55e1WGG8bDpM(|GHmlrgsXQ=Mt@!Sy
zQA>H)UIv0n=~Q7W#d3g)H8!YTwDXm>uOW(G*soxOoeW(t-IYoPv1<3w<_*dn&Y9t<
z7z5a89DGV*rNK5JGGDIHlPwIrH2#Eej~RZ1%!7+gUMDkbE)8<&Oxh7?>Pyxa4!SuL
zIJhaUILT%jPAz?qPx-=WV@lbhZa2!O!p6+XJa>fdo?=p0cnc}!n}VK^*dGQh@Ej8M
zV&uCD>E$1vbjQa>w*rch)>pptI^Kk}#HuYxP3Bo1@rrRfUiGBy5Le{iI2Z{17Yd@&
z0udOhE8@HpoMJ!gRdMb>pBKMT0XbAqT-pY*HIYl;PrbgUQBi86b`>Y7%GplV=ajtX
z^KY-G_K8FtZX|UzUxLV%p+plP*==bt>6L`>tKX(yFk=Mk;+dMOB&V?F8D|-xk4RsE
zI2$+nhSZ^FjgYrD^yANLAadO|^*%VBzAUy+M6oA`q};rz;!aty;%Ui!1#PQbhA(gs
zf~{K}C<o@Ar;@iN8NBxitU;l3b?z^LsZPbtoWWH={(-21*>nv_Ml$g#s<7qp9F9jS
z!5CQXV*k`aIO$*H=fkaQw$l9_Dbb+)Yu}h1H8j##U+=>dT+<n4%>9s_kBb-%gLZ{x
zg}Kxko|KI%ue&6<@g7(O#u6*hw5&-qeF}%@bpHO;wV}L?S?k!T%6Af(wPUNtMG{p<
zl_<9&%f~O|FSKihz9myFE`1CQiSa3neh{I2oIz6_Q%b+O7D;TYEU$T8TM?cF(b2=b
zhW*?We;=4YOjCkyCN{D{O<j4w`*>jMA}44pW05{zPMAua4sT6t@sokB(S}c<FBc8b
z$$jkGx}!YtKTQV@*duA3+v)aIczu{lhosE)ModC2Jcu7DZS{MoxL25nUafYjMr!QG
zX|&u$VPrt&+M^LbJdw;}Ablyb2c@PpD`-afCwTa7AEZvjwAF6j4q`<uK0l;Ht3?F?
zXK}^au+8`~$dqhxst_VfHUNnuU7Ku;!<yNw3N=*!4nyI;`UeqNW##6cyVg0<MBa0Q
zm(65T<u0iQ)FAm%60~NxAZLLUnH_^6C=2}*mMyW$SB&f(pu{2?*{V9hd0RpQAra8M
zY1@m1@f-a!wwcQhml}`|PKhSPr<T5g%1Y$KiGDh!E&b97AHS@1256UC(j@#~VwE6f
zE#jS=PsJf69NnMe9NG2(zPf_r><OTZ@BOqe^#}~(I2T!7{6rBNZW|9ZRk_da<|Xc~
z@qVVZaQm8VAThoEw|H&)7b^$g;sL51EF+;5U?7l4$1Wx~Vh3%GhWh69d?wIv0@sRi
zu|9GN^*p|BnZI>J!obnEl`<pS@x3xAqDE}jc)YIus8|DXaEb~kr1MgT7>N0zHaD~!
z1R?oAc3UTK?%t)(><^LR397;Pzg^@b{FKm{j%_}{;B!OZKEm(VbK9tJx2B^*s!ml%
zIHj!xI;1Hj8XX9B2O<r5`R?_!{34eo6*xy*LYeG0Lso_}C?)4Rbu*ZYsn#rpNDVIh
zi*M^c=<|&PMw8^GgIml!LUmig$mxtgFsZS-V&<DPeF9%&TTyFRVx$xZ$G3<|%7u3?
z7;rC8_1Sd6qdI@$5!9LX_wdEyan}NYZe>(%=~JNlrIMmC8_qF31p`ac;2wbb(1Ijf
zt5kwhTmIS!@jhcInmNdpFX#`yOP6Sm7e!rm3pt~^VFKvOy;Kg7@G+j*!F{o(OJ?NA
z9oz%!*=ri36tm$QeL`5b>>s^3vx@~d8>)+&0D{is7X^R!8^i5;xG|Ic)nhk~mUHBw
zl%xQx8(jzb3|b;IT6YRh$c^IPeMfu=3LTl5xOZY=Onek>Qf*fLovgK;3zPbsDKp$2
zkVoNzKcntnU~^_AvRgG397a8}NC|1Al!1aU$)BlIb@F=#WxeG5n-BAd>2CW_4ut6z
z@0%A@`0vlQqX&n?KU2$eO77iRyi7nJ;9#8J%k-zga;Y;J%|dbW;uFyp;q|U_xo;*W
z*YXwX!Oc;A2%(i%D;;;gMRRHunn0}ugitl==9}cpma-hzoHjD+BAo;pfKH9Nw+M*5
zMfg<$S3*v}55KX$8`dZGuY`l2C|-B_%`9&?v#}-)Y0Y$|VYww-;eAf5jXMfP$X*{T
zvT%K~fKBwS+Ww0GKar5O$R(d1=n9`+og?EMy9a<EV&PW#m=ALMTGe_;z!VLLoQ%8v
ziz=Lx7hErXzCF_7yC(CljfV4UJUmfen95M|oW~EvS#kH)(6i7LJTBTfn8W-CtMX8n
z*R0=T>$5)F9gEVKWOnI@s4{;nuNc)Q(<NTEb_*ZD#m`U-%th+^X>=m?F_rxHrz6pD
zzrU2S5Bl$tz5ZzMYridGr=Nv2ZRq|qzUuZ?+x(z0BS7n1PVW`iR5Azz$uO4P3jKf0
z#J>aCHFw7m!_}LPzz@ZyVy-U1?BLJk2xr{MlKB+A3(ST%T)YLvf<7JA-m&<rrC_6J
zJ5H0wGvKs<Sr5zX5fHa$3q=iphC#2W)U~6o>}t4E^Vp-K)ES`(?Wx$(=k$k9G?q(d
z=WF{w0YlGHnn4<CLOJ%^EP+EsCa1M?)Yl<mt6?B*PitlON{^^xkHhK!&(|b&E}TI=
zE&1SE@Bj;=eqZa-;_^pje($@|i|?g(&ZFA|!TQ95@>pJ8Q|Q1vj1ele9Oeo>`5fu{
zn5%v*rEchiSSLE4j{TV4wV*L_L_)^^c8nD_&Fo#Iv7);$ipHXW&U)^(_^U0H@agUt
zI~s`ZdXp~MI}@CjH;0FhqYB^C^;Gj5{BbyVba3Ut2qR%Ev+zp`!$E9SUi?HJifwnD
zznm7Pp`h)2f#h0=ngfNn0P)A;Qs8(brc58*JEtXJrN1C~kG;k>&;?1-3_nSEBfM55
z1m4hJLdCmmhyRDla(?4z!&buCRj?^iy~`{ukTJSjel8bQAaaqlTafS}Ne%zi7ea3T
z8f`v^^FZFfY5obwkSC5*76j?w+x>H8QMD1MwZHuvI96?|ZTbZ!SIoVq+oVA64rlzV
zYsHO@8Yt<z?-k$@bj;^9y(e2zC6}of;BJcOCMv0W;+4P4Qyd;U#sK2@RX&q(i5GkP
z`7OahB|H#dbv_@l4nE-l0O40nRSZ*bUF-Hy*F(-y{~zr|sX_@?E&-Vjh>!vxJ<yL0
zH}=!vUhxMS1W>gs#S~DsG+vEat!>Kv&1K1N{&mQO5JEz65k7kg0t7a|!rC%E=zBPf
z3tB<zS_gx<g&YjMXAvNa2wK0zH0rPd3iiF;&X=073%--dTq+b=9^%;FpWA-DB4wfT
z{Z{^tc`(!pvNZoz6UGC?<Os8j;NnCz9YbBwGvRD<uIGNBf&dNY8|Ofpiv*fi?PQd+
zL^c-`hKGS!im_+8iNV8*<1|Zu@gu|kt`X+q5?tEj$*e&RR@$EEM1SAtPvHk6fHru7
zk2#j!X5)|aEa9VA9>v7Q)-^SSa)vr*Z(f9}2i6oJ9rqF|b%yC6>}*|5ZPKhfyCa+E
z|NGOb6A{~0v5`-M(-gXBmO8t9L}B$0^%VTqsd6V<guea0T9J8#!Y{0>k31RrOtL4}
z^2oHq$6M!@9y$@)+_R&I3eU%2ld1zwX;wOG!kZ>XGdnAQ{{Vc-I>LXV;R!u49o)Oh
zxnKqVYwnc}AM_MP286z4vyEx-kpdKVSFW2!^T~PP6%o-1lA6P8_ky+>n}@$p8n`wV
z>?y4O6qlO@$*6a`$Jh5EIqezZvOSgn=MF5?JPxtGr763z^KWWb_(y9ENg?hN;+9q{
z6yJyA-}(yFe(Kq<&u=ady54n`luGR<Voh((6flKBw#V=5#jrBt!vt4w7#{(!FR^m_
zI3E^*4$10P2a?+Mxp)&ts73DZFv~z^X43i&#>noJqgzwzD;cg`xSYmTlFMPfxF-&S
z>60Pbx%WDGIZ;9C58yB`Fs6<F?jyx|CH?EWN>wpfe<t3_WK+so2u=>?c=ZQZh_4bC
z5y|qIt30mqARbmR%s729Gl~33g>2?k+aT!Qw05+J?r0z56K{Y6s-?9THTVDX5;XM~
z+>a-|{EP&vlgv+71|4eY@dpCh_+t>EnU$3m>`O*+=|Uxf<zAl*b!Xoowdi7nS*kG7
z!ua6hHf-||wS>D67XMG#3LzxPXGb3nBEiL-p9$|XW6(stmgB5Aet>K(zb3FQ29Syp
zf|Bw~-cc84<q3|WF(0Djv}B~WKnu`JK?@_seAVZ?U8`j>ddAR@PwHuG^+{*HdXDPw
z{kDieL%vEBQ1i!~FOkds5N}!e)i!^!K?{fKy6`4mrTk7}4|Dvtf3GW#NJMv%f<wcw
zd#)KoLt~$+Byr}J;?~&>Raohh*-LNpvR#GY=N0QBAK09&`A5SYMimqGNn_9unk;Sh
z%o;&FAW|;_b#zbWN>1UEpE|SYHdz#ps@3!k&Lu~W&_IMD)%MK|Z?RheQqRTzYu`M1
z>?58wAAXi?I8jPT9pbU6IUEhzmos)HdcX^EOh?oS`u|L9WxCsgu)<=ijFzmO%|GZm
zspaJlL8t*u9LQ6=?GLN5XqDW7xJ`}DnWJ1?)P%-<y90xsp+2x+NKBlxSOTa%ZB?CY
z_~=BZt@otrJH8!pFuitw_TjgjncF-n^!3GmxrGrJUS6FZ!@*p4?}?sVHiqnw%uVg=
z%n$6nG3Dd_U^@5ugEmF<BNgA;g4ppcq&JWp=$rRU^|`lA)HrFu8g(w+9N_xhH><cg
zmee07yxE-z7L)dQ!AA~HtI@XpN^Y1LheRe}nmO{ZLIehZ_2;df(w4N5_fxJSy4(Gz
z2zn^sG`tIXNXl$U06qoIP#_5>1zTD~j}&@v8%~6pUL=~|U3q*`1VKIhSCFd2O^D3y
z!9LeFZ^UQSD=F4(>Z*TO=GM3ZqBybK5K`3h4tql}lG}TFLG=LNre9;*n_^(-2ST)Y
z+(iBMkq-0P5_)!_xE#)A3@6j4m8!V2r6+B5H(#S9E_^DA^QQ{w&G_&RTOILF4)d01
z?l6e4_rIzap9%sP7BF=qm>U&S$cH$-LT_NQ`<P0ht846$n?k@JkJSHLi-9%f{G@PK
zynqC*bejkyFQv{;8w0*zqb(24<Li1Ewi%p7&JP{Uai_02qGI7yd^z1bbH1@6|2|+U
zDXMR6@>srdE!fK;`4f0|w}*e6pNMJSeG4h_s03q|I~QkcFw>CJpX0-cU@KYl<7xab
z@A@7QD!Zha>@K%qw7WU;tZ$p{FVB7J!~LIuMUFvD{SSidk3^*<P2sa>xTvUT6>eoY
z%=whOkS7Fh^qrB9rOi1j*jgrgD?05)!Y{iu<zGB&m+llw$#0lyK)!(Vdozx*O6Ar6
z4vha_9QJYU0Dc}gu!Y%iv@C_TYy~x8qMmwSBgSj(AGv#%fjp>X?&jt#(Cg!^-P%50
z+{re|nkLlMqIDhj$2PA0b2b(EB&5l~%JIudfW(=X2XloT3zELgC{L2j|Ebuw!xE5`
zy8YXfbQ$MO^4|(Cn2JUU*D|PtkaiH?&qm!$yOb+;wTAnT|6V5ZLq>rsIM@sTWE8PA
zV$py+1Qdw?efpMLClJ-S_YPOU21;NEl_M%B&)%j?@S9*gN=gefCS&2*j6JSz9>g;f
zd#={;^rf~y2N<K~)KYNYxOsp&#EFM6bZ)bH7vR28CH^}YLN$PJ7;TOum^?Rl$W<PY
zhYtNxNC5o=5bV7`69I5~n4Rf|H^hm(PTyE+c?MPqZcU%Lh<)|V`#OJ^sF01Aa2`%p
z`1lmXo~;I2-GVv_UOv%ue3QX&!Ig>qb!5MN<N}K+C>(78dv2aDqr1UQ_)JZcrzGNl
z2gm9|o6$Uj=-z23o@(Xzu-ww$b$c;vr}g<Z5(5hpv)%T|PkF1eS!)m-l(=ilNxfNO
z_2C)!=OTA*Pe85uE~<g6gYH~^K;Z$x8t(_Ii8}{FiG~IU#?7BIV?p4Xi-hu5enWr!
z#K%;_biY^>7)?50E@Ac0c{~V6*zh*lNct301qZ5SM|8(%BplD#O+KI0?7|!ucRWw7
zGdtOeyTN2o{qNLdi*=zDZ!tdsi;Yx^FQKjf)>E&9T=*8*4(bUqvX#yI8j-;4E^EyM
zEL~yFlDW_{F4qOFa_<ZlfP6#u5oKV$;<<Ad!8#N)ys3?Bqp8W&K#N||k$bk-IvGCN
zYU@S{Ur(^Qtq9%h7<%cDzB`h831H{udm8^m+j;-w%JcOShsZ0Bl6VX?vAIpxI+Tv5
z2HSXHk<s;l?qXQkqX1(kNiH|!q3*U4TzghAk#n=H1$SOo{!~vQU3_G9%9@mjMD?AR
z)zdon1N$^xlg_n7v{1Z6FPDqLV!U&aG~E$=x7!~VR3P|Ln(%T@T=KCJ^<8Mo68{IC
zlC>2!*7jP?QO3F0Ur@67#go`JIV^pge~X?%M}6+{Gx<9<WD`<vb}-JHj%StR-Os`J
zv4I20a5!~DYlXAaf^RMJ(QQ5#93ti(nQ(45iBhewJa4yurrT)ODW`>*+X%)41i>2b
z-U4#&)D;If?UO-q2!3B~Ccm=Drv2R=H0$=5zu%nzetqh0a9?7DzqLa!wqpB)76P4r
z;S`Gi-*6Ospj(IZsJo@=u#y^6)HG()N|7{>#$1*K7OD{7^7;GLHB(yaqbR0{SLPEF
zZ+Tn_No<l$UY6gZ+BeeH6hS=~@7IRqS9p7z`CFoUEUcGg8U2*sq~*D$+}X?C4!=_S
zXWZOIF)oRT_U9<jg~p_uAG_n)NN5zdoP8wMqU2-$2H-R5<3OY7Op+Z4ieeZ|%$K#n
znO#p~0hkteO!DOe>2ncbohv$Nyh>OEt_Y`9#9DaKfAltHJT_BNnHS`<Q&?{ct%q_w
z|GZJ(%Jav$%b9KVyz1N&6GC6RyR)P92JTe$ZA0-teqe@$#C$IF#K3n;p&8LPUO;<o
zEN>dVuc)N_c1?Ww&yk{MW*`qp?jOlNy<aodSaSV)8=e1)VpC%|rbO_f@e5eF*urwM
zMp@Sbs<{J5;<2rEV%bV?vMLLFM*byAz2tvS$G;dOM?TinyALg-=i6-U<M|;Y%VN^e
zgd>-ze+~RfOGe`CX(iK}v(MWi>>0;%L}?JT^>V(`2{fwZOjyUqtNOdEShUUqThb5U
zhK;4dm*p@$X>l%h1<rY%ziAVb@=MELB%#%y1f}dRo1FpMZ9Zf&j3OGCf}jE@%_lHi
z9&wh2wVU9)gCv4PQ_{Ty<52+T9!P_!synRui+d}TvWdyS;J~(6GtbKvlV@b@5_d}Z
z!`HXYq*Znf`9{Y-fU8(j(h+mLNa!Ydl^vH&@zSg1Nu|K|cQ>>jThryf#|LNTK42$4
zX5HBl^l#W=Dh*9kxvPFP|7kaAFqh~HNR#B5t~p$%KpwT<TYzZ9tU@k}O|k3TFW6=3
z&Z)^rqLtTfjV|LPb$c5e2k!#>sciVD=|h3d^p*^5Oa0khJPckNHc6E}!Nhj;W7Fw<
zad_#=JO;du<xlN{Qe4~{8YdHOy*YCj>B*U1SfM98!r4pxd2Bfs+gkF_vy?7PnSel2
z*3*|f?iX>uUcNtkC}a4?PL@c~*?_pmwcPReq1c=|C+kL<4S;W=@BnWAsuUmc)sx_W
z1N42K<x(pC02AwD7$M&0q-drHdiD_|?UHwB!7H<rv5$BhYr9Lpu>?t}{L+9F(VxA4
zVdD9rt2^#_tCyWD^v`7xS{HJc8V7;?DsWu>3WzUFi5D=$EKC^Pvro`sWG+iDXN>G_
zHWAVp7aSiTX&0;M@FQp0J(=!s(b>t~<-PTDU0ojl0E|V&$Cw3dXPiz)P0gJTga?2+
zmNIRJ2mZOHo{e-TsFkXfk134u7hT>iWUMVLH-7eKCFjf<+fNx~yOpsKo?F8s?D=XY
zY3@V%PYnZ@%)yj})*ZH>H2=6PN&xH!qibBFqARadl#N|Kn8VlgoriQ?+=I6~z!M8I
z4_;UtLDG{j_z|l6TfyQP!SBTEHs~cU>vsnod`Xx}Yrn-LuA;G$)RahC1*+XFG_O|f
ze$D~Vr(fwe0t4|AM0OH2%{2qHc-~6K5AThI0PO{+_PuG+A8hOpgR*2g`cxy>oo3;#
z<@Vfs^!{RV1ak#P*LcfIr4Gk%kp&sWTYnWVCNwc#5|mEhS1f^)4{|RF>%V%#dp_Ou
zJGkIB8sZb)B`xSlvOF+Z=<X3L$YMV5{>2BnhWNlVeWMpp8^Q(DUV+`t>)0fV+sf$C
zr5pb(wC-i8IbCi0P5bdCH#IdXcm`MWJB(j4M{Dh3Q$0=t?G;OJV-g-VT94s`1@co5
z4bH-4uF{)JYTy&7X1)<s%yQv)!jYp|4;-$aYax7XBUCb?Sy-Sb55lT}#`%Ui+N;{u
zv!RP<jx+&(mH1N`^9fmb)ozPE_lBO&CQ%bkgF6*Eo$R7)=eI;1*ct&A7CK#nah@Ic
zpVGs`v_;U!mIahKT5kM-!XdnN<pBose*IMf@E#=7HVAm}LWv2!m{hfLPOHDsfK*r!
zor@FuQEp0*9!*w!Cl0I4vkeNJzXx6j!2KLbxyq^%ihp1s&~mIMaIfFJ+jTO?5hc)L
zPoNc7gL5!ri|W>+1aL8D?YYsspqtuJzbJ2=8S!$EvYlMz?K%quX6O+7yaiv$+Fe~;
z%QcZ+z^j%JPr>Xun#9}#O+!Pg1Ow)Wyofc`MB(a!hPz_!+WadCHo02tv7n&=i~=nB
zGtt(uRX0E7=O&C^=q;b=>8^Dr#hZ*5&1JjvE;u-97}<=yLy)NcP^>qxyj{n};`zz!
zXCPOI%GWdFh9N3H>ys!%!<}TiDb#W=sg&aYr$09dv^q++K37Oa|1eGg;AG3?6~X&v
z{Mr&DGx3@656nn5J%DHS))X)HK|5`pWnOUaVhalnv6>r=W)~fWEw@i1g^`i%#o;ao
zaMjjWH~2MXCZ+?AiS*D+uF+J=ryfi-u|k{|<7dZxAaC6kz`z^u!)W_Dw!#g=tEM)Q
zLW_iq2crJ^n88g-Om3R^%dLUwp<Mi50po+Lu{-#rc=DO)G*AD@M2X=cO>JTt5FfT~
zAF1I0vVJS}VUDP62DPra*G}jCd()1eMm&s#wt2g}*g@K;u1*BTf&2mxU*zHNH=ZE6
zSM7RBp=)V%(bJgf(JGF_lB~PtD3*+i&|*W+{WB%k4PP%(cFr&LN42=oe{LUW1axoF
z5}Ir+Hvy#xOl6xFhAd~%6>O?TEon1UZutPAeY#Q6qiPq+Rn=5{0?`wR%eiC2Sv}`4
z0eF<Lgw0@6R0iJCw&f*o<0(@~XVbNUXGwVps)1l}^~{$tZ2THYP{!mSc~zkmUlI`I
z-g*j-I3pgqpNno1IMBK6m~$$B^}MsA57YnEm&tFDQ4StfIpTjN4y&tJ$*S0QM0b6l
zae@Ydr35V6w#|w_<(1HYu@TR@*(<^rC1>IbifO|E3*-k*T&6lzi$M!$m$g}kvIX0t
z{Sw{?sjrS^ogfFeQ8$jd{k@YyvMzU|pG^oTOeIE&YmT5i2Tzg2_E~H=uZ9}d>GKE8
z5j7XQbg1-8iNTq~a5y97!M0zlgymEDl}<$^e=!9&=HrQIO}kiNW9}9ea8Xm%!X)Ho
z;&0$n*h}ySqMsL0%4<e-C0S5d&Zn=mi_y9~nK6Svo3hZr%7UImJ*Y5^I@tXN8C&{D
zFL=$Qjs7VSW>AR<fjYVqXpG>ykvU6VBf`P*9g^tiGq<1J$=&sDn|G4R2U+c8%T*Q=
z_$qW^sa4Wk;N$t5V2hwd@zilW*t!i9d(jFgHt;F~teM)1vztiw@_y1%9K>55ov#0i
z+JRvP<ANjNx-EsCyH8L@<8+S#SdhXnAq*EvhiRK>%6%5=3U}W7f-D+<-DKMkCqA_R
z9tr%n$A}=JN3GM!B0YCOfHP!q#qOyTXeb<!!N4oWf`x~1741E~r2}(|o*so*z+#aq
zgm{}3<{LGKjrq7C{!{)fk=~{LG7MkTDe=>eu43HbJn5aaQsTbHO_S3eL^h+AOXD=)
z<b2gTm;qVX6G+v`<a@v6Ux4Rj!2zd?aB<ZHd;Mbyh31NT;S+3hANz90498z7V)4G-
zg8}F<H)#B@@cBgro+Vb__L1Jk9fbGE`ipd-=QNA<EW<Td2uhs`>n!)e)q?^1w9eN-
z<CX^H9WfD!P+z^!iW9IQ6dbL-lmAFs$dNz0J}^%*c3@Lf;d*^vb^WUYd8z&>2UZ&G
z=du<IV5fY(kTAqI=JsOV7FXOCn<FtKgYBGqrO?*c15~eKD*4~TziX!>>3EXV-G@HA
zR28lgXfW=k(E*Re3&R!APdv3gYF=k%az><J-Qtfv)`MM{q*Z{cFNA2mbYAFx^s51-
z-`Ws~x6;XK{zBm4r1h86f|O=&B084dhOThpKD4wgglRdW^|8pSJd(O^;%UF+{>Xv(
z&A7k1T?qHh(7$Ew0O>pEZ$AkuG(H;fbGS+D?ZcMR=I+t>oK#C7&<YE0R@Nt)%iqD}
zXl1dZe+I`<{sn@VpZ}|&St&8UHv=HP<`j!*U?rpluBK{~aM-Rfb-Z3X$fYqEWB*8@
z`J9g;Z5zMZ@2#^WGAP?`#?2cmmR^G-`BH!<8Vq_7`Xzk(Vlc#$Z24Hq(s(AlTbaBp
zQA|X#<#;Dosx{&ls=Lj-uQQI{qh<+SYWb7#P$L9-5HF)Cm)jZ2=PPesLzDB*em@&z
zDzR_s?=BUuFs~C?E2JKgyDNEw;{Ul<X;L#XY-Qw9XxLTYP^HW|#F1A-#w(Z~&p?~q
zgRQDcv7BWqz#mHWr@f=MeB^UchF7p)ne`728NnZs2?&>0uG#a)8*!1^*M8fRvYN7z
zR&SmzA1}dzL!z}j(GM%HB-i)dw}sQ>7Dkh(&52@0?WnQrFGy?GFK^yzt4oL5+?}~$
zq`%S^zhG(;mHxqbC?<&*WUr5Z`_ib~aj;TQ2UsU)!X<QIpXZbB?H6UwG{0)6?wK2?
zHj6gV$4>K($5LG2U5@w^`&K`q3{9!)6}<pw65QM0q-hV4OWvoTZ@2ajq7L7v@4L{7
zIvubIwJQIrhxMJ@C}7fo@+GxJlazssI9weagBAxEg=kOetN%dxMTJUBGr=QbeX{6$
zf>*|FeIrX(Q~t?HJ~dC?436Ob4r3lzHDPopZp;-_$eG$)udKGlQ~Qw@nm;<$K3UvZ
zaaYv*2k;;FgNHPuH@n1d+%Kr8>1k}VVD;anYee_Pe>^DX$Q>Ws?0J>b(be;UZ}~-D
zuJ^hoqye6_Oi**c13#AHa^=?|Bsil&9J_>t(&DH=H>EC~)9%q3EYO~K7LA0Q_J@_K
ze(vKqF3McR5@pG-car_C&ju_TB0CUYsZeqeG+0xgY?t;wyI`Z~>{<lPyYl&}FH-^K
z)(bgtHAZ}kY%xdf_$8Hx6KQP`wIa%ScH<DCN@js=A^DOBFPG6klzP=#-1@Og@8HO<
z%B*bmGrWp8)jgy)vB6<sp-S7h-q?AEuJ*Y6dQ~IYe2yc|_}xS~?b3drU5q8DFr8>I
znR^nKb^tAqe7s$if!)6qqTy=*uT;<0Y>&;Z;Nb)@OnAA`+B!?vyeLK&p_SC{z_AqN
z56wCl(x#(Z=yfmT?}3tLz;OGz@Mqgk-_27yYL330pWqFjT^`f<yf&s8lSu1QeuBwF
z9uDwy09(5#kSNi7(zG;`bTOd4HO^JBu%CbB01uASuKAe%cbm3ZYwU6S`U{okIfvOt
z;HFubun@BF$gl$c=c0adh~w6MkcU!C{l4;v?K)$unjxf%m$?|i%zUIvW-Ra>Onk*t
zD9#&+K?4~CsYZt-3;^If#6ce{fz$b@VYm~^Ozi0dbqIJQM9H_!OgUOxpN62r3}PDG
z$7b&ju<kTzbX5ey$#g7gj(K0?$v0i_3~O<e>z6^$plgkWii*+5i_O<h>gq~jE(0`W
zGwa*#smDg<NiryltK=Z1hT&p29dkg7hcZZxhG)ahCCOz*iKF+;xZZ$Q&)|JvYOhoB
z7yHX_nY$9|`QOMs@Op7cztD2>;oH7MfQ}Q;&|lSOpA><ISe*fM)SZ+-xZU4Uhnj<Q
zBA;Qo`*J0kfp*qv&>3aTB|ne*RUC*;FFh|lz*1e(OcXKcK+_!cbz1^iCy}1v$G0o|
zDWEHj5s3~-#eQI+<wLLc6yS&dIX3Gpo7pKKoVzmXElyCKokIEoo!Rd;aYXti&Q4js
zHzM2Au^!UpEVF?Js=Pj#7Gms2n@2XBsoa3{ALNImJ(6-MnUC2K4!SvIox9!uOaQz$
zk_M&)(8rZG*PK66h?`*bZ#JPEKYGJK7;|+0=Lpy%HTMejz}0}8NkPr2GE9JK!Y^!W
z81VLzK*nh5pNgWWsW$m3K7D#sp6Etm_UfvK(Z%ui({b3`HS~q$LyOcr{q)C{uQzQo
zTHc~|W{(N416!u~JnnnX%1<Gu3DdsQ$Pi3Y+^K<?xnxdPo|LAU%QmS~$xdm#C4EQw
z#OJ9)WZ_OQ0ySo6w-9N)bE3C*SU3Mgjvix?eqxEvLV`u&3sUAyzZUh_;}mDh)v|a(
ztqb}&x9vA+qpHxfsQDngd~j~lz>sHuLT(JH_$?*yq0K+zb+`p4KF%z$lCG>P+S1LN
z%`C+PdO88Wwq~Q*C^A{OdAI`{y_jr{7nWf`U#rIUCrpNzVtd%kn_62FMu}YD3ocgL
z7?r~{h*o^1Y&xX)eZ+H@qg9~1!s%2|ZJ{*zr$ze^n_rQPMos+YH@9Z9vwIpYqOCF)
zDuG+(#5auF+SgAMP(7X1(Otg{^li1)O&4@gIKUH5A*^w1(R%f0dyS4hWnY40{M1UT
z+~!0_aJZ-nC%top7cSEE8EJ)SH6+nQa{J!tH&}D-O<A~f*`<Hh{@Y$+wXeBh2sJwd
z?rDvAkledq<>J*pLGznKgxW{HMmAytob7K|QHmynZtzn2C;qU%`7z2<%dRfAJ()G<
zIl3FX8xvLq7ec@L0SR|Q!s!h)3h~vq<bX?B980<vi1wrz5)%6g!PcftqRROetct(E
z8o=S6j+<qaj8gjCiN+W*vAef&V;*)AP@OKUbcxf&n`M8^4twm0-Jb9JMV1XK%OiuX
zERu-|ce&gxooiPF(72sSH7s>=AfXTpZ(;gHsNFH+DT6@MI(>o8%Fad^{kD;ZS-(?3
zF&yXjTv*``9DUX5_~dQ{({o#1ij)Kw{aUYBZ;L=PGt;fT!$<eJznxTgyDQWcLyn93
z?;@>G6AIxJX@>E)K2Zi#B{-Gr3Yugi{Jgn>sb98Ni{~b_#qac+ubS2eq@-im(;v|`
z4UCIi8$9FYAwZ*tqK91Ee>?IOb+m6M>W8@NvdrY(TV*p-#Sk&77@X3*K0P_<MDG)+
zheO#IHCEbHMZuku60?6wn~*qW3%yYnOq-9u`So+a$XERmdMV6M;;0lR*B7)y^75iv
z`yRLFfZ2XM1*+n0Y?<#2y?^U+vkJ_nV))dc(z!bDW(aB+?WpwqdV4HxX3^m;_s`<G
zq&w{sYHBbT+$s?jX`ZEE7cF0X!2HxQHd!|yE7qNuU$hry6(uDyDFhwQ)l$x6vg8@b
z)C#W5a43Ao?(!RR4t-jdc8t}nADRpqv~1K+(MlS$n+@UeEp@1?a6K&wI~~8`tV9}c
zm{eg4n#>imTr`|ScexifP@im}U1B_KGM#5Js+%#4;Xb6;crj&ibY4RN`Dna$*$@5g
z^T)RQ(Tm~)nT^-2btll13`NrmeHytsKL4cO#iMrO2lA>ndxp~HH;67AloSvbL43{8
z5>ZTJP~S~zR`I>o)3*Y&l;jHVGNo_=8qyX5BJ&9tQqNLK?!9QjJve9W*2jqDulp+4
zrPGj+pJ5k2A9sJ$o$(8c3)Il@C)_u&=WR-uxZN#5Eq8zSMdkMYKa{r$o9VoHV~+S-
zt8HOAS*7Oup((`TZiX0=@T_m{D%RMHV8Vp6uci(>WR+UZB)a5JC*rAvKL2Q<&7|sW
zG-Xf9fAyS!+^Z>jX7S{v__^~MVYfa@yvyU2+#P#0%rKFZr8D&F_p@{k3NcW`$PU_v
zX>T%L7+d0C^+!ljgil0}yk#QY_s=;wEo`&vSKs!}tenqQC)Zym7~E9!wx$nyWkklD
zx={?x2w(B2LPJi6gvIP=rrKP=S>hXA%9{S&A{YZGWk~?Uvv1cOtqbOk4w_YmVg_p$
zlrS}!e77g7Vp=<BE3hh9XkRZo+d5MwRKG^%@|9cd@`vh8e<i>49@g6!zZ#Nscd5n7
z6djwLXbeWe;q!MXI2oCF02knlfE2lTKOTR(&{XJ#+}1cME63%1L#aqW-iH^IQ_Vyq
zQlci=3$y!?XIlcfzGnw^)?w($`Bbf@>~eB(EfZwf<6Ju*+;3zEnCDAp|FHWY;Rty9
zCtiTRef$3(-uj4;>uB9J*2l`P{wTvenOR9VcSx3hI&(n`egU+HtSe`v-p1klXC`3%
zJ-S?j7_^j)<=gG_@kz9iZ`|Y=Lc0C6Wgab?e$iU;v3mUD6TUP06iR^)N~-Lo$Gk`D
zGPWB$eu)B!g6oBLsl*y8L@ZD?9^n-98{vd+kA3<ArfdMC2n)#|f|8X@j0W_Qm8`Vd
zdprC;qmjMiFApHQ;7}{xGSONO5fZoY-!tKRLYZISe1d`LM8M5s(7VYUR9pRfVmQ}~
zC5L6`#?d2i=c;wDWPnifni(mw4pr0H&BV;^&}A3<yssKdBYj!U_#Uaz4fQAJ(42u|
zw|}HA<(hbR4sX7UxZxusdu(M1k@<$NOzdm++6x{^j+Q4iwrB$qD{zqsWbrk0S0wm?
zb!%iG)!aS79*BB*SA=UWJd>dgF1si(SCta0WB&^G&QCyB&V>Q?zE;)U3tU;(n+5?N
z0b<YEtb0Whpe%cvI||LtVYIjZF|cey=o{hrKc2oatg5bS7Xd+#5a|}_knRqpyGua2
zyKB)YAktmZ-Q6HvN_RJGy5lT--tYY2mtLFATyxHG*BFZ)Os!>VL2@A{&72J$$lyqn
z52wmVw_JM#Q+SK54+u6Z{yH#y@8>LV%xjGh3R(_X!SvfH(otlDzgFu8%!vhG$fo}g
z_-pM&$Xu&sQ(t$~`_80_T7=8p$bgG3rlW|}X627h+9JA!w%z?TW0VZw;g3K+K>69Q
z0~@$^tBeu1wt3{|QNe-c&;7^QRgIukyUe*Ep`SNBo6gZ(&+-b-^yssZB~zf8TSG8w
z6xtxvGTJNYCwulJN=44fr*{$_8&reC`BlH5DH*VNg*`)X7DL&)n^#wl9c#P)MgB~g
zf3BA1pPktwA|jSK;{cbc3g296MNFM9-1*@A*UW=S%08aICVh43&gbqm;w0XC33<4b
zS5WY<>f8RnK#5!{8l0xo{hr~<`jsRwXQq1OB0nVCa<bQ?xxhMY_x+IXXa6!=6n$CI
zADH;1<|MNLDDa|W=cEsI=%c44-AZmSHdp$l=JIg6*9gVgzy>T=4)LW#ScGrK@bZpv
z&HAUKBIoRgbTg(%Cyrz)7Sq!_D=5r8jJkajO|M#=dUNT+7l+o-Sg0|3&&|=&F1kJ6
zm9qTTChz(p#lrIM?#`~?HPXoGLpbe_C7izwh~iu3vqC78XLf*?y>d$U!NEZ)1{;2(
z6dGDuhA_D|cEDpatPNkGG7W3;_GyTwSHYB`_if9$1c6aa5<vof(tu$T)d1~Gcf`>{
za7|tt%7qH1vTE$w?xBuJ>uj<qtL8ApMq_mK<#YRhC9Cqa3#s<AX6YGwzoF$#f`@TT
zR@2xW5l@bgoQ4@%<ft`Pl(02oymoN1L)2cz3M7aV03O&hWm+M=w4$ZIEm^Zv3*#jE
zlbLF_7oAL5XO_p`@7Sx(ba64O;ug>p_l)+wifh`_Z#^kszc%=ooGuBr)N7Un9%z5f
zcz%B3%I0cKFqf@-Ly{as=6WzNI*MYlv(4exsmMNJV83JBY00jnqZ6rGT>4bOU@fhp
zqB6B&2^=^!S=>nxOxVW7#X-%oo(Ux06(QD@-o|-p#iMzc!jpd)q~a+Z4eA(YcZ2}=
z{x)fnwnJA~*L!Q2V3Y;R=(QQA`{Xm)^&j4?Xlqo6kLU_yyT4BW*5^;lCfv2t)2!Pq
zoR;U|Fa`R3eYxKC?-kA#fD}AYi6h2f?0p${%5psLc>5?Oo#!?43ORBNua4bmhRW&u
z8VYZ3b}OU#7`pY-{MD2pBVGZs>Nf)m|Dq%l*ClA)xWJOZAeZ>ki#4?IhtcQ((~r(h
z7?jh5WM;YqxsAO(KUbYmPqU_0WgYPr@Sh{IBAS-Y!p`;;T1Ji%L3G>Ff)pX@FS`Kv
zxX)zI6(5`f8XCApMn*nA87gY6_@CK#9l0gW5be@pEaP7QN1&zF`P*}WCyqu9$wx)-
z5X7kAXyU^UBL0yG6A)H>K#L#sNa?JX*?6K{>FsSj7Ms>`I{eZ#Ao7cFMX=Z#b&{v8
zX7&>5a}7iT!e2&io$l>=3l|gP4#ZY<ITC0;9xSQtHyX?o`5X1oU#&W~xm_c?mnDLZ
zmB>0=g084^$*Cd>9-d?=(Rsiv#s)+|q;lTh&QHNuV%54~VfiW=4DE$4jc0maVD=iO
zEk~4A^AUaHk3dW2+<D;o*!1ogxd;#obPrM6M`=!#I^@94uOl#C!Vm8UWcro8vlM*m
zkvUBnF?R=K`af`x`17rl=P!j=e@|s8oICOd|9+Cs20#8GuZf=TqA)A1&#c8*UWkG(
zxCbRE5Lh5Fbp)n9GYs-d2v6!qmU?82(C*mLxzmmT6{+JOLAqM8SXG~Xe8S9k_LM7b
zyQUor(;Tp5w=)N3P$dEXK8@C=TJc=P35RiYRfc4#hf(5?mRfK0PE3;wlJiZ~RBc|<
zg+|@9BQ=KPAr1&%=zw7lZ3#S~D6fcop8~ebg{o?z>{+J(XVW&J{#DtI?qB!Q;e+;Z
zgN|MI-q~4C&DX0~n_sNYgw7OE*3cGnbskgau4ySmFwp=QkL-zog+ISBaOR-J7_cRy
zFuUjM-&B?BiXAwB*kJ0%sEp=W2aPd{mVid*w45rHfLK!ptibMeku;)<jxqURzklx@
zfiAYpnXi&`Do^<8F`62s^*;_p@e%S11pFZshU_V~cSmz2kB?H8X}q|+k805Ky51i@
zV6RLAj%`N*HTY*ZJUe{S>%J?AEMAO|#|Gk+j(9ffoFN%f1SBhw{lX)%K03-@;g<;W
zH($>O1Uv#d`mtiK1%3;uluUVPhimd~2AhY5kruDyUHSx9IMmr=6wXOc+KlNl<L!Fu
z3ZQsA$XlGlv|DeDxp)hgPE5G$YtJL4dd~#_E&jy<R`3Fn2T!Q>wzoqd2cocfFd(>o
zI$`}}Z4O83rb2Va%b3bmPOta84lu3nfnoA#zPmyOhV}6?HK<*$=aamUv!Fc1;$1!~
zoD_a_#1*n`DXWrld~HLY*M*sicjokUo-KY>8D&sB$Pd-zNGB~1ZMXAXgip3944y!=
z_+Am6chj>7k@^Ll_tC!YsOSsQ&u)zVNcXbu(C2H`J*J<Y$el-*zu%56<Ha;%SJ~Mb
zoaBU?^4us$hMl>d6K^o>7Moj^PnafM^(;Ug-W<OQt6DD&V*)`bpMU^d#|f~aQI6Iz
zLsK5_?U=&(1EhiMUe1&hA|pWGt$n0Tx|IH;K3Hpa&-IRW*x4w2r^+9~G;@r4l#0z*
zOhwcBG)rr<?vHG2|F&iruqUaX-C$#wO!J3d$NlkIhnfWQmX-J+8mC1n2gx#`b>3ba
zFV|uN587nzHUs15$ICVVXSB__f|l4@uHM?nA6JJ5;R_tMTmwP|!0s!9{wq$_n!kk1
z<%sP>$OA`R#Qs{?+1#~?TpC$eSE08{Odz+a;da=X;h(4_|MTC3kz{c<4W@GeH!L3#
zmW0(qLhI8enz-a;CH~*_D!|Y6s0!~nj@mj1n<+s~KCgt427mDik+^dge|TLC$j~g7
z@x!u;zIVz2k>oXM7U2dhI^SKN=pW`^$GDmESO^eEe36h{qD>kh(0F9`@b-ePRR!V7
zaL;sMyx8i9@pNnDyO;^_atY$~mC>e&A1@ndp)*QRM}06pDnYB(T50ZS3r|`U9sElE
zL%Q1?He=t|aSycRbai@uH=>JkJy3%GQuUxX3J~!+y#CmBx_fAq#OZ>ChyTel3AX*0
zU*qv56?;c2x^2OmjA`HR?sk$fedY);o&8N5STq6;ECe43KIv)%Duy?d!-HVx@X7gW
zOwYd_4LdfF)__~c^y5K-w~T@}`zH^X&hksa;u$py$BXm_G$^Sy55npJWsCd!40j-k
zBnOgH!kDIFb7?R(R!u!3)n>VnUE=EeaPrP#g;>_ejHq75J@r*}vY6Klt&EAy&}KPN
zU%t!-TstlGJ7JGVn+I60FQ?>IBgW`^l<em_3os#fM;bMC+{ZVj+$4PF<$FmD9FeiR
ze;ZyqW`fGeWmxq>z)WSO%hvc$!Vy3p1CkDjh1E!4V#us^Sv%RUS3h-U07Z(x0mFu0
zt>8n7S!g%f3nvs0)$k3MgDr*Ug-*OZ&&?>az2J!K82!|64zP^R&2^wspG+zkt+V|#
zVs70MRgXr9>JfA#Sw?%3zTLO8BsikC^1$pVi~PvI^p?9>@Y!8DJ`b7*8F*b9wzN_z
zrmHiz%k<Cf(?(YO=Si4;^k8Rro%wh|xxW-_V(Q|lh5X@s(pi>pLF_=*%iOYf!qn{{
z7h*P-M=JpE2XCb8*n6I9^ZN0CY7hu@QE4RY`{nqAg_0EwW+vt$z_pHA+aOxcY`RkD
znqk2fa3RvZASecOo|3wyT&j0wC}-<$GHwE{d*^78B1R*t{2Z%fA$i%HZ=oO|P`}&n
zKRgkN92`F1H2Fi-;B({0X)1{w3{^W_d}x8Nf)61eCPjaj-$=go&J1a<azHswWqlV-
z7+h7fX@~a<ZAp@#%r*TCVYejYI|<zl_Sb>}WsyHEfFZNoQAa^J)Q6gdql;#U#rHeJ
z^*S$!s{v)kB*9MxsL8@9YhPV$Ppj!3cj^rZ<pNEj{EJcMd3^JccV&JkAoPXB{Q-qZ
zrq;%z_UlQ0=0`*xyJLQRzl_VvfU)!Hz?8)ao3V}I(Ty5hopuE1{{$fP{{$czL0v#)
zAhf%=!r`o|_)9t5B_@U&ZBl-n1DUx)DS{3ZNJn<5<(Tz0F`jPryqvAnrHv(#+^Ctb
zYu|!pp*@9?dHXou&fd;O-gD+HKD^fN_VW**ur+#YGh=5y^bKT>KhUn#z_t1Us$7n(
z#RAbAcbR|G<9|rc7dyTuaMSV!^pTDoY5O?5g}c(x)z~^cOb2Ai)BClAa=S*_2@SRn
zNtw)<>=xQ6SgdKOnP}-X#Y_%9+g|hRGOpu#cHOH!FSGojZ5)7?Sab=i;I}H<^<EVB
zT!3TQh{EQo(r19jhJ3gc<lSDn#jdS%sv^~wr?hrAg?*LQRVJHRD?aN0K-bXhZckT4
zx;J)T@xH%K{SRi4l)e4mH75w_62>tICK<NMML`jgj;Cg?1+ATndJh`hZy1k4>yz$o
zv<&moNq9(;erEG(FSeXqz3rv^P$I<ib?kvB!1f*<)qojifHz4}$DIAQBn-W`^-UdO
z*YZ?YQ^pR45wMvvF~7b`G?_W8?HOmWW}aY5B|ZT6)Wj}Sh`*G*L#BC=`dP(|BoNmw
z7CR4|08!y`+4m`v53lry+^g}X?jy(T2aSA1^3H`GWjwa%4V#bl4>HOa@zE)X5_YPu
zH1P>X+sa%6?{g;D#-iYZ5c#%V=dM~5OD9ji^OYP3%?7?daO3@V?(ma$csO4~E<80=
zO<14N(cDh2xmjAT*aQrX)-rIy2?G^D`-dS)S9HW$Rez8hYj?LmjqlgY?pl^W`YrRS
zei3MiG00C{H`@2!9u!vnY7u&5oxD9dIAj347G|5*t=9T4&+nDv)&6jHB81O2a=)7e
z=l@NibnB}aNxGpD9H^~tengQZR>}SE26#6E|G$0OG186gbCx{}cderHYtqx{{KCps
zcc$sg?>$J^$v1j5!dTwQTTF7^H>|Hve-C*@h-bx2B@&n(z}F`c8J5pTiYk&zWJ0Xw
zsRjSnT%0o1tHkxQc5W9aIRk%EwWTRw%$*)rywl2`4>rvnvEdsLl6q`-Btf>E+xTUc
zdv0TWM^-(99o8jW(CS+w0u-f<TBcnY<ote<=x}Pa?en>!it4}~VV^ipsyMjl6#f#8
zQS-29ObzHL$Ff-0Mv#3_dG1Orw{g{bElG@$K8g&H`Uf&_e*piXT3B%7RJb^1F>b)#
z4Ze4*i>G9MqUWGSTn!=$Oi76C-HwV!Yo<Z09`KxF$4iT+ceyYaM>i-!|1n)eMPsm|
zW--eFpx>P(i(P*~I_w_@q&v#JLcTn@2i(d3{3@<<M$*#dK1s{&llpl}(-!r#((-k6
z(qA5fxd)(=`A;O7PTjzCR42F`-YnKxJA5MJvy+ck<MmdImBpw(QXD(}j9YwX*wH2y
z`f9UyaPO{^tkhH2NE@nCLue4UR%NAZj(H^NY|{UC<J3~&Izf#l0`FSS2Gklq<RiB1
zGbKN@#N_(e>;AXRQ2!H4SYLr+UVR%-U*Yihag^vO28)k!s+hQXKpe!;P`lGE$_7EM
zBOy8sX(|lQ-v{(yO+H;*`iPsH)V+?ut)eAqdb5%4i9$C!7`^3@c=FV(DwiH>{~`TL
zwLm^={+rtApda)bZw@)%o0MajIy%Q<=EXRfQ%b-myl+3EI$ell-VLop^0Z_HXq=96
zu|L8&k#X%~NWgnr$op|n-?`+rcc$%IS`<uHp9K<gwK#G>m3dpeIpTgbR&hMbb)Q*I
zl})ulMjiZ2#%IlPu&mzG`GqI(qSwHQqRoAoC18a3Q3%wzc4@;V=B8}O3_;v8c&1co
zS5ZqfK$+aGq=j;R|5sN@L-7^hjEA#{u%zADhs8=m&gJpz&3PHGzM7bw{#RDg9qGIs
zKK}!{Z||83(C@cx0tnq0m&?6hCHtniLDk?XUwe_gKr*{e|KmoV;Ubz6(hjgeUxTfy
z2~1L3c^&oNED0#4#t7b*!JJ&7TKQ3>)Q=F;8{C@!G@)6ALV#_)z^v{zXXJ&iFwBEz
ze320@T`E_?!jlKXQa)0ZDUTh#fjaNKgESl|h+{w<-ezsrZf|!nz7_25|8w7C_4pH}
zU0`fFDj_B#b>zl{sDO=YYOOBa^465Ak7F#iyC2ZoG4$f{1Koq4sGs^eyOZL*7N+K>
z<+FDB?gc&oTY|hOWltB$4<z6Cj+-32+Qnb4_A^EO+2&OVuz;M6fUtLE?75@f$~NM0
zp^+P8QW8RvD1Fmjf(exj8m$6Q!Ky9Lf;SzJmzuh)48ie-dbj82ot8mNZG(IVdCfIz
zwhW25aP9d9cYAcjy0?K{liESg_oJHk5Z1DK;OFYsShyU|jyj^4(l3~T+6d{NJWnC6
z8;>b3(|kh?lwk5R|Ml3&xFg}nfXD*`6_Ue693?vKWE`Od<j?ROtx+27PqoJ<RZEig
zGYW5QYL$L|Nsu~D-zf^z_;w;}F<zu`ImkeE$gQ*ET(z!v`?y!mg!}y>{Y(Ik4>&tF
zifP_4tEsCYzJkrkScu5+`0;*cr`UWc2$bQzQeKJaDh!w1vM{E|IQ@1a5mCVfnHo|@
zN1td;b@|q*V|CKj*rVW}uPd)eGt+h;6<9AE&FrtCi8|A8Z=;YwsFD(vt%x;z6I<1?
zu4w=c4Ss|o(0Rmal$cOGLKOra%5GEns_;O3-M|5<clVlTAv@w37~~S5$)HaU_vzEp
z=5O>!2eoIE0GE0HnE0<NC0ncOgtgYBj8%cm%mVp{ZIV^|{U2dGe?oQqFrk>_I{Vkz
zvaU&j@G@jII}{vWY0UpjUgwNLpndg`vfj(;v?<q(+VaMg6|)zZ``+ROchq!Y(<Ppg
zQPrZa+WHXAOmiiGHa-_7(F^C=5%VoEg5t^b3LYu0R(jBoy?XuDZFh)vkNJum^gOah
z`+CM4Q1^RIC{;6S?evD8pFZ&Lmrh@aq!UGHWnxQAv{$=yDuG^0cwJ68-M1m^n0^_4
zaq*}(O>(&E7WV);K8l^=7C*s~f5UC4fSv8z3R52Q+IW5?eu|R5uNZk-l}|_HwRNZm
ze}hh*2pRWjXHN#y5&uML)DZf_CSo3@UcZ6z&mahoP}-1=q%1?pS?)M90Xhs>I}aIg
z^n2f2Mvk87TgVrypYNB)8e&h+dv2x<%r<j%g-PF#t55@tdZ8a-S}@>o<7x;e;;u-W
zHHiM}U}vHHZ~YJgYr{4R)c{Z|u5fsAbRP8x2wN{nR%1+&;H?P2EWacZxTH#RN7-5f
zygav{ttp&68<UIvAmSXs@GX`=^nRgUIOkCDMS<+Qv~n&KIar)8)yG4g?Cb9q(@SH3
z*ROfpXgEAk#<<S&My%9|a4{|MVaxJRn}D<J6n)x@HLUA<gXYWtcWb!7%AwfcT3kc0
ztHVrw_4uxNS%ndM(%i+9B9wLV9oJQv>NK@V;V$TnS`sdDwOG2ONgQ~zc|n7<U=UGJ
z<J8>-M7Dtgj<*uEP#0TY2yTwJ*a1&&l$gpe=X%Hd=bU;?>>N{P6#8W$Z^=4Ei|ab~
zNyX!J=nc<_1U2fMD~mO(-GQFQ>hbyg)nE+QBQnD&UXqW-u~kS>#DNCAH?^qW9-2`#
z9==4rC`LWCg698hPMMOi9|$eHE`*SWdwS5<3JUOg`{u|SUWm}9a{y>2f05DNvh;IT
zr}Pw%Z)SjptD+K>d1G#Dd>@Zd>w&G`lGV<2j!3?EH9hmrh4mB871q0IM~+qqvG*Ls
z4&{df?yFo)DV)i?rHVeTs%^q`_TTaf%0X%2A`w<z^RQ>*KX)`3H<bS9&s1OZqQHtp
zQZ$_7EIusgZ>-9G0qoe_Obo=sbR<O=v7m%KYPZYeOx9PAl`ZJZ#Lx%0(jwK=QT=$9
zRLdVL?QC<pLrg++uAs%@Ax66_=;qcXs^_y!Gl>G1Uw`JT&#?h;Vyf*T;pylv1_^vU
z6liXdHxB&jEhGK*tWP85Ae5VZ?bVILWn$l|k1--O3~2^p1I>^dbRj4Km`Ma0l?l-f
z6gktAo&rHv<|B{oPsDFcMsr;~hz>`CpNPhX2%#<Gs_1XsIW@AS`<=64ed;(|j2=8t
zzisGgJ#+rYElB>uY-)n}=_0p=5_D2+UTJLYUP2FLZat7>m$|x6nl9h=4>|Vi|B5Sf
z0IO8hH8~QJk2B)fCk44dCSN}ibk4+opEU#_o$XbtD_Mnt&FaOK0cgIx+2<@;PMMhJ
zd<VKd`7Y1>NraU(?X_a1M|@g0%#TsnAP6GmY15WD%MuXhcf6fwr}sWa^0udoM!%a*
zKH-K;Zj(X<IJH*`kz#baFG^fgVRHhtx8g?#I)bGt4$tM!6z&tw>@Bs4|C#X=!D_8;
zb63`#I^ThNAz}BrA{Wo>TdE;iic8y%u5*GsJ9jRdMgHR0G}&(@LL(__Ob#?t;2jYS
zwu{teFGn=5o+@5F#%s_1^q8vys|xPfh3ovvpRc0FFF<I$iR|Tk1Mk)m?aohL4)5?L
zZuz@b3${x5`k0I~QKK)o);iv>L3LSuJ`(}8`<@~c&SOJ-F<EpqrCv4WY082;l>H@S
z98BjwVQuV;U)_b#cy=}CR94!I=Xq*gtz8@w56BGH6ts>m2YIuIPTaSxksWf4CZ+XJ
zOOEAa7a+l-nk@POEg?9Q(W};ISmfZR%P2>txeP5|&PSbVdo6&?>qUoAbCwE<B?~>T
zK;6pc??P17bKeMfgJot{AjOGBq!q&mrD=`tbfVqVS#?m!O&o4g7*o%PYf)~cTZGif
zz5~IhuyGnlC3&#ngZNo|pIAig+{NfZL7!;!#U}!w#pFPv$jg+i0UcyzhQj}5ifhmO
z&r$m(OMi^ivSkgoY=1q4lB+>8@JSI1kJ<g27d5x#i<$F8?!thAw~15Q{1E4gn4PMx
z%7NLDV+|gGL~&;{a^$P0`7t<14A}c#Objw8<#=Vi14}n(UQa=>E_g@F*{fcFE$3p<
zE5Ku8seA__BQ?i4KVfaiF+b2__RpdglK7|Pyc-FNcx=2&#LNqFQMHB@+;KG<8985I
zQjTYLWw}Gl{hc4-|Fv~`Om2YU(eypJMqmS+`+HE(r6g7>zMo**=2wXahX;=?-{Fl@
z_(ylu@Jd#_Rj^<K`3kVUzC$0?$3ZIEy0*wgBW#e<IoF4bt;<n77CpMQcsUVrZQ)QJ
zt)-zTE340K<VHT*$ho@k5sGIf*}~>uGc+sPe+MpGqfpi|PldIm+A$Z^gt%Tx$NoyC
zPAS%@sWXLol*^%POsZZf1_VE<yaG|69JEVYMDMl)jLxNp2&LgX?kiiS(n%5U+*2<w
z401<+f=jr;Esuy1^;RG*r#olXv^kv)!?zhb6uQOVdsbqFbd5I;T=xwV?xrU0OB@)G
z5FhtG3cL=>?aApJ0?iixo1?z*1T~$Ds_=Grt01R5@ZLFnb6#06wC_RNY6h7p9+5BH
zS<MEH2()T5z^{`ivmLH;4Z)GT*J}E^k={AW%~rA*{&JBA)c&^_LX63Oc?2+rF&04$
zPqvqtB%wuO85~$f*`#zx>D0M6bn$%jIlTFcVU*s}TYsFK!6pIX8Xx($lqgzP<q*!Z
zKBD1m2Z@imT5sy$#gH*@SEX?=1A#4quZ>Z&Yzj#3o^&wAOR6UAS$vvkt$O&N4?mv0
z)8h`2n?FT<!4LbH&@XTRgdeGMxWx~``9W8stA#K*xXi^={7N0ZueCrz6jA1KX|f4W
zs6=sM$uJ0TrEHBo3I2JC8Dy}W8t>Gg4%BB9F(&_FayI+?*ErjI36LD~fMa=hKK^W>
z<+1k~Vha!~rcs4|@_i4(Id#4fh&x*z%V}nfev-E6nW)0>D6Qb0e)g7wu*Mb&yR>R-
zVCxsL%tvs~HMeODh3ruLxxH#r<1hv97vV+u|68_R5On`IMk5nT^8*P)M~*jP=>W8U
zlBz$c1pyE^zr9Uw*q3kTd2Rx#lM1cuF)<ApU>vFe&UJN$zG1p9j|N6J(j)HhhLzmT
ztD0kR5A_=gHf5EJamXiMT_vU?2EFs(A@K+ay|aXBiz5MjTgpjIWCVrqUArBvSuQPh
z#B)fbP4ZhoP{-Q0^tc+w73&m4`M)dyfW?1D4f<uPHgA>uvod&QpN|r;cfVP&ZwH!v
zW>}P+U;9h}Ts9~+39%OTwA#@nU*ckZF?wf+`qyuj!i9XmBWsaHJto4<4iDr<ie-%q
z#v8j^iUun*5y{gcJPu&k?8F31x@c$ju=V!G*W6gj)4DcZ38lH)Epzs^cI6~MysVx_
z8o|eT6nmxI_(K*FR=z?Tb+EwGBcGXAOoP#=VbbEi;i+l?WZ)|t)@gWl??$FjgtSGC
za?f_FN=v^pw&2~3;N+6Iwo3&Ol_gNTJ@}a)tg8r?+La3}bIWOqefcZzc3*TY{``?l
z+=kowYI7Ef7+<fq8HJIJWx!qTij23G#BtO9YVP#-#O)dK6wOxc%tBWlf$K987(J+X
zn1kTnPs6ZX*+BNiuc`fKpEoE_9-ep~kE@?n5<m<J*7rWX*_5mLtAbi99eu2J-Uk61
z#38*{)L|4QQHV}timb#_#6{3^I@yLPVmtpC6Dpl&-Q0|JYrGMVCT-MtAqTp<0SMlV
zd>E{3+WG11wsM*Fkvd!w=F-q$x-9nI`$?^o)a{<1`DbEoGb4kv|2LggBcr8lNp#N`
zB*AtuF^zg_^sqalxA-gfUnK&bS)k$EblS2IP4}RlBYm_ySEphQpY71=WSx<`jJx8H
zG+xvXjsm=8U*O*zrp;FhGVm4vUcA)DMV2<|R;%a_>@y!|QZy}35wFw<v9UXyIp;nP
zYcb&#&E5>CgW@jTO61&dTj1CaH*<LYQkXq<JHgTsj@I_r8q!cn+#pn*$<)Qi<}95P
zO?>%c&Wi<l*%ASDi*w(oh~LX2p`fU})e}GV30TDaB6Me0=f#5Qvo^LIDynv_4iz)e
zlRcL~J+QR!xK<d?M(M(^ndk2^FY@bc66I*sU7T)7n1cGKx{=;Q$yn^>F`q9ZOUd@;
zabSz(YU|bb@6gFa*y;AxUsu~}PmUSP;?!y<_Nh+jWl6&)%Vy&AlqD|s^O}t250d?y
z=|ETV^HUK)`OL!f%hw&HKk<gQ9k19pS4o|>|78aAFEb(d{@<){(@13+3eOWpYayKq
z7StyP;X)a$gc3>DZJZt~^l|D~1qSIthd8jpPPmJcebG2A@oeR|H93&&$Dxs$yE=MN
zsbP7r#VEX<)JtQcaf^^Z*8(U#>wf+s_`@hBPvu}A@tQVtUqJV!a5^tCeRT11FQ<SF
z4@@J5`w!{#7#7{n<`QZ371iG<sOX5V1dIK#{#>uQ(DzUmQ_`TEJAF(_GCd_ZU9=iF
zy;?+Syl2fU_e;I9szP~^L91VC(9gPzvGnSUB%wm3d#JQiDCZ(3Q3bF#M;ntN14E0P
z8i;a)f!_zgc?OcgH-Vh9$>Ct>aCxv1P0Lv=kBxvl1Pz$_1&e&B+J29jfN738Rs*Zo
z%_GH#Un=hU-&hd;mm>=1uIacMc$l1)SwP)W#F1nX5T|ZFx$cR^6{lAtI`w9RlZ|`7
z_$ifIS0Hp6D7!+)7Az!jzZd&!#?tb(ch2Nyqy6g(lm;2H<kHAnFv@7_xk34G<w+|G
zxkWJ1b;%p`z`K42(X6k$1OUBB{J04(uYC=ND`()OE{nB31Vl;G^Q_=gK2GLA2jfD)
zs%?YB3}5Ws$K=3xq|W51A@dKk`nU)O%8Fy4J6bt@`|thJ8q)pFkyB9h{eshA*TuUb
z(ZHv)V%0(ubN1@{=F91F26Gd3UE4_E0}hR>eB?z;pE*-~NDFc;`_%k?;=|O&V2`KC
z;A$4m4GqH2!8nf#;iR~V56F3hevhwRgKn9@)!=m3<Sc3pHR{c^ckg35sSdRBzB~X7
zeJE>ODDIW<PMsFZU#jgp<S9co*9cKktvbtZTZ3_*#l&LL)1PLfCu%vJ_LY?aUo@u>
zi|-yG9H^h%A}?(|ZG(D1A|UQdtk^N5D#4#L3|_R-D6dT|Yc9MEVMWsMf`9)$tLI`z
zse3V~LAS@?eo-KF;R+Mf^?|O@Ief}7c`(ciW*MkQ`M<WxHb*bvT>e?Yl6m9gX;o9j
zIGFDp$_|CkhIAdIt274+oog`M)AKDO6{Ea0n_Q}`7c2t`l6$XfMZ>FIV6W);!9`X`
zb%677LxUP^J6?Q%@a%Htxuj)nQW*Ylpb)_=oNaq0(TKKu;>4T}uoM3_WX+oHpY{L!
zlPE=4L6pD7NpZh>dWMQPF+q6oH;!m4XYMpw9jN3k@<7b$ysP~9R!aLS$Oz+q`3Ix4
zSj90w+ZV^eGJs*BT~9FqQC<Nwo&>m_+5UBQB1J4wSK6GB1Cc=cL;dj31so9&mPTM@
z{Ntx9Y!ZIgz~(><5)p6W#l=P5?myv#<DC|zhiE%{1V3uz2lq90S?e$bBJ8mT3$k)h
zbDu{#5WXIOu{*skm~jSmS1)2Bi5Z@IJH1ZuKYm}cl!nIaotv`Kepoc)t6}+iyelSX
znAq!uowr5vc4=uJQ>&J!?8Jfww{ifB?b?j6A9!qk2nVqmXO$<zsaB+XqMcvYcXX(N
zT0G4wE3pN%rCob5fpAe?m{dMjQ4Q1GCr-1hQuM99umP!IWiVJKYRAs4qp_=Vjs33n
zY+OhRK5TSTAvRif=P`G}4EJsr_C*eTG5$C6aF`?WsnChjK@(2qAS-Q;-3$zm0k4q5
z%Mrj2KMqg*-Jt7RJ&}K4`r@Nb%Guj9qoJWeeg5(V4VPM%YPkT9uTx81UA*3UiR#m*
z(x<Q2$4h(5t<A%q>0frF`?>jE9|(b%*Be7|Xf7~JurpPQ?arKQ&1<Z`9xkQsKJS)Z
z02itkaVnd+TY4p|tLu=+7{KCDv=|Rrv_gs22m!B4%#A@sL-E@!Hw`8l49|O}3ev<8
zWsC%+WUcR|F@%vsqXqj+=5?IKf*L0l0oTrLb+#pm>m|YxC+yklB@SDhcn1V&)H$5Z
z<LeCasa{Js50jj50&JitBq*UuLXm?2ItPwd?8s4t2$&}6&23)zfa@WzrC{;hfj2k+
zYC^9sir8e}CwYl(dwPx-=x2kqI|YAjZO!puI>P($ZX#cfSXfwCo-L;7SO(0m<jfo#
zSW}gGM5M2#@wiw!+1DEl#UCy<q6@UUPnZ3h-_?Lrnuc<{I)AXt4hPyc)#c1#a>d0v
z3luRDg+fIZK)@LcyCiYmo$UNC#cw<7<EwpZ5h^<rR)5*L;PV~&dgZB!W$|LN_7_hr
zdk)I(TC^l`{;uQg&Yf_Uz3aYxtT9`?EQNM7M~b`Tu?DY?C-7}O!#$zs8r8-l<mE<&
zhCiaCJ4e%aZZ~7(pl-GjbQ76#j|ZFs8+2IR{an_%fWWCCsd&2R&9av>VQp#7W*&j3
zRQc*6FC5+a)_pbx75*(rF`^{$YBmg@wY!)1U^o6S#V=RLX1U?&lo3``#*;epMq<w4
ze)I5eha$AhFV1407<>cJ+VcZ>7ERco?U{2gh>k*p2K!C%gM%`nKy7Tjo{-%8K;cZO
z3ZNdM?HuhT!Uj+^FU+J4zNLLCy7ts@T38h26x%V^=5b>ibC{e>ZTSjnF=Yfi`EZ`v
zG#tW>XglK;l#Sp^`bHE~R4`lJFW138T8$^xn39!sGW)iy=dyePy@gUPy+r4DvQWvu
z<zINhZluz|E|abuR$nT!`Yp!(T0WqB!-9?k07^P2Vm?lDlJ+k8P>XA*b>^I!NIn)F
zQ!+SiUs7pvZA%)y5#kQpf^Q{L^%pCU>xQxTOg<NAdF$H9+_<tLD!r3F@WIT6093y_
zMyGtAqLwAg3q>a8HC4NBghAKO!&`l9w0v4#L)$wEE1&CX$9t0DS-M^V2Ee_z!2y#e
z`-k;Sbw`!@4aKi=8E9l|i!i{#Rrcz#*v6~$H1dSfZVX{-HgJBXJ;&j?*d-vA2n~GK
z!p(X(Ls3jKcUq*&idVwvMj!gScRs4cy<pH!RaMm3m?9XJ*cdD!vvYG_w+aFqs=yPB
zkAd;ovxT|PUo@vMr}iBHKJ}hUhQqwNp0Y5%5iZH>t1V8Wi{dmq>r>z5<9GU~vUpUi
z9eawiaj04jYrz;>kc0&)04DACWn<Aj`r;>O9yNFx{jk`_lM4Py1S--5DWrf)i>L^L
z^V<*cH4w6J6QGQ2i;&f}n)t%m|AU9^Bo)G)Qf>uHM%a;?r7PfpP+5G%vUff=Mb1a3
zkQm<w#$8k>NR$foeo$VV`@Dg-of0ZUQylyz$DfDH_4%<XaKi$X@wB?{%`uja#mc9<
znlC0c{!fChFz`DX({1jrH$`$oL(mBEYNwqXra^Xm^<0Bnx7ksOe6zl`_&D!wx+5AU
z4Q_R5V$TzyOW@2G2A&>Qx<Ike*51BrU?BA5#Fj!j(Kc5EW4*{CAX^)|`Fxb8I3YY7
zpR?${;nduWH=&i1e4}i0C&NtS)ZCMAx19nhHK_(f04;I^Y{PuAJ5af|;ie_^Mju?Q
zB(<%<iI5eEYKGUHdQ`BObZl%T#u+*JJ9p=#FA7NLz<h-quDEuxjVGm5K5!m+kvH-T
z!f(3-_H|_jr!ap@s#cY_Ksoi}DqSrWH{JO>A5GTQdX5}UIY}cqwz0soBLB0QFl{YG
z{6>xz=dkbsQle5f^B0-O=GA#~$H;Q*HWthCvP&FVUhVk&9$1L$nW0^T%pXyLCXfVZ
zc7Xb!uaEmPeQZ=Ea`=nTd~5P(Js4BvF|qcuKfpE_LMt(6i$y4qc){s2;)qL1LT8Fa
z5h1%PKCE{7&(6;385oF|7lBh$LeVK*x06iXynWj}F%ix0b$4>wN31sX$tbj{irM?=
z*5tSW?EZdytgNrkw_2zhPG$>#yua331mxV@xWcg+j0a<>0|EoTSy)I33PR<GMy&sp
z6DW<m6&@HEn7VhoIa#BKP%59PcHC3UlTH#IivLtuTiZJ|6<7APnc3^kQME+<+ms!8
zL@{{k!K<%GN2l4@>i?hmQ^ae^0x8bKVMz@?c)zt3v{jwH{+SOvuGIkH_miOaM%j7C
zF>u*daHQ1tFZpyzp4{P{W6uMg#0=<s$y3NX;d?C)R&SrnAE3$hld?^oqA~>&POcs{
zfjiQvfUjkS2+mzGHhL^ODdXGZ41ZyKmf~1fk^}Vsz4H4<NKNo6x8$pa{nd=M6LBih
z*67~q_RJ4#_0uY#|HLm!jen{Y#H=~5sY@wh&qc!b(UfiXH+5?sSijGP8z*!j=|Rx6
z2RLfUt0S(ANqZb#7*}PH5yZbhiQ&WKe)}KQx;o@yHyiepiG$1COLu7sbY0+)?P$Js
zqZ?I#jGFr830xg74aLauOMq+J7fD!LJr+l+wtjRp(h6KJ4?)_WWkp5dFflQEdV5cA
zcQXDKt1<kJB$RTy-5O+kG`c<8a&>cKvRnU*MW-&PsQ4a?4*$x}2bUR-2zapb^Yf#Z
z$kb_&lfgA{cjqP{AzAdkKcZ8w@CgWjEh+*Ha+>w=ED_Jg%bD+lJn_+_V$*KJpZ}U(
zz}nc_0*v6JgweaZ*fTUS;kf#j=O-!S0)^wju2OzRE_vJu3xra{baRAg$t4vFig|NM
z5D9*Fd*nAFJ5D*RnCT)b$5l^+6Uf)!&>L!SA^MzKccI5x>SROrE4zTe`a1YjRT`s2
zsrxxl1fNn7mC|-~#;TTm1?2F5qR2%&026Yv;#%8h6{<_ezPpzITTp$%<osgqSelk@
zqu8zfT07+@xwB;7Ta&eG$3tyGocnDUNrX^*@qDIMQ-LCQLzS*`$`AR2RhUvpCyE3;
z3Ge@q`lS#NP1N`zlJq#qma6{EmBNN{K3<ftx3|x^xSuo0VDh@Vz|&=*p~<!YX3D^&
zoSEHjhO`+BCo;vx#Z5JpV$!NITFx>`Q~Wj3&ERpN-}oI-T%B=eZklRbowU{+91|Tq
zL3SGYTLA)V<L>Sr0k~i8_a2nD%L)m-0@vU3ZZDrez&o|3rbct7_~`e)*_I8eMt%8c
z&D&57=HJ{4wg#5YYiL}HfY${ohY#+^apbj8(2a`oEA#%}p27F|d_U1A@Qhc#H|^WB
zC%?V4V%F3m1i2tL$n?wATWTf1qy>o0PwEsz)wPC1k-=spy@ka%92y3I;9^VV@J7GZ
z9+`Kl>kMS&L@2La2rG+o`gx-jE&jelCpt<NV%^)u(tOHefY{^~HZCQ&G2-Bi(d_E_
zwK6&5b+qXgijL-w$~3=Qy#&Am@u3aFn9!F^Lvk)`hzVxlJXVorA`m2(|Jh(=Po^GV
z6VvqLb^CLvPGm^PD!9bt+}uq<2ov?o?r$SRyX#lHrNE`3hzT7Wl+4b~E*cU9PYRLi
zsxL59mXVx1D1yod>jOf)^X(B*5Zk&t+Je^-JPLtc{|^$V!{yeL-f(Q)KXSst(A6em
zWlwHh0SG-kJ@Ozv95C$CWBielGL$EilE~{;HFZCSV`pq=_zoZctzf=vT1Z$}c0&VB
z%z*Nen7;l;fu|eN=;&zGJ}Nx1))l(|ASh^>3zi@=JKL|17+75Q9@)lQX8qT$uAQT=
z_QiJ&Y2z_YqIsXRp9s!XzL6~fWVh5gZ&Q3hPhkyjcOYzMCrdZ}RdH2jaWV7ytWrK%
zPgT*X;41=JIF^PR4@SFVX|RXx>(NOup_WtG)gs~3;AhuMw6!;_OH>P5g1~w@Cz!ry
z`85DmpIGd8nwY{su1OmnXY<5W)D?0EOBIYZD+@VsR5y3t<4&Ct_gieF@W0>OC>&js
z%%9!&1#N<cB|~U@HprnsUy^07y6?p7FYo&kHUH7&_`>t2&Y}e1g`b{u;^taA61$l0
z6?lGAvXpOY0HGyroDaZUI3I9L)HU50S%nW`5jtk$VVrdLeQN%@T~e?m#isp{{pAPh
z>FH_F^L*sx2Z@k7*2f1b{GQMCWc3T1mzAaE?s6*+__$t*69K<xVy(r@&!0bqZBKfi
zXGW7);GtYxTt-v5{8JWsV@l@d<|azDIM^*`I|>wl9QAr@LWb6h_c0VQO$AHUWBgjJ
z51sJDCl`O^bT67jb2_`b{YgZElqtWJPT<(-sAK8RRT<S-&Yk__7ZwpgKt|qL_JGuK
zi%xLjwSNAWl@Dl8f2hkD6xO{Ca+)05&<31l0-AqvM;`ZX;ICxvw2F+IIZ8GLmB*}>
zx&paDdo388nqXt)+wLbP_^48^S`Bg#m^*AvR7BpT3y6sw%*w))2R;qnooX?(r?ns0
zn*hc(u4&dWaG5G*dTK<Talc|`Tp=~gN%N=gQ@s{(ZZjl?wJ!&e(-!&U404Qy@vl!i
zx=hUMl(hOesC4x`lcCBp{G!nHG~f)Cf%orle=n`@4s0?I#k`OJ)+$tihaWKTD2)0&
zy}34j{{Bq@<DE+RvgC}6KR~SXo|stI1VH6I5sW~u+jHb@d%RlM&zLFIa+x@AZ%w~F
z-+?(fZieXmWNrPSS-Rbetqc2sp8h!7(9Fn)JRIv2bf#1SJ@4hDlGAUih~HrV3nwQ|
zgeY!ih0RKvXoLc5x_i{>Qlo>3!S5Fl{f*rc4QC3BxyqxMW%|sYweqKAT#&<wwmNcf
z2zX$FHda4>lBAt830^52B{2syPyGkRsxQ@r^&V)|8eDl-cqbOKu=Gx8Ir}TY9JcGF
zf*#|ArA=5MJ;!~VXSGgZ0Rq8{+CZdSK3Va*hzKkbVh+Eix}J6s({o-iH4WQ(+q}_d
z_*||ppyeRttPl6k0ODT7To(y%JF-JIT}^-iH?8>;xSyuWzDh|*AiP`}&rgq*D{W~J
z_#CQ~Wc4;Hvc;;U>aFf<^^5jW5neJW?4_etY1o8>U%zexP_e(xZn<k&Ohz|DDbpJY
ztdDND&gBa5BYyACfCa<WQF$5WS8lG#^V42_hUirP<uN|3td^FR_rs~!aofE$h=#iE
zA?;(W*UM5F{9Yj;Asv7JRx4~^T`GaaVJw{wp9SS*B|{Ek=A<>e5jZVM%UkbV%DABh
z&%*GD05$Z9`TY=_+2|18gm_BRGNd>-@B&AcJBDtU@{Sxli+lg3fwh9Y7}0i0F<J?f
z0EJ5{A3AwU(;`9t!i-IQe^X;<Vkl5{<4#^jA#(1Q=Lct7(^YH^a9K*$H#d7Sh5T_?
zP2W8~><ECp;XQqQ=WD^lXCTOcVRXMRtpcj6?OfU)&{@E;RckF$s)~+=CgkbKn=hBq
zwp;slC+g3^B~}~*UwP4fxdaaj3rlgrLY<X`9dW#Q3FoU9D40P1)paM$B@x`E3JMBS
z<!<ioXW-dAKj}u@I5=21>G;wrJ_V3`t~O%E4ULNUNc+h*+m9eMb?Nl%ZI%iUtg~y0
zw<gli;G8&Om?lZeK@#&P^&=<}aNg69LR+Pkjr8f!h7WP!I|ThK5eAor)4g#O^ElY0
z0mvCuwVA{RotIx!d<kcMiib;Swr|fqW-2ABAY|byfFY@1wL=mP5_GRGn4o4HaBU^#
z+J2Q;w?12_Jj=x+ucMhK4fIU@_^XvI9ci?q?<G87$sA+H8~CtE2R2!PdD(6dhJX?8
ziMa*V4+X{fbnO8(-JuvJn0m}Ob{F=Y`A#2*vZiGa^J=HED<|~nn5LH+=yxa>s~uV&
zxQh5)az#(fSL-B^W;y2vTIoQ~8*I}kG@UBK%<#Ix_4V}~2KPUF&P0Fqm+(X;1JQ^B
z^*@j;ZO%WnB=o8NMU`R@;f-E55J(^0Jl#WXw^MAumpG8Hv&XYpOrI>dtT3635^N7A
zfhhB#!|Qr!reFax-s}PvGQ`lX?S<BQw_hw%7x`}eE-ogi>nCJcZziZ01AANNtSQYO
z{iT1kEO`4JFqyri5+bin9oM^c_+-t;B;}%z=)PB*Ic;tVURa>B^F<xPu0|q!Kz_*I
zSyn#%qvD>EB4_XC%n~%;b4KfZWZ1enG3V1)=Wlz2zrx<oe1r~b&C-v;V^cuSk&vG1
z!H`-`yp!UX!qAOiYMU_CFQ35ifP7Vk{j4Ck*J|?!JTish?KQvIe@yM>g7MfHWigt$
zOyU_M->fWi#9Qgh+;lSbq`BXIkQu&Gip7duyliSrBmZ<CJh2fcR&kCIr4l!>H_Ky!
z9*k|7Lk2lcN<tl530qW_jfs4P^nC0_M`%?vRhkw|+KY`y(g@2tY(cCm%=AG<QT6K=
zjoZU#V2gOvic?s~efewo3W42%)p{nJ8g{+5JKw~DoWo7M6<PXsTKq3=JUe;THMmo0
zR|Xzg^2w@8`O-mY&7>M?e?LeeM^35de^f~&ZB7B?sNjfUtmSiyGd~BXwNt(Q@W$`M
z8xFqv!)hb+@S~OX_AXc(5X7{#wP9Z6{M~DC1x`P9e2ocfY+7!LZZR1na=+MBC|0*2
zKEgvsH@KkOioo1yAK0=_W;GKA3*^g57I-*;dAY4h>zCI-pp^@j+&ChwENdLT&U^5-
zk43&Ov#zRRKX~JJvI;Vi65WSSQrq4>nkr$#pCdg483<bZnChKHRboYC?J)*`iRRg6
zH-!5-dV~HLo3o1pU6O=w`1eGt6IybWYppBPqm2ry?Maxh>i#*nv8E)A7)t}%^hZ2n
zxn4U6y9+^Cy8K$j(Gl0ecGe^Z&(%J&X(qn})M-BqzgmpTj1`GTq{sMyo-mM}@w7g>
zo`#fkO84obuz3w(t7_k9k_Fcd&|bP*%PrrZnbsOBUZo(I4!0go6*{~z8Rqr5j88mD
zLS&cy+qj6X3kx58QlOl-(?N2zil7#cEekwLQtS;+MihHnh}8Yq`PZGoi*)YPG6v+5
zDw~xu-(T9|0o`+qU9DF_M5AyW<yv~R$r+lB8J{vfiKr;kL=d3V_-?Bz0=T}ypR{R3
z9hjqbt@1;box{fanwWQU^5qlpf2@{>+!k_`csmdw^yNO=ezT!T%Mq2_(;QRoy!B+J
z&#yQdiX*u7f+sa@6WnkfDLP!Jmy)9H^dQ5Bo3?~KA+%1-$@@AaASTmXw|TJ-qu6{5
zM`mJU*;gOlFhTuqXP5!3$yHm~+08c7D;GN%z<VMhowWlHpc)NX&DS(;urV-{S8e4=
zB}{YM?eDo+F#jb69-{Np-3JRu-6AkeRHUV)6Pb+=ZCWp}pbn->4bC<PoT@%$>3<~s
z_(s@2^RL<MUXICJ^&aXbB6%r%o6NzJO;$~thhbJzjn8gcbttgN>fip9IuI9b;=0p(
znW%-s`7L5xk07x=iEyPP-aFJbD8DuL^KQG@MFkH!7RCG4BaA)lau(;cO*oE^Y?Rdg
z0_l%Bl0<BZU9L87T%g^!bXUs~I<m>aoQ*3y*Ddeb9y!x8ATBP=_NJ1PC9X~=b{=EC
zxdfta-L@QN>jeNnJKj-bSW_G`clpSzm=lA;bpBczW2?sNjrO8L{1(Hr8lK9L8GTxg
zb>yEO)8}}3q;|^em4x0%o9RVPX9m@IVAY2aWn=~VKvAzCCH8wsp)}^<u%I<fzQ)*@
zi5Gp&O!ZbcCV@u7*ZE+Q2)*x=oN-EFLb?8*_9{3-{}uzNKd@4Il-gTE16!R&3`#@_
zCqf@?7o#wK?MQOarCT7Sfv+@kvuk{VMfTH$Ec}Z>yV34<69l5=#y?M5C~r1;{_s^$
z;NCUI0>{NawGr>D8EfTL69cXx6L^7*)X7UpC*QFfrn5JHh%)pd)DVo<h5ZI^1izD$
zn}_qUq4;i~tBc0+8l;ZQMuV~T7Axkgcsj0YfmJK_m(AzR4ULUlpcbRw8-_WY#*@Hg
z@O!+~;~KB8NSOwg%btu@t!$IDUh(z+g$Zs>Hy+N07^D*!vwlA+_o&;Y^Sb?PY2oea
z?xvL%TuGOflaMHM8Wuic3xS1<Of=`^Ap;IIAs;5=l$_5Gcmm)I7SxJ4Ys}uNqUd~<
z)(no3>wU&ap(!=vJF!Zx!5_HN5DDdN@|P1p-8s$BVJ~nllq?m_i9f!fiIdzs3uc*6
zSs8tD`Yd@&BeN>-6c=Ty)Snr};#d`oYiCF#Wi%J?%<O&ZI828xnr5u>4M9pbnH(T@
z;o<J!c~U&9{YbRUxJ1m&(@~!=d$lELxkO-kDUKOsxp!&l3UQ%sJTQ8dQ3F*jDfvpB
zc2&b>cMbL#TVzt8uhO~Fr~_VZ+FA@^Fc-Q<m%*qdd3<H-XXdS^BO&wmH`a-e!)!dt
zithGeP!8uEYxWMl2(5N6OAmJYaVgMH1L37>hBIiTvDBjHW4xw`-8Ti_OB9z&g7R1=
zBY8Ye^-EpEiq_iF_#?SbGj~@XleU0ITwM8M*X(N>_QU|-n<$Z#ZYnvs%Y9P^ew*SV
zrkOcHMx76G;qJEfw@040m39st^$xx0iL6HiJ|IjArCisuLgZELj-Gbc8wD`Icd7Aa
z=?0Rk4j<9&-8jO+Z(9Xrly?P>Sm-3UsW5BPW=Z5jc@{BoYrKOMdUtF>$m49d*&pqA
zIEw}V009AkJWX7`Av4I;UcY_~nw96E@OOT_0wk1jSk6{R-XkL*d|T@Z1W6kPEiG-0
z<|h!P@m?=Fq;lFp`-A`?Z#+Fck2PPmpfR}3pe}YN{5e)UV-V0O#(znm$4c%!lW3O4
zKRkHS>$LQPPKE*%hIAVDWu&+9aAN0R9PO(_PP>oyS4SI<*DHN1E5u4lO5k_7y#-^{
z9`P&Q_YoQ1cL=n_`>2-8qN#3fZWsGg;T08(FZ`0*X&(rPi9u&h`u^vSqQ1Vq>t1fc
zF=ts<s~Uub85E=vK5<x3v}Gc8b?=_}Qh}%3FDZ}^Chw?#wb6qIKOqmv(*k>1mjAOA
zf9Jr~{XVB+R@s)$n0|M2yHw~MYKq$PsgcEN&qZiID~SkM{d=|IefDv@*|T26KMXnv
zM5NCJZU896EClj&dW;Cd0j<?P9wt*IQ^d-6hUeNNeX4Oi@zvezubpK+tsMwfsMtDA
zjB1gS5q%8*?Hw+g6Zdl;!dA@N=+WFp|3~JH9^^0r9Q~}i?0F|%c+r_nHtXObvxm6n
zEAY$75^wJBW1LVNyIkXs{;P5|-$DdWG<oo&0q4lP6C<gU+Rec=)2rt*xl4l*1N`O3
zw3b-Q-7PxuRa&XzWovO#*ClMQWbSk9C}_%S+i4|ze}Y7aTGIsUqW@Avp2~&a)(s?`
z<8*BvzrAgXV~5<G3GPYw!)R16DrmMhp%`}h!EPEO-nH@!7Q`?kf;W(`h$DzJmfeAa
zPax#*4F+ekVqqmFCVq$+U}R$IoSu$HA>>|+Ui3O{+<Ajk((r2vG-_YoK7`8xXJKJs
z$c&2^=(oD67fCFDB3^5ZO{Xq@e8<nvPp92bad=Jaed7x%1j8dE(F2A?k6fIbpa2P<
zCzmmLc`yUn?;9J7g0$V29bS_IQt|I_a6*!jH0GY>{u2g3@v=Udh&!hj!FIg|AqWZA
zFFago?v#UrgOZFafK)6J3RJz7jP$=2;KPx-Zmh32xSp;P3e=DN!A3`Sm^t78z@7B}
zXu9gKs<yX@f`A~PpmeBocXvp)<e>%W?%qlY64DJK-5}i|(v5U?hjjBTyua^xynlF)
z;y!Dyn0MZpS(DAsJyH42t;b~jYh(<8M)|?Bk}Qe8yy4e)GvfUg6-~9Fs7s9WPayqs
z1t^_C4Ub=ZX@2T<#hoq79Cr(SZ_lFYrwHN!C%g<dW$Xd@jx=lzK@%P?E|R;^ygB-l
zcaG*IIUdbu_apI|3<k9_lX1OQ9HaScyH^G&C%rRw>}juibe_J?>?U>gtQXIpyZ**i
zf5%iiQQp$3!sE=|A9L#HBV7D_q8u_lDv9B6vfA%1ElG>=zPx!$C;!;#RcD<dDl22_
zGTF=ZY(8ce<(MoiXU}q2qU)}kjj7Tv7F29|v-T40j#jrM>Yknw(8^fcF4Oqn@=kh9
zE}O0Bk{f>$r7;<Nv@vvFr8@m5!wCx!I%C^BIW(XvtYL@|&6Q0>Y^p(~g`$|tSvG?7
zC{($Hf7BL{EA)B2dXLJg;f~f67S)rjBAyX9?Tv}*j?1&6ISX@FZwh;Ww0wRK@3O(|
zn8u3L?a0XcVkr<suez*k2{Zv+J=AY(_E-n`ex2>j*xB0VIu*jb&z3dV8p(1xQ>jn4
zb3T|)sdd=+V5WJgO$ikj7bo*L<@Y#?%rWJPiR$a?Yr6R#MXUE_vXnGk<-b`%_OT_A
z=J861&P5<>6#Z)@Sf}PybLEt6mt;Ve43hw50Rq~V=_-}KM><<`eEANBZwz7TVJ>iv
zWFOY>UcHk_)@X`rvJteNeQmJ`j%(!~R++q$V@GMT%&CV85WdMc;g1F045sor4<qZ{
zVo7SLsd-|Nd~VqB0uQgqC-$x3iCTNMsu77P5;6}Hn`rFRlft{rZHQpw$;^mUn6!|p
ziciAF-x)dZNL4F;HCe_RPX*b&M(JDdJadSOyd+*%K2k0Sq<v@mnqWLfs0q;mM@zil
z`^owDdXg6nI*_wSoeE-`D)rb^5gMZd?q;yLAx*>%`sh||k3`0Jr6IH%iQVBYm8E5F
z*uc5ly?PNj=cj-lBCBgB_=dC&EZ51y`2*7p%9d=lEkY+pIL<7`FgEwRfrC4`3)&~a
z6_a(mNA0ufc4lgggC*jR=J}30wdm5gl5U9bY{`%UAdX0L!E2Z_r{9T?eGpsv^wmpA
zhIi>~A2;I$G1+QF_rm-iUihpDh^ygp*(U_uixsezwQFoYUtG8z9UU>Tvi<_g8EgfJ
zX!z}*D^*rq;{N@MEX7JF=lp9vzGS!K^)l-@o{^CeWWd#RQ>LwE#p$+nWFJLpypFMj
zKuK!=ekh$*RfsPV24J_smGeLw#Q*#E*8sL`O_fuor>DO<J8s8PIP4+~FEt;d886c6
zZ1qI~0V^G?t)GX66cN!0yS}_2HCIV()v)VXv)A>${Fv&tjyGPSH+ZzxJJsmHSzBAn
zW;Ws{pTffoXb<ogsH98AGZ2*^U(VOBKLOTk<={YXAstM%@HtPy+2&&|qo31)DvJwl
z$<d2*cH(5N2k!xLm10-q<(47UwV-W#f=&LSe#Nno($&yqKHiJ7t@bQslInXS|1^Js
z_B6@6-a8kCTopC7)^B@Vp9C15>j>?Jjbgm>QO_Bcs*Lz38Gbhr8-)i|l=)_G6Y5k%
zS!2;Kra$L;{R~*MtGsHJ^Am1*RktJ<qO5IKEv)%}GDf{%`GMM)gWlOuxqV+4t%JHG
zUzyakmyK*P886)zHhP-4IF`?@J?$Ys9Y;JyH>FI>NnY_YebbUy%4GSk^^UsP@#0d7
zoz+rz6{9*PL<P$);@yCxYd0h*IMe&F_%~12nF^Eh@4^@N_<WKMm+Q`_xE+YV@(^lX
z#iy&K(23$eF+-?68%JKw)`8#R^&$&>Y_9R^KUiX@rltlUNkkYCFN@g-S;iR=K7LO!
zw?jOuiG0yDm?VaaEuWteat(dgu(Y(q74+s?=?pJ91Np}S0-L8}1*b-)$?NOuA;kQN
z!x<KwOftDzU~^XhjM{|%OG3gBc}jk3S3q{4P*O*B8)?8`fZI|js(%aiqIl&2O1MSr
zq2BM_v+(XRx9`Yt38NGRdj@oJQw)e|Y11Pj`z}=e{HED@vFWY|Q<Qc7P{&3yI@a=Z
zVR{m3O_f#tf!QN>h`%>Emphhd|7N(<dmjsw0VBv3P{DL~X%n@I^dSXdJ@Q!m&16;r
zlHuJs;>6pPpn3bqUZfvgGTqr+)yT9s9=nanhZr_;i!Dsao&|!KoVB~K+%?JlBpA$h
zBy(Qyy+b4m6+PAG*93W<es4sqmMn>>m7YSX6t0hE9}}?dpHq{i*2$iXN41kQ2<b|5
zkYE1RsGVL}+!HBJEqY2gi5}R&P>J3<{W~n3H8p*v7e44}g9}W_$dqrw25JK9&f}AH
z+(Q*IGLNsSf74c&6;+cyl$c<EYMwC+AKKX0Hqf;a5E5ESo`e5~a5S0~lu<|sJtGJ;
zGByhrF?;6?D`sxP+!mgybzl||8J~KlEJH0h$WHb519HZ};GWl=hgKf<NS6o_JV{X9
z3k%_^Uw;Lp!|Rzp`<1^KP@)e{F)3wCZEX^3IE`OohI_(Exx#u<#~SfDgF365{M0=o
z)6!lT2i_)9Xm21wHo!%z`KMp0I8-34n%~wo@>nzU>m&7nvLM<y3iKZ{?L88oQ2#wW
za$7TAqu8&p0`<_jEO3-FC6pQ8kMCrWEZsRBD>HSN(Rnn?ePVEQ_oc{*8*9B&!3`gt
zm+079_f|T)H2h;RP$SihU+~q7CG%_2LB=$3+|OHPai#zlJH{<1cOweO3F61|3<Jm-
z{PJ@*IYB6?udgqwe~Ga4$av3Ef=244G5H7U_tz1z`OAcC`EQf?2=JJBmUMsG)hg+}
z&iHdUz@ox=Raj@dC71Q&x7e+N&m>Wflt9|+(4(mCQs=7Q{9S?AFWw-aTdMDqnXHvD
z%#f&t{%Y<GHB=toXr6>>rj?>&Fs96H+4Nu9)a8%t@tG#yT6>yEakjC?c^@YhEEt;X
zxz9LlShPIr;NdsQQ<~QJV|rq(<osykUO;?zfHhN`QYk(rJZKid0*ALjBx&UBTfa}<
zR!UR_E-BMB_{Uz3D}<^Uv36dP;q?QT%{I}dHHq?ot}#|y2mW*3<r(r%;E$V)q}R3v
zO}2&b4cvN#eB~Yd=KPn(Kf$d_?$rpl`HF;4nj-83?Va`X>+J1^u>!yHg*_dx5YlyP
ztIw83?vBYM4%)jn_NRZ0Io-CQ=QXRv3^FmYY3~bj4$S58Z63g`!;0gCVAKP0bBgim
zDPki#Cxq9KA*IKCeZ-GlA1%g7>3gyDETgEnNI>?DTHYI@Ekz-Y?&s;Mh84btX)SK9
z(8w*vz)(V;xwhXp5-a{a-J9<T{If~Mm4e!c7EiaUVx2YPHpGT<C~o{#ltd4q*1VMU
zHMcj6$2pX@ib>hAa^=OU=<2knGT$DWq>=cmPtt*q$IP}4Dy%ZVx7m8BrJ+9;MJ7|Q
z2aflB1wMc61WR{gZ1-hzxWUGf<N@_BASpDA>6S*a!HMR;juyueDjP55Y2Yljsa6ZM
zSdbv<Qs_6ho+#j|U)n5+=t0R73&>Y|=RWH+vQq5eI4qA}Z6SVtF1~QpQ20nLDA7ys
zzJCdNGNXvIAQ`{FC4nEBnw<H}K9Ys2dI=eQo<ZzpgZVL%AD)fGlmBKX4=HAY&+9)0
z>y4t5_s0~===48So2k2uC@Qxp_gfB>p%-sDo+;D>-qcmMiM^3<;nGS>pU;lH7qgN<
z%Csn;0rRsk<x22_OKJorQDbs?I4i;o<a5+fsgdu?>Q+jh@`FF&=D1sV8RH%6Zqi0R
zVgF2n3o;;9efRJVqfPeCDAnO5?7t;Zn@;cTPoHyEJGh@dnoQKaMkpbk^j?|dCp#d;
z@_#BXhaX|b{lwX*<qJ%}KRP`!GLQvtFAas$`WyQn@B@<sbr9=Wmhfzt3g%3p?V@v~
zv196&#_Q$wBZQb?ol=z^NOjT^2DAiVdsNKykE$)@4xP==EKE5QEG7&x7?MzU!&z|A
zX#L~c3JG!hWoW0qOZFcZD{aQl$4@l4rxx*cwdw`7g_gTz9_S{2Jhu_F00Zlxu}7YC
zS%YN!Bd1PW1moG?Y;WE+hbBU;i-V~YMq+X1Hk5+wfooXzy{F~s!Um5;@ukvS9*tKI
zD18>^kw_W=eXfE0iS?1Ct@#rV7dDC`n6AXb1~v4l!fFP)k!pvM?|Vq<A`z!GA%1AK
z_zoz^cXxLWkB>8)cE%ryb&u<S%ZudK$Q^61mcL5uPyiYswl@&EZ!bBn!9S01w<8IP
zASiQkc>P@=iRX)Q4NmkpwqM9*Ti&Z{{cd*GTI>wT@;{{%Pq6%xOPOORrz_0SdUdZD
z)x&@L1adblJi96CdD@4-GbtlT%D-5y@uNEB!<64th|VDeMR2gh%K(17frxj!LnRI9
z;Wa*9^x|Lc5)jB?&~(TaxFYp3<5SY9d1zw}7B?kp7J}RiEc)qW+|x3Du$bC%yp9t~
zptSAF^^8LITb!-LosGxil^5nG9&b%gToJE77;83S5<C5#?@ch~_pPnr#T?IR>w=LL
zVFeq=@JL%Yhy=J4SLEfF`u$dbBw)MjTerv7tcNrkxEeiBdQ!%T0}n=1O;=vbwbHAY
zqsendHJ-@oUp2_ot=AHf(a~%r?|%NR78}|z6D&UG&$G}X>fR{+d69516rG!ugH3TI
z*}3J+lG(i!zy@k6FhYAb5B3AN?=Y$P$l15v&^E!AnNRHap**WC?M8V!>=A!*W?BaK
zp(>#S4zwqxn)zW9N8u}e<A3K3T0D?2^~Nzu6sZx+4z7d?4k!rTl9^wjwE4<YkS-T>
zYTTT-5MBX;0=RSaoC=Em;1=xgev<z{N=W*T%0`4|`qK90l_KaW__)jcWDBk;7d=&1
z+gXg#pc)_9-$EvApe@;I0L-zneRUo=<cX!x_~KW8(PDFv0qmtZx<vWy&KQ_sD*pW`
zo&>|rDlYPLbMR?8evDf4`1I;T*X8;;D8{kU?y~sW2CTJtKJ)Jb%Z+<HFL>mGhik8#
z-q)%<7Ho%q&b(KK7zJ04^BAe>*SrTicMu*EDY}5wn^2LWVymrEa-sy;jxQv-P+`50
z#p@5#$jfIbqRp>gz%)g_%blt-j7T4lrleY9AsQIi=$IWhXER1e*1xGbG!r_FEJ?3o
zs%LL-3p<8Leu3kSf$^lIwAMvS@bOjE8}}LN3kvjk8{P7!FF7;Y=E1!{mRoF1HfU;v
zyuWMxLFob5If1<=3T~>63ib(5bg7Lc`aMo>$2rc@^lbT&J(@eeMfMF1;033D1@^aA
z;BVym^7o<mbZ>ytTadWO%ak+QM9x|1zeA`dFyyX5b)r~jYr2vaXs+s44vdYBE6hd%
zIIO1it-Uw8!T{L>gaXX$jWO|>mFBMi-~=og0k4?wb9~$dXyC!viR>gLkktJ3_3hD;
zc~M*(ey`xdNhZ}bzOX0{Cd{`})jyy+JAyyln_btP4qP)-N}J(MyKb>~i~FUg6AZu?
zGg2C>RBVsXjIS<Iq~ii+O6B&?t2oEYA{|!G_eOw#Wb=GL690*-uo1B*E-62v4iT?4
z8R2Jx1wKe`t=&Y}m-qRVhPv6y;|i)8n0Dh{J9e9vlH&LD<Kx>y42=AFcZ_c)*a}e;
z(LAy5ouj_PII>#%df`qISqgiSP|kxbP!+D4dGXnJTp1C=HdJhWKZ|jsCwsXvQS>T>
zR;uoCJ4rBw`SDX*)i$Jl3yz=Tzh&J?sEu|nw;0VJ<zLdz?kow<Qsqk-QH`~urE^r?
zVH8RQ-0QZAy#X{Yg&G&io$d@cZ;q6IW)1%wliHsz&VX+i!YzXJ#tz0U<j-h@#S@u)
znejm9>(7d)fA|Bd2I%*ca@$qM^_9`9*ki9H{~cfmOS&vT^bB6s?qnIfiAf%!{sdI4
z69g1GnVEi3lnVAot5U(i!C!)c(oF_a;U;Hh95#pENXy7@LwBB<_8LIT1(b0rG^BmW
zbfCL2zSD0|fj1SpJo}QvYs{ClcRrJ=`ldZ0pDJ*aKwzmvp{+|Us8x{fOS&E(DKUwc
z5CCBk-~jf8F_Lo$)>HAb=n_Mf=yy`Sac`GY7#RBbaBq(b;$bQR?oJDXkc?CWbE>j^
ze)eoM(=--fkeaw5_31x<M7pt>8v-2Ob5akVDU(@BMSWl#9Q$y=hTRk_nmxvP_>1!C
zrh4cIDGOe>Crd2Ae#(u*QYKj=k1k)4MLYaC+)<79BxX^m@UFn}nupvAxA*uy?eeAI
zvfex8&33d}x^gQ^ms-$BChO4nVSHE&?EOB5b)|t*Cx`23<6A5eXxF$2=sl`Gl2Y>t
zaarAWJ@0?`j&+ZC>(R9rgp!2c9xBH;sh{7u_*NVyyG&*Ju`ioE@g)EpmmO$l#&A?R
zRxEYv>2;3x3+Mmlh2CJE8T(WvJ7gk#Kv_&d1q8Ug|I=U1ZmAu6Hx|T(L{tOxtDDOy
z%3iBa1=s->$u^7DKrhMV9YHH$NNcI)9HvW?b0g}6FfAmABYhOib=5flL!kMu@|Z<s
zX^fziwWx#vrnepFqqk;jSpiin%D{M!nkAn@`t~j4Ggt3|MM&5O1gwfwFLTGtVb#`h
zva-+MEXE5Bfb6v6S@}5NllNw-Hw5o5F*3weZ#1k$tM|pcB!+B(h%ET)S7ugL(e#??
zWAxOYpaWYyvAA_gsM8Py`d7j_KOQGMo5)VFJ3;T5o%SF3VQC-9_zk}vDMigUNLxM*
z2f7~J>6eJWGkA;}1M?TBC;t^K{n0;7r>*ftJ#8e~!5xE~0K!naa^mx@VHIgf@;J##
zKAVs#%Tq}YM2G`DtMXmet-17==cqS}c?!N-Re=t_!ndmHNxNw5by;4P_xO)sK_fRG
zI2x|&*_L0lh@e|pw)SfWzmY-1IqDUkhe7p|ERG!)_~E%@J14sdm7i}NIMYW>vpVa1
zEn7N8@bi*$K3}}$(t~s7N6Dh*U+|1(7(Dmbw23s*k^9)XSwh{{VArqNogPZWrYzlE
zL-#6qFfbDnPIi^zBtak&u%3CHxwJU7`_BRDl9rkeO(N?r%p~!6pGWwslHh=}5<)j(
zX-nN~{dAs|>$uGNjY93Z_*lhOv1O!mnw(l#nkSssU(^sscjnG5W<nqIIxKPqHjY|R
zM@R}^ja=tODC4Ij=J_~Wj>w_6w9fc1E%1k{6EwpRT`5D=eDhWO{j;9@gzcY9od#<1
z?m<HfPqOi!PjskUP^X@)xP!F<Q69K+Z@?%7Pm>u!+jXA9t{ogJadr88wPb8`uXcV-
zcV2&@W}p@?iMk{BXlAB4IE;K+1a6P^&!f&pQKH!uJ}bLNjcL+VQ4v#$IH~Xr4Mmgw
z_{-HCVX>g3X?b%vqfoc8u1WL-LV6f3e^RCUsrdtA-NM*>qip5_B!rM3pdvA!KNm7G
zA_GcTL$EW&CnmB0WyRaKZ{alR9L;=zz1&AJ(1?YHk4Sj2#hLH|NCs?p<fpTIGj(`e
zC0HGJLCNiyInyr8b@j(y+$p`)VxBgB!r8NRc)`Wicu$yR<^57mN;ooYboRwNgyGY*
z&tvk70RBC3vde>1!hcqE>wLqkcvBUf+C_djJ4RDoQc<?}R})K;??vZlx1?In*C_q1
zyY<uDoKB=hVmRPAl-49ykGf=yPCN~%(!-q$?q}t+C+#DbS>0?Y4#J$(2C5#my6QNj
z52KQ@mMDvxqhIm@n^z8PW+v}@1z&z#lX3cX4_Ln6C6EU#@vS-EdKDUP85Un5YrluT
zt?eE5fHjxM{%_eD4Oa;QRvGgCE2PoAHDP7$2=f|uGI9JT3&ZJWL89L<%bQ|>cRI>}
zUxwzF;!|1^tqz7{+1aLJ!n*}E0M>=85B42a$48PlGU`U19nWcOnr1CNc5j8T=Q8W2
z%A+5?hxq;?{pj)HzFT+Z!E+76*ZPzc8I2nB@*|G^mZF=E@&68vdDbh=dBOXea!c;l
zufO-4-Q86a$6X%}3gYoP)4>t(I8t3--`!q^;-<ewMoX#uN_4vb`FJ0MQLs}4fubt7
z9F?iMyls5$(Ycy{ssIliA8sFA3^@3|K-O4D5l-ZxeYDh|<!fM^-d-0$L@X?4o{RvJ
zC?lR7<$Gq@RJv-l1Zn5R8at^!J$@j|(-|1nu4)TjX0HEWp#I%6g-xl)nK4&UdTwug
zXrjQnxl)#ytniwX{A*;g2;FW`*}44>rBCOg><KVANB3gJl++y=VHkr6C-EC-MH(gX
z1N+(hz=acfrroviDba_}_!-R+k8N9b1w3(gZ5Ghe1l3VS1xIIg1U#{@fNI#}ZMC*x
zQ+Pw)o2D-ftGhH<4RBpX7w3Q~k$9eU`FrG=iXW#4<_m^_;f{FHjSHVxorwCn;gI8S
zk~`^ht(1S7V70(bU$0l=G-M=MmK_}#7fJt6TRv>tXx{+-t#>(S8gpYOc4_wVJJ{QS
zX==vM=VWK&lXHL;BqIwJo;7`=oyC?8U`)|I8DgmuCgG;u(0QoLO(WkU%x`9JL}4Vr
zsQ#KQ%pT(A>%Zk}C9Qc%+$y*2sfB@Wq0Y<0hb3&5r&6zG0OhyzRjg;^6Ff_J7H3Rc
z!*<M6QxMreFws!?LV9X8ezlaWq%=6d<^8nA;!N%KqbF!mQ)T5WZDr+Z-+QRu<&4ca
zt(22BqYLaDwToH&?;AvbZ;+i`y^X%Mwq{_Wep$HS?_b^2x6rh|XK7^>iUq1LpTnVY
z#|Gvrh)&(g)r4D_Bvzgy;G7)UsgQ{^O~zv|BIcXCfibbP1?oc|8r(8j+|OsO;ybmy
z&Ti9w9uf1tzv*DBk_S^khhpo5{OG;IkXgkD^1LxI+aCUxgpa@Syj*!{J7>aD${%q)
zI}r-Qci`={Pcx?`FiSFQcUYwo*!hn`f$-mEP`Y}Y@Ap5nQXjOigM7!bS$mw;v#y<W
zoUd>e5}6x;sZBgg830~a4Q|(ZNmM&_5_;D-8Gp_zM~((Y;at)Q(vs4hYbR}<=j3Dk
z^nu`o)G2uSElTjJT8v0v1KvvpF>LeC!gdC14G&ILiCO~NdZVjt53DjMdPdwj`?aYp
zl6#ZDOEm7-+y566)SI7qluRz%G}Nddc&o*57R!EMHa+2Tjk&*E8Jd*bkjx3}eb=>0
zfmV0&uSI~|_N#C6={)EgN0YnidH#JjU1=c!gqJ;51Ox=V54zs7Ymob`T;&NIQ9sv!
z&Brj18W{pyWF}}V8Lk<LE?qu0Q%UrqH)$A95q5mUQG8azQ8M4I;$zFWRl!r$uhAN@
z;35gxvc*u$np>ltp-}y4tHk7r6dliE0|now6&Yt%^D|mi`G$#1L)6u5l!{LK?<z<=
zu<>`z+Q@lXHPz-JF%Dy-t|jj}%9mrfS98`sK|#+%tlsqLCKk#u_=3~3SHgoXkkta+
z-g-p|@pb`sI%He7{$l$R6#hEwzLo&pNLVm#+Qm^QzifwOzHT-y$r()ykWCO70h_zx
zpP%c_Dvwt4&1TcNjnxpxk6-=wlDGZ9gajVB4=tY_S<W}$Rl?*H6`uoBf#EJRS;l5P
zi<=-T4#ere2x!pi^W@{sJ29~#puqgOaDTM}^gqCHK)@>tw!)=1<WWgO+Yx7=VSp8@
z6Zll1a>sPaX-Kp{V)|=$j~}8$!FZ`>o8|IZYSea4*m=O`;x)?4)YE`2zIcxq3zOwb
zrwbFl;|(pYOOE_f2<+?-kYjzB_I!LN{W+qtaHoD1AFc?|CHKc0?OCtbS0w|&1tUT@
zNUp6pH&5G%oqQjo71o3~zhfbo*AjD9Mr_Ps3pt3At4H{-r7t(UWLcw)AV+U8FI9#;
zo(VT(u6S*qq`{*B>W~;mw3{eIQ5u1?m^!;KpM1N2jB9Ir1*COfeDK^{cK#sZ3SHP$
z5|6)?HmX~r-kDxB;F$ryig0wg4e>zzuU;b${H{}_2JjRV)#A`Yg$G<3a6`p9yoQ07
zL%bh3o0~*UnNBvTIcPg%%>qQMPpbXEZl+lM)R(K$UU0iv`FkKK!gy)%YOUvFd~iZm
zWHBQ~A(Zs`Nh=%xM-=2|4Rnj;*o#WPyL$DtF4A9lLJ2-Zcyu0@&3w(<M15AY@&W<P
z%+aV?3P0|T6WLfMXMPw96L<M|X=>C@4&UYJ#&F0ryQYlU+`J`s(X)IO?jiK?T($7k
zp|+u^1x8x?&-)EIuB#f%D~*Z7-x^4uv+D@0K`*L2n+ThlpDjG`(L@s$+WD+kEK;1P
z<~y@8-B#X~&A;HPQiqOBA>!d-TVSBG47zrFNFIzPA|xgEhLr!o?C6r%!BN$L+1cbG
zt?HjvGaPLXq9Pp7fB`4wuT*}I($bn^!6(4)WdejFA%xsVVeKUkdp&N?H1s>{i&?l-
zI6Xv&`j<Ud$t_P2CQ*=ca1v2Xk%Z}oxD1HhWQ{tcU0DNhp|S3{q(0e2dO=eg1!GSX
z-exUu^DJ<V)-|y`53V>AewG=nCfL=*cd?BL=PTfj^8RZuCBf*<Ng?F%DF27DYtMUp
zdhI#k<MgUSVOFMyvaJk3s9T{o&pX5^cP>5^P)Q=Pk33%?(deA2{ehP$Y0YK>`pwPj
zv`$*9{TH)-=nCfno@k!W(eU!Ok=*!@jCBFgyAq+$@s{%yXX}3~DSkdrZ*q?j3ve|3
z0u@JhZ!akiPa;rF{d76dWzen>vbJUbwvdOV*k#US0hRw72vJZ~nex{08Wk*s9+z^<
zCCcW$h4EWT`*IG77AHC@oRaNIDUs@$>YuoSKD;<A^^aJ-oE8=Z%SAYeMh|zbC*#!9
zEugzN1?Ti87RrQ-{)pye969+L+8^Vs!8?a%PpF1RERL8SUsdKg65v#&u3k;iF4{a%
zfIKEFA-L(uR$ERsa`Ur+X$_}q4<YMp{Xc^x`aA2diOTbolxqWejPmt6-f6^t5i#q@
zodTceo@Scd&f~osWuRX&FLMk-H&OQvlXg0kz-+*QHzxug`Rkn@ZH)eps>cuTu&15x
z=m@~3hYbO_u}2yA1G36Zgk(EN(?WiAjHU9IbF}pJma^czP0mt+28@+4$wZ&z9P3N)
zkj20afoHE%AXI7oiz>pc{3A*-NE%L0W^+lUULMShc!E0ei7(nv{NY4^%A^Vfr@?CB
zQ_DNc1U9?LoHFINRdG=&XRY0`_|$1?D3tbD5_hB0%n?aUi!01f1;J=}kF9*kG$TFV
zp6YEx{ry=crKRNBu3jMCVJlWw|2D*yA6mdaSmGVGhGpWwe}p0%;x~XkH1^`AF!Ska
zRQ7KgZ^=Kpd=$z<Cs5}Vzvp`zzDO}P9Tp@q5Y^l?(1C&NVB9Z!!Gs9YklDzM33Uwn
z<1R@bBfhua^-1;>WdEL9Q~(M9oza~WuK(5(azMeoP4xqoP*gCYU}@7M3ATw70`EPz
zJt{5}7W2w<b|CFbHF0Ns(E@*n^7H(I<zh**H?w(X-9t$`jI|F<nHUOJVd>ZCSDVy{
z=O_l9)dfI+<WtjzGQ+tu1ybTAhNxN!dYJzFhG>>C#C@{}r>7=dIeR8)zK0U!lTK-K
z<NNAW2Qa@K125*U%)&b8`tPPRMlAbI*)rA7R)L4!>paf_B?s7ECro+mS)YV^U((TE
z`TYcxi%~n7%-S28=R+lEjXBRcrv@9vXI}j8Bnp%S)eLB(HG=XUuX)!B0x$E+oLfNj
z@zd2TX({X(ar7Fb%b^ArrcjvcXDV7UHhm!XltE=r;W}JM)bR074w32pKB2x-caDM4
zR^mZhoYp7;n$O^J&5YNiIJF{S0ClFO%bC4+;S*J&Rr#w+UKSimLf?p}o_QWr{`^cj
zWuz@_#`IM8TVx4Q+8!i2GtB=*(DC!v`oYMk_<x<us@VT^GH*PpitRp}qqE>W@|qd?
z;esrc!-!RX{0uzN`)?6}v$CH<zY^(M-I~655e$mdmA&r|iy6GO4KW|}hnWv&!~(Fz
z%D8g=UriW|9B!7%cxgc1>pvMjA$(@7!+%P{AR}dxZ42_;U}@&^{CP1e0Buv>G`R*`
z;8)!egj5MupDKMzwQ$vQYWdR^0va!-^Sqhz)ud5Li$v?;F7Z%`wZwl*tj1-VuC-vy
z{6TC$z-he-<Wgjmlo=E192vf=%fG)my^4HzI5w(8O!mty?%c1c8oV<f3?Q4(zV-`J
zKX!7O5e8jKPzj^!4%D!niw(n>e;ZHKRgm&#ezH~DB`aHMS6Sfv_((WMah|J$CS@sQ
zZZkt*XN;y#_TbC!*PK_1RH5JKu>X6_eu$a!O=X=O9V6u?3XxG!J3l-gmV1rOtBseo
zp54SBs?yfM2ISIU3~lL$LWAReVVfdR@{bU$Bve(}1WD+ZmiNnFeq7M8813nf{nV6x
za=+MpO)-^(?GTF9dyWa-0cegLxb}BXgaU%gFZ5|=>-CaX!)(W?c*=CNvf0n_Y2~fg
zGND<%?DnHq^#VZuCWT6!1ul%CIaP8F`R^$~i`4)+vDpRM4A4V~ZEMqtvH%|B7ueVo
z#10S8i5TXdlKQLJ>LXCr0qwse<jD5K#X&JqC^qqjpUB$@;<xsDZxWr+RwEQT$7wWc
zuREJ31zoUsLe;Lol4_uDfuXzMOVXq>v4&&yp4q#f3cD*1$!~02<JsAS7LZkc=}&{`
zN$qb9@`&j^#c^{beMXmLBxe=y_dR~k)nLoJm$og{UeT!-FX&0i-uUZ=8{i1S!OzxV
zl?R)S#HOdGceu@sIjrLh&M({7M8fcQ(YSIt?com5v7*QxR@uWs!v^pkkeNEMvp~>I
z4ctjB{Y6(%W)GjW`HJg6g&*cB?(GbAq?jUFU#sJ8BR0h<7mX&NumQc$o$NA@J+5Qi
zCr6Z1j$|5s-L{g<G5mW$@<q^9_=>*|UEEE`VuqJ8S9xbqg_y2B;7nU!rb%0BWUDtM
znZ4EY+LA8eFEyQ)8MdbSNQ>H|4F2!ctXv6mQ%tolt1{EIqS_*$1($EG+n5S{DSnk-
zb#4)K6%tRMJ+x1oIZ<^@pN&`E<e$6Hz|p_mt|GQn1-l_k%+LPA6!8Tm8qh)MNZ_C>
z+P^>#)v^8S0Q!6XeP<uVLD`VGC7`_Nja4s#kw?})d|6SslF{=0B7*KYaq0L`{CP_h
zx%-$|=m|C6#KIGARDT2OmmAnWPyEv)L*)y{t74n4%A;ph6I(p`^~GgO4P8v^aa#U>
z{?;#-B7O7sid{9Omk2;8Img9oMFyIDbl&S8)U`9i#Ml9CpwSJkh}@TU9VhTc{S0SO
zETL=AeJuEcfuk+s_WT7cC4q>zV~jO6oqRr|LuqofsK`-ejgEk}Z#oNSkv{6+^{rb@
z|I(>TWvl_dh<IJ+$@iVF@IuBnJQM%EAe*-4<t5$Z&rqebf!B_0WPIY`S(oLELk`<B
zo;9}2l#Cr?E?<1~nZTPjG&Jmu)vC6(+TDMECreHC7u7|8sv)iArInQ4ReJ}$@l{W3
z;bQzlP7|vTA6i#))IJ_WULQ2&jW7H0P*n!mCEuNw<$E)&=g4DOqhwfR+9o|^LAjQb
zLWxO#v|hK&K;+rl3icR5=(Z)|0;0$_G-@2osgvQzy70=+TbCV*W+Szg)SzKuOq&zi
zW-{Ve4HQH+&p$FGd(N4j8*xD|&6YAWsES(sEZi3-5V+F9EgTd?B%BM%x&GlPQ;lF5
zRnMbN%)~H)!LO{4QU4u$^4De&achRSsP~i5KA#?)j(nh$4a;H*#+L>{y@fZZB1Hhj
z@2st<R?)@vhR!uUEQDFk;Gd|IydD`Uu}dr1>?I^T<ET8-Q4SD&RK{0lm09^5!<mVK
z%r*5yq3?42RigvC__z6i&G!f4iA7oxK-l<CutdbyufP*9_R;w9{RH8zEO+_(YtSTi
z!ZC$6+F5V;a-bJ2p6~7Ubev2kXdqvk>J#SxKyRg#ZBM-`-h)eIbi8uH+kx);1B(8(
zTUTy}q;3p^Zmb)?-VGQR7V`nx_^!~qv41CrA0ibqzc2Co`Q@nJGE~&6%nZ5Y79<2B
zBbhi<@n<xeQIz(cH@E_*2O?O!T2*O$?{X$6%1yP%7ivj_bn2ad_w|MR`Sa%r{T<V{
z;pR1gyNi{Ny+9M)v9J4ZT<jB2#_mT@s1Phd%8dZ>P{8Y+J;K66g68ZkMk*lgs|{-v
z|E!^QOvJizhRM#^7F8(&V{c#F-yZt&#%4=yHhHUJa#IOy@0A<JLAVgd2cz@A^+q4O
zvoY6I6s&lUSMyv*681vJC9);Npe9?;@c}qCHW+{$yu&3dwEXDL07pJ_G1MW`_h0D!
zHU@xWLy|TJK_VTyD)ZN|iM|XzAQ3|n?h(TTDm`r;b}JuSJ*obI^v+eY<luTS69tz6
zKEYT$q*H_MdRWQy^h^=mrEIFtSC`u>!VJWkm|_Rp>c0=0yY0^$T&^8J+a5Ukc;Ln_
zSW)o-69OsI6`4gz!x4%e<b6pF_mH#&DwB?RP_Tiif&-4HSnL3N#qjyw$0AnL0tqs0
zK{fW2+Pvjg-zyHkSoDk6%@6z$lNMu?UY?AcLpZU~x{gOviBC`fO@HO0xdE_2=QgAE
z)l1Lrmz$2^uV#ziPmg|of^iU%VM7XX55H7wS{w?qY4l`41dq<L6*E6ukbh$c4PdXX
z%$!&?sDScKe5Lp2jn<5QjY6`9oz)gxXPpd=5LqjU)CkpqxO5($RJOFV#)72L^*Qo~
zLz9(KC(cD4%!hMrsVN<N?VG4UDcM1NR|i3C^sVrp!8yJCtM=JW&qyc75?&N=9(C;2
zRvekhRzu&_&Z*3WnUaSk4&6OG%7Ic6PS<^l9Qb}i8=pQL+NOt~FaRj)aai`E*4N?s
z>K%Oyp|-zY*P4=Fq`jrDsOGff<s4TC&7aS-b|U{PM!0#ZyoD@!6oPG!T$NSv-LE3v
zLj3E2_wuXh=(q`}>q`RTP@xGEO3_oXwO_3e9ajp{7R{`$=O()wZplHi@O4=uzLEVo
zh5tx2(3r1}6{U@sdFl~1=)5k&4cll*xBNoiKHK9n9`3xnC78{;4%vKJEq-`!d}7#E
zGp*^lsD2#wOVzEn`Ldu>m{x)zc=A*tW9;Z^w|4VnEYFS{8O?V%@FxIAqd*f53EbZ}
z00TEaKBsjb@EwSXi6J;UJ9p(@DEK}Iu|yxn)X#%KkY563%gRY7^>i^i`bxH+`uw9@
z5NukUG|OPdjuS~pReQkj;9SvrI`r0&(-vABFO%tQB_jECU60YdN&ytYl0)CZOjYrZ
zM(wHk`XtP(!az+SP4E@ELNDt8UvBe!pX#cSHL|VnG0KR;P=19*{j|Z#hd|js(+^0*
zFEC6kXJu3{^QkkmJA;M#uK&I6U?Ae(>nW-*oiLNR&%Pm><71NsYElJbKY>IQ1a-6w
zXGl=f)BgtI*hlb)Ox`!HuqsQ<rMiuQ6hq*J2Ru;1>Z%cRb8~|WtUM)+XZT@Z)z<DL
zyv`vYQzbUW<nZLAy?u53JO{YxkAU<w`nGst=Ms306F?#b5(cqY!PrRNdrDnEwTXw|
z+I7|i?0$t683UH25LOusHX*ycY4+rllpv7E@Sw#dBg<6lJTcG*61VX1L84P$X=!O!
zXs>c2zp+Vo_RT|uKn%CNOb{iMK=5WbGg_QDSB55DLmFK0B(>Me8f_&JteP{1Hy-}E
z=WLX!Q+_*R*OPA**b0E~5is1v2Z|x06u)}95ovSq<>{y%_KY_84Gmekpe;PdzenFQ
zB=)2;yFXDA{c^RRqfpzMg9fo0AmA+lR3JJbmu=?)a7*`2m76>`UAy}F8b~!!l^FnP
zGV4p`J`QV66NFNaHwKS^$<cPP<q@D4)z7B)YI%#)zN!^8W;7?M(7bx!IKTk7oHRP^
zs8w0dNdxojWVs1OY^;kwrACrh0*@07$W5us@B%JORBSqZ2=Hwn_QQ8;G-ttS){lp3
z_+cs|KL(VJ6l#LFB%QI87Xa!p*FS?q(9MZnn(uq2s~{njFE#%?G<YT}fN_rl?@-)Q
zMyPm88<9ls)%~7I&f;aNZUO0%D<N-_&aWOR!SciI63wHwa7)*7mw{qISvzZk(5rIi
zpE1}hp%BxOXzZWBnDU|}uc!ae1v1+BIM#?*tM9EB+IEgjR7ZUG+Q-{D8DP6!Y55F~
zfQX3s>XkpZJT+%jNm*G$Mdc+32cSRiz5tP9%ns|akKy6D9XFr<{{0(Vp=j^mkicm}
zOcN6f1{p}W;Y=5r_}v?u=$1TgubfWYy`ZT<pxN%d@;)6?e<)q7t*t?T18yZOR6Jwp
zWNY-|>dOA+!mc9(&oJ8S@9F777)@qoCWw~nJu{fAbJEb%{5!yV0Ma%r`b6(qn68lL
zA(QogSjWJ~%*G!S0PlkQ6~8~SfWPZ*eC_qZ!FoFP#)?}bA}yA-(!vH}BzK-5`xU&S
zm`m?MG@kV-<#FnX-<!}(sqr62k4lS;SyW0!9lS@GoxPgFcKwbl_JN6$q&6?Pl+t75
zHbx?!ePm?nK)f~^^>QpYV;Qb?ZxWAbIOX3<^ch+t&&+-_)HBun-Mg?~*38mERNEQ-
z^7rpajoEs8H6S~OhxGSnYh9ulfr$!aN<@OFi?d=UFi;>$ghgIn9;CPsew|satgKuE
z(+;dS7LdXdOQ+c;!`P^DcGW+E@+w9{QcmvU-Oc64GT?5CX~f3FYy-rTos-kY+pAL`
zWnVftuo-HSP35~RZl0Q+KHAI(KN=LgAIgxxSnr3D)c5N!cmY(LskX7%W%xI8ccf8E
z3+3z)78tFcEjJI8>%IT*n2FG}2sSv?%01LOebb_)oLGE)SxR+Je=BzfSceTw0eM?1
z4E+-S3uEL5o<uZ<lD4yF7d8GB0)PjQi!sPUk<s9F5@6;Xkc3P>!p{+<`?GV!i$G=A
z#c^#BmK7c>+;yf%8yv(R|Lm#*f}?q)Y0lTP)V?a~&{M!Dw`Hmj5owpB?8!H^V2^Ku
zMK8wwo2+f{)U<kaWe27P*0y?K*z4cW65QO**;+jubo0BqB<ld_R^;ToH|;N<v$9vc
z&_oK{Qk}#eA)wuVG@%F0K@p2ZJwP-L=eit3xi1N%Bb%FDz>78j9BhLi(!~{|7NJeY
z^*C-0J^Iw2#Q7D$hFdHM@!??1B~2%-lSG9mK4-_Wuiwl2(y`wy3BRJC_GsWhFk+dE
zeH1ky-91<3%p`~_V2SRf^Sw4E`o_ut4PJX`6#LMD31}9yp*Mg{@|^;`R7W>Ep{zG~
z!P1&s>jc1UXn?L)>35*jitIu?41VLSrrhR|NOCiecvY(y9W(gDZUM9UTs!NwSJuey
z*$&+p_8n=Sb0^xOi`m@I<NxO0YraxKtHMWJGz`o&B}rafQ}Z({dZgkZEr0$*OGQx(
zHv<EoS`q9Wp#FJuemfY=5>oSmEUmu2zMk9K1k2Fi;Cv-MU<QT*!uirR4v<>(Xc0JU
z5D$9$adYI7#3dz>$H&J%Uh9AS_(aIj(J}OF$>7a@eLO{pL`Res)(Yy(=(dIe_Jc)p
z=9^eJp@Te98q4*?Js$;LXlN@d@);zG-@5T=QT4Z}Z+7gVoXqClGdJbid2En=zooD9
zm&cudXW=RocyFi^cxjQRRhW6kfezxNX#^ofymy<^`wgo=)`)6d6CYew=o3KkOZ=og
zXyzk$zvPMzBt;cOa~bP9{}XbbzartxOP;Ddk@damR?n+RObYa^giiU18c|2SOaeDa
zM<_uj@OC^nTS0UOLxEsYY^?JH4>k43xp|6wAgqYtd}AHKyjmesjVK{uq|DK|*wtq~
zDJm1hEaEmW0%tUH5v+0<Bsr0?utc`CiTIu`8Q3XcHh{g5jDo`c^5`RjUXv#y$3O4&
zsX~dw`$8x(dqpSi@z<!#y3#SbEadsG0hn9pE>gNmS^YYRZqlekMEa48T?eV2>Dt*8
z%17)P_$iA(r#7;VQ<bvZHXb!3`IMv*E(u@71AiTnlpiSkm(s%H)TaNm%_p{|g+zr-
zgm7_1Xx;AZ_I;XCXNpKV+?*JE!^!IYj^<$!L(dV?KW|N_(XduA50o~Hj$a`B;UT)6
zu4EV2U|-+b>IP&j6$8UM$WZJ{<xc{px~7{W89qL~{hk6~tAi^A-e6t|Y1)I0?Xg5O
zp4x<l#)M6J1$lXNIC*)O34!A7rgo4%J1jwBU~N4B&P*v88QLUMxsr(tWjuqUZ3syt
zh1?ZfHVry;Z>gw|Wn^Rk5(`O7Bk}h3cB!}qX3_)!FP_mH1&NGdm7%1Ei9HJeaVhAm
zKb`FtHhW9%Di{*bz8tKsyaRl#4R*vM(&<@jiOv&!m9L5;(-MCHU1z41tThdJw1vnU
zUurpFh6f=6Rd><rgu)4^++v-_%q_BnO-f5H^c>NJJPt$5?$mURb+HP6HZ|2o0OcLb
zl@o^FudUea;!nqZV~f`m0#EdySrZ^ioZu6G7{v$FJu2}1NvMtvmm__GuhfLaSfG5#
zMm?!=-sesaM^<B3c4KR6c~euWhpOX&{~s^ZG_8aQKp>idciP|I|52&kig-rnP+cV(
zQWtQp(OQ-}Y{<h+SLC@8hi~up?$6afdWhk`<FLeKU|^tniDJaECCI6%sTr_80AD1J
z*+5QC?iYA3=y6U&teU`Yd&A309W^g0g8W=WM25+a=AE#)m`d=ELRy;1=Y`ks!V|jA
zZ(jd|Xuf=Sq9leOER-enBJ&*1Ec&?!&6d;jL3onbhpRra&$a@CQGrPb=cyk1ecy}*
z8qOPhAKWv*#LFH=Per8<pd^S92?S}W9AZ2m^hGR+>+)FO<H~T=VRLa$;e?mw;Usgs
z6TX^Gz?}M`KDMR2?|q7Iki_E3anIY27|_M9Yn}9%*sp;r-E}Iblp8q^ub;5Vrw9RQ
zUId4Vkv4rs>7%b-=DZtv-gM_MMiN7KJ}KYB>)krblv;0l=$5cPhAW^U%GXCRM??)z
zn(M#A<Z~6HUbL3=ztse1l(+n6PQ%3|(o*JE9ymzE8b@?X=Nx+^Yv4)2-AI3PAypdP
zqzeVO)R_if&y?bMW&wVCpI_a1en^!jj`!Z<^22X|EetU_p;(=O`2zwR2rnWR5NOh6
zo|}tl-a0-y%5Q)+htk)|-YYB9%%!EI^!q$Pg~MexS4j!i8mo<7)3?w%y-qlv_b3_o
zYa@oHyqLmd@ED-N$I+Bl002O_W2z+iyT2bpsyiS%`^!!WxXIdDJ$mGEdUgHC!CARA
zQ*F`6_@ZsqNZffp?nmeCS=Y&Lj}v%4-gE+0gH`mxpWEo(;$`RL+#?M;UTGZq`ro5J
zGiA=h=X3*C{<y>*j$L_tLRrq|9$4@X-Rm4pnopmpxyzMNgfY=^p_V|&pC^7S{c<S1
zM&Y1A;}qnJHmPcmIs9IXP4@3L)i=cwSb=CuvBl}hk}@VES8^}6cIHc*3~$`BZD#yI
zv|6Bopg-(mH%Y^~+xeV}W|Jp(P*BjCdM!byI3_0M!>I~~E8yws?=?F&m%#7A0S7>v
zA?U!buCK=~=&3*|?Ev~deBg1mn^$V%>dFQ85MNkFgR%^E0KSVwUpN2*1A$vBu3Gj>
zSXeg5i2}e9sG~L~%a&stPWj_N+9Vv4VaKyDVt!Z?t`>+g1Q7c}a78TQ|7!^V#&Yr|
zv=LLYk#+%!P7u|E+5)gF2=anZ%`5Rk#ow<<yQw$BRoN^IfPbwhB@#qW4R4|puPR6C
zWxjN-%p$R$E-(<G#XUe$s<maoVUdm&s3)!(>JU!A+xc6qP{d0`+hI(Rsp?ip`!w%W
zM*@9K?25Kz5z~O>e$}Ftg1kLN0H#-p`)b{#2ebPCnKk6}_;39WNp2`g(%PJZrV<6m
zh7;yrnc;Q++_J)WKp*?x1fAdyh4uHS>;1SdzyF}Ks~G@QCk{%WWn|<?R9?R%c-FT7
z#aBD;{YIekzQeoncQk9rEWsoq>VMEV?A^AuwyKWzKh$Ty{OEencwTdVcU>?Cfsngb
zO%U+R28Dz`s{j?Pga5=ch5@Js7+&r&WR@JbqJe0l$?JL(9Ee%*zyb+sx}IeRJSKfO
zkPT)8ib&Ne4~-hzH>25d1swCC1KX^w%^;8VXb*T9)p;GNL&}f!Xr0pd+(xFi9@d)N
zTfvNz_YUVYC}H?9KXa$x?+#^O_W}Jy=d_5Q%i8k3D)Z5ZL%v{a3O*rQEV+(*@=Zv7
z_<P%|0EhhZvgjYycQLatL^Y#>U7@Zn-Dlng&s^@$4ffs7-&lKivk?V4MlJt~F!pQd
zoJ&bL7|c-bL?^6CmOUSgV>SClG5$Lh_%a3DGa`1EF~c%5nXd=o9d&+&K%6dnzX!%+
zvsiPEC%e8-^xlHp(Ffoxo5&W_(jvr$kG8+R|BQem1cabG6h2Y!*SfJmJp*i|FR-wF
z0Y7@pR(8^mDO*iV%@9xr8xJIR10@(9i;<-NEy#*~sQy4YFDEfEG43-&M;;=IQ9W8w
zAW1#jo%*#s1{L6eZ3Q5Ga2QzE?{UO3>J5|`cGCPg>*L1(*S~_{*94;sUI37$g8Wps
z)d&eL?;9=!fy>NSBwOp9VBC8XX?>M{GiKx)o7ac->rrQBsq!02f<ZT><BltNetm5>
zGN6kXgL2Rk>xfGjP@s1^O)^=}MMZGw)C|!T!m^kvio;ari=6Dfb5$#9e(xXm8MKxO
zH#|^2wcAbZ9mq@c9~hBdRDNN!)gp|h>->B(_hm%^cWl+JT^-`qS0`<O6ruGj#}jgQ
zZLhh%JClex*u`r6VPGSpX9!B}O?bJ6y|}p);<1D6T6G#gQNyPAFd>%RCwh9LpuK~B
zgVycJKAlVWg?I0cbyR>$>R@v;>0aOpIeqUjK4Uix7#i_VeAN=~D)pVAba((r#l*$O
z-O@lHcZ^<;9#JK%f3nOFDDf;)RuUZ;I5;>!KtxR5`y4g5`!RnLrHC=qhK<2A#)Yct
z>XL~jfX3ART{4~vpAi8S9Yf29!6`3|H`JT%PuIY$-NVP9XK;^Z=M;O#sNn|Yi*wn`
zWu@NfvwfcS5t+*pfz#Gf^D;}YkiqyFRZR7J%vlx3WXC!A2P_<3=)(&Ti&*tFw3Go4
z(3rJ8J-pmARQ!Gi7Ed0bLwiwbK8@wryp%<kmmJ^PGfXaz%NL+8GkE^RrAG@CcLtyR
z!bybAJ#2)n4DSlJc4hE(b&qIGIEW}1i;fD-!mH*T8_O|nDZ+(zj9ZT%ZbwA6gCCim
zVzy{=&7O5b+Bx4dTxo)D*Oh)t&WeSDByrQghle{J!+d-FVqBm!Z}e5$TQ$bRy88O4
z<m6|cTJ@>Y(zPI+f&dVec(nKFQ}%797zl$7rCbuCTM)SWJi>?*{bb16{)r9m+I0mc
zSE?wmc|vs4EuXWrE;lv?hwCJ4L+#kBI4bsD^!dXx+frbNX!j}D>tVy~T{-6;?K^w%
zI{dY5U>S&fe9$Bd_xReW_-MbyJ7ubTRedt;smo>C9UT-;f}t*rV`Q4;KQJJT8u->%
znJKKtzu&Bs9Bu!^pZbw48<KPg0=jwPyzv+{(1LHM$J(2(2(KY}u%Flm#-2%|hJ;{#
z=~)~3>!8i!9VRF^(;n{@KEl)B)fi^Tk~mpw%MW=rJn|j*-Zz3~r{vDWYw=nF<L{+W
zs+{4(4>XAI!kB<mw<+z=1Cgyo^S>LNZ1xA_`M4aBG0=&Wul3&)zX!v1wlfh9a?jwT
zd!GmjHsg2d@vpeF!0j~Kb@%nH0(kq|h-GPax9{i2-a_wZV0usFh|)U${CjG{geNC$
z;HXUSF`~S=`Xy6*^zud=4t0}aSKK`yE+H2ZiktaF$i^vbGySEjI%SgHV(5TxuC}(8
z`%u0`eSSFW52Si*t52uYOvPDlR`@8-4lVF=RUGlLb=ST*bGnG^Y)cN`SzhoxQlt3%
zwB|P!v1L((;l;JsnkT%a?B|2X#f_I=7~u#$DsC*r;3$~VOvSNi+-=@VhBp^h#KuFD
z_yJXl!42xHk8<pncD`zJC-zKWh_Ue4vw&FY_c<$@U!86)ZgRwIxM6jhsm({ps$&ye
zdB|#HN+(+QJH}q`%umS08`yNGIUW`@IJIa>uijb!Z5i#te4Ud;(TX65o<G_HvOEoT
zoTzRwpmTO_B=&6u^Uq)4iwK1LCei=cBV+0UW@dM8t*7A_H6r~z-@6R|)ycVE_%Hov
zMv9D}wEJ83r-94npwo42t#nnn-=}C?G#)pP@O1#*cBeuqHy?P*ai(Jzg>~L!;WjDc
z-`l~(QixQACGBtfN@yL>1e)$zZr-~h#LM>BS&kjrBe@olR+RjxL4_7GKmRRS9A(;E
z%zP;kzOgvmkyPw5Rzx*D1aRlYeh@~?vb6%I1}37*g$l#Ik9}A=0}~}eHiG^T!KK;N
zM)HEXmqN8Y1=fz9X7)=LEzr$M(zj3A#woJM@`ow|r&3y{0vh$Kjn5+&=y=rs8zQk|
z!_&tHV9y#hhhn_l2L-O_K=w$_=X}%QQjo#6KA48_2_&M32BKG*GXSjkmWC!6WG`v3
zgA?w{PLh(%YCWZZV&X}-*;56i*g=}2h)|SNG{5@jw+GcxIG=_!_H4C~JOTzA)Gyal
za3$#qJt8R<eI0Ch!eif9ZFwKiv!jy;IPh`6v^1btWJo*j+78ZJ#(ylX%*9Usu-ROf
zEB*Ko1Tg2(ERKcQCaKSb4Y#4M`$wNIogQGDD&4~@uJ+~*q(W;E=Cqfrvf^R1ub2if
zQc>GYnSO3hDMb74o7-CUrxD(7DSQk#^nm!WE4TL%9qK-%Bo-a3MY*0>&YfZX12`hk
z?DIn1MJIdvhm5IDBgAjGxZ+Ei@9<OP)dh?JAIo>#$NCD!Qm9e(WB%*UF)~KR@F%F)
z58=oG7$nw}HQ=WdLrHh_m^ube5qMjpPIvE%AE%ll?GaH^QyITs4QEHkAQ%l7z(7;F
z$`>F|o!_VupK^EQ$M=@|4endMh6Gd9ESUe)i_(e@re?kjSO`-!p2RoT{HSK*Q^L^Z
zVWO0$I$OT@boSPQVe+o*0w;^fTaUJ8s`1RBHqjSxeiMO%<_EX~TW~H!vtKP1WO)sf
zxX`})Kd!zys><j48c_^FR8rC)6a=Ij6cwaPI+gD3cu)i^KuPH?>AZ9!(jna;BHdl@
zxqkfq{&-!BwIuGnGxN-xIkC^)<0^au+1*Gnzq{^p9B?BG<_i~%1{dKhou10Ja<g1w
zd^T|iTW%X+;ffd{J4j;tsQ&N#7<*x|)I#Gxn+6LoLvKOi{l&fYtE3OT0)-CKkB`7L
z17_CCnZNUN>o5UL>;fKkNQ5Y?!_$l;;nM_jfsvBMFx+AYkA9V4P#ky~WjCPhn`dBv
zmw;d;8D61G!z!bkv(nNn0_&1T$o*0w_Y1p*GdHIi?oM{rW``QZ^kP%?N9C^x`hHhV
z+Wz4&!(%RWVs^cVZGkqA8H<kE$BO0Uym6NYPTCEDJTw0d9JVc<e9K$~+?Xkyv}lHn
z!Fo!eKiHBy1&jDwuT*<Z4Msz_C<)YGM@2{yNSU4|Bf3<5K$%p%=qLww8$w6VOJKPx
zN4&w_{G8f2d#?U9X;2qUB*|Z@eQ?5;RXx5Gv&jl<Tma^sUm#!s$@%IG+xAkM*{@KC
zSQ#C(+F2Mulm0^PeQU%lHx%_(wGN~A6|;PhPF4}_pJ_wS_=(l2my$My*Kf`%tYdFy
zJElt{8SZCo6d|irjbXf*KLnyL`W1!_*HV>8<PI|O-Qx-iTw{HC?MZ52Q(Z@fw#m5t
zUWfCgqm89e#^*ewI5-k4^G^Bl)wbIb31u(3vlE^kA9<=@=&!uT@Mvt8K!SCfST+Tf
zQ+CH4x+PgP0&Utt_}c^Z?gzjBYSA1r|I)eN!*8i^kg6OhQ?TcWdh4BEv@}}Q`}t3H
zbBO<CoPZ?xIytNT!#nWO-U3PTExo^t&|oY5FEXxUqQR$t$wpJP2Zw|-f@k`W6c;k(
zy(JbJ6cmd?ZaGxY#s<_ESE0EnO(j>`u<5ys%x~aLQJOGfsG?EG9aPlRG<S@ZmPV|)
z2ZjDzl;4epC}{h~tnFQP?S9FAV~$rM%m6e=^kh6am>ADqd{(z8f%UecZfzm@aOKts
zRbaY9_|=AbW8r&m*=s0o8!oXjQ+>KCeD_jt&)b+JL2WIK8}6ghZzuZ6yD+G0E8_;4
z_=;!h3|Oo^b#RS5`P<9J(^_Q8->@R6%l<KVmMiX8$~oUM-*xLP7n6$#40g^cQZ*NU
zNf9zY+Q~%|!xJczmg%@r>KhgwCT%GZR`4iVj1woOUkRU^Ct7-D!>(wRQgdN7nP5KD
zf0bCsaa?QKtknLv@|f8Q)1Zca+qFvbwqsJoH|Y<@I)rkE+9KLG9qJZqXFINlN9gDv
zrgxp(X|>Ab9sy3{*Qif1F_h5BnFJLqrHQp)zg|q#CnP5~Ch9Q(5;!sO&go7r5<xi&
zS&z!o!zIpu`wR?sxVZ<H^*D-yUbRup|9+lZQQ~>!W2%FL#?78e{b*TvuF9q1^TiA)
ztZbcW155L!e$(TB9#)cLh<YQ}y=4NkOI`^L?DLL)bGwyYeu-i7s}!7{Jcm5*Ol-v+
zUKtq)lQUgOYW~n04bC}U%Y9N5g3(MLjZ)hsZ?^gp>EH~vs4xhp(giC#ckEN;+PYaQ
zrj~n)n^V#=ocp%c`{RIX$d|K^n6>fTy6LIXfyPveTc<U7c~m${T$nM4jLJ*4kJl0p
zOhPcNEbRGWls?=y>WqE~)gGxRE<A2-9HSrTNFq$iw^!IQo7Jp{L)1mePQ<t`om2VV
z!vWEh`hlK;CaJbbQq_(8h9UbxaVb1&p;S}i67~owhyr@v3`V2f0?4+wl+j^V4#fXl
zaE465mCKj&oa;&uh_O;-V$|n6fSf&0_>`2NpPzz~l8H}7v7oqkU7^pg@$Ho=nLyE`
z$Jv&j9WFIF)a!DDkVtcIk@bCBTHn!S`{KM!k*^W>#rk2`SjYo=tR|jf>!!jO{Kwwq
zy(X$o6oOHP-+vX?7H0PzdT}tem$u5ia=x(NcQXI2rLpM=ve9TR*xK}D<5#L-0?UKC
zmQlRiM>=mVIi$?4ue<I|Sm5ReQMHPe`vk^_zfb5*x7l$NOEHN&Gh6l4{=4`&)vHR?
z#L-pbKdM6#w$(~dQ|^+~ZX|Kv0`HQA^%ytmKGfAS^KMO~VwrJG8(eCw;P?TZPd^vr
z%=(P`h_#WF+c_r1RcAG>Gm%wQf@B@VTdsNe`Cs<RML6s_Qfq;WY%(Y&Mp|D#xk<1G
z6B82(L$;xN;;&H$_gU6QjPrUpqRwvA<I=@sRMwg$<(H+N-pl+52Vq#?=o0gCV1~0_
zc4VU6ef?<3+q-)N<}q7$J?Qb~!|vbBcgQmrokw=kXQwQWH#S6hz9bQnb)1C0#0;3j
z=r(zd;ycl-a^{KcC`4{;JU`aq{st0Y&y0dwc%r1*2lUMSkH4G-PRfAzpspUWa$LEK
zW07eN=rj$%4*U?g%fo!`6-9M2qxQLv@r7An$KL0O!Y^|*xXx!6T1~9(Q<Aw^++84v
zkf+qT35co1ch%4C`xJ#v&PhMA>ehQmnpa#Lin)d-W7QqLHywKRQ6zUTp~pC{_{*S(
zo$~0>8K_QdV&=k8-Q6ARoE(Orua{r@ong-Ki^2d3UQD^U_R#`K2}sb9qC2!e;=;37
z@HOa|xY{{4$#7NYYu=Yb4b;aV)!ogM>ZXsA-e8`8I$zg`Yai7+RBU{nbu@tW<;Ks_
z12;~GL;r6Md%Ht`x0hvj!|u}E)IQuUU-nu3Q?@sy=WPo!Dl@kHy-X(a*pu@hF@;H?
z8<MN=XRBBxHpyKuy3?ETCZU!1{N5gN2dw1kkuw9c5HaC}Z1lRDPa)((q2~gNIT5<W
z0ejks&z8#VIj1kkI4Ljs*_iK84aJ^wI*25>Cwa-Q03KMH)})Nm)dikp@*a%<hIvcN
z*Mk?1wReYnBL2HCXZR9svVO}@1a;#cbL>h3T;E9Y(Lu+WYa*IfyDa0axmLMyAKofr
z>e@>h9GS|H{h3Bm-A@{*@63wxe)&PN*Y$gYGk0p)LvM#^O(2n<9+|+`p?ReL{KRM@
zQf?PZM0EF?!#?##<mKowr*4w3(Ma9%xYiTZ7>kg49V;hL46vI4vrC~P{q9$OKg;aY
z#I^;eru+B;M-dD-ux>=ZG*{`wTeGJwvMt?$<85F~6LKg%pxA}rDqvSmYQLGDi04c)
zt~$OELV5+sLyFmRo`~-1Oa5ZOMr2i}P`iejSVipTXEsfSQ}9m5M+OlD?NGm4`mtma
z$LWB*vun7nCmr*q^f6}?-2-%a9g>Yoo=lph*D&bm(>6!-cXO^K^77Q|;nq|gRVnZs
z%OXak^D1SfM^4qd57TgPH}62xT&N<+xHS&f_(_}}WRyeN##J=GAzoHDruCW0_{l}9
z1oOWq<c|h2HfgzWwl5{Nk7ZlrXAh`5vs^fI{p`Z(uhiw4*nV7facp<YIOW!9|1*Q{
zHXeVZ*JW@C$rm^tQ0$rM?m7b|-=3;R3t31w(ymmf%nck@(Y0W64A{kub~Gw(SdO7A
zby&Pdd&-~xIW2e6Cg{TyGLHc)cvLS2RP+8SqJneufFlaiO1jmx<9XN3+fVIYv5_9k
zWv_+QPui`-y9pXjkEH9mUc0~#Ch~!p%8F&4K<+axb$(GX8D;EH9qqS7;tmMoa)g|r
zE3@PUqE8-ctmez#YDmAw|J-Ven!G<Lh5zPScsadEns4}<+qJLlkl5+I@7@lRsB>BZ
zn&9n3O@*@03BEK7welKFg4MyeX%w6rk+_qNqMI3LB>)DNO6PkSvzMs&*>xQ$Fp}=l
z&qBx4_i1inj*zL>TDSABX(N1!drga-#I@}|Kc}UwB~M=%CsRkA{EZ|*qVz3!Vl5+x
zeD!4L;C~^GN+!e;+wT+hsv?WEyj0s6==t5=yJKd>)ri{C#WY0W*U6o8+I)C62mNc;
zg7Mj9@3X5YJ^Tag-epxr{>?nqzhw(<{0to+^=|)~y%_&GZRA`Ya^hV)hjYk5lvMYl
z(`PBv%C|_4m@dkNj5x+mN{f>}LRW)NiudZJf{@{s6-o$s8Y7s*$e8vbd0fiY17+is
z)I*`7ktSPznzi`x^nTMrZtaNo8s_?ql6yV0iDKE!@+N(s+|F#&{ZJ~duY?^iJ!RxS
z%qiV}Slj#5d6gy^AtbOBWAW{%96ToYt2GJ=43Zr=mvTYf=bp@o0$<XWP|e}qXC{If
zOf9V+yb&!av-8H|f<H_TwD5f^A6{JBWm4l`V*KP+_;T@z!MP%9ZsV6}g!{c8r;V@D
z5MsvX_dhgTO?n%!sAYDnSVN`2)xMylk?KbdyW)Ddh$$j<Ln!9BBhVlLDlk7J+k8PJ
zlu`A-smEpi{>_1uT50C)mYV|Rp)<ec?(lrZ8%}AO^rU)t_wrSv2LAl$)46L}qJ7$=
z7P#;xF76$SS+1)D%v2u~4e3#N-rB3(y;SD?)+tZbG+FrYrKYBSH;xxlC;ZG?=c5)|
zTKw^3QVN+5M^(@MXw_Pu=G_-~%qv6AR66u1jH>f@!xxI~n;@5k>+Z>c65^ClXa8S<
zAwRwL2$av%4SLvOaw64)m3tx40EyDU_+~H$q6Ta>j(hi0{`CDEFHk)ZfPHD?6jwft
zU5UZ%zbcuQFrDSG!1$_!s=+$p8Z$zuOqP~kobYqqdi8XzVTjk%%-@WIl(&YPjDqu;
z@>a@G@#6eBn}g_ak8k2Ql}2YgJ1Ro)4~U%#{d$|7HOhN)rq`Ir`xsKL9ebVWSUh#D
z@b>?(FPuoma=>T3;b7G3#A56E^ZM?CgHXxl&br);(B^|u0?##!Lt*0ARUe-qRe^eX
zR&82qLQlJ{oF6{uah-8`_Hrmf-XuCEIE>k8M|bL{(@f9z(JZ^aa)Piu5Mm-{;5+S}
zYElF17fLiwBp<DE!Ga=087J(r9QKQ0ah0sk!#gdVYHj=YX&BdP7(bL!(gTA3Qf;9>
zH(V^&hiW9NUj0s+*X>?FZg-0>xR}2$G3b(?yQ8befcO{L9WRN>YJ45JntGu&LIBz?
z)%a4Xp5pLdc91tr(e8MBUTBw)ZZ3)OLES)IPETioG6Rv;9oKH!#7izF!!#5h?y}#f
zeDa=(M91lvuQ<UVoJ-fI`ZU|pwE3-UY${F7ySGnlera|l-=n7%lSN3zG*I!<fwwqH
zZ1Y3^If{mxIgcT(f8sNoQuXRgn}=P@<WqeX*<shjqc@g#w0{QVfOX+q9ty?HD048o
zU|uz!9{)ZriL6UJTS48k&*64i!JwAoX|d8&)rUe9d5o`Lcnz_+xkX-Kv1g8>jJm<b
zUzK~<PY)6AXKQ>vdwa7&{b297T3b6QaM@;A6FoD*y@YSnrl&P8B}yPp-xci6@H#6Q
zS1!YPZ>y1o2BRkD+|DAhI3>SJvA`so#o?<*AS{N&1QTgHrFghdj?X#J|FdCXdFi!F
zzkeHM3&0QhysLtVv!m`lYw?jYmSc9|Mh43;F~&}+vO>mp(vsu9LaONZ)aGQ1WLt*q
zCm6d2-PP&eV!p4I<c_km=~-0Fmo0keiC=f7Z%c9wa&A(%6j>&iM>D=4#$s2Ah0?e6
zrfr?5nOV%Qf1f^c(ctB*DC`qyk(4{IaNSW;kHQS{cl9wP%-)dq_#o$5-8q&GN?24S
zwOw<1*>}$&&dhQLuC>%6nm;7F4B?`O-xI-^&g9HdPfGPTW0uqRN>U>6UuwLtEblsb
zk#ob~crN#gFw^p2r7#X|DxEMj`cg&*4Of%V#(OZMo8Og@oG~cmPCReG9Be6p5$9PF
zGTq{_91?_naLs7AaW6!z0_9pCK0b9K=bVi9X;Z11%H%wb$n=(VX;bi6--BvTYpOIZ
zsw_^=ACy%fM$8;>MzPZN7RP7belzV^d~U?!tOnr+*HFr_)@&*VY<@i#Ch5;zV=D+J
zD*TsedPg4opaAs61~X)yG;H8lw9TboP>n)x_0$Jg9wnR;;=5F9(KW7>ds@&@*Yw2Z
zm}3^_6CHs7!DXj@1p$j}9~J)ka{W&qnez<4&x2byJ;{_eqE$ZOzz*42npu4|a|xfz
zafh@KX?{&47^Hxzl;d2nOYTj-xxGA|6}1z(Iil80RJ3ypq+o2(jJ3iKJ_n%2`;c2M
z*ngKxUQ{RfuIJwUnUc6$v+mf?qBF!hGTYl5<26C~*J&tF))&!C)4vEegJv2i&)%Kf
zr#B_{SGX{y7})mvqyKuGYN_0F-I`9Q638~ty5g~g{82kWK#-1HSK`r1(>NOgxP8uv
z**9i=Xi4q+SrStR(#w$94kWQReHxjcmUA2re!x+y!TR~d)8@LD_GBa3t4Cb%mU@TR
zyDRQSGkFc?x0+?-IBLD7y#nb}(u2eE(%O0d6)HQzdvt$IZLS<!tv{%a+q~|lygYS0
z!JgbiYqPM=+iA{MZ?)#xb;(=j)lQAjT}IZb3`OF4b7}1<q<?9Ls2Y7Tw7?@8JLr5m
zSFp#er>pjA1q)?|wVr!#DBMR(!)o<pN<0O_<vj2Hiemwb!6PfD6kQFS+2eRW@(7Zj
zJjW+FWPTTSm!ulo_Pi%9p9v)u^>z1eQZ&sq6w<>@I6<k->UgR*D8>3BMV|dM>3-`2
zKf#EHCxnRRy)2isS9jeVu1^G<3w~cQ;qgxS<q8?C&flkEGkd?ibv8<D!f2Mftwpx$
zb(Em4=r#8NJkMpUNqfJt+Ycx3$IN*F@jx8t>J7P_NWL9I+B<hJA&f<1eV)vBlSWoP
z^G-^-%NLV#>=xWIM&|={8$IkABa9Mt{sE-(@Wx5O(|uYZ%e~h^t5a!MmpJtor=+Nk
zZLk~-y{?RG<NmG=jIFO5FUyck=m{k;pf8p;oEr%LW<S|5OmXT-QbkRnIlB%C&brQ)
zSiD^3hV^Hd=FjD0le8^c?eujHDPmfEVpLj<^e*kL)mKpqyss+szA4T{Atp;gmm7^V
zwPt*LF5pTW%X9U-V5rddnCQ9&<%rkOFL008r^7i)72aeI1h8&VrAR&vOclcVyDB|W
zM014eb0>Dsg`;p+C|<f=v$y}o^P}ma`*K1QqrebZwRrL`zqYKe1)tr13(Ne8sfwaI
zFceMusg!;&p3ckROUJSfyFO3zPdYNO?`YXE;F1Y+ri9hvFRH(xy#yxN@yBs~zVZ20
z8=iRj+ERC!n6HfRJjzpC;eU~q#9q`JWc&LL+;bsK&8F8pyD*ML&JXdei2(}YZ7vIg
zjuKeGms3Uj3kS6p>A5|_{hbq*0+sF;kHReH0-MP1k9;}D&-RUS(%>mR;wwe{Q6>H+
zdn%wWz-FbN2l4qulCXPC<5ZUGu42~ReU-u97iE`U?fwr;K!xvhi~cJ?^N5<x%*{U^
zb&N;9i>qH|G~RUD1+eg!0bG&GFR_^39G9NWn!o|LUNv`*WViWqROB;p?dvcS=aQ|I
z*{O$+V`%CWM@IV=xFY(#(iqIrk+flEc;zs|@l!Wf$i~vfnC#4UnJS!r&KvxEdmxF1
z;vFqFCk`aK(Qkrv*lY8g`^_b|8g#AHJpCW(89g~`zVnCo<t>|0*&Br?JSwfkIo1O+
zi%X9mQ5#HuAb#vA9rIs_-1v7Q7u7Axd2!<vB>ur?O%0np(<kbWOs^cW+}NM<pOez&
zU5YdbcB~qKgrZ`ab%XIHrWMfiE2iB7U;KlUHDjf>20En~$roy-psgNOFl}JulGjT*
z$GY?bWoPEjk19FdAH3{lbV+1wnJK*BZ}#$CsvcXuN1eLC(SqaDi@+Jnh@)d&M1$|W
zSl5k;*$b4GIEAJkI9s~9nj0F<XDhdXL}gxnM!7y{DifpCL+NCqe}F`Z_QEMpbrl%*
z-0?gCqBxyuDKwWsbKrc0QEc7M!bS1>TIKe)g@j^@iZ}`i3cdpTn4HV}7Axx)9nS-w
zgeQI=WU2~0Z15fJ(H^pvR$BTM$et~0P8{i^BcFnt6ho6gK-K{OI$!UDvwxrzyMCRB
zsEIz#y&eFet<BKD9t7V6<>j>oR<-EAYVgHFa|8&w?dm=x{WFSUfixBm6u_Qj=bZHi
z{%*`_o9!QptVCahA{?okC4`4$uGLXq4`SFP2Ro3GL!qh^R?9|ZlD|nZ5lu;EF<R?6
zk#7ZbPw~08-lQu%Odx6xxNF)_C(5+W$IFZ6j|R|Y6ZF-S@>uJJ&H~oo3(&2g9+qxh
z0po$r^Y%F5n)(&MXn}b5M%>x4ALN07x-3_M6_9oy7d+P%&JKGOd+#QKdlC2IdGCvO
zuK>1Aw90QYbAeta8VyR;*Vli+a7$(IY`~af6U0K$x?RepCu3!H57dj^J^M)fB|7?3
zZm!jV2LR8WK7E?!akPtlWIj^#9t6&M014r^JoGv#NiStqsd?;cv7ol}^oaQ~+AHF?
zMndgma5trNhH?R+m6)6{p`Qxyus*w8y7sx?Wc!M$=M^>yhSlJ2jn=zdTlmlaxf4o_
zD*_qj+k1Hr<fxsx(@#GPzKiudG{GQd)>NH4;pOE8xruWmBqXkTInFy{e$<{njt{o3
zT)8q0VCUq#ylY84&mA4PKn>zP3>cS;<6UKCr7039Y%#(?<?`z~?4(bSuhVB^h4KoQ
z)kHfW?$ECW<YP!kh%N^i<auY{o8!r|6Y6iSMTPG^efkC3SB;&W0g!=*B%ObcF+kJb
ze*H>n*b%8@@_QNb^h4Di{7f1p`fC$afLa0~F_23_E9n8KY_s+}ejmbn$*eJJb;A^!
zZ4MwCjd)GDW*7hm`uWA&AT_7+a@&oU(rh85p^urlrnE-tZ)ab4wHcaA6t0^^Ewz2g
zA3H!lX&`+A2hX@*>#xe^HIYZkugVL0Cf@jjoQt-o9gKaGtZDDl<X_J9-Rs=>Zqwo{
z&9Wet>U~W5v=8$Ip2&$S20;7Du(7&9s0E~7uwYeqoY!w%y?S+SudKX$bAL7v(s9m<
z8EN~wqgIR^r%dcneX9nnk3`c5V1Y35K>h$bn{;_)1t|7$09=&0#cOC}wACxg0`c{k
zXF0S&c*bRCGXYJ~RL$9mm5+ylg2MiCVOOi22{L||>*xe&nptQA*x%pJb6ioMnVGqK
z^=b|${{L3STjlS{^fW|p3BI|lcuC`@?)Urq)%MfH8=iSQxXL>Ur&X?Q(bS~!A1c{p
z8p7;<^!zMw5-YNFsKG_!V8F7!^uUTpVOJc_dKjM1A$l6RXjH*<0f=D`NIqlNLd8*i
z;S`v-FxNpgPNfv2M381<4mP{r?buD@nd5d{f*4sEsGO16nvL(|O2mw9hx1gzo|RXP
z<O|{HVmvhMONT#<G#@V~@`6(Li+hb>8{6BzKpzWB7fgXaabbCL=KZfH--I0&3F#E^
zPH;L1$;eE1zW&wU?mv{FlBus#RkfChC?0LOsl9?Bq~y+*i#ftyAA&(TdSl|#YsIo_
z;rNeg#QeA|H~+H~?<9(HQi7Q`x5F;ADIKQsm2#av4G^RD2mu5ZyVL5}X5s@7>s4@B
z@~Y{gV~TpoZ#MAY^kfZDxjAV3B1c)kVev(AGqMU*p-@yHId=d6X3)ZUh0p?qmi4u)
zJ}?Bp(Nb03Fz~;eV~EzUXvJ=WB3zbZd)~_i=d+VN=<|?EYWll-4N06E8X7XqL(DUL
z?~$Ql6e>#s4=CtmNj7tTiF{?O9KE@`wC?vlSFn;An<d6KxQxfGqUU8dB8?Z9_b==d
zwe(EEai=8|VR70zPecxte13YS-A?r-8Wnu7>2=V_1UBYNWaNjyKzxiz<V+9)gvhdr
zWNmHP;t6fqu*>>0Os+rV=dWO}+Jn^xQ&qX$D~YD|2a|LzzTed4^kCj)+J~fdD@$5k
zo$Qeh7Sd-5V!G!ZfimScO2v{|4Up6QH>KW02tph2=r@LVQ%aXiaB>~9G;C0xs>rmL
zE>o5KN-+qw-xO(}{R_ka$04}(TuRl;JwtWkKh|7-K3%}O{F<)xSUW`lD)+b;V6n08
z1>bHPVSmB7)!*<$P1O=WkYA;cRI(ao&Ggd~N$Hh~u-t0p_V)n|Ou=jW)(3~A7lbgO
z3N8)cXdbs+^O(-Awl<vo%x(=51D}C{L~|v$;20XMK7AdOaJ!fnUrSC&DedlF*<S-&
z{+;h`b90Jd2$iXxb|ABphO^wl&+X=!@*U^@B@&|HG$A&-jy_4e*P^VnuMw<!7YK=k
zkGiR{l(|L%Oceu-{f`#%hy*%XyQOTDGVg2=^pe-=e9Ysd$ySyGuqO2W%>DRs*B#J=
z0B@UcnUCRHnN0HAEQKDjlFsJJ1-iL!Ed{4lO2Z`<<QQ-bSrvShZdoSA1Np&$_9vGi
zfon`yUjJ(5>>Z{U-luggVY#=^Wruc7wx98t7nKZ1D*ShX<ylPqBQ@g4npmtnF5z@F
zR9?%=n8$6iSg>XbCtlIGVJo(Fv89k1`q}<~XoKHX$~&M`2ofIHUH~vv;q`2McOBz_
zR(Z&vwwr%^eEfHbP|$EVk6a@lP)yrs@MARlnx$6g3qW(wYYtc6y9u0b@zFYP3#RN|
z1#|}JS!2i%TV;OqeK#DE{0xL%-;uYb=*>S2`IAtgadI|Q5g9{Puhm1c@SM+O!t~%@
z!M6RUUi~9=-y$XAXWG`ppLSH|7xbAc5#kpNmdFZqIwvBn!NzFz^P$y*0Ni7jkCiA+
z_(7}|9y;D5l;yZ;nJq2Q6U^<A5E0oNZkxc*JK*A9jA?@|gEr`J7nhQHnyW&b&raq~
zLqoF;DiwWSHI&M#cLp>twpYia&~j~W-e5jjz;~an$C?0b8*C;DSR?24x=ZK&{7gge
zn**uA)V*Sh`2WGRdNG^E@6ZqHW+fP$#+`fgWkvKVjEoPy%8}jW=55<l{m%|v{%CA&
zfp&J?xmGDRxl<meUOi=buI84+CrV&0<@1c>k(K^XyPJvdt>lfM9MR#d9w0O(^nwnY
znW-s6Ze^cD$NdX9Yv?Y_)ta+Y(`Ji_%Cd<vw3bvWW9|z##M;`rxbS1<<{EMq1)AbQ
zN6YN1pm)m996}V(_CT%R?a+`aK<33@!5IAU@$i_|)HGG_t`}zfz|{sz#2pg!=9NK1
z9d-?-q#ofsi5ivUze|A?g($~Lsk45B4OIX7f%E?52i5WI-TgO1AJg3`Cgn=5Rip>t
zO!_A}G+_6HnzASPAl$~$F-PtgL}N_-XO$R;(2~(~biOy)4XuAlb3={F+QudUYVOc9
z7z~gEcKsipUFSDJytTmd#1-vf>gwJi^S=lW*%y!eJJOpX2?VtDEiD<H8-1pnK*2Q)
zHz68*3K`9$9%JLt5~e`GT~c6fNbh5N4DFqO5K_+7?f}6NYg=29%zoQZ3sk=Qfh}|~
z`4)ooXJ_W^S6rc@QFWO=7h2P729r}|@1n0jNnV#S2J#5LHt*2y)~3_yx#J;1wMbg2
z^SyxR5UFh+S{|2*o<t8xzA<87m3XU_$LJTq@+oba&o?AxY8y8D3ENFYS&%C0`1Xf~
z^2b=j|G+q`8R6dd!(yA6J2L*sUizLbX9VOx<<>m-?Ee6#rTzh?g~N4na*f#&Xtb11
z678=bu^7H%(2o{YL6fv0F|!%{rXV)o25_kCFmGN~Gq=h}5pN=AX*vSf4E$(LGVrj8
zT#>$#I2cfmoWOba5Pwp3m4Pbtq8kmFF|skfBnBkw&Ga>S$guMEGbNI*+$Tn~*%g;q
z?GJe{4>gafV*i^;De-Ux&qEmMAAT2}G*8~=%ex4HCU2aGw)$1a&VbVWa)RzOf*3y^
zf}vdY^arYOPu|fD+`^g|8`DHTD}eQq->HN#yyTjcIV~;OzD3VRpzxt?aH8^2_{ZBq
z)O!O=A#F9LHXGIgjA=}Fgat2|_do2$ajkeP$z<IBIl-H~EN=<>z1YmbVVvE9pv~p$
zXGnET3yJf83q9jHH`41}9qjwlNnxD$9KnpM{*KcpsEdQB@8*nUu67mhf!;UW(eXQk
zMvw?ry^FAqydWfs1z=*HVF#I68u~%MNLIMcl^R!2bsThHQ!Resmbv|;y6{TixaXS`
z{7$706;IS_2dCF1F1=fSZHeSSe*e&|_E*We3-{OS7nr#yR7?EV(jEYoD>5$azk!Lo
zGwfKznuy_d+JsNTe5hNCvAa6n2CZOiAImVHm`@EC1xWiQ^_!|yI<vxE{Staw2V}%x
z`=lqI>0LM0qMSy(7>ZLm!$1t{!#F6$V*aSK-N9YhK*A3;4rE#X;69|qFueE#r()4b
zE>pLZ+ufd^Xy8W02~U*KV9j3G|HeV?4Om|2!C;z<<+jrHsb8JMx^MwZ8MPb`u)=i&
zQNX{CeY*7f2c^}iqMG2IaK3dGRSgoL->}T{ur3`%)i{yj+w@qPZlI<B`4n_~2MU*v
zNy^EtI$oZR>6|iesLn01UKlDXpZ~810<E0;atCt_?zC3alP6CyDeQJcObs&oh=_<b
z0FtE6PYLHFQ34Q?4Yg@&P)cs+%Pumz_bq`)B;i~bZM~#Xd^cnl1L^ejbIAg)qZmH`
z;-~xjIVwv;>x>3KgZU*G=5AIFCwu>*K)^kz0q*G|2}_t2q>{p|GIJJXjoBhvYxXub
z8EQ__;yh={=3@&TCE$oY4YhPzwFK<-@cwTUX-)NIv*2lB*~?BRnf@Z`S13AO3$xwQ
z2pkb@h*=YQzF&pvBRkudbqn=J4!yr4?Mg0pD_^?YnLiZ(K~}mPCc*ZPf6WDYMrsAF
zuLoON-X|s|N}!Go4s75K!GP}6){iwRB%Tm4nyuKwj(bP8^Y!6puPFY_qvJT9ou%8V
zaf$9HF%r)pS4+BCuPQKl&__0?&v6|!dtoR&J{Rkmnm_hM=X%+eMr4IYJn+bhQ~3WL
zv0*c8HIQ>+V`IaB&NI=>e#~pc#LC)JRkqNi@fwwvm*=iiJhDiM9_2fA7p#(B6ZmPL
z0T{HrOQ0E0{s?A)Vd)vW?^BY}VDS>rAlY{9^;rTRY_V^3U#G}i6VsuO37L#g=~R(-
zCPw?;Xyah>q|o&>3OPFk4OM)vPoHqX7dux9+U5rUvN{L0=@J3G%!R4->4v$Z-G!gc
z|1~u@{9axW-P=HQdmJ+nl8~I&wX(8O&H|^?XH>u(T^0bW%-^?1FJ?%8T04l^6}caQ
zfu9?NmLRKxT_ohbBARux=Jb$J@B63wH7C&VhH`{XDtUTZtfpj9otfh}qvrw?vW6mL
zbV(Vzls@UcF~PQ2I6fEKyopF9^di;FW!irC=G#GdEZ)1Z^~>m&dTj9j4vg8Gej_w`
zFy8$LWSR9rT@eGKto6mr+)aq)6y*>(IZw9;(6EA=*YmIa=x4hr9X;awIZ1f+_t*qG
z=?)Q&4WQudDad4MS{gR6&4-_=eOL<ph$2sa@%aTVU2Qw(07hxnc8SwYwj0Z2Vl?Ki
zM^)(08mc>-zZFuBG<=I_w*$1dV|gy{i)~>y|6A|dk+E=+S67i$CFX*Q>j3IQgV$n_
zQ<u%Htu`y@8({+xzuRQ~K7!aj*AfF1I%x_Ve}~h=Nyu4Rm%I+JoHDr7B>J5B4hxrG
z1EZtuFF)ZwJAsdJ<2b3v-5513Su8I@tsbhb^c&9w8h<|#qu=sg>B0nYi)KDN;R0Kd
zsJkliAzJasShobu3Ps4EO8`|T(Pyag{8GHKd}7Ovs@cSc<dub|=l_lP6O35SqecYU
zGV}rkzXQXTxU_Uia`FWo9i74>a>!`%0u)ql2PZC(@qzpYOZ?*3RF=A1sQY2Ve8Bi<
zVsHbHk1>GFiCAx8hx##x^vn!W{?o;|RAEXh3YYy&JC1D$M!@`3Uw!Dn^okjU>3{Wn
z9>{qc>{`F&{$zz^L3zq`jYkE8k@-uFpf|P7e#kF&lADK!Ug*D?x$!Hji^2na_b=Ow
z8-E5!7x)xR8m)7kBJaM3JJZA}b-90n_BucH24M{qWT6s?VmfZ17IkhQXVv6$3;^iE
zW_Q7Y@p9y^=Csnz?dXT^#^a79ro%W&f0C5*46=CjT=>}KeZ|tnmWDyPBYO)~br!5J
z=1}M}$X6?!qlEwB+zRFN+|`F;<P-h|d-OF^Q>#xQEx8z;^xuxen@W2k9rO6ae_^=q
z<~#CGq@P614W!zfq5t0nzR2yNdX%0EMp;>gk!e-tiq|Exd#l<VZGb<1lt(~<-(0v;
z)sGjOX-u`OML~L8PipXeIg0hOy5?StbZy!dXcx$>4ck6$jwkbLUg2q3mGwL)=D+Eh
zodXz;s_~aHBzcO3w3}kgQ~|ke0V3%513j?FXn3|~2VDqSN*?`1_TM;!g(shCt8<Ml
zUeP;QL={$%O)Ov7H3cY*sPZe<2AcbkG1~aJcg)R&OBFD?V&A5ZjW{u|)sOM0dfbF|
zLppc1?C+;tblzV&>^$YFT@nN@@VwwJk%)<&&cJvWF?6F08<9#qTw(b;1~3sVs#Sk0
zW_OY%YO7N;0F@**I$-g8b<$<DPSBHd=i*iOt%`m-!j2G$CUnVoU-10YhpXD}-Z<$&
z!!zlr@^fjH*Ztws(C<h~NkYv{=Xt_XiD_nOMp$!1Xp%23f%ERQ_U}2D3j896shI>G
z<`{&>N*mp-@OP8@uSOc*$@sZ|kyjkrzu<EX@9wl5#j|0o)>o$4MZYhVk6)^;db%B!
zgd69Q``|h4;!{1dhVe6xuMkHLaUpmI{A{{n7kP|pn`I(ydp=7ebe8n_wEV1*ac_2J
zR@?0OD;nH7>d}6*jv0b4413P#%2ULO_XDCDAMyI%5FkSe#Q~7e2(yAty@V_5*F`vt
z!UJNJxLgOcA7si)x~&h|&WSMi3?J~64%+oRu$fiJ_~SX5{dQv5PfvK2&b3Yc#?_{T
z8aY*=#-zXH?j8$s;p6g)eZ1Q#dMUYURF+|LLv55olKV?%AFuth|L@FVpJ7dumhwm;
zOFxYslD49fl9J>zGgmQRa*#pGGBk7-9G6_bE2<h403H@2peu*hiwgz51`28d`jb-w
zyE1WNnPLU))K$@6g+fw&G^Rb;r7nxl=Uf}KwKOeZvl|?(A4CO?>umgQvYI5w1BP|q
zw{JsGD(Mo?RJTNA1$-l6$#A#zv;C0}b}z^mS+XHlYWgbjRWhZ%uH(ujZu*b)q!QZ_
zAF(7gd_zvp9{$-5Yx<Y_yWC@(^ZwZYpVMkP1b&YpZc<FU^VT>ux=NBotLt8di>_A4
zck7M(bzA4#S&hTZc-siwkNmucjJX#7_CV>XDu?$Tz|TX~Ie-ffW`B;yahVh(g<3rz
z*a-*-aCUK-o|$<iP6k&LW1P%s>-QGT(ze0lM=q{BI`5MF*k!bH2E=79;~dy;(=%~h
zk@;PVA-i=xru|$k|InmF#J|A!t+6pN%6K2)hvrM|DLfpULK<+bRi(SVu5pxbk?tCZ
zRG$Me!1WAe{#>t~r+#%W(*Z7dmE~=yJq3kmpYQo~H(vv+iAnYl=O-0v_u2oqhw(ZV
znaI#Qs|pG?#Kpx2t}?y7W>{#~d&AYp>fXaze8W?B2Cn+erx@jdHT@rkkt;7FWE$#i
zwyk|*Ze9zw(*CT+Yt2LMgs_a@|9Lfj;lHX7<g7_Q8m=n?nNHWpZJ%rP43=(|9P&$h
zdiz3G$}1kDT#Hm6o;%~mF`-3|ZdALMT~m{PIl2hf{ZPi!!wu`0`>N0%^%}KBHNLOp
z_-I#9RL^Wj82>Hu_v9JhpiBDlHTa09$nCE96;^$N<ah`B?`_R--K*ypSaoSSbBHyu
z=o>lX<tOThh>3$@V*_N;&c{PG%-=>8>}Bov<i{_$g0+V#;dN20>!z1dx;4nXuMq2a
zF)cuzjdd~RJziYawVl4DIPdom2kdLfudlH~d21cX%l2sQd*#eTrttB~&Eb!EJs#vj
zXj};x1hTI!`ZHbGnX1))h=Ojs`bJS`vXeG6G;-ahi*VI#n@CfzlAKrac2)zdc<Waw
zIy998h^yGx+b3I$mR!KXx=u*=;^o-c>7Fno#G4?!(!Io_d7^RowrLM!MJ@{PDySh3
zR#Mp4Td;m)OdH9*U_NHV9-y0r3Lr=|fzg<7zJen!26-R12?P@$7uF0Usj*%iCb}F-
zRo)X9mK1eI#CbFxJz5L+6uWgfJsDN>cR5R|iHLr)JN<V4-B})({FbzGbW&<<!|vw6
z%+>EzcD@xlf<G)|J!P%W{2n6)4ZcLw>+K#1(L#q}ifNW5@7`So;jB^!s0(=<6=S=0
z(jObH6B7}k%K|_oqO>#w`4musJYNeb2uTRa$nMbIetVA_V$I5tgS?FntDZCGo6vW@
znsaRiGqyy|{7ncNIyH}jvA!}Ec|<JTl;rQ8$e5$0&MiiT9PB=WFeHRvY4^SbiIA*U
zl;Ew}bDHM2H1OxKb&=EpJoFtrsuOi9`{v(=(&&(z{xD7@jzVfI<2(<_SnTZm#|`|O
z@+Nn@RUXM=x=o7Zx*6K7H>)+S97L+5;ZCRQc_z+#@kics+gh+M-(;MRYRHe;bk7mP
z+mx}aN8VnUYZ`}4*w%M)tB20(x}QY#Qu3kmh-q@7+M|Nd+2@-7o#P3O%Va~LG1S@f
zjU!bd%Wc3p9s1n`>yT~g{pc0)QR?E<I%d~v^`_>O%cMsWdfN~UP*NYaA9(bvOPZ$G
zF!)Etzt$>SDeEc1HXGc3{~zMvr2&MtNZ>>ydM3`OnuwQO$DamLev9*6wPY%D*i<RH
z{HJANEq1Od{H$&)cW-(!$L_8H)7v>{7!m}AA+y08EL~9bG#@U&`pXxTQBXSyy|bGu
zAlAYTJSJvoG?2Zp-LNt0)4;bqVk9WFzU3P-^wc(}Ft2}P=J))w3L%6N?a}s`V809l
zpoNR^@Y$ERFKq|HU7tO8z3a2w0k~{qHvR^>7HzEUOcy!+vj=&R=PNu7_kM@+xRe+S
zu`ii2t!ihtbsOFLAu+ShQh30zX)>W%lyZ)}xEf9ZXQm$a+<|6bmJsF2N0#b)ld-~x
zkf}q5pS7bSa*FJ(@>jMLoE`1cO9!uB2`Lb3i^clDhV{bw?mM^X`nWsqPy{>5(L8c5
z-X6(cr)N-nrWtR3?WVz6m83?w(Fjc!bpnw&jzw@#Dlil^G<?<^bny&MHQ8E-SW~li
zSsKXBVZMaX*=D?}yrpf?zq9MP$bLwsCGhGqAO6y_QskrQ?Re$SBU%KRvG|$^l5v%f
z1y`w)OH}M>$FMLk9u>brjqY@W%+rdKOzysR_rB9ADB8Zm!pL(+J!<ecgK3GlqP)?r
zwY4>9*j(A#dYyMK{SY+E^tqxOCO!6I98Sty2TDwSiW(NGnV6VBQs^_Z@}~Umh-x$E
zT7tqH@_?`NRsJ7ud?7g?TPM19E#7jx++?QleQR6WPpGyTa7A%KT=Z%-oNF&)3u-Q!
zz%-#8OeakUe+mhiCDf!R<Bn%o%B8gui%|}!jTT#us{?a@DdZ4E11K_pvxkJAOITP3
zhlIHgJ|;bkNrU0mtw++496fM=kxVEJg!1{NSs!3euoLXU2`AFx3Mnx8h}sjT1=g1G
zataKp_DG%hB=TLx9pj*aAogD*H{w4ZPMh}8N&whtP+A|w=-GU|X^k&&`|S&tqpyU@
zH6>(D@0A~|=ni(A$AEw7gnoM#{RveXgadg=(VxwC{)Z(XL!8QUMD(0*N-sM1t7h*X
z3o#uzl=Imyt<5k<lE4(<WST*zxO%Z^=;7{)#*7WL@y9}l#QCc6U15%}O4kTjqq+Hc
zGpPEK@!I}UL#bseiz^r!rb5{*6<Sz`#!DdYn5|Qz4VBLf&`3^(?C%ZX7@d(ZpRLXa
zgk1~>f$l|0+PZ&0T7qa@f$xYDDz|Kb_r<xQq+#iMlEs5N%TIwWA(GADL#k9nB2eY@
z1JnDTl4fe8a%jU3pL)P3&Lb7t7^R>NKgixCpg7|mTRV4d4MzWw`Dj0c#S3`Q?QA$P
z1ms(%pE5-W=<s_62GcU4hLDvBA-ZQez1kksWy)VH$dsI8$F|K+b`~%#q6-yn(tq5W
z46Tb7W$ozSdd=9k*elZ%x%qcVzxnPkB421Y7uXsz_ALGRbm+b7GL0*77QTp4^(+q;
za=L8(x^aGDd)xj5ir0{{PypTEvSw;t+hph*%mIe1OxS5g2-EjUh{S>7jv2jipx`GD
z#TI?kXffo|w4BBrTp_*1=w7p0u_h$A-Tef}X`&bVGLoQJ+7ZpG2%Y_?SdU#b^slfX
z@x0}1RkafaSzemL;VdYqnXXUOp+Eca<JVWI1kz1@rV=gP>&{YR@A%64BwW+>J!0Db
zjW($#9WoBkUZe`UxdM`*nE+FW?Uxt`E)|%<!^2a8nhs5~&N*zAKVbvk>R#m&YFYLs
zzt3ztH}!a#t0GJ`BG-VrGrYqzSV~)APcEaZhR{5S4JWyQmq9fyH3o;vc>WFEqN9c|
zp6jXMP3bu|-qD?A(GIuYc5_Iwf%rvn3(OR*#|lirAx6P7Tgf{nd^b?C>{O9isU*d{
z-S6@5<;gss8*@SL<VQYD)Z_Hx-^P<O6=(M^4k+tp(@BW)-ICM}XczL#xtA;Dy0b_N
z0Wo@YfEz;bQSGorKXkORh`Vn0h;@zLD5eS5a*O~Zh4QlEZZi@{JqV70K3hZ)Xe0T&
zJ>IHDL5WGd%r-?Gm9w8}hnjHBgiWWgZ3S;S?l7oz2*uf*4(>@~sGpS|E@rlID66Zh
z_k5UyBxpL6qn@E6r09x`dqiL_vN%paq^`VN4*4WJDgjl<vSh+B$^?XlDgm9;y|r;9
z^M>YxHwvoq&%)b!ZKvCbYCtH%)`1s8;B9v8H{L+cWZJc^;0}%67`zlUtrQgv$vv(x
z-~M1<?jUXBh0@dMRE0d$fJfm|o$(#Z_lzErTZ4U*oU;in0oHq)e7)hUy+<o4XJ&WC
zRdvdf61Ja)Y)g|Yq$n%s2~!@sg{LA{J*3~e1*!}@L9Z@GdVJBZ3*z78U|?H#0X$~y
zI8CFY8jw-XqX`{~+l6MJFq(0Xy>TKkWUSPNwnR>=a;V438OTeGd@6jvN;oJ_sA$7U
zG=D#c4-*b5a$1{U3*2d@4(M3RsXkmRtk*KKUz^ZUXbCJ!vv#RP_<`9hvCW9fl(_?H
z$1m&;!Za*x(9>^xv%y|`g4$8gqF(Uc4e%PARNbsIWLY?TZX$8%KHnq1!Ihr9K-ti7
zL*X@nQBUnmAFIwV_sZlm(2@FN+pc!QJDKWb@2{&0XCo4UvOP|Bg<i;*8aA!nL(j8z
z?)W^#QQEJ0B$Z!K*)u>7x5f*nhF)^|w+Rq2C%~gaw~(6-WJM$<B{6b1LU~rA(h~%-
zf|4bf%aZ%-a!Y4EV0YXX-&q=D**FA=MR5h7?Tly!U~n+axOD^GT#g=GD6Y^Q3%3Y8
z*2Rmh=3H>+kYtBUKvA`<Ig3G)YW?a>gzaf{b#+iOkYzG5F?F!YUy*DE&S<5+hPw|Q
zNQG;=W}r{mx!niDiJUsjL2oKkP$Lo@$8C<U-Ue=ql4r!`I-Zn?FV$$db~dmR{N%7&
zo)4JXEhAQlTQBICouw5dVA{b|Z2j`|WV4ssfFB<{dF#i00rtyXL5{X4MPybLXEVfQ
z@w2=83DLC|yVZF)y{XdVAU7?GoSmR<+&EQxEC2;$@m}C1(VEWhFtmF=NKa2Mc}Q=~
zB>wbilZq(0KP}4yg9+!x0ih!fkKOs=0_TO4@IDYGsoLt~AE|UHgevg|E?dU2<CS*l
zDFOes^Ao`11-r;8C>aR}A$%q#wgi3&kJ)^MQl_#=U%I?evHm_-yw<Aw<QQE>sNtRY
zx4Z$!Clls=#`&d3d8qBI+1zkR`i@}QitP@vaG68#1770h#mSG7XLU+<ypz)sQ{;k6
zI&?A#H1I>$+|iSMlwvZOT|=|*=XYET8NKJA0+DJ#|7}O)kq_o`qno%F1%W0YC>hu{
zC4-h7-FiMue8|!iv|A1$yjYQuL2EALsjfDXr+vBDv_I%YwX5UWM7GWB_m;sy6~F;h
z0n<=OTpUYO2VKRu=fJ6MZ0p>)++TkZ<cC~u-@h;3;D;~FIJMHB87_3Z-jIkOn=SNY
zfSoc3!9?Oe@Po0tQV}4|bUt~)KvVw6DjPFJXu6dzG$96OMnZr_xH(aJGS@cidsX(!
zrr#Uf5-85kG}&jXW0^kZ&@&^EsrXurjeB)4Kf8u;XMYzvu+U=qsd#VGiC^irB3nlD
zmxluyR^mZ$t0q7aIUPjhzd>=lZ63%>yb&4`uwyAyh1Sy|;awWnp?30x<^(V|20}(~
z9S0>tnRF3EcvNEe9Dc2=Xc0P-<mm3|hu&D1=!FWP{r>vPahqn!6e@PWVu0T(3&<73
zAoN`PhtZRH(PkgXB_8D#<-QaLidm)+&6$o%ry~h(#@sX9(g|m@He3#o{vh16`TV$=
zc^33s;x}EQ+R1G96bnWz4v#9t3{atGv%7xWoj+l6_Y<`hcdBM~CDIGb7yC2Qpd7LY
z=#h~^qd+u^u6QqjtbqLeEjW_0LEBKFVq`Dek7YmFTN{KzHKW1mX8G}E8?89{fx%kH
zr`ln!1tTr`3>&*n6E4l1O=a>aSd~Gr{qA}|CJ^lazCM4c=v5OolUDf)ey7zJj*cY|
zcqM`NsA6~p<6G%Ir5FVXAMbGG>Gr_}|J67*2#Y%jOoJl#8QEE-)za9a$#KxLnqpI#
z6p+{<NA+#2AjT_M^lioW=;!_zGDR&Ss>#k>dTT%Q9j(ZF>bdaXR_KPrF6{|{Ww|=k
z9XTN+QhPwWvofjz6A2%;4DT?6-sx!9y;ZIITgp$3jnlO~4-CY?yU2pEDHe%G^_%v?
z^}P+H>*)Ua1d!GWq6l07Zb5^>VH$Cue(Oa$Naj!YWq&oYl^_M|Q5~kh@)pTu;d(pu
z{M2K%U<z{9Z^|tl6KxM_?;YF>jrgp*DeyimH+SWiK+uck2bAr~qOT$`O?CFQl>TnO
zri1;%rQ!Em+kN9CCZqN5f$}wj%1P?WFiY}w@UkQ2r>IP#pT3O3EQ(M3gp)pA!mYA?
z7Cr+dfG`>l<anbwD4Bn)QW1RSP_9W89gmdG7=AODI<%1UEtNn}Z!(nkGLpk2sIrp5
zA=@}YU~kNh&wW1+kW=5FG#l#N4jb_U^WI{=BqSvIfdPdOFcr%~Ey46=a504{ulY>%
z=C(p)R;1r@1N?DWsOk26xhtm0E8vBgrl+@56s}xd4l;MVSt+r)CC3riQD#CTu6#7C
zhlzfA-`-x7R$mA`Lp?9PLuPt5IcdB*bkE}OfHMVL?xoV>F<eSbVX+z+()gw`^3>x>
zsd4AYt@7FF!Lb(y!@x${h0Ow`glUNwmV5V}1Iy{)+baZp-;@}E4&31n()}2~U*%ku
zW9i_yh#wjSvpa2WZuSDeGZBQd7XTY44{Kq-LDpYtqc;vVS-rwBdl72$_LX(I-Eqii
z2Fx=P9Bp`&0^le)-1n^kzo3RMv>#8t=5JQ2kF^F5-=U2OWemxYCQCC2Osb)<lsMOo
zVV#Q9^4;bu+e!~HwkZ-$Pyd8@zbm!&Fd?P$uDC>d=01hb4N89Vi_fBGb8FFSfq^of
zDE~<3UL~VbxtAPvdRRPDjcD6U3bO?3<Q>1ctMoZ4A$mKwV3d;4YmwVBV_J{yUddXq
zWzMlVaH276<-<kFuu}q9)esY0YHDgh1eoDnVGn#9S1xT)hdZ}TVDI2})p9|bK8r#A
zVA-l{0vvPLpeFNzGC`G9RXYHX3nvQ+SB8+u6>AL;vPAH-UzIZQlCkR8_dGiUm}qni
z_9M<0HBS_qeeh^e%J&L?&JV5)75#F-&Yk?cIwl}ayzo3ggsd+JQAOhI^!{#QhG)jk
zT7rk0;jUKSiN~$GR5>kr<S%Me_sXs+%eYpG)ZguMo9b{q-m%)L%d${n9?O=1xF`vf
zxzM+|*`(LEx(uR9sQ;1y^HxvI$;rtMP-mpWGaRXML($2`W_`kKrh5t$--R+Ud?~e$
zFt1~r89N{*Moe*=g+&kL$&_UoQkv>rn5;SEtT^Kp-*a=XOi293&`}#m;}aSAusfsm
z=WE)Wm&^?&GW3&(moOWvj9Dc;C`W%SJ66~hu>YD}VoBO~V=g(zdV_#~)1;RMk6K6*
z$V9_XC=_rZ7sCM<cO10{N{0BO&6uOS3{ePagmD9iL{V8ejHWrj4e&UFXo?E}Ws0H0
zDL6NMiO+DvgPLC5AMCZo;Cym&Rl4sXodjQqB4KaX=y!V)i*K-)3Yl$u;m+@Rma{EU
z+INVCQo0z2s0BE6B3(2uvUtB*`=gW3iXJzH<;#bpzo@TCm1<x{bs(uK+AhxkDa;lj
z#YTz?k>$@&73~c03bY*hmC!ExxE7tn^ov@4U<F8Qi}Tox5E1EI6}w4DXs^uj3GQot
zc1o|sAG+NJvg1>w0tfL$!WHiG^lHi66doQI1stnB{q9BN_6}<p=SXf$T@Gp^rz06k
zQQ3R0rrb2vCx#S9iG}^=Az*w<hnUqAf(9~9(-8F{<J&AOEdAe<BteO-2Z)hV0oPdQ
zio&*In9{28^yrzR5_Ao5mV&UZXAT}SFO0+Vn+=M1aqo|VLael_t6x#dCl6bEL&>vY
zI@<ZnpWU&G*5l9S1;$ut9zB_Kdj!j}jjVbuqy;Oe9i|%*KJQu!;Y!Q2T=FH`UY0h<
z{Ohb?;P>R~iFp9V0KJd0lrw=Xi^FME6OWAJ`9*a?WfK{!N{Akq%9q#HbY@sV-FtJ>
zRx0@O=c#skjBoiu_&T%3ZlA+g1k0Ydo_X1`<^J;ixtU_(mm;r`vM*4KR11g>R@xdx
zxXy^~Sf<rkjm<i1McO}db@VG@Ec_ME-9>j2bW1Q#?SrIZ%pceCxg>eEg99EXqrZm+
zL$fg>^{(TQ@cSQSLGDRC_o6y}x4*#n#_JvaSf$!IYO%yL9o%w-i7xt-I9Z9Fa|-Op
z{XUJ>JPAL$WXg!3ab_fGSi|GHW~LCn){1JVhA5@qww!Si-{EwNY-qWVxOyC2|IU|@
zi{dZoT3*?`ZHxLzc{?}Raj<bQQe`znHeSa-@WS6sg)t|pVuGP05w3CgZ6}`j&}?yS
zWOqwFWmS-NrGszjtJg@KDSO^WS3&0#kwrfh<Jj-xiJZ1|jL3&0%LuMctrYMSi8NTv
zH@D=sYaJJPXRtZ{^kvoWUHSX&@dek`G1G028!#Ofg1P_n()%v<)dfokx=IYU+}y;w
zBKN$?y5B^;Mw8J|Qjf!hakul_hblWSwi^mWHAm^avx@qgoQy@DO#N1aPa)qxGIsax
z+j_N5BL!Z2`OQ>}i^WG)ppRj|Cldr9MoN{?3$;IQ;(qIYpimt7O!@G|<Le2VSGH6h
zd%mU-JF|AY47oA*rxeqMIRWW|F52en>whmRaCC=5szZqX{u%>AE`_)F4L2)mkCMZ`
zpJ24MWO#AY&@}${69j%7l)O|?UV&L_*PZy^Q^mj#AUyZiAA{dI-^0}Y_u6&23zh%=
zJC$(R`QIPcKKZ}T_VT+k_20-aF!1MARxvT+Me|2PL!&d=8($1tIART4k$;q}d$SUE
zH3#?aJ0Z@u;O$E>ZtPJFmM=&aEp<st2!C-Ouz2n8rauv(cc|F8(Qh`{C{6U=^kkpb
z;$VdS2nx-d)QL*)w^MGlEacBIpP_%Y&%;cw7UJ?yi`<ZT&^My3e|1q+i*uFhz~^Fl
zuJPXwn_`7Q#|I@-@~2pdNh#E?>M0=cD0@W@6J5&tW81msw^EwYlC1(vM!U=$+_xT&
zAGX}<5BN48TmN^eUd_6}oEYgBTyFA4dak(zSfdp4w$+Rl!Va4nw8XJl#RtVC;+(=<
zeecgbPpqysN-PNKWpSVW`_Z*Yu!A1a?k|+F{6E^h0<6lQTNgzP5Ca6HRZ2lXx<Nqc
z?(UF{fOPX2gtUNkOXsGxn=S>Uk*-a5cgLNr|NozJ?>+aP^W5iL9^mnjZ_hWgX3bjf
zde@qn$TA_l6+-%88!Df#(-0>9h1hbF8ahf93j4+r>iJ}3JyQSqpAT(m0C`UtiO(EZ
zoWfVrN6nT;-n5P>97N!TU-ap~=CVHJIl~TTIwRHS{yy*V=Xpr>9A!J*<r^|lH9tQ~
zgBp?uFI$P2)X^aR^HjDqkV3r=w%snm`500YmvFaO54hV;rGw42EHN%<N@reP`0~$_
z3I(9k?{S(ox$^;LZ)t{^(dilChFAImp|37ApELc#M5!<n9Id^eZZ#@?%Evc`GF8#=
z@~W+p&SqV+duwFMSkz{1ESeYBVOC0@%Tz>|u)kR%CE(R&o0<gQg|J?qNQC#wQSS<@
zUgxZuUF|m3FC`GXOWk65t=liIU~5qXv~`<a9CB0<GJpHAZS^X3s~O_iWS4y^FW=<p
zC7V9ZbHc!UHA~ZI6*DX>tdde!Hid%wTWSh`^8zjpr7Pw`@L;RGr4A6pBX3m~{a%c>
zn<3gVcZ4W+?Bt+2rWoC==F2_osp>VdT65Ir77Cv=&UkG4Ek8;Xp}~(=oul8=dne-;
zjkz)_;wNmLU&da^Z88dLP;k7s2Tq2(<?;a*@*epS)awvXP;6|l(swYAOrmI1h@D0v
zBNC}1MxNVHP2zFH(oBB)=+P4IR5~IqmR3A&>^%UVEq@2KVc8snGrl%HIo=j<PX!nD
zN``}-PUAYa##><sLII+ni3!&D4(r038!85jUue3Y=nDy{Z4JKTso;5?=W<LOoRIL=
zHC~3*HGW(doBsFl#HG2pIZ?q5?(N(8N_C<qM@PTf1MeihbZmC!#wb@<UkD4q8Dpb1
zzm%vI=wmU1RMI0b|7!P2z<VMLou}NaGGSJv(%Y`R8yx|69iGUKM&?YpP3Ml~rmkZD
zHMRrTz&ey`ERwqMFnPF^&3y83{mC-elcYqjGWmD;=MsWd{jRtWxu^Q(lUDDj3UG4s
zlm^Ssmb<CJg3NC5#MIS)1e6TepKMOFJ~joJ+2{&Q3%U~5q_^J4@bNbla#by1H%vBp
z@n$Rj@E6@Lc<;*J*C&O7j?xRyS~EJVuR8M`Xh81CRAsUyU%f=MYT4&g*sYOY&HL($
zxpDzoBFfhKqsvn{)hJyI3>Ec6-vbqcSDzO`j$iTM;6WL$C6%J@wLp#@WH%B^B;tEB
z9CLv8rWGo^UgroO)%!hA(}aluD<e7>vUk3vswHP#2-JhRb#ijDbD$noVs8hyLzA6_
zbw>o@AA{#`cJ1nWsRsj9i4@1X90g}fTm`3Q{?9FyGB|RYLUs>0o-Kq(_dCyVU}uwi
zOhY(-1xDeI9#gEo#Cs<9mFxSIY#KL))_VO5{Jw^Lp{?r7O!=%ZVAUtl>0TCwgtwwH
z{#wJkqQDvwp)ptX#5($wRnmWr^1!wD{J;Og_@8X@Yw-NvFADT*|B?THkG22bT1LO-
zKX)V0f9^;AUCV7_D6Dz`y<z?Ce_LRsp`q2mV$RW+TqSebDqr*;N$zZKhlMSQ21i9l
z)3t8vyzDeZXay$WOwiiI%9?ShUvs7}0w+edM<mxJ--cMrtH?l4U*a+R%bywl$Nuc!
z_2aMX{%a%rzYh@YfA9W(Y5Bj_F!cZE-2c0}j2{1gqN3<0{&Pk8zl#|(@gGYW`v2DQ
zf34v^;|F~7{{ch&cZnG-!~f6_fcO8y5}+UXuk{)I<p0v;9Vh22g<s?1azQrJ0-Rnw
zT5RkG_I^FU!}|tGPT%yT2!psPxJvpTU>zemhx3Y!n_J}uRJmGcG(<;B3;X@;HHf*w
zz>O(|AmnTV!D*_<-#1~rD}do|k=(w%4=5h6r9oXq<*SIODDaguflX+Z#G=vrmR6}*
zaP8>^9|~+S*<eDOHuP75nw8tubu6r9u>U~~)C&WCnGx_VfbTs%I?4vSBp^4iC5uK9
z?Bn`60CTiQ{QVflR5f^Gl02x0qLFl}(j_J)wk`yxx&Qe}P&dA+7vHJoFw)Y}5*+!D
z|9x+%5>HXRh`uyIQVB3NASdeFWz0wz$jZhhUEOJoBs#zB=o#)CHgyJW#JBUHd+2LR
zh8a|vygiQ$+>D4TT-r}s6b^P;P@lA3JXYu<sAU>{>*P8RV~WR3sYmoka;D}L!nyYF
z@_=nZMg>tl)iH{pBG2(;3Zd~FY?k^+Z2^jF7X`dP#EQ;3bh*lafF9g^s;iXDcVx0y
zUgb?sH)*6;+k{e0$t{|juc*4Fa-RpP)|+9s1V{-RdtCQ=?)&O-oqqtHFCbCt0Y*_(
zbz9|IwO#ckvsFV()!h7iG0}a6PSsPy4UnHjb5~%V9niOEfr;2)`Wr7{VCa0)f=?mF
z*3mr!`ziaueq1zpWy;G;P5lax%+K}@K(!ljFeC=DW$KY`4X)c2jitkHIf<*QP2M>_
zrbwxW4Xlt?mAze6BbVpR*h?UL?d?mStCxP=asAuE{u7k0k<ZB5!8bwN$Zd~bzq)=?
zl5)~pZwdkc|I4iO87sdGc|1p7eIW}r&TWV7pm)COd-e_O^$eo2>U;O4KYdsf#WjyM
zK|&)(0@&u~UG=+Hs-v@syACfV0qi>O0{d<A`Yk_2M@Qd2yMw7>Ilze?4LTW%0EDn3
z$Y6mJ&`o0XdyI^X_CU#WG_N&aU_ojdsQgS@IP=pyIQ$O+E@7bK6Q_-koDfT^YR@UL
zV<bG^%+lW*4APX6k!IM2V(5foGdBvfml5t8N|#-UOz(WUO3{vN^XLdMViNxQb+@0f
zWY*rwp3VO8N1rz4r^jiEV$95_aX7pg<>)8wH#Z<TvCK*SwnMX2vtaPs;zuM>(JW%5
zlb|6HJ6}z6X9uuCRG5l#!BG+M+WmdK(UW<W)oTf4uI}#bbC9mpq@HN_Hvi^jX=!O>
zVj`pD&fy^(Aji1Q1$qK*%y%`gWBlwF<6zEm&?$X%fs4^~Zo|Xh#w5HO1QME$L0<L|
zhU6<@m34kQmFtq{@$9{kin0Xcgt-Q#l60l^Wse1gh1MprIQX2Gce0ZE1U@kh`K3`r
zh2DTWI72=a3~DP$NIH53E_GqEQtLOQ85go7Ig0PT+zt;hH?vrtPCK#a_VO%!<8Q{S
zo!XivzLF#yT~+1qXWdO*$<?iQH!CN_$4^^Ka_=(D+j<}O=EsC*<ynq-WL#v1mn;da
z@qH~!KTEB6*7^-WLG<ZCPky5NUjMiGleI%1Xw-Se+sk#`^f=CPAprSocfoRR?vhTD
zWT2q5`QWo^x&F4QAQEX=uqVwJcjTNfp;aSyxB9d;gqCKx4ms(TD*e|iP-JQ`TB0=s
z)b5r!EGs##^h+{VZ(W^cn(~f#mP4{L!ywh!)|+jexW6kq6#a06aEJAwgMB;{rR6x(
zx$;m^yPeZ!Ky(w**t_2xXVOr~9UQw8<`E>;`Oq$Mf}$tu=R2fA(MUegsZRbe7p~fu
zk^Jb|ghFEJDBb#izxbUl7uZz8@~XEZtd2`P!hlPq8plf5`zfyM$o@={3u%#qyuKG!
z9ORi#dJSvE&v1yI^~Fm_|GZeDu25IKHByan#MQO(5_M=&I<{ZOaQ)4@z--p!u{udA
zPuulJ#MA<34SrHxzp|@E{=TmzkCTxEZ2HPHZ70h^fy?Ipu<>qdHz++)K#`4Zb}vR~
z*?zsSXO)XT?D*KCH|1h}LMqeCAZ4Oz4G*>RgW1h`6e6l*{jME6Qn3#So#)SHg*)m7
zE>sriuG%F-sw&F3o5OO+Ruh>>^RB_8(~TT&`akb}>8N0(ww2UsjckV@qE7T>&&s(K
z^z<()G{W~Z4j*Ev7x`^WMJb}is(pv7f8qRwj;3DZ!F;0-jpXIJr=m}bOQE_mF}3G8
z)M7N^()kxnXJm6SxPY@7vccU76;-!WMvvw<5#i~n1>YF>eT93!sQn&I_>z7M!4<Qz
z7npJjS9rgENwK<dH|MoWHgA89?zAn*b*x<XbBS|Le!cjEb#)nskum;1dip*tO{;JU
zYl<u$RfuW{c{bjy-YF`YK(Y#bI|&O<2fOmCzLnA2bpJLEPak#*lM7Wn8eSI^p#AgV
z`Z@qepVHD$DPyVSJ;6Lnq{!P?W|>h|N}y=RPmqy}q$Uz@x)Pp_-`Xk{`j7aF4&b_`
zqthda@RA*z1PX@saAryDt$ka!=kimV+DF6lAnmPI`#KzYu(Y_ir}5`KDQ<whQ3y&7
z)inI@fj2BRBjM_z$i#Wha(Ma)0#*S!2ya+Etq|hqTrR%zK)`AGe|j6uN?&TtbG5uE
z3c;0r?z&B(r;;mU#Wgs&xhK4@In+~<;Zc-48~yZpd8k6H8+Y25<})a)#-Q?|zg}J;
z5pWGpe@`V5NrzwPk4IsL-cOFrpmf@d2yIROpj!n}ne?44CcTlx#chmFd!n_SR9q5x
z_zss;<U&py-oT6MUlOU^{4<VRd@y*L$_jmbj*#rM$pgy%kC=Y+?9IFF7fN!%A1Eph
zU_J>GFY;+!Ru&+`@bn?|0!1Y!xr%Oiq{iON&Oms&xpl?w48&hGU-5ZQ^wQs>)2zn!
zvL-ye3Odns4_9t?yttg<TwN{TQ=y}N)je*fZ2DEhwCR=ja$!5Os*zKO^~E&NJfTLr
zhjnddru!mu&xd|?-2nl$Xh;8?pHhFD&Z3%vuoQCRnU>Rjixw(NStl{Y3Q7mwCFj}X
z!Qzb9(>aKyI5p0ZnGoyrn&ug}bE_6(K#ehi!c`*iTb$+m@7c<;F&^HiFxLtD@Dh86
z^?XVWX!FYJA$V4I@<}&V_Jcp-aA5`FSlk=LJhYf|9D2$_r>@?0B#b=U3RENGfikbg
zujkJy?e4I9j>$F2#20d@gUjWUlNYLws@)JatC?*mhWa6Op}K?cN~CoLf=({?oKAna
zu3gQ-*-on1_-=aU^1iC&z(E8vVy;IKVLAG$E(B7Y78bV7Mh8}aK*Y0gCc;R8?Q)Xr
zU;MgMe!Ch)!ne<MUgVzaH4N3rKNd9Oz`dpTB~;UYWMnitT<yWh!E%qDvRJ4}d~@))
zPL2C(UmuI#W3&5>i~7qyD52i5yrI?gfmKzZ7YewaFipQW@Uuya_p(t-0>2nsej-(K
zTwjB|vukg&{ig%<RA|+T;A0|Bd0gYRvWo>RwWvZC$592NU!gA=4(<h0kI++t#9e~7
zS2{#B7`!t^sQST2@{wZWSz$x^c{z1wnXb@z0a)zIE(=18q*|PBQY@J^3Qr43($#QD
zD~ta@GFSjA<E!R-I9yE6IT-zu&l!)`*ZVaovY^s175m_ju3x~9_Wpf0(<5E<PaD+U
zJ>)-UGV<1G0OxLWa_9(Z{*P|_30G1?g7owPZsRDMXv9=WZ*vRMM#K#ktZuOW8P-dV
zm+0vKJ<58>x8`)OPam>GB?<6`B|M4m%ygTcULhOM8PXZoB|=y*GOGy_MI<E_94=mD
z^R%lk4d*Dgb#+D{e?!?*7*Nyqw>gSH0WamqQMzW4U%y5bCSU5M3C(b>D2RMiVq+zs
zAWP2#xe)^-Q)P)emk-wU$|@OjC*I@#VN?soD|O}N6Eg?+oCvF|b}ukbJW*~nJjPLA
z<qsXX`&ohyT4$H66hC;d)ZkCK6#rV<M4;w_erh#z=3zR&0GrSUcl5yrU@BQ5f}Tb}
z*Bv(IGGTd9Upr)Kw%27mk7niD+Pj&ZP6G<Y_V$wlA55+0wOE(yg-*I)lAfXk_9wV`
zFNLlaUE!Ks&S1`x4%ZsO9<o~3%7G%pU}R*}pVEzEHMN|MvTISb(xiv$L#o|1DG-Y&
ze2DaLlFBCSca3v>@SQ}Np4Vy3PoFXBhO-xG;-h+awTkzj5D-tn^UplUC1i&-Z`@hQ
zW8`zBW>Nf<<f?aBVfcB%V@i)peG_f-hrZatm)?nd^Qfues5aKt#YgE_S^6>b4~e((
zRE);CNbMf14ds(p?XT=@vRf`=L!Ott&_IbHrgQY4bk9??;m)QGl{=PDUg(UjYt`P=
zbMxO=>k(^)5C)%bZmG+lcuB}f(?<9X-}HZMa<kdE3#L3Dp6ps8X07A=ghz9I<5Fl|
z>p6^=B2&JwQif<Clz%nmosd2U_8q0kW?c5~;?6Y)w?s6O+3bLl@_>*%&xYS{sc!1o
zx*z0T;&(p<SZoLx)LRw3uX+w%d-HH#)j=b>(1tjJ$W1?85r@!n`ooWv{iRv0i-~Ip
z>;4|ufj>o&^o{K(ER2_qd#%iM7H!qXx$}!zG|0*9Y4R~b9OrJ6FAm15gj%+Xvap8_
zvH^Bk4u&?(bjv^H-2^JTLDlkf$Vd_HyN_7wA`SGZx+HwVxjbJhu-7@oV9Ne*Z^Q<N
zeSK(>S)oZ)Tm+Wu8{)32SyAD(GGl#maFQ1QJ}}5mts!)~N{-5i_bCmVvZ$ViriG>^
zQMS<^U1dPC*?}|}8LK&=LXp1>dV*HrkV?zvQBy%K-sdXz*5zVe6BBOpt2Pc>bFEVO
zWh<`R?aOHCW<G9+<Cuf_oE&{U9tU_}f8o7giL~1uM430G@9CabWveeY^gEpz)t{N6
zw=yPbs(Wh1{*ZmUM4CO@6@Q(_Nynx;K{v<o56;jDMDwg`s1f_fum~v4oe!bBDK@`p
zG5YLV2t{ykSVqcAxa?&i9({h?Th4geu4BD$_UYbQ?89Bg!-sdBm$R9|?zgqJ_YLPH
zS5*F_U!7`=IN2MTBv>;Z{MmjI#WT)m0)#bPf5A!`>;`RwzOD)VV^UNp(X8zkU@n4a
z+NmYwb;k=DXQ7~F3aZlI*#Lz<t4pa@omaIGf@|jK1ioH}4fPKE?xbh;jc!obEAQ5+
z?1%(^wO7v%A291XPaMsQO4j#je)E7VeQXq=6w>q!AR}8_pFg5*69J})sciR`BDs`=
z0#I4l12UCqTM<u>exxF1hN-DVXBUbozi0)DJ6OFc+3Uy{7wwL<63b*Ws3ABZMlKYo
zV?4@!R_jzl-D12L)WG<STC%P{chN!iPxCmR&>~zis7^DtZu!g$;L>o_VSTGhR?E&c
zpL1WnuW9DCl2>JDU;BYn1=8|z8^7>rK^ud|1kdR(fxzO7rP`)Bx}hx-2%Rwzkz5X$
z$UvO8PYNAO!cO@FVaYLCY$vIirlykNF!iFQ`dDnrDuI>|K7lAY2`Tw#I8yNnZG`gd
z13~MorG`J0D#Z`LO=7%D^zK(_S|3?GdQ&-&3yd&qGCwEo^`PT2zkiNJT<x1;CrW|b
zuR$^#`O4zO1zIrZO|st3!A^A@DX5#4J@riW(k)dJ$BUS=;bf>Z^b98EeMKI}!8oJ=
zAX#{NzmBv&vIRs%KU!ZknVYtnAsJX7y(FiCMGtp9=HTjPF;g(8fNiOK$ohopujbEc
zq4!uLsWPBJ(A~tn*^`?893YbURjd3hZl6%?H@L1{SJ3>^OGn^HT)3-Hg}^ne)Kmy9
zZ6;QM668SgzVDL8XorgaiLGI{u)1veD6}_juD@hr7ZA|>gt(^q0X}&KbF!BRBPnP0
z7C%wj1^#yB&d4PU$h>KzyKS9s4}4_aihU(S`o>ZkOX#mR{Ls6>?2uWi4L*4qB1qS~
zbd1CJ)LPbmEFD27b+P0Q79`2<iccFG!)lp*kB<wLpP2`1qByd3wa4y~4|XZH%I$|E
zGzvze<LD(}1KFv0ZF#*Uuls+edS6?Yl2fCj6IpLuw2}pb``6ftL~q@y+(&RB%odp>
zl}rrYUuU;SZ*CCYIe=J~T+E^XGR<7~yJOk$HFU4xPzZK1ob_NV*R*1RCo-v!BJDEm
zoQ|x<08@rm7aXwn?S}}#oUxjS+V`0m{IUg?>cRDiG^fMM<c%7m6hRWUw)w*m)eTqo
zZ?zY*-QNt5#;>e&RfLqnK3qWq*M@$FbJwNM;r4duaQ*}``)$PNwJ59GUpc_jVg<Bi
z#oK!cZcO(zAtG;XTnELn$z{bK`&1Hk??E;?Ft&Y{a+`JOjaX(BIZtNiElD?%1|!cR
z>$KY14Q#~iyM4)iciM;Es)c^HqCafNeqy~mq>eMxvK;MsW52%%MbF<Akk%4u(W<`w
z#USe<Wg3#a{?nls?N-}sKY>yLp-q8Dl9lY4rFl|+EV@cb+;$gHwl8{&mdTV^Kg-x1
zgR%?MnU{0g<2a*$X;>RBKe|>^w8ngHyH#`|a8Xw4vSN$7I#wmKyT)(#3MC$*=%knQ
z<m%b}q#n_7;T@|v<wn?`F>$QhEF5tHIrM>heOEU}?WW$V$s|02(+_@69s4DYu-z1t
z>@6B(+!NUh>}TcAAGTGi(zSP99|&x}=)LG?->2C1&b%-W+MsZ;$T23Zsb)`imiIi{
zg}cFi+jWBKiX2zh6*YLF-tw5}c^rFhU7d)Nq-99U>TyP1+1W3a{dhrX<aaBmKt{UO
zcW>nxL`g|Bahkh&(0+ZX;gHs1zZhT3bIqH{S^+LHE2D!?CkX=8na9Oh6{*h5HD?1{
zCvScb25$^jh}5`5*xT<nMUkT;dfm`KhEyrEZFy;$t(AlGo|Pv&CcS~VIvq$3HCa)k
zHtrg6ozqeYR4hsAKB0p4`eZ-tn_W%-Vy8#fMy_XtoUfZt-psntEyPO)1mr8fQ<4TJ
z?c*ty#%q7RQ@6{POJpyq^QyEh67ldTGh4JqosE*pq14q$4|Xg^%bk-u#L@(-+zfuh
zM#3NBUPldA`V8K#Q~jj<3Pj7R7tWJ?QJ6Qnd0LdY?gIFi4K0s3KjUDQPpfhq3?K8$
zfo@F~QWI076&A;uJQqzjtestz9jlfN-#c4Jay$D8d`3n`Ta)&H8p~ly=}2t@;FTsf
zC$fA%OlQIl>&^Dl)ixj)EVRJKT}Mf8$7;Gc&d1p$+eGiJg-1$(qDlev+xaXtf#}tl
z!V5i`!r~7CK1sl)E)PwVOcWQ;;;u2<6TAH&p%x$^O+%9wz$GH9Pkilmq$Y0W_c(Tk
zD!Vy-=LKv2EdF|4NS%Hxth&sCs<UdFH#SJ`#H)$^ynF@Dny5aa_7;^%sm=>Ko*5Oc
zM4&3h7$0D>$M1FStHTs_X-me|p_Sd?h(-Qd$yeS_L_~qn<H*%Eui;3MkL65elOP$R
zQe!@-&;S(h<11HSW9hcBuzaS>Ac~xqLi$$RvDx?QE>b<>3py4H1#FcL&#PaC@BNhh
z@nC6H8{FU|37_QBX}GwEE2ovye><XAXfLEuC7f(CQONuFvD<dn7(>>}Km7m8473zF
zHdhBNTQfUh_bGA@nK#vE0($I;M-UN(A=a5`l9F-!n^{VsV|G(1Q8p;ri0J6rnu|6&
zXzmm+2yQv@cO`p&P|7fwM@AYCx5HbRmK$R3ko2<V>MWArUib9seB^U_OwUFqzS$9{
zy;5u+1@GTh4(Tj}nh#f1H+Ykm@8^Mwk5#nPq6E*age$raO=`_b3FTdsK}gj(UaURT
zai-Y9%G&K3(rG=UuI;>3Vzcm+ukK~E!%B*}@1*Mo`a<4st3!Q5#>ro!8&YmS=gQs8
zx)}){KiVfnShNnM;Y2SjH!+y(aKZ+0x^uk>ws2I?<zNs-Jpl+I?y{Lqbcd%Cx$En|
z^1SU?B;5AHld@Xkn5$5FGq&9xC$VWX!ixtm3KFGMEy|5v$^EOC4y-f?8OF!-eXU<r
z0H|vp8q@CHU(-uvY4q?6;JkD*ZthSDjo<0>knENAWM=AXLqVs9L)n@B{vo}THo?*G
zk@O5dC^UDyMSn#IdfwLt45PVYnN^WAZ&GYdH>ZvACfsc0gZ_m=(m<|ou`XgOcxc@H
z#`se1M)Aq?Jkf)>#jINpT>VtOwwDiTu^7sEpeYZ*wbz}jWcf3;p#s&_(7GOR((l=7
z@4k4EUgo+XZ59{Y(Y68ODKn=gCclJ}9+X6AHJ<*mRC4J18@*}U(SB_8=wCORn&{rN
zfB3XAT&-DbtY$U#SQ1T%VRz52TJ-~)$ZmlWtKmAl;dwaDMnKW7DerZbLv3MD+8l6m
z3%<n-%vy#yAzT3_mI>;`f-bQ~BG_2V>{guy4i5X~JL6Pbl7`(J2cGb&jNOIE2Z6_i
z=`oiy8oPcIBp<byj83-rRUVCSTKwaapV`J!^L#v5JNx*pNIm?_(_?REi;u%>X@7&h
zqOCmFJ^$ijOF)2apRM!M5|N}`snpG=`z);?<(@ue#==6YQ!}NL^~R7odMU_0c`~lf
zjvjtWBYMtDR(~l%AW|V2&X@F2+f2&QpY;t0l<xu2Yyeg@$Sx3%jDXeJ&(T}alLh`3
zS7x2YxhXW&&)ua}^W(n8p=EDRj_u0zqdcqicUJOJgZd!+)ZQ-NuYas*#i`qy{UIph
zt(&^qISE&P(wvIbI1=6r3cOXgo~z_dK)_&b^jSia9X%zT+Fz3rGro@Vqw_*BG69A9
z^2oi-EIJzswbX&$0o8V2Tz2R2-O(cF<b^_w9IREYB?*y_;UF%)Pm)U;=E#w4NrdXj
ziXP=CIHG)l9w5esIDk}NX0zlZ!%41T@d~I-Kcov+4nR4r?eXtpV$RQ-NN94)OD#YZ
zr`c-X`#YXX7P_x}r=wZjB(DO4rwfnJ<qb9Nn9)y*qo(DRD%E!xM47sz$1FWXb4-!b
z&J|UE*qxU&h86hlu7FI4lKVuI&$PoZOueb)sS3A<<Z(fCL;L(LcV4}jz4iOP!#;n%
zx?dH|M~jwoA6>Cq9E@&*YCY)8SbeZUXL~A2_(Lb{7*?hHK&Q_M$)A9MiA-(lYUGlM
zRIVA5Zsu{`(3{iLAH1>QPfA<U-KGsI%VkM*hjyRfZj6YUZI*a44B|MiSwqi%91j%v
z7(WPcJ>WNFmEcVZqlrfW-fyDOwee~@@|)vEKqNY91-`1z?m0v_vVQO_bRKCjE^(;O
zBXL*HUNyD_2lu9eottk+<yLW3wrrje5->oL;USsk<!_VR5CKrIl(IqT^8WIqMRTty
z@4F{@$Dr^>8u(wpW@ZJBeIG1EqVgx!6DDOCth@C(wZ7}%4BGTeN8FNX?NE7R@Sa5c
z)zS_V3&|i7LK<l!9NdxsbFqQ|?F}4|xBY63>NJZM4hJ=}fNse%^mJdXa~{`P*i4I?
z+a%}XFh7=J3>r)Mam!z-y-i8Oz(6t3m-O^NZE@+TbM&7C*t@Hnz8|E%SQak@Z<?*3
zMkX2fnB~Kb3Cye@p?Pg9-g}VJjT!lsu_kG#zgjO{ug<yLcYob;!d9CD_Ef)AGcP1b
z!TN)0?F7N*kE*$z`=CSxuaO*-qhcMOG6?W#ug=QT5Bgug1LeE>cyiyyHx`W6iXv++
zGDB!-33Sal-2lO?Z&BgeR;*EGxG5N%!2bI3{EWeq-R!IH-iGUht$Udjz_c@nIjHBp
zKrceMX(K58sN?xMcjR-hZOSUZ=H*bTt1^K1WqZF^G}jNX52q2U$t}S3T^aw$Gr8*1
zT?ZmQB?sLvBhR4-Bw!!G3M#-_ohdWWO&H~}%-w7`SMPkX8ddeZrbT`U#M=Yab!tSQ
zYa|534^zG-=vYlaFXJzTAa`eC^7#P{gQFz>DpJYMX;-dNmseGNe>sWW6BR|GDiv0#
zII84BXcJgfbwl>*EfQzv#2^nL+G$MISn7yJjEh=%P>knX@p4<z?ixP%4`lrJ7v-Xl
z^t)X?m~A-1Jrll!H7p%d;u8{hnsbz@JZWg{_&Q_9gT+dWj0itE+l2N#kghnpNnQcY
zbiwdl?#tIhGqbLzNTFe9yEhSH8&ua%TMNe%p-Q_M95mCuPf9$1VAGd{F8uR%pWpoW
z=+F0-{v*8Vk3IF_r6t5wTlh@l)jl6o+Je3p#PoE8h7>YARC#+fQk`~I;VnPp>toN&
z`9U(-tASzq$FUi-fU{&d!r$5d>$BJT)nNjUq;+Ik%&-n}AnbmurgysO)bgCh`gLj4
znRe0Q+3P{?$sV2t?I+Qt$70kN@K9o1<FmWkNVY}f{8LqWvQ8OSgeL~oS+#{xmY!n%
z)tTL$5pseby@7_W78h8iBf4V*+^8qJ5(*KU@JylZ${*qo&qGpqnBo1d1v6gqzRJ(>
z7{eMdX<5@+BQf!PHIYIsjfF+b^iip`da6dSSr0pUMIP6(y}kNANq0Q892M5HG@rG)
z6uAYgO73l=F}2%!2R|G7uVMH|y?&#0UE4-%r+uEmyX9x^%yI17lQ&GFeAfru^iG-e
zW^%e~so~i%kl4aPHOa8=wAaMZu*PJC7)yogT0qcrOJM4n02CIX{zG$jW9;_kc=6yY
zStVK9uTVlIHN6o3T0!RTmYp$~wPp5B)e-F7^v?_-2eLX}oq5nlz){Q?NZh<3czJCZ
zzj}J~d!M=MLY~btdd8+7ZQ<&pNW&O7QPQBAbD<(+;>5&fEnd2(^mI%vd0u0Q;No~o
z-~?fLnw`b`&-s1S|2yK8sOS+t29J-=p696mnBmu70@>8F<mde^YL*r@+~_KB3`|Xa
z4QJM_CA|n(pA|E(eB05x&gW{|8p<)al*nPkjS(n^rA)8h?;1jHuu;smQY#ixt_-mq
zc<EPfk>UA!3;lRfo|ve;2(FIRyG@12zDcC5&CA}G6m(%VL@{yVNGuf@Ks1I}I0>?-
zM&$^|GK5#9=wYoJf>NgS(R(LneI1(<w^X*yqt>hU$5;!_`9?Hm-&nnl)oEzS>Rxr=
zAEC+c@Z5NUv75=SYBio8_5&jZ+JT<W^Y#a*=BjiE6xNEtgUIxZ&*~2si!U2#g7)?_
z1k)+$vg8Z`pk2OJT?_$#?}M%okKR%oE;GR&_bBDt=Iz9~^?&Ak{;bux`cC30)dC}W
zQn`4`sEnFbCYE~cOa{y`Wy8ae2!c{f%<_{)m5JpC5{8n`j0wFOlXi(<5J#7fHu53w
zAyr#j`|{>M3@Kak^sb)nEj!yv^VemG{<q{2Z=1KD&1&+zeMEb5LWWZ#0_WKrge6>l
zBRBs<e7SA@ATnx4IQmgx^$Q<$Ow6rZiH(mZB*!CKyJ{^?uVj*&^FCaeBP3v1`+-G<
zQ?>K--%ab%OKQjkEA>OR3Vo*6HZHIT#7pZ94XwsoWhME=sFyZ(<z(aEr^P;FxJx!O
zr0ZVRd}vAZa`O9?SMzf$mX~3?HVcTkGt-0DEi=<gzodWU!41b6(^v$?%0voPSY1yk
zuubP3NfuQ^@NndF^rtV65t>@Y@aCU8EAn;Ut_p4rMACrsuZ_$MQV_wZNfevTcZ@ne
z3uBxWney><C-se*?KZkfyccS0)P&B}3PQ2^5ms4i!{XK!#I(M?)x5A*h~hN*6?#8&
zlkJ1B2uG3|Q+<7lJdt0#lT_MDF;6Xnb;}l*LwtP=zrnjxvzlKt)wu@8m+-{I71?&Y
zNU}oVWk^9yDTfPT`a1atSr51C5rW58eZbaSOt)QEcwtPx!tS7ke&}lf-rdtFFSL|A
zE;tmQJ<}Td+Geunev0`(iUvJBgT+5=S=k0g#jeW0etT2NhyEDvdCzu#$kpZ;yTIxq
z(3*ypWxd@-Ax9<uA@BRpO2<ZtcG4>tnb%J|AXFA#;OYL$LWLL8vsxo0D*r4d=$7p0
z=pPlkcTM>3w@dZsYw$<xUvwK5N1Ugh2y&|BQwV0u?c3HA6by5_Jyr7ec4bhDP}VpX
zGXRT;z^2YWuz+#(iS-Nn&BtvIIN)=tSM;&;=RVD5v9P`{IIlir;biY0h0IzSk2T`s
z@sq-uruMZZ8K374Y_A1|nC>m;{)&{MJy>bjk{i|Ugidzh3iOdO`!-#x<f8+yAgS57
z<88yCknH!%lb&!6PEMMAIgztt)D2C?eXWA#U(u~GMUGtJ^st#Y5@IC{B`vEneeqDL
zi1siE>NB@vY&a#aUS4E11QxXq-KXQd4%c+NTi6-l@QV=*D-T->;^fQHEALH3PX=<n
z`X;?<yo>4fyQ$!Q1}kwlRQ1a8tU7Y#bSJ~9AeLfEf7KQv69MZYB_9Pd*)cApZV{YV
zF()1%B@LF*Oti3P4_q|I=eyjUd@{~^?`gEZ55$yRz|pgV4M|J6Rw=vFBJ(S+q;N7L
zEr0A`KqWZX8bz>m%(&FQ{tf)zdh}QPp+o1m^+Yw5hNkHPwz-lmVqG$p0}12bo_LLa
z7qxs##i#A$#ZPHYN~)$`BbOC<S+wiKj*UA{u2Nqyr*i$G)F)Ifp=r4OOP!!gbUm#X
z2h`QtKm2)$vi3*Ea172fi!%LfB57tHMIEN0hm`Tr-9a^e4{I#wfWHPFuw>)hxh&*9
z9&r<*+@%8-_voAR5Q%y+Y{x0<z(qqdJT|&}<&z&rGtg8)&AeK%&b7B+7$Q5y-D9@3
zoxu!e6bH+sPQgEuUb{FZa{p!&CeaTKJ<Ch^V(NaN%9lJM4PBXvRqHI~U|ARfNwP{e
zfzz~E1vUF%Ia=!9?kHM8A@?ixqM!IA`@8A!%Hyh!$5?>{baO9fE7a^9+V#4w^epi7
z0K?!kw)E=+f@ci%BtrRU^%u*UjEu%`ntfu+m7^ld8wnjUBMvhU0M6iO{sl!!UGTV(
zy>B{q>n}f1i7>3yyS;Pps!(AI*5v4drxWaN+A?B55*xoO9DRp(zc^c*C_gw@!*lCJ
zv4sULzvw?!$GvOQ!uR9zZ;4=jmX=@KS(m6TD0sNASYR&%zkx8FkF<O%;OHA#*6)IK
zb8G*x%d)e~@tLKZkN6A7Yt<TbZ)ttnw<-e*3_udgCqYKyiC`VZd)mUn{L?A&(UX&Y
zH@VtErk@4cuaQHOGh!l7?C&HSY;5=KS+9kJ1`|a2`lb^_zvJNu{!(@u1L!gHUCOq@
zQoa1*%cgkb)fWSGJWjN<9Iq_R(z9Bm*R^&hek-5Vx-<8k96|+n_5zPT#QRH@Ih8G*
zUDEEhT~M9fCv4~%`@L38F00H($_w`CjC6d(t4JuXz8dnAMnNHLkG;2VcCP={*JlqL
z^h-Y!WOZ*f4=buRKMQO9#Xv72|75yUPuCT0e^1$YZB|4I92dP_|GVSsL@?@IDZBVb
z@AO6yxu0}lDk*vp0@XVa!a;R?+Ss^hbc^&stI5xdjM?cq4byyG0-9%fIM_YQ)mN@@
z^Yo1UNKCW6byn;wRLb6ehVoi?P9GVtI9gU;;;TA2z{2*z!0jc2AdhE=RaSTJH``fy
z7Rx&nVyP;<BLlA&gU?@{ldS{0qTiHhFaNP1tDhs@ybRprTkIg__%<l`NI#G8dBwfP
z5Vpk_fyxtB>VH7XM>iX1MX7<4x<ZDlY9qrw+@hRCkR2H}Ix=Hw)ie;J!Md~auBfD!
zWYY{(kJ!{2sBgJOy&Y-J*}BPzD%YhcEP&r-^S!eGVV)SKi0-tEjt6bc*Og2kNls`U
z+>nfgG=Z@zz{`SXiMN=^YMyV91>cgCmD8!fuD?eK=j~p#7WJ%l*<5;7_){g3sD8RV
z^)RT^!GLnXX}x}Gf2=$<(tfe-lz3%%TZ9xMJXD4|04%{B6Zh_5)DhSp7B<dWx(h$X
zu`5e!gEzH}!L_ZyJS>h^wXMZ=`Z8X8jDVEb2QP8$Y>mXm)c6NVMiPU6p15<EjTJl<
z)IGdgSI2Xj<dUHzsU>#@IR8D^nw}q<>Ht6<A<Z;J%X80yZ%t4=HYBj!uM!jW;6I+*
z<GG`MXZ}+wex66|#jyW%;;cM(s^_<HTid$DX`_v^jT?6-=iV*V93l&SPuz;<M4+2~
zq9U(9s{hl^OzY3oU8=($@n5?|yb_rwYh|@E5PI#6k)^DRHH9Q`EC;QNux<o6qA>(u
z(y+CUS3zVT+0`XdsL=asFe)mEqI`60z)~-gc5kiAI|<dw_}u8ccy?S$>Iq-n`{t04
zYn=PWZ~PvmJ7&jXe^E~_gjUD;3%wS*#+tEnUWw%7%giqN6R!lPZPbb}Ei8Bo3Sqt9
zAF?iCrVSY$_vpBdyM_2}=A?izF?gtlne*ynase)}e=D`Q>Fna@Qc}0*w{N*YR9Big
z_f6j_>z7DL-IVU8&M3N>A>|WTL40xG#2p{68fzaBAZSfPqk8<DNdUqmm@+ig%yR3F
zQwfQP)8%Nu#dU_%Ln2f<Y4ut)SCokNYlv>AD*<}~+L|&Gc~Vf#zdA5kCu~%ClE7(^
zVoA-Zh5t_{eRO#5mY(n%3|E5qP)%aefhy+s0Kqb=GBt8-9GsIfE9Fg|{+~bvV}JOv
zOX#y!R3<_2NX7H6#H3drBf}o^WW?cDNBYmL6#o86C>uNc&52Z5RWGega5E}0`R;o}
z@q*HIU`Y11vrz%st8MP7DJgB{k+&>$R|F>EJ1)6n+=V;bjH^6Rv8h*1lq?X3PKJ`V
zFb4OliKZ^Qbv@jg@x!J%8R;7}^x+e&7j3&y>g^CN0oFflsJFO0wTn+nfK<keo5<A`
z5%`ywFiAbd`*#M_T&l`nJSrmZ;@sCxJV+T<&oAqI&XlvGsinvwq;q@P4tAZmsp+9a
zMtMv;%WzAexfZg1tCj%1{CY)@te}AM?5AUj3jK%|ZA@J#IUNhtr^aw4vi$rv3mZ^Z
zUYcKDM~(IT)gav4VUg{vpxy9%8TzMlAX+T0I5q)yzgWHv0>-&UV0;Tvad`IY0k)k~
zdrzJjR&h-vP0BIJd}J#(8EJQLK3Q)-z>D8QEluM=Ox8?g&Q2oq704~Id5+w*UF+br
z{=5t}_8?{TNKbBy@=Z@&U5sDPp`1(ZWrA4^3ys&eS5En!u?6ObD*+GXNI{1zyIrBA
zxn5&sWi1D=Au;Yd88Ao;c1Qa><*Je$yAVHJ8fyB0pL@5d@op8~IVjuR&k(nSf^Oel
z5kTH(XlM)nHTMJ{;26>s!In06xxO04e+5+hV8Yd1o<CKE=m#nJy0TU;ro}{z5VoG9
z-=}BIGkc3&OZ~X3;#}3=-Xw@hvR$0)qNd$#OcDiQtg8QA$#w&ImFIABelbdUXHb6g
zInkd5;HW&}SP2@uadL8s(KAa?+Gw^_zhn*frJM8EuOqud^()$JstyfoOk16qJ5B61
zlMpLI6KzoV*^E(RFWXkJB3+yOPrBkMl^5oC5qVZaZ4F6xs~ioylkeh9!F*5ly>?-o
zuwA>_Qxc9FkLc+_eX>4|xUq%XWhfc*ncK5X+=R7vd@o=b91_h`ct^!RN`;B3^VTZw
zr?bsEI?%2N*Sl07|3S}LV7aR8I4cN(JK!j7Ay=%qsK^1!0n2&MvGO?guCbBWqU<-I
zM-QjwQ+K)5>6^D5oO+KcqD$G-zZsW@M>T*LlQ&fPDFYoV-Ch4zACZ9z#T(H8`Vd{d
zH8`H|94IvT{({Bu<cn}MpT$+DC_>)uP2nv&TUu5gABJUO2U|-O`%f^H3H!+Iz(+Ts
ze^MAn__y??!=&3=O&jTwZF$c~FfQ&m-QM2*=3l@OB_?J<7=151s;ltV!w_$8TWac{
zXy3c0+kt(Kqc7Hkm&Q0Ta$YqiWowvVW3e4jn4KfyvoY*NF|WJ3_X5xg$OWLZGdnvx
z4@kRnyiM*j{hZzIqkK-;O9|a&z+nxLURs#MlU7ElOL3OrAgzCIuHSu1uS!>YHGvG=
z>*EhE;2=ap3<CR#8Ip{}dA9vP__0?zr<}*3V=$=>?()5#Ak-eQ<qLQ$sMj$Hc{|?c
z!g%x6qTz+UjTX<Sp}CQWo+j}STj>ZsLHj_E46QXv6oNN{0OB)5j_gO+@No|5m>C)<
zxMMg?ge<fRqK=Q19#WF=1?%Me9R2mVK`_bT?Oaf41u=~sPqFJUS&;wYTO-y@3W280
zRP_;Vw}<{%SPy=g+}^mCZOgG(dVady>nwy97bUEtCrH6t=-u&*an?ix4{z#I7OiE_
zwu9XhUXI9`uwof&NV94O?@&2akPf38mn?<F40PgqlHiX6?ce)STN~LA%@q_<n!EKM
zvVK`N?<$#fV#jfF7<B#7h{<9LWI9ex22;g%JFM81Io%=M2?${6?KA!yN+&>CH4QVo
zldT?387M#Z?(Nk;E}Qvo&l?)27h?kyEom=a6bam5_9;4toE&QKs>f{dlGB6<&<gP{
zeQ@V5F>b_Xpydb_w4b1POob2K;H%x}OPnBCjrcxS<t3o$xz9_P#QlSr&)J@hB`lGW
z+2y%j?Dx{L6zJ4xt<J??t$heDIy$sgr7a<3-8D^=FW6COa{0lp&U~^7A*TQV#bW*{
z$nr9=1QXwV-(}cOp#ZhW^uny6<>^<cQ9iyM$K;Ix+AmRi%3J0d*C-SD+ncnJ+i^QW
z*6=e-Wk!VmB#Dr3veTN;?EaWed&h#58!b>YkBL$Sd@Xiic_og{LkoD}T42C@Vj3OK
zLwxz@fz{T07aFMRsF<Y1wiKC_2jM_+e)`5geb>qSrRyc+L+@5P8aWxfZyp+*+$4MD
zB1Q=2xOIB?np-~Vy<Y~ep*D)6x34cPtcF}NR?~ynBP{vGLC(Q2Let{{Z(fRK(9BFj
zRhiIm^nLAfjhYa6w{HG{&VUKz@I2Lm_#w#>Tc$QqFbfs0^W1(iWNLNfI00?qd27F(
zDP>Z^3DmHqxhm@EDdxDG5A@*FDdr)God82TIWq0x@}lyh?N@pN{iK?k_32vqPcE$v
z_4JOn%~j|Lx)Bzw;<|jW@vRL+%6Doa!ji2y9yQ&X=FhG;$MXnv7q6{)-tKMThkXSM
zn~eGml!ntC%fl2mlsg$l5RiX+`{L&3_!tsmmVD)w8!x2)=?y_dkv-;g__60bwv^OE
z4(R?Z;@BXom3IGU__>lBxy#ZS{6m>E=2^1oJ~fQuB8@J`e#s!hziT@<puN}e;AJ9p
zF(;?d=1;}vOB0f-v>LMOZ-@DvxN+Pa#IRm1%*~isb=rSohv4ZR!$wMd50OjCfK5<7
zi?^R!w$)O<K>{G9QPp5V67PL|Qp;XDyXBJ;ZLOAf&hJvxUs^t(Cs6NYv)Ejd*edgv
zu^uTK_0G8VaBa4A*#e(<Ze0?9HLMal57LSrx|9>y1wk(}5)APfymv1+rhAl)k5Z(+
z48~;u1Gff|356f>-cWqXu&8B57@SBJ+n;wbH$CXH`vrpmt~0Blvri+QYyXTRag*8m
z$(4(kT``0$rA*T23jbR}tfj}iM@ES)txSybouYIYE{=1bVd_=9TH0*x{Ss{tu+7$m
zK7I@7PEg928Ti!llqq_*cPkYjM2nyZsf=m?Loy5qL{k0Wz9ZU6{hN=U3t%c68w+|l
z#SIRNMvJ=aw<NQCk<t|KcqDXpEsdo@J14Ayrw(_cJs=gcCqe=(ac&Y4wM8Dmo5AGd
zF@h%dd`j%hD7mF{>j;20GH(irR4BEQ8jA}#(LDVpM!V;|;(rak_3OD#O~~hAh}+Gz
zyx8vOr3@|%nKBOO(ZpHt856U%TueW}Mqga%E^}%lz@~l=C?#J-(dCT|tvn~ZDU{p?
zv<?w`(boz3kTmfr0ak!}`c81(jt33HJ>TgE!FZU3v@M-NC_e-j-&Z#_I=0?bE7-Pn
zI(UCFy75p*#4>!{8pMsznE>p?2LeYmE0~6+mKGCczyA`WvEThUr7GSu!0BM2_*hD<
z7*t+b3qL;5>UxdO#q~u(Ay-9awAQaZXBh!FFEKUFS>&2Xro#JIA_n+%D~mLdy`$|o
zw~D4rURXB5hujjzOIU%)x6cyWTPloGA5e31?sbVPuP71X4S~Qhb)tH*oinG;ROlxJ
z?ZOk0Y1?*AMjL$*WzEph9`!icn&dBLgS9AKlW16>Uta6UDl*3X>^7dkG6QoeJcD0U
z(c)RRSIxr=sitmBceYg52(3`I3R7jThdnreODolf#3f_%2dya6kCl@%i_<W*ZC-C*
z{8oNy8T&b!0g9}4n|kPHjQ?h0IFSdR9sZH%UlT<k|D=I|A8&h^_^jHP2(V@OWV{%k
zwEA1C<MVuYq;Kq9H`oYJMaY@(OLTkglG9J^r(n3T7wXy_T7y-VnNdSd!%9q9pf$FI
zpPz;q&1BE^_`WoiXw=>1(6SKtrMsqfaXAEci(LIUu&SUIaqsza#`(>(77Pu}a!RLx
zx8Z8!_@16<cm*tYlXlp+luRzC(}J_m)h(~+6fHXE%>!VeT{N}Pq!W@9sY1$iK1Kt$
zm)rY$zZe<Q4l8BP2<;G~!Q(G$+V%nK0z&n$xyC1Z8A`xv_0^tbj&d48ioX*ooN#fd
zL;)J=J0zO{>*yCFL!TR&d4$qZu^AN|e)t=eFqvfa9|Gp4X3bz>7O`-dYrf$IG5sU@
ztd;8@odY}D%{vwn&m5lJW?)nP&UkiH8)uzCG(+hbQIuC`PT2j$aOLIiZkTlVxpkQs
z$E?Jg<N{4>e1~Z&IWZ8A$5TK<41lv04ieK%7GLE0-vvg<GI(HaWRQ_~VfD3<8LiOf
zxa#7;-H!lnYZG<+X#D9-LnQLB&XH^T7)xtxysOJ}xOqLl9PdM729`GcD9*oP_e+XD
zGYC-%_=3_5v#~*hUV&?!q1Ob!=5ty){6sd;1Z;|UP}WKKF%mxlV0UvlaTKRsZ-hk5
zkgV>Z-TcKTTpoc$kID^746_^Gn+AVsWC_xKe<K2T5fEOFcwT&Lj-f<u0R9y46S`SH
zx!(gm1`HroDd)%qHA!MT)7vZg>sS5g@Mrb08%yjVfM>R^x^`u&y%06fye<f32@Yb)
z2WR*f;F@A%p&@lWr6oF!27ntsS^uSo*A*bWWBQ{f!VDJ+6{s7}zO=qWyjW~3T?rnL
z6Mv-=Pp0oWbq~Y7%E4f)d|HE;cxQM1Mcn200Is`<Q!i!6!gwmu2?_7VMK$d7yz{jl
z7Ta`;FON+80myYG{?i9gFe4+dVPZO!2a(OPsB!MyA4YK%Rkk2U#tG%m$0nYRxXf~b
zR9HFzfjuKQtBi}v)#>nx7Nho$izV(e9a7tq)26b$eAIO-5T=dmRNMyK<2U==rOQJ?
zKsQECiWcI9M+r+A-Ul54xXrw48lU~tia?sl`%#!g%qi4Q9aXy%K6n8RmuEO2$1w0Z
zltaWt&_-IRuM4HyQ{ol`hjL}rT@u<5nLCL$Pu0D}?OqrB6%`+*)7?f}kR9{6<N2U|
zT~j~#>CGc!#+Dg8i|OAnr9vKl$$1;3Ww-vN3ZcHkCbBNn^!Ns|ajKUmuCmni!)9JO
zkd&-Bplo>AEk|FL=2azs-+YC2duyj&tlig$19&q3oDWwlczEUr2@MO3cd8!iRFdDS
zKXTKd0(iST)62tcfxm8Tyz?C{DOZ6|Vc`|N)QXMS>@hAw!1<BYlI3pi0=oCgi~bgA
z%T83>?Ze%Mmu}wWWez5GJiaG~Ucr!FCAp-nCUKHll!JrqOdF1}hRX_>{tbO<5aP(o
zB`-$w#kB{3uqyM$(XQvE9tK)-m0-MMT?wW#$U@Z_=VawWw!&D-Lc#?dw+#!zL~Jvw
zBV4aqvgYn+jv+rspet$C<gqi8WFUK_C<`v?GkRL}se5<lv!Pb#WH*Gk-yp<f*cxT!
zQEH373ZFDOckC1@iXkT5-56G^=CQsVl^MqskQV~T{9>rU=%6?G*nUztq=n9ozxKqf
zd7`W9?6`~6!E&Uexz>h`20u38W8Ek6&h@ve?C>gtT~Ss}A}*j>{pBQ^wEa}4?~34r
zHtQTMyVs)1eSm17ZS|kH7Xjs|EQ7uMUwQDGcSe`OVu)7{L__`!3%@}-ObMOPz!nAh
zP(Y*auVj_x1y<G_dHcphbumpS(UtR73^F$u=flPM%*EwssPB8sUjKA;jO*v#icF4*
z5`BcB@!MDUeDQAk135D!uW4y4&-huWc9XcF#1@Nm>tx-O<H$V__f1`H66<Ob6ZiHu
zg!Gi%E^PkUJJEfZ)Xl(!T3K5WF;xsaM)B?#B16s`Sib6u7JkuD_$-sE7isw@6796v
z3beQMqVJ6BjvyoQaH5s^cw9NYNqt1B&{FZz5I9TY=rF+UR%ynd(>WcRgZ=`FP%)bA
zttk2|5__ADP1(aLpLk~Wwez9a!wFT<a?l8%H|E7e);Yb|xL6i`WYMv>@7+Amum0H}
z{TL9X5wXdkz~}{&lPyiwl4D|y4Xuh}d**yiHm_^sV$0lZ??Ro~uW|dw#y*JpYLo)^
zi_9TT)MlUEXGvS`$v@z5fEenSSKPZloYYpuM*ax)_HLnVz*z|=3B2Hn<Z1~vob(AK
zt0ofx{KE(q*08qL%<|gL_Xq0_?+wd+HTfrLCwvr*j$Tt47?|m@itaf}2l5U%)+Cq4
zx~Uq#<m5zUWX*F%DlmAX<ICfvn^;hCGHxcfCSuCiF%qE8j_gT-tLFCm+j9g27H8Gz
z-}Le$G*N0U6W#qUDWye4cTD5Bs@s0G=f%jM!^28riN7>-9(mK>niwu~2%cTK0`i(M
z<Qk_t_f~8lr9Qdkg)AE%cQ@KtfEjjyR0>!&Lol-FfQn4Bp>>`Tb`DdI(2bzY-!*dU
zD{Ub20U?`1rwg`?Nz&`uHgRMZ3kw|tobLuADu^T84o#FclS19-|7AlOF`wX7H;S64
zz(bQ%q;hohtYlL26MzL$!VcU*6leM;;`)+s8ZoFo#dradTNX?ppR`x4gQwhAj4W;*
zn5b=II7ArgY_~zNI;nm98)~jH&;oB^F2i$hB{3gqg~$i6%GAs<C01f2kcAQ+s$Cg^
zQVd=G!W@gL@R`cN!|@!2Ze*n~2kVRM82SA}U8_j?eou`z5>oNR#2mxirUbJw4^?aG
zf^lze`sakURges&IW+x9QOfz%#{7tp?gl`&fD>_WI|ydjz_MCo(x-)}GE#2(ze>zU
z=9?v43OQGL-uO40DdLG<cs-IdcnMZk5?vkN6x6)3C}3)m&){U-jlauZ*-5u=SWpSq
zC|N57W<Fo}O?)1ZZUAC#jNlaoapfGP;{ASNkX}gu;73_S%WaWOAm^*~1tW1^4MX>x
zwa^?zQ2)7HIXZUbo;FMJNj86G6a`}J3-%+X|IyS}fK}CWZ67QYM3fc*AN2szN;e7u
z0@B?`N_WS`0Hl==r9nCc1f)?)x<R@@K)U;1$LD?D@Ara>Bb>AM?3r0>-Rqt;a|^^x
zIITW6SudlXqaUUGb91=hJj+$8r{4LpxraqXU!UB4uM|NbH@d}(Xe(2N=KA<FcD74D
zyU)yUkD2AH!|bVc3NgtY#rvyhyR`*<bEy|f4(GM+Io`aC$ac~v9F%2p1(?6J)n{nj
zK%DXpmug-Y6Cb*{)h`eyDrGJelUHxLkPi(ljyAn^0H&oM8^37b*oU5)`&Be26mrP4
zQE+}6cK_oQ*>p3;eT}0W3}^F;jgx5O_3#+yPLuHgm8)Aj`mh*rY>~TmT){l?_<u9#
zRoA;LW`CkLeu+|nBpKdyw#Wvm(rcYNIEvY1x&1!$=a+|^G(Q{stK(O{+B($A>4pMg
zBT1+Un%S!<$kVS&Jm0;vtF2Lz(f^IqYCO3vXYk{G^MZwd=Ov%YSk91Z+dB@lKO3${
ziJT7s!p&UhM6$>vy$GmyjU7$(+dF#Cv?x34t6;u<3yb%MNsQ<bPLR*EwzOhaBcp}*
zUgvXKOuKoU16c5TY2a;q@ECCKBEn)cHLJrSlW@1g1}?)n^E`QXA$JBTNU>ySUgX7T
zYGTuONH`<ybLY-^RAF(G5$bl@_N`=Rzvj(R^t*HusIsV!qx?WkuG=|fZ65s2i+hi1
znh9UcL39Ru6jICWg``&%cXr%>gHNAt2LL*wFB#o2Q+uZSnVBhOFgwAQ<WuYq@4b1}
z$*Qs(0gpGw$#*q|N4lHp;zELBZQ0r8eza@0X_k@Kn8cevFw0QNB0s{jSn{imr}CR2
zkjqd4dZ@DU?6T#VR#Pach<|f?-N<kN=#S+&I6F7M%ip8?Uf;+h0e!jd)XD>G8o>~W
zAtmH=6LzJSN+lBi7moh<EtxY(Oyz$=o40VIUcw6>Z()m#^s;m-O3#7twlQ{bs>nT)
z?NyEUYFf+-fWUNPm7A)+9Q~6{7(V31K<@wKM_0&Wj*zm|b$MeCWbeGu+#D3q%#L-u
z+}_)+?VCKieICkpY!q;Wv@RPRt=x}4J!W7YD{m+5P3SEbwtH5}{Cj!P7BXVUzu<#?
zi<QT&)aShCse<2|-IuT}-5%9yWIm248q70*x&ywrbDsvXc&a%E_UcnfND8JJEvs>=
ze@=*p-tai#9YKWRWdj?7i+6<*J^2PG1zR?|vSAJlF1o<##V<QH0&lLykO4AG-c0e$
z<&&~%{)SWfqoAGTY7MoEjJSB%dcvJl<OcDS?B1$Tr$3tCpsfszdY?5BtIGWH{ckNn
z6!IoBUw*_)MPx)>W7^PM#LVg1nN~&fnZ_Wp+T?e%;vYbrn?Z)Fs2E*lE_imY++kYD
z<wC1)-D20%QoQOgTwEXy1O1ckc#aHI3$DL*VrLJ!avPD6d430s&7+I@EkJs;iQ+7s
zi*Hhm)-``BW9XOIi4XRxhtj99TfU8jB#{<v?Tg*R39wvsOl222mo)Xn(D=C#V)!$L
za%4*>FzuV>cm+APV2yY26qlOaDJ$Imij3qOlB*e}N<(-n^8egoXoMOjFEl83Z0K$F
zuY0)hCe29C(xN`^%Dn^kzU#ad<)h)F7k4gce2(6?F&=&*xRR;EbOaolIyLpP)iFXY
zclX?x6@K@SGtVmIqg`1ni<|RIU*DCW?X5DXV@jlb-_@UcV_6*kdDH$4K=cJJTPh?u
z8rj82rcz5rUL2Oo-fudhVl`2hnv_?<BncH)Yo|%m&4Zcf!XN673wP*L$N&n$)Z;+q
zh^oZut8%a3e`-;A?H?O$gsd@S&-S<<tXY{;&cq-$SrTa@3wKnA(g__k8!CSH@xa`N
zW6VSHvjZDHPQ~#~>sgBGKfVJvq}wz$c+lGR4mGiX$ocU&b{&JDq-w#F;Zo5|g^kmK
z-o6u@$4W9(LfSYY=RAa#8_Z_r1q5z-hsq}0%YU|&W<4D7bI!x%_$sPM>}odWYT3*4
zUmhY{CffCQ!t`F~%xu@DP6hhE6b-=7<vwu{T>l{b`bwW!B9curJycSsP&?7t{2to*
zv35CiZSzarOFL8BU;V$Se=Gt!j6d70AXnY&xXDH0?Hm%1zRMN}-s|m0!C+^;iz*VU
z=h#cWPQGS-{G{!pn6tBLp0VNPctJRVsub|DsH6Dy|C#qAScMY*@w3*OlQ@4~6G*`A
zMceJIzUpM6zWmgh{sKzm8YL=&kl&!(xqp}jVpXQDSFmKNe!~~6N~*pRl^scJ7gwA8
z%mOts^39=%<sr}2%7Ohk>#7jgg9Q2XN9W^baRhHg=jMVYkn))1O}ld*m5YR3V&=!1
zHB2+pxg!oQX4{gQ4LE30R!vM_{5+Hw`8Zlg*Q`GD>~I@5SK<M-IYa#F$Sv>mk~Qjr
z#4p#xFCe0$lcn#nAyH7N;S)Fi)0y8Pe2$U-W*d;FxdnT;oj<2jJ;hP`dYkCnm{}(G
z84g%9a{sK492~d#@CJvg(in_>&;~}EoR7C>JpYh@oXrpKd;fR{%&@f8Uw&GdIXm32
z#uK$mIS+)7im@@Ho3UhZ!N9z6Tg)aA{N;+KbGVL@R=UBVz$`t~{peYnHdlnx^kuHR
zhVq-1j~8z%-VJ|P<4y}Cc9jZ8;s`H~joA`%VU)3L$mSR=WHpk>QH>mxaBFDF=gG1D
z90}JZ6w|lX*rM0MRwQ(iC6j<9g9wHQ?vdfH3f~3g|AEizROt5@e>~m8x*MxV`|Ffq
zIAPw=L)*))HLbFz8jpVRWRBr$V~1|p2Uzyg!8dEIpO;=@4}mnL1|M0(uH~yad#NSw
zDXna8V8Gd)5cuTo^NoX6;Dzq9GHpZ>Y~ubt8Z>53PcwLK_T#;Ebr<>V8@HMY1&$z8
z>(^!6QC7KQelPOrzLB1;2G1{hnrU)PnYeE9n&bAtb<4iDUz7FSa)xSC`?krO<gg&`
zU59lvln47xrdjJvC4_t|7q|D)wn&n~L$@uN)2xgFt#FZojqMY2Hkcp|*8O4l@zEI^
zUNipTK<#X!2_VCpkGn@}9_bO>(GDa}FeJiaci+EzA0@6KlQ+OgOM(Q0Sz_x?lU3Vd
zL(!qe3>J<dD~<I?tI(iT5=*4rpAi;w#{P|^jH#eyq&xV{@wxK+v7h}xZRAbzSxO#G
z^XH0yl+%?B4ZiK^t<B_xb|I@CDU5s*^vw6jbxUU<S}K3EeP}c&*ueKr<Wse;#*{oh
z{b%=Wk6V;`?6=mx6dLZleM#156TfDh0eeVpytXNNK+D1B*E==1#4h_ke*5u@r+2?2
z;7Ea(Ow(dckMg@!cU<YXG*f3q3DotplLG7W?rw4e0yZ{v=h^!uG>w%Lsjp>bg3P8x
zxSUn3XlM#%EWQl<`u_MOqvsrTio%svkvh|@Cf)&GU0Pk09gY)tGLrbKY3;YouSmDn
z{^*M+5@KxZ=~G%JgXI{}R0RlS$fMZy_9YMg#zQxc0ja8`gFuAyMR25KYV4Wm6)G7S
zp!2&8+Arx0JHL6`mP5}F(Mp7~`=p`Av;ERIpBoNHflA%;-Fz-u_A2^#V_-Sx_vn>R
z!r$#H@L3H{I3Y_`k*K9tre{t|+1TE#x;pYDuCkDI_~^iHJ-f4bH(x0e**C5;M%RX(
zcQ>#!tj8x6G2L6x&8hJU6jpbQ<b+~e51T*iE2mmb#cdW)=(Ob#^MNQTa$_q*Tcc^q
z#%{J&H9_A{ueST!1r27+_UWxCj=aKlX1Y13pIX-~@r7dM2VK1eGvfB9b93swrXqLW
zuP?-Mk&PAO+PWS%=oBOIgZPvS#ws$A=1Nf2eBwkDr<!ZhovHUMIB4Hw)!o55)%AR(
zpEONWk>QpKP>0pO`-7tghgACqnqOv~wl_1;>kq|y_BVPKKPN=tLTO>omqWLz{sK-`
zeV}qs9%WviudxyTk5hf)dxuk<kjfuwbk04vml%o2NKnYps?l&-|FFKJ3M4hBO7)3x
z<VPqol+RRpX4Or^z!bx5dJoI^!oEHI9e+xbj)!S6yJ6SCJtak!l760Nxl6+pO1ew&
zLvP}S2H)_EzLnS~;m00cSX@PRt6xpm7MiB41)C~;$ul4ugUfbBNxGIeO2vg>Cnl&G
zApE(ESNwzS!up!C-;z0*W^}Zj`y4Bc*WpE|M>cARysV<-Upv-w?)EJPMm#(T7Un54
zHyY8oC3Gg>5ux_@hOeidKy@(T3eL77oP~Pryig)#gJ#CMWnd6JGBT~FFD#=@2W@$p
z{WNSzBK6sA-jAvIl}ZK^q*kM74usY(&nnm@&r!<lTg7Hi^*x{irh#>Xb;Qrbp9N%$
z(;#Tp&k<i&FOVFB1u|GGfvU%APyk!N&mOyC@>)G3)#Kcw(hp4jS|`|^^Oa32DTadF
zPSIseGQB<DuW{kxq6wZd-=8)UqiG4SyJ>Z8mtucf5!m;E!^%(OH<1Fb@gtyq*b+{5
z@N0Ccf8<ByPnsI}L&5O8@1+zHVkp~_uq~tT-jzO9+ikSaaCh%_i^X24kIaUAo$coc
zA#_aK(WmYb;_N&fgQib<<DVYnwJaPoG*IB-)iwTlYukR}KxLx8MN{ee`vMgEKoz*B
zg&Em@NLtMLw=BtD|6Lz2`18oMU&~oF21|zusOw0!{>r=B2TR*1me9lCnOrM`o<CV-
zLtBHL)^ldUqrw&!h+!z3b&K4*bji)_E1a}7LTb`grMwbW%tc7h+`^I=sCf^HyUeWQ
z@UATV4NU2E_dfPjqz||NlxlRneI1Dm`_V=%WCuVI9R1UuEZ@>)#m&tz<wIO7GU7GY
z@+&7+#w2DwahP%EAMdkObjH!9C%$9bv)nX(t}oolnY+KTGsJEBOHC#`oOm#AQ{!-0
z&0cSl>{?@e!QqnQu6Ppwrj>Y3X$woT`Q?V3T}z;ERnKDL@F2yy9E$i;=Vx4IU|Y$~
z(6+EFJ6%|_5VHT^$}nidcIW+i5aSQF!qs|+>cF&aoX+|?$^=Y|HSrNPZq#6%?7KEG
zm&9G6;iviVtMZudVVam&&>w=Iy`JuD<>sX5;tbJ~2V<@}Rm*!-T?5Tne1wD@W8eIr
zG3P5p9YDGi(Mr6%fox<cn)+f2$ht@Bf?=t%%w?HVK(jP!u=0O(ITs$2VrX;+yHo-n
z1^=o{f0=x-iS|~Fsuq^?r}nw+DLHKc+^ufG;KAO&VpFraI;3A3w2kAiJ`D6qk`m2~
zqyyvN%nVecHBPj?hPp<M0o5v9g*09|<hPM>%}iNE=<J_A8l@Yywo_eE30T*oK6FHW
zBxAL++pxF)92}||JV$NTI{ItBBjC@USHi-dlivcL*#149b6n@!uYp`+wnvg54!Nsx
z&k&=G3yzzJcN8`(z-9V~!VG?%Z0#LbM(0GM%rzF4SiLsAh5Iz`*X3MJF{i~g<R^2S
zJbmPo`UXO(TYkbxiK*$>xs_E8GH)k3yN9DM_NIN&8z!>D!IfxkNlW}jMyq)Xr~%Wl
zH>>heEm@T{j=EL2ONOOU#nkF_)G2ebD(XpI3p|B+6Wq>w?_(pCMN<9`Sn`vFuIV2B
zKvZlq-HPaXFFj>fwn84zM46Ab<$U{^bGIXL+~CviTkjM#e6^Bl6GQB|IN4J~if9~O
zsFX^Nuc~UPxboxKuJ$dYY*O?``i0-4k_p*~7ninRY6JA>D~JWPa-DJZs_n&W?<8$Y
zzt&cf!~`qHpM@2cC-7w|Niwi>YV;50vfQ$_hh*sXJ~v+K9c0ZqJYZQ%OU2Q?qQ&+Z
z`7}=!g?Q7#)Bi(POHu%b{W^O3i2R=jL%kJ!SJ@>Nn{<-wGSmLZ52%h3PE-mJ$&_)D
zd$Lrf!h0R9psG}(xbO<R=CWjp$63C?<xBLr8BfPd8h8%#?aS>~1D3{A=R6nGS~sr?
zuOQqbN`m=#wg~A&Z+e=o0`4-xq9hn7+hVUo>7yXAUVQ6MmCZ{{|K^m1d3`JWSJ_pR
z0vZQwu*Il^gyYQR?tkkg{^^T*x{ee2=2O8#A8KmI0<TZbCfW;qm!LzraGU8!47ZKN
zf%E4{h?RU77~V0XB_l1Kr_@uAj7cvDlE2E&W`Lsgt>#SF#gnp=kwhLFf%B-B*J6Ih
zdZ2BL(%Zf{lYgR7cfkvg6qw`}c8>m!{6r3eU#Aav6JlFZNXqk_U=Pe4R$5ZpTjs$=
zBBw~~SNeqh{xZWIXF0-B@uRWz47fHY1>g5$WpjT?iEgs3rtuRpWNDer{@3;K1b!@W
z^NqT<|D7$uB4!_xLtn{>N2cPGj0W|U;S+x;OFrrI@?>wkx_JG9bHiemT|8zxPj<p#
zA=G>*ZW&{iXgDS8SyMToqKEs_yEnmQJV5qZ*WMJYzI35PM3&yj*^rUiuB_waPe_nx
zn%sl?8HfN7UwGUi&u#o3p-Lt~lC?7X$`UL~lVOeb>dLRj233EISFK!l_-=fNe{3j|
z*D*8tvt?exm{VFyd&Ax`#a`%4<svn8kmf#}V)cG>Mg$S0zw62gs}!zgj8nbo1jlXY
zfnbU@rL+0pQj<-Vd@tAx2IAVc{|g9^>wpM=RK53bFH-^qONc>1(*$vc+}A0a4rZ0E
z?(Dcab)kYEYHM&dH8uFK`8QL9NOzgW0;u-FD|_{_nhI#zzRXbH?KW>O6eB-zkYtmf
zppgAbyVS7sqg^lku!s^85=^ORP8fk#nTKe);H)HdLL3i%EyV3xc5cYINn>kS*iwQf
zCMvwkD4Z?`oK`ir`;ClKatSm+nlium?F8QR5yO|F#FpuTFZYCD^Gwx*TNG@5*(fa0
z@2r=n@N}a968iV_z-6eEz?4jl^hLqO0b9>Twr3mGGGaPZC*z(8C!{O_OCtZXtNl(W
z)tq{d{2nSNKGlgQii~3O?7TyiRh~-U8yN7Z`0O&I$=8~`M3;Yq5%M_x!;S(F`mv3d
zCLB{*@Q3HAZCNa-A6{+=QGMO^v$N{`GS2BI9t1+VINA0`kU<=i+g)TcFo2o0fz`IY
zQT!WGgsl%A9!nI@7jHUR39Z$uTNy1N*C%J%YioD$^?ZM%cM6=Z^`q^uVPi<|SXrPD
zd<?6PulX+YwAUCF+rnlypVyJZm?ff_;1Nfvg*9`oAs~ogz0Ww5N`7JY%oysiO=@(W
zy-;G0)7It|a6IS9KU!P@TT_gN-g$Y9lw!}>T?`jG35)Uf$2V>KONt^G13t6F#23~P
zQm;3Vxf~@Md@9>}V1y%@JDf^H^3bmL8UZejt+UN!D`|W`xlpw@SEJLKBp^Ikg3j1!
zl!S%T-aV>my=VOGqho3hRu(mi-Mg-C@2GF=#*i&7Z^7Q2oGUdt%O$ka;Vy1Rf>6gJ
zRK3{~7u*z2gt+>MLS7{~YbRzs+z`3lrctc;cb`hdR@8e`gOL$$emu!nT#IQ;(0#8<
zV^E()2^Ns2%m?)bufo3r1=cj)6PB7^ox?E>3)4o{)xMBFHpYv<F1pMeiqeUY>4qvl
z=y^tDnWHTz*nDiaMsMU(5r(qGXOKcTgKq}pO00zFo@+Ck@a5wBdavf&$3d^dmG3f2
z|D7yTSN@s^RK@@N;Iq{gE~$RYE6xkCocFLZE8c+I8_63>-pWEQOJK5`H-SP(s8ZGm
z=Lgv7!n1(K%W2+{+ULNVuhZ$fTq<X#&f&q#6DhdNx#3(djl^17{3>6+m($kfAUNPV
zkm<_yq3VDf1*HJ`^W$|XFC?@US`8U_iQC@cjr?R_m^o=CjQ>e+9?BYUAJM9k0yno<
zY;_N+XOfh3PC{{Ed*ZXv@)*o66E$6YA<bK`_o!fFzuAvtg|?%qQ^o)mE$ipzdJR^@
zysA7denLLn_6Ej~&8$}jtPabb5R-vxX7}&b@j6asFLo;4jbig^Wgh3T3;Pq+V>VMH
zDe1Djeg#uO2zeIJUTpX8%cUC}mnvi`d3?3QLrER()OXaUba+D9(-jtThHKi_rs^Zt
zedCk=>so&hiYF`dZ;b8J#!99lckkbS3-V)rR|skNmll6Nv(?-C9RGuI!Fn`#f^SVx
zOXbod*kp5cW9LU_Be}W3y`DhH%hFONCWh?rs3?~TZhNoAJ++uY!9@)Pnwf?*#Bglt
zGirZWjcv3u1_*iX=uhu7H5zmOP(a!^F2Rn?xnZz%jBi_9YCiediVX2%_Iq&9%hF88
zSg)p=fTa_>I5QQ3*$U#O5QZDT{r+A3cm^5N8$ak8oM}?knfLjA{`OAa?ef_i3vwY2
zJXKtKt2Pf`>ke)GZV5+9193J*-y^r)q5hk+Y-}*AF-4!(37cT82K*kZ5nE6QMD!)=
zH(HE0f03d5Rf4R47WY&;Q-|CrR3&c;HrQg>JmfNXlfXqu&NE|ZEGh5G^RVpTPiV-D
zi^pBbNE}#)FKIiwj<9s{1{eRO-Hfk*nS8uO{@~Y0xDJ)oH;s><6&@=*zSA^KICP6x
z3NB#W*qGyTgxBb>JiLy5`Vluc=&sFxIqd1M*+o7{D2@Hdb2_~mI5~GBj;W!lsoa%(
z-BFN4M}*Tte|nZ;ofw+*a_V0E`}O<H#5jg<uaD9HbRS%~eu^t8DJ8FLaoNnw?C`D0
zJ$|zWYHBhvGCeuzA?~S~$JVuSxe27Cq@dS!)!WDCqL`Q%zJW&Z=3JDn-RZF_r}g-8
z29kdF^511WM>IhtzF6k7MyUxFOcl<1bM;IUEBznGnX(&=zq)xt9l6mF_^LVcfSmgq
zUATE!_ZB6dEl=zfOuS)asPWDpu%3;It5`!f!@&iStOct0B+p`?ICx7Wero3V+)I(l
z^%Vl+63d3Y6L+lZu-ol8++>LGCZcHBe9`>gJ)PyQ4^SYYXZ7L{WUrrZe@KdW`g6EE
zY51^HxbB+=Ul_h>-h6FCYk%V}b&Urm$J{zMc_;|)jfGib3Ly(utyL(w-#mj&TL!=e
zoL$<UsiO<GEhz3cjYTRt&&F#jBeOqER1v}!$DXnT2(jcW+9BCh0tLvj3g@q5stb^T
zt<n%w3RC~*h3qAGabe1q<`f^zPZETOhjFHi?#qs+Ja$2tB6=RSz#m}*7EFxmPPX$6
zQzDB)<;qn}OCy}#)t2YqN=iIG^O1Uzo)#AN$tO3J8p`7qM@sN4Jw4w&J*7mx4XJCa
z&uuTLJwPmR#Or>1vzZ}j9vb>dj050$dTp^8B9f!V%CRm{WbD;Pd<$oY<HW@Gkah##
z_wrgls(-6l8PB~_P<xa4QV>c`9ULv6sKSc*sPd(HNRdf)OSGO!+@t0@D){BHR+AH#
zz~U~b{Y27<)4fx7)=7bYPMzmt4B3pniBzZxa_V5!T{$jP4^@F1t>*db{Xq*n4m@w(
zjmV>oGq|ShogGqO3(v`LaYN+?I43BZKH>A<3WO%uX3ZsqOTfxT$H3C`vo`Tg8ts3<
z==u`^i0uW*ERn3Ky0_D|m$z!0HX*f{yHO{Xl$`SIg2=tkO<u)$oHSyHFQB>lch}1E
zvyNC=e=0}q*>rtrY5d;z2FMljv8dOX&2fUFulA;F3X12-9-pV$=O&?{30fWY>DI0o
zRWCl^&aHOoW8K`uYVVZS=82Ae_fGc_AP%eD-JMn0{{;|O!mSj}jf`l%92!i`!*J_U
zB?T0~N0(P1Q+h@_8R;~aISO;0KAnqd25?@Ta~v;taY{nowUd<nr+v!mjO}o(!vC%{
z@e~pQ@;IDOy^XY7ib{u11VdFwpA?z^6KHTnE<H@OhHd>gYEw1s{g11J*PWbXst)C`
zp#BHNUg*xxH9`0WyP^HDXXivqM-md{7t?VP(L#)#CDk`I3-Ixm5Rv}BcKf4qd7ii^
z7?Bb^aGIjsc3akVip3L4R$%QKdA*2riV2wyr9A;*Q12h&>FhgQvDEeO>H3)=E@1h%
zD?|3tE!3`9(?E6t)N(+9@9(ahMMB+BWtYuutKJ8l-N2e5lCXgg`>U0=I%a+6$62xS
z!jyF6m7}J$@RCKzq^~M1cdyBj3|&;efEyGLU}430_wA2P)wu2(<F+>k<;Af_SuQ$G
z`p7?t4PfPEeDz9+!JIDUK2d-L6BA#mWqC^T!O6>ZxeC;nOs|K~VE=MJvwi$1Pw<^G
z_f0sQ)#H{pvYz}b(~pgP8;;9%;}&VJyY+j>tBNB_MDyZ76u<DGWwb`PlUgTr-r(?S
zKZo`RagH;VP}BN;Pe#dW5xKo{O#QqxVli;M49&`yLk??=OxN-8ucBe>g}M_<LD()n
zSlWEq@-H$}j_n@N(aL0M8$L;$cXjVok?8s9rK)Pe|M&c(bqR{^feJEnvoH495ap$*
zt7D_{&chySanW66BAIroVHUng<{Iq1S$x*LTehB&Ym9wyEV6@*+H@pn>=qL;-}p-V
zujs{YYr;s~+4A2!18I~4HC~ePhA!vFgdCnaieFf&to25nxc05s#iX4&_wU(#niuWd
zSQ~RkiVth=6o>eCe|%}n@s0p<bc+Y>MHb0}8NM|$XN__zGtV-?sySnyk(Gt9#&_E;
z3GvZ_zsZTY<H~1zq9S7wbhn<r473?^tQk~Z88)+@mlyb=-!B@eSQFdP^P8nQjBe3-
zZ?yf&z+@3swb3{I7>@py6YL$OALQE48Y<>~A`ee$j3%Cj=^Q0A<E$?PKEbp=PMN6k
z{vFpnss>t*d?-QL{@!(K)`X0-_(VjY!c@(IAvJ+J$+=ohON8!a%SO@A_E3>5U7D?d
z=cZzS)qzDp2;XK$#TeYD|Ln7@W7|~MY>XA=JHzMeqcI`^vJ6qgQ)~*(pkgpZ0$0@w
zh1D<m5L+2YYJ4*f&C^^jv%`sVa~mbY9OCA54s#6LBLVU6fdm0kLp9q~l6dDkk=a%2
zn_;IU&c#<H*lM&GovEv<ar;t?<=<=UEKT(1x!QYHt$7c39?Erk*2s*6QnA>glk3z|
zs56h)r;E;B{7%<hGSsO)ec*v)j33OLXJy{Rzx~LwcdC#?<DjM0jEXoFIjPj7H1^Ce
zT@L$tG@D5<(?LYCv=yPGT`_4NeAa0<#~bHn^@Cf0%PUs2`R`d`r&ARWqV3AQS{Q^K
zTXF`k{HpHR`WhG{w`TlMBOyL%cFf&DxdQe+6J(T^w8xmPjN~tWLhuE9a#D3o4-ciI
ztG&X7&mB9OzQ}-$6hIhFssBhaa`>tf&o2L2orP%q8*54MK&Oi$a!`Yqoo4kPnoql(
zaM)In|5^#ShNt3`q}XIf87~y;@GZwjr^-6asxm7*wd`-(s*@&tnojSDJF@a=Rz-7W
z_H|b*l$RLc=0ZV87Y&Y1pFEz3$j<KV)xmqEA$_l`3BMYiTuP;o?5*xqXT2&C#cq*t
z8pmt4-(Te@KFWb}_3{^9R=EXs|F>O2x8t(J#mDkwRcK!<`l3_QZV!wrsu$k$jTRH7
z&$BlP{#G6;w!6%L)b!9U|CQfR!n?@A&ngr94y{cxjwH6&bKv8Lqm54Im&+e%_tK~N
z$5U~M$%fwgENqh0VYtvjHJRV_?)rhe7#8Vlz=x>C27kis>G~_I(u!pV7jdqAZS8E7
zl>0kvd+)nUP1#jYFDA?8sE`QN?JmdW(RW_#OM932DEl5tr~ScIpNrV)lz&ea^tc4g
z&`0jxk_zi7w9n#Kb?X@7iDtLbnk%auy6GnpIB3}wwsUI-M0k1amLmR~Vg>kH<Wy4*
zoA38~vS(<*z8lm3$b8Vhd__uCW8U-R`S-_?9sKluMNjWwMIvq<$OG-LzT*D1EQWYh
zI=eFVnbwBGBiapKTbdDjPw)1$yQPTcO<GyAt$3yO<4yuc%Q4CP;~IBI%RMU+ntu=W
z_vT8T^kF@E<nrHQ0>7~dU9RmvNu7j)zgx%LIZ6yFB-_J3#C@(8Su`6>e0iwTofNG&
z;7MCl#Ic(Sxu8+@RSD;>q5YJU!&$mpWE|J7eP~@#(Qpb;S>?+>%0`f$`Kp@VzO?ZJ
zMb2ZH>1*Vz-IbW`M4`Uv#hzTC9u;zUi(d5MxTQ3iHF{Xaa{NU{)T)$bjQ!U_$H(~a
zsQhIrcZKW{i+*DH;Ns`gYM$)JVtApJsXr(6Zek2Ud;Eb`c!gt6^p(#$ok_2W$*(7p
z&a-qG6wRxhT>tBTL7`amzKi^iL@RAa&(PIjXI?2pt3YF3)aooh!P$H~l#loZNm`TG
zg#5jm%Z1I>m8mN=e-vxxr6X+yDVGQ+m8V=pdVVns(@o`#5Z@rmC?)YsI>@fhjr;QN
zbF908Kbt#C6HJDquMb}|$p{?pnDLCQ;tpFO-(@QtCR7=W$b9DoPEADL`1ITKlpFPb
z$UXdXbi0qg@}*#v(c2#@ySU>eTh#K0F7?u3DdJ}j#o=f~4r0Q8ufKB{JhmwbaVCqd
zi;DczRCJZ2-?ZSNQZ_yX{oadlT^W<b)dmKKwH|AwSts*G5&Wez9RYI=d+qgOL>w0_
z!GiZq<Nf#Vq*W(tnk8MouUR@WGDxj4>1d4odw@7f4&OG6BO+6LL@hUABTekgGCIt8
zty_9SH^ZpTM%($os66NS`HP1KboXdg&3mqyD>mMc=%L2DwN%%1s`}bp$JF+4ILhnz
z&A{n)k;WUYF3J@H27%7A-PI&yKdxeDXiw;Ofqbc|!M{i0qx!_5zXs=`bg~?+Ym#IM
zs|S}8n0h|+RYezBif7uke@<pM{fXx`JMi92&i|h3@@wl-t8g@)sb`smul%Abuedl!
z5}daj56>brDI~r*o>+eO@5#fcM5^S;Ty#_`b$YL<V~`}9b{>LD<u0y;jNKU2HEjwp
zH)ydkis#Qfm#+Bm{re>wbzH}Tu{l;RCJt@Vht}HimD_I?ow5rT@eShz0uBN)e$pE~
zL_5B<$PqCqbX@;=g98UIv&Z8bciK@6?)z`jikm+c*s?-R<2?mkW|qmbLPw6*mR~J$
zx(e=dZw+n{eABW*lutX$GVt5%b#A$6d!s+4#ViO5EC_ZkoY^C1j=3y06=T#orKmn3
zqRPmKsHpUu9L8(|B@MmD-$kwMB_O_gD2FNxl9G_9h@)GT=Wb6LsHam)1iw0$9sAjK
zeTauuHeRX1SUQ{kHhq)2Z2VwG`76mvon$Y&GA$1mKXfMYZt!C4YlQY63DJ(b#F;;r
z$TA*ZB;7cen6l~93ky_LQ@g)+F`l2&+ZlQ;<=wmM9A<s+rpWngIEB2zW6z`MA^(5$
z$lC(ctE2Do-W3jWU}dRku?^wt=(s^hochR1wm9qO;&ucVmQ#fVXNSmQYk<@#s$y&!
z{UvN<Wf$Qm1<@Fg!_~9BtJZw7d_(=^%a;no9}HjZG2FlZ&P#wiWY7FQysD{5I6mHY
z(N4ISu&{-ICgSE+r`HGED_7oEU90l@I<m5vJ!_(k8h?qrq;EDZ0)0An=Y+uNmbzod
z_m}F7!*!|mTGyi4-=kN!ost98QxZZF-aRKXLtITAQnwQAv^Yf?mK}aweSNYv6X3;U
z?LxKx&Az(h#V;4_b;kx>>*9kU!!J)?%d%aq+wz!n#0oPci#%h*TAOr^?>t+eqL^M-
zP)$EPcTEdxN$8XXJL`M#SVi-Lt4t#mvU&75mJzYGPj2#*m1k*df{82t-$X}<Dj#ut
zV^4m6$zd@kC|1j}(m*wJ!PU{T5N+D$ZJxzI{;!EIJ;c#TjOG`bVwUJ(o5)udlUWtD
zu0cB{RF4*^O5eM+TpeSbwF*ue^Do2iKBzMG<o6PnyRSR)hAnrk1b)!!yE4d0)zMc^
z&48A$7T6cUxpFlp6Z!a`kBDPF!oHP+-JdSvPHjTf-@i3o%q;Nv#|Ajt(}Fp1?X%{)
zb40v5ZQlyYi;)@)$^R(nS*tqBEMM+B|JbRb`tT?)dcS4)&~C*I?{J?D?^IhVc<Fj}
zb_|#DGqnsHEoRELmI1D#YfGSFFt8e)ZpemQt=bg|-$~aL184kiQ!u{xrP;tg+-kT9
zo|0SD>ZuD0?Sq3y*M2X1Vz*Et(hD-tf(KMPW)pcNivO`|!NkdnX5Q@&*lpJx$=M#r
zFmPG-Pw(8?-g@~O6FkqIKi8LHYp}D@rBj!``kAW0(vZxz|2;qxjQ^kXfg6$&8J3#E
zVbtWq=~4ai|F%R9Uw-}Ye$96-P0FMWuta)pSB2e6c(bHrFB~b<eny6Wes%02p&(c1
zyNP@G3-^s2C5QPSG##3M^zZCmRC)}O(#@A~!gSR>Hu&qMCcipIY-qyp)z)ChA~yS1
z6pw7&?XcFlFf_;8fI@D(92J-8H^(Bvv!oOjjzlQ$(CIk3aO8YS)I>9uOnA+CL(B_Y
zB%-DttFj+o=r}Q)R#t=Ze19#w^xWJW-Gc{mpdqi)^wI|h7gt$H2`zzxgA>-?I8p1N
z4qBjScth8}R;L2ygHdqQbpktl@&%fp0B?Bp`ZYwMep|gH5WXrKvyxDF`t)sssa|nq
zCI7_>7k*A8gbr4>Eg}vriw=?saYq(q22Ixh)?_?C##Op<<Hn8QGW&F=)lpMNA|=9J
z_JhPraFI`GbP|qBcsFz8kF_R?zn6YWTcoqSS~rO=G%0W%y%CQ^4c`jQgBL+iY#|<>
zN#pAm-gDmMh94X9AE72^;*nEVCY?&Pzmm_6i#-j~Pef5UC9Ye>W6l$vVJV<uzXZ}|
zCEPc>%k}Et-jI})jY>#J;I#Oq=SYEatc~Td-yU~glL*ssnYxH~leK>O3CiP&UlJ<n
zdo$Lf7rv&^Y5jLliOf%>q<S+I)6o*Snx!h3$M#35F>!IJ)*NpJqBCu08qEfBKEW^h
zF4ED_(K9keJSSV~%M9<gMfG8U`gA|E11-@W!=?J<$s2gvJ^{}EG=miwyHVj-SoFxL
z3N{Lrji#Yl3g#Nmuxa~h+S>=(?i!L867zW6wlh93X{M#5peY61l%llk{111Q)0gvW
zbh^8{RhrJr;l|3OgQ}3@V*~4|9hKnQLe;*6SQ!k@dX*Fu!m1DEZ>@}1X_l3h1%!mC
zgi;Bx8~(iBP+zY`dBZytY-6(M8PXfkI>)<4hN^C2m-YnJw`eYF9slLto*tsxxAXV+
zU7+uuo}N}lp7Z#AFwR0`K2l<dM~fw=${i;o04nv8($Y+}o*Ej_Jv}|!K;bQ&9<2?&
zyCcFVdAT-fJ@bNhr~BkfY=>THE&#ZXC<|?-gxeE@bWcxD-Dcm5JM<&Us&x17KNS@f
z<@MPA@_|uZ-|-2m82!&bkIoJ{k(R2H-`$_1`CSXI*2#fTo%zF?C<f2XW?HzdG7Lz(
z$EE~E)d?=gYd;?ZA5VrRSDl?6m2b8%^ruJ=X3w0<SWtUS>fKp1<U)60T^A$detfV+
zEwNU8ls7$%OcPO1FxEXFx_&(ox_Sv8**`D7cJ3XT8jXGmxypvv(2A3J9V&yP^#fBl
zS~`wb88pA-A+|w_K+ulEYLwgY&oB9~h=>S7qh~0=S1N$xCn<1AnIjyWn509#VaeU-
zWiNiweOF?7<dHZ2Lutipej;Igpc=;7w7rUqX@*6Fhbx+wq4t?E)5BZ$Q(7hn@=1i{
z)ttHo_M$SDK}v4chMVZ-%}i*a1|{|*riWKW1l)Hc&?_UQ`E3uIo14Q@Wa4=HHJDM=
zfomk^*5S;roGw*e&(yD9*&#GPrQ~z!a~#x8hc`j^$)3T=5LH)K$IMs+&0*T}q{4Y^
z;D--xvE>M-kFRf*`CuM7r-fRX_%dppN!RNj6}`ASV!a@6e$s$OIyz-`*0AbeKE`eS
zlL^tSTMR-%LOc}LJa_t4hf6GDhQFH5sAplJ%)*}A1-PIDmyXtI2F>%UR908w_?!pq
zI|Z8D+`RFJYLFT#ul?r;jJ#>v$A{4m4a7dpSivNgj^!?Js>8}qq`2nS7<BWV#Nqx-
zXn*44^MNq4E#VP*M$7DXz|bSV39_>*!74$z@70@kp%sBWSnJdj-*IbAeBiYBP~`Zt
z+wPxCFb{ZUXdyK<wJBWi5OnEw4wv(nS&)OkencLtFYMJT5L;*tr{yqcyc$fwqu@Mh
z>-Xcw4<$Hc0A`5{$RPG5iw5=;o2!Dv-Pm+JDyPuv#7#m0W+a_6*6qj-AJQ<r)VBkc
zhMAc%G9l1^FCMd^s(oL<)T7toTGmQRN+RdBNre9S3|e4v%LBQJ#>O-Vt;RWzCmZsd
zR;8L%?t24Od*jPxOkfdQqc*QWsCnl*_U#v|qh%NcwVVK)TK$cUjj#@?Zv5K49JK;r
zXqu7(B;&4g#Kl%)8ej)@+!T^epLPkIZs3F0k|{E7SNJRzoD+hZ7(atHN6Golx5v2S
zOlns+smES=WQknC$KTVU67+ESWN-<O94*0XGwB^GW0E_2POda1J)Pm+JrSphIKul_
zg7<}lbTNvMXz2t&ZJ2qg{~E!I7a3O72gYMCvMrsRgD^NuGPt#*a_i?F@dU^wJ8aHM
z1cswBm=?T(3&CX?g<QeI+tFecymJdRQK*kJ;J|_Q&+F853D+NEU!!=3<}e?4_Wu2Q
zqmadknwpfNg_FrXag-VPMHW<q!9=xZSiNc%>NWuxS%z_Yv_zN;v}K;muTEZ{A2l%z
zG>2g`fke;b_t#7^=WB5qCMF+V{XF`xp|oslRipVrZK*so>n4a#7TeB<NhlocA}jF5
zL?|gvSJ#?6R?*i_MrVRr=jPsm!zJ(hA(+qs0ReqE>TG}ACbsv*fX>eMA3mt8-hMNf
zqkd1qY^c!4NAv+o2^)-8kdA>t%6_4v|McWYSx#=UOc1&Z2D&%lbemS9CsI`>Gfg^1
zgP73!;-#x(iC?}v$VJqP#I_bXRdsaYQ;Niwb9AaSicC5`q!7^RouA|rJUh0R=*f`(
z%qVlNmR??7-mE_>61@mBcyPSRea@~79e2XN(atsTq9c|k%q15k$9FZnK@YW`*qbhE
z4*H+OA*K`6RXQFUqSdeP_7m&;@XaAYQGYPt9@KUoYry72YO#$#cjS0~BP<16=S6I6
z&DhJ{%FwO3epGHx!As1A+<c(tohB34a-$MuMgilH5c%m-CPrWs#Po(strIr4wlJ<1
zBkuS55?&9ns@C_IhqZoyK1>&TQV&-v)*OC!-4T8ABm;sX`9u9TplJHsksn2(R_P*K
zVB8s>3;nfR&_q?ldkq!H6DLCr!l$5W3Z@iUHzf!2o7Z_Y=azM)yyZwqG4EZ}?Z^MB
zivS+e3E&7%IB$<+q`bz)2UU|+5H>m5T~TS;aMdia*f<u1gDG-DFw{yP4#vI8C4m=O
zY9Q5B($b10=d#L~oYaG%%|Omh-S(&ah#}~m$G-NHT{`>o0B<k*oC7a>F$jn|IhCYG
z#IhPma-?l}L9~ze3=a#l8{b$s74+CRxQqde6SLNi@I#Kv$i&3FFGF6V+QT(-1m+s1
zyI@xTj&mM=k7iQ)^2nemBzNWGj<D_HKXzq|G76<cq$Lee7)mTc4D)1?bc~Ej`8ri!
z%?FvYKf|w4DEdc#|KNIdvL`TS1(v7dxucBnRWMm2YrvU4>y4<J>H@^^)$_=SNxksN
z<HwIl(*LqG6C(e9b|#pMKPo&tyuZwTfuQu3ZxT#EA0HpgP?pbq&@|)wdr0@ASVKG%
z>R3!({{1xy?tdhfhKfuw?Mmsv9SffB8;m)PSO)m}%iGv|fw9lWxH1U|353|H!l;#`
z^GYf7H?>5C(@IZYrlRA)ToeQNXwSp#?h)&n+|kj|E#u~vmYzbRR(8iFC2a^AU|tHC
zYn51LON7(V-MRC9dB8Jw^&}=fJ}*7qO%jG&8U9}Gw2~F2<0g?Vn^2v|pJqyb=g#BL
z;-M-adJC>L1DUGxV-OS^FB)IR;~M=3L|eK)i@TTb+0DNF<5a=x^yhb39`t&=M}v^8
zz)qhlo6-*EoPCrdrTF?elsADz$=T@%7o7hxJg;47=r0yZ#RNWO`Y<AbC^aPoPDUH=
z1t*1pM3a+~3L^>bKbT5Ka-nCIL0y*t1mgh;41akJ(pU30bG0i<o;-HR`|(2<VsA#<
z$nKw^&k(LLZhUF1qO|(-(A>w*Pj&vzVAj%TO`540SP40=!wdh1Bc;}ZyDK9*o2@M7
zy=l@*T}dKku7KAaMl4G)wnUw$Tj^3j!lIM2I#yx6GF)uO8wX$X^awru!#r-OhovLX
zF74^^t+<H1tiJ+B>z5!R!j$?lWU(}m8-q4&kLKWYS=WnPh~;z6^ToTRlI3xn+Ym^)
zJZzrNq&#~vP_b66nKddf_=(M=ugcxo$k0%U-nszq7Uzq}+Lw-}Psf+WYi>j6&eY%5
z0w@FK!l;b2kg!QB-V+PDue=&>4p0dO<s<k7hBgItWzVb7)83^3I8mt7%Jm<(@vWA_
zU%Lrq`I7(~CzL*7lmgLXKo)2T=&$#J$Lk`Lf*xuz>532(U0q!zWn=~tAHPORU|cD{
z6oy!Ph@x=yUI6kAhufxQV~7<o5z!0ivUqB0fW)h13-NXfWYC7wlQm=+y0N9Tb&KLo
zbX;66Jjm~8^St*CC2zIL?B+y;R)-Y`kM=N9$SyA5re|gZjs_PU%-~)D!NHM#f%k;h
zk>#fmS~U7aLjZADM|{fK=!D|)=Sg128{#|5gR%^&NFD~zNm?R{sZlNL(kI)!vhmAa
zhl{kI($uF-ptT&kWSIGu{~S`&h^hDDR1c{m5B>f9AJRVisOPoaEpF`QcIdkBiQ6`r
zt~u!K+Xu3MG*`K!T8*wz@&Af2ECEc`w*p9y+Rg$zdYda`bM4I9<t~ks{4VQ1klWkV
zo7*cRgAs<~Zp*nwAsBqM;8kKdqLO*U1JKwc1l;-W<kEcq6Z6uM_O`Y*)&(B04wKQ>
z_=0NlAE!@Ly1i`6w2Z+mnrNvXxq!%X{D|}FsQRiwwto|*?Jl?>m=QE`+iK^EjE&9t
z{u(>#n;_tWt@(DOcYkB19}^}U8t6DUiraW#6aZ<^O2FQ3)mH6272XQ=`l3!jhQ&YS
z1;lapSK{4Q>Go!=8^nW1jsT6d#--l{Z3BFP?RRL?gUz|f=+9~WeW!7pmeI^r+dZt-
zfsnyK_@|c8EVW{Cbhom~TgVs%+1)IuDCW6&<dcBAQ`ya|Ec$;ZJq;0%#fe&~I<BcI
zIciCN{#XE>|J|>u+rK_lM@&w>bTj<p$BbXUez8h*1qEI0bj!!*U-t(2Y+?$EKL=N?
zURBc5O9amWU!c>>KSCws1sc$VG|#xLSM_7KteKrL_uhtN#~kuYvwkEi5)x$rb$I`3
z`7*PF$)GtCBV*3GV6L3v(OzllsWk>U+Mm@p?yo=ZoZEn+5pG$KMiv$prl(L;S9bkt
zB3v7#ywK4ZM*u$A^=eUgWFSR*x|LuQou;6uxHBJPy#$^JI891krKS){MMx`6D(qX?
z(t8oQ^j_ERd+PE@ZbL><%HML^jTK`!Eiu~TfGR)m%tOT6rs#IN1H<yi>F8o;J|0Ig
zQdd{UFv<yFGi=tbOmfW0%61eyGtL5}7je9fJ&nP)7${V4{<m>cmmz-IbVJ?)4IlU1
z(>__TM#3FR?t{uAA}bY?ow?ceAc#KrV7(5XC2KS{J7=_J?xA$i0G!O@3c=009KbLN
zrPN&SPbd(DK(r+&6%`kEXGTCjnpJx16{2ro&|1|Afn4QdGh|(z@d9BfD!Cew<+(1F
zV~L6Xc&>lP9)^isSX2bqXV$4g;3f%^hQs5dT7(`RC;KiveT)5pM-8n6D6NYJ&thh+
z$3;LasEn}W<hz|l7@uB(0l>@%B$5423Ad4BTc1+viRv-?GMC(beaH?~A=INyPmd4E
zk5()C3yq{8L;`gf$-P-b^Wj>C4CY43roCzLt+*tN88&|ah%xXZLnCDYgXEZSD@X9}
z4Ix@J+m`|2p|--s!NGtyw8UD?Sv3J6VfLJ$37{Up;E}GH5NBy*<NqA2O~hL)k2;zK
z4nAjUO%n0j2CGngXxg5Yd7?@D{P}Z;!|CY7&!K`nokC|?^V|@BAhoLbQPv@VHG)96
z%Vt4iZXXKB8|AIwwzrx)sN*gTKBnd8PvC&6O)Qmr*+I)X7w81c3<8|FYr$nV`%}Tb
z3bN(+R$E)!XjY>DNH-z#nJgQhvs>I=?3I+1?13LF8EO!3vIP%j2|I5jGDk@Xo$h}=
z++E3rY*3;HDIV0(TzaC5p(qwMAvw#fuYUpn(3NV$pOBW_q=TZ^dg3%$Mnfh38YS<~
zzc@|RMZc)!@6)l=92YLwK{i#a_x)Vw8fNU_Kji!_UsUq6)Tj6%9f)rQL^}vMnmNq_
zc*@C!`z~0hM|8Kitooo!vJj9>=dc`B`u_cU{2VR_1pcV2>j!mpChhXRqINKvY=*pV
z$$eT(nk9VzStbwGO2^%#FiDJ@+`FrF!1;o{H-@Hh+gVDbjdv0D!@r&PU?PG}QAtfL
zgKw=8-2L--o^pr9WB^<eVZcsrcOs`neI-Da+0Xt);3<hnNz;J}?=%Pu2<Vw@4hOO=
zg-$U|$#^eP4tO^((8op_aH$!(*PVNcIW)W0%7DQg$=P4LLuY>ZA_r)C>o^{+4EzsY
zu12vWj0XlUpqq^uPD<DTM*(#GfM(LI)`AXUy;@g<AMdu5N$v?X3kb@a>u-Ej-mNE;
zeKC-uJ~!?KW@SFt5*hIJZN`VA7#LYhXt1-k?n-<a1}G3xT2^ZW0}30StG{~n>Z{|j
z%E{?oO}^bgpt9DbgtY~&{wgH91^8;slAOLxvW2Bz1)o2#=&tgx^7Hdow%IOF-O<T$
zx&2Km4_Uf`@n9|wk!--)-sQBkwBmeDf8>CL8pzd5hlZPh^Tu#A;2||it#ToBz4-a=
z4#1uaat<?xai09DT@52+V@Pn64+m>bQ;W?9F(1LXdbN)_N-G(d3)5O}EKCLsfmUTo
z%lG+jS>!MrlSio+86%5<EY)3L1pT?1T)~ulD)8(H(6e%@yD%k&3ylIvn6)zCjivz9
z=;`Sd5(GWj?Pi}!N3%;zO-&600V2A_m#wsLntKg^GeGw;a6GBNCaPwu<SIa{$J_|g
zSJOYg3IKU*#a?IO<Sc1(#h4j*Si}sN1&soIAHaHwOj3Vk6nA`{licP6$0z4!IbJzr
zPf#l2cP$fCv`J`%p(>(Sijvpi9U24Gxh6b{fFFc^Yjp$wzEK&DK#N2J7JyTPw-*hx
zG1#Ft5OP-tYQwo&WeUjI(ZtSHr%>yZ4dg1H(vZy9SPBSn8IXNqa46ao!css`&{u8@
z%$<w_1`K}oGsCdR3Lc~YZ`O>{bO7?>^KGAsBXBUcM7S~D8TH|9NXP(SP)soYH}}v9
zya%?1l3Vcwn+U$ewS?*FrT*+<`vsXe9(ysMBA3`9SBD<>41S9q2<4>=xRSRj+p}Kw
z+-sP;Mt$Zhrivs(;>lw;%O2KnUH;`irJ`(r#U?pHb`1Ba#kJc<F@Ka0m3fjUtoQ2w
E0hZp2NdN!<

diff --git a/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-46-1.png b/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-46-1.png
index 390d696c843063ca4ceef468fafba0b967d4f477..eaee34f8a40cb7d7b3a8bd23753d28c45e397eae 100644
GIT binary patch
delta 92007
zcmXtfWmr{R*Dgwl2q*|hi*yLmogzr5AYIbk%~BDN?oMf>K{__w-QC^YaVF1uzJFZ0
zFIao5ImWnaHnkzAwICN;x<Xdu2;qamR(~XlhTD9e`=>d}o3P8t(SOD|#CnA&B6vhw
zFL6L5yvY|yo^hc?8FzKh8A>M{-bMZcZpVX|1S8JlTs515#q3;F{jM1PrSLOSk*O^Q
z?uLZJ=P@!Qty7gK9s;O;9DN^N$7Kx1s>xN(e7%<94E8c2k|Cb&4$+2S-daVBwgiTT
z(WCje2#rey2Xdh?n%kU=d-0Nbm*PsMZ2B-Y<!`mP^S3XeB-*xuv?GY2g3%d2m`z}h
zn`iUvaQyv8e)>|)iTf67`=ehYTcLk54Qn~J`0hU>J8&;>RHN#wef4JlxIS_<R~{Ry
z-cpDFhlKW!HfqD)y}t_`l&AN|nMgTrpHcjS>{P?LG~qZNlzH!lw6-rUL+f^xER*Jb
zMg{TJYZIQOHZv&btYjO%9F04@G)ug4lgOQzz&7T(y?J)(N(8S>ptAbx40-Rc@QZ})
zga#UxS@X^PCn~2W_>)b$HG>mLr=#SQAzPAqvjxbw)=c=7+d_TN^?UJP-i|Axg(%m%
zpSK$5jBd!^H+-c<^!3!{NuQJcdMV8NI(!iu6qXo%X1sS}vf*s9=6b}Vm)*<k_jEPl
zpvSIb{x09t$%Pf+Y(w4TdBpmjV8M9k8=U5bPeo}x11y~FzgY;lt|#6IeuAmYi|H%d
zhv)u<_^w~zOKUkov^?vl6g+J0i+dSuS_TsscSG35E>{BDD~qdI-_%*`BopWAp77nj
z*Usu&OhF^ARxOto;C}Idr9?Z5A8{k|=7KYCSv}9X)oEtJlj$<dB}(mMnNhZ4_DJn_
zVmGmgW|V)#6&&*q=ys@dGKDFv<zz5|x@0Yt#-#<ZmmVe@5WlCx*4c9B^T1*2TydM;
zI98>rjw(3k?#7yON?PY^ZhJcl3MQXwZM3|qw~|7c3P<a{3;KGdqKEz=&rntM4{K_k
zzDM@-NV?n+JHjv*t$bqjMJEGP>(@(+XmV%om%_irG-Mnpp?(!n#R_N#qt`-$gQYG?
zAA7o?;mvmeXkEJMAj910tJ?J86^6QC=Ca}8^NQYGDK!n_k~`eh4O`cdZ;t(2zFI1R
z1xW8*99B;;SMZrlxzW_^DJZs|5;R!K^se!>6{SvK;grUsMR;{e1ZC+Lk)9GO-F6O2
zW@BKwE^&?r&;KK4hIPu<emXK4Ola&U8^T^Kb6--)gw!tND-E|^{Bz;Dti%t5dQ>Fp
zJqH8IY!083bH9sRx#3U}{FSLP1|v>y(D0YY)LUq7LMXx)^RzT9+4)BKsrc?@USkUO
zQr*pnKdGMB4mA(?wat<8{WaQ@>egTf(R8G(0yFl9zueBTcE^Xx{YVFep>HJV-WGF_
zQPK+*O+Y1P^MW!34?mw;$dv44P|fP><XdV8Tq6fFBnLh?qETjLGB!P)!e{cT;y+bD
zU7cGI4;BO^^j>@+xacT%4VCv$LfQ2fD>xSs`(?X+S*SO|{%pJXGR{mscdS*5uW=j2
zf%9-j?p?n*ITaszy_PRKhgZh$5zsx)^8^Z@Po427n`f^0NDjK&X&;f=GKOLgyYe0W
z?HpmobyZICrjyMWI#HbtDlkVhgQH2)IY7a8j~{b!-`h@8Fm<*fcwDNmdC`4ScXO5Y
z&g@{`#)!Gb$AeN&&rn05Q1<{OHdeL7Z3Xk#R#LlSSw3R~yVzszsS+)l^tFeH5{W5P
zjv{?cCw^p+=gBZ|*q!ku`$u3xH4dsILBC~k`(Ig8YtuVi^4S-n2}ww<d|Y@aBFPKw
zOA8Mz;J;qFSfYM$H)A@R_*kVvQzO#m6<O4$zfto^=dVu;8o>~Rp~&_^d3>&-<~T7u
zxZdaRo&(Y|G;HcRi}-qGA}wPTFQo@vivQ!<+C1a79bTz&IJZJ)z-!-@HG5^}g3-W+
z|G{DC&rJF`IK`m-NtnCh!4F5jc^YK#?lu>W4OV#Ma}TYKn3-ZOY~2^GF^ys{OebQl
z2rQYG`p#QL=T^6y+BEp8E*(`&PhICU_oSuzoE=7AxW*h0Ie^X<h0b^&X|;=QQ+1DD
zOLX<h!kg)&NW_$a5)SSyoVbXPQhXEWPjKj2<16;Q=q+$-K50T`|D&8BS3h-<>A#ag
zZ?19pcP8jC!i>w!Jw|B8bJROoK*m&okeF>ocSWIZ)3c7?w0X2Snx(@jJR(WeWP2Q`
ze5WVhfb{mf#TorT4Qd0`;H>t<v_3eqFf|l{sR$S@?xg4#SX_j@YnZr-xQQI96+&NB
z@=+_ZNYXh+lOddy5^AUz9xbq#*+X=P-#(eaSmA82f+sbjd|7Jv)cSynx$<Au0^JJU
zht+fmyC9Uu(;G=8e)u>{L#h%~2H+Ou%QWMzk8;L{zeB}?Wk}^tHWy3TXA>y()`{9i
z^|G67I-noHOHrc5HZ4b@s-(4t3_0O_gE_bvq9i`uWNZI%h}qzDaYpU9uvlCl8zH9>
zNmn)l>!($q7*A;LdiZS)i)nd#s%OM<`e#{bum))il@0Qq?8uk#Em2Ki;<%pseLkG8
zUY~_}ocm0qdTmauK7sPLiu_V2Fl;$~R;1yqT+k#oqJn92_f+g6f2sFKle|?lG%opP
zdKvQmms{iEEUNe1r;HW-CoQ%o-6jTqJn<s>qw#Z0q|d!sm-U;E4$m3gtk+sQ7(-<Z
z3pUoHMhNH1qV1^8L*I~yJBs3?_wPGjA`C+R`bH4S4}6ondRovQ+xul&^XH!vpQom#
zGBPrFcRvJ1N9Z3Ff6TG}r}qBp>I#GB#nH>H2PRCpys@+~98F&(-urzG3@)y=3CH{w
z;h1T1W>O`^Ht!V$qS3V^pd<OrX)7z(wbZSydWl8%4kx?uP!iI6e`3wVm~6xswN8+J
zdh?TF$9#qLs(ee5+N#*!V$1O~d6%)z;PMH(4<_40=33KD3z#od<~8bXH}8gTPCd6g
zPF?w2>xs*+_i2ryjuNR_=M``?y8n#lmob}iu+zRkjwww!IbN49$GP)ymx7U!;7(=L
z4Gak^uqd^kFcQ<(YcZx!B*W3rv@Jk8+;b1fgq_{p?|5pDBQ3i7`s(TUNn2OMFzZAW
za+Qo(eS%*84ZOc=S`~NPK&xS%mJ05PkjfsJJNqGB=u4$8B=kh?XyOMuqwj1YeZB>e
zWInF?>!5ao`ohKYPSuqM_AS(7)EhU<LI?|M%_6igb(f^`tmd|vKZwr}@k$LUWgaQ6
zaQNW-`nbh*gXxavMia*=80P1vnhD$AzfF?MRT5n&!tfWhyRq>6(QxVGBj-t^r)R3R
zH4vtpASt7a#9k4cL;GQz2O<-kXMFY?V(ctGbf()nd4LIsDP`^aVQzc(E9Xu4B3rn(
zH)2Ia1$cw$ab2xV-k^w6Q_xgwZ0u{#NnF{iy1Ke%s|smpX_75Q8<DBC_4TJZ%Kfkp
z<t|*iGsoJ&d#XL&ipZUxC7(v(mxH-C^v{Bko0snm2~5!I#@_bm9UDz>ek65%l6zqA
zdlqrLsIArQFZ(x|c-1b-h;LbF>UurqxEvYw`H@I?>Sa2P?g%*R>>zBT-~)0l((Ch0
zIkdv$23C_fuZlS>^5P>!gW#gNYFFwArKp|~N#)?A1eGxfXl`v|BO)foH!e;!Q}(dN
zb@pEKWsLfH|8f-nEHCW>7cF(?#$Yfu;Z(iP38|a7v_E_#LZxq|fb<NnbZs)@igvb1
zwIv4?RrZHC#i?LFXXtRXtTm!(yyDW0kB%-NKc8N(3pX|)LEv}s3%%=S?m|?*s0pr6
z$scjp+nb8prr`qtN+uZ>d$j_lw)RYXN<-yunIc#?uF+tS<f}3Vp@TU;or5>Y6|Ld;
zN3{N}r3Tlh(Kez{W0uj)%QOaa<hO5&tA{*zB?uGwtZgnJQqp5W)!d=4Q*QWkDyw_?
zD(Qu~9sA8)cl3!&;RGr&Pr5VYKr6+na)%~_37W4uuoqp1_0f4|=mr$lT`gTG<too?
zUUsVo)d<0fXimy!d4he?s>u@wFsl|nTQ?c2%}kP=$Y<KRWeaz)T(xI?v~}9F5m?CV
zCC1}l)1c-eYJtI=#<3UqtgkbS{a(~ceVa5Ge_%?8PD)-1x`5@`-@N^A{S%Wn(SaY7
zihure$LS+%S?Gb_8ncD0Oq~dS80`fcOFgJy@x4EP&hb;l+H*$A(MqHFDzx66*^IB@
z$yvajK#rz%MM3hw7(7fN{me|KwNtoWt78oM7!Y;OeM)zB9r6y0){}FOsXBLtKghTv
zizX|SKP5w8GP_arlapNXwYmwIGgeysG2RQo!r63<nQAQCjsyWt^~p}sr_oXP(wAKh
zi8IVeBVv^&OQCPuW$wKUWzX1Y+mF{*RB)Y&wp)Et$Z3{e%kT#UZf{(dJOFH*E)MNH
ziH+^1?mSjXw|t2%v!9U|;-x;dXknruHiIQtlzpXGPxo(C$OJ%?d$mJEsP94eN(i3k
z&Ega3-6-*;{oj4w@Z!bq;>l{2?T$u0+xx!i4|>o(7VL;Rm+)X1{PepcQme?Q>7Ub;
z+Z9Ia4Cjc)E_+Sj?02WTWG4#9>B;V_*>LehHZRW__gFkSIm0$3*8Dbj(`R~>LpfiZ
z{$7s1JKy$fmM!9-?H~LoDc;}^jql)<^92?q@4@Tu%J;xPTOMDQkR295ta6Qh>iSb9
z=%Y%gY!U9ljrQEgJ9Eu5tdndQDlJ<9@`Ff<Y{(KW>t^99!}q9qs&XEvu8&1qJFRJC
zrK?}UsQcy)>~WPX{d_puD1Uy&Af&}he~h6WNdcPxcAg-`ohSs&6ruDUHvT8zZ)!0H
zBh(Y)_;@j8I*%ML4Nxo5Mh&Y_9Btf+1O6ink7e2zC}_N5?_`VHI7f)&+Po}!_>Ul0
z{K)-YW}-Rl<PoJqFGEoOx0<nyCw#Q0YOamw%{8}9w&d%c)<#za$3?*t&1=c*p<6|S
zMYl#?+(Z=h(UsPV>Nkgmjlgjh8&jLu{K@oZqFH7Yi>`-n%uCNwxt>m2%atgfw%>W6
zbeYQCy*^U4zxqV7aZ_F8#=zmFvbeY^vk=pPGjrZu4W-k^-V>V!q0EU|Whvh*Q3O{g
zC&Uh&##tGkb@STSJk5e`3H@xsGW|=^?6)2$P0M|9P<c4(m+}+}@+Pp0oth*;|Lbln
z3in;Vr=#wsQa=z%x3<lD>Km@>V85l@KhZLw@>W{1W4<W*lDx?avkoHb!FA6iw=+T%
z<}sp%wp60M@kOtibgYhTORo&g1xlF}NIuv7kHP%>AA@-wo)pzT%(Viyd6^Og*dIY%
zp+sg@Tk3&L(a_`EG1smwo;(-YyBTsDq_!;!pW-bbI^qb%`<JvSSvz%XL3Mmm`iAuu
zb(`J9$~rJ4V4}S@!?jPeRcmel8lRf$>iSIWLQ$U#&XCP2jBF@2_IT(t>Z|Gu_K3zn
zQCYi!T&V0IYS~f#NVO8v`2L}zZu3LYRfI&j<tNp8XIcOXX`|vJcBpLh#+*-Ueqv6H
z7-pB)?ZW}cFPy0m5gp#SRwhO)*eXkmDrv@_-0&QW#qN^p#r<zqb8^Vam0zj~A$vGy
zTRTLT#8Z2z1V8Pqc2C+=CG7K6BgxZzG;LLHg@CGZEEJ}t{2LEL?^f#n+UvvT|L920
z+nEXax9_=#VW~pSqvk5xXJhz^YIT=HaCErRqD6G~OqUa)&}03>qwxC{6lbizP1Sy*
z^j@g6P}87BhHzj_<8hjueNt%?s0T;01XvP8D=Ur7!|}4PneSGD9(oK1x_zyJ4Eu>*
z>{NHj6|VKck40tOqXB)^(%bv<0(QmdfM5RJkzhk+cWF58Bd7CNCJ;Q{U>ij|iC`-G
zXVh{1DN674!L>mdobjm9yR2x(#3|k=KotSthpg;tZr!FgHVkYG9vJhl@)=%bbLec~
z2=-;qBb?r{uI(Wc-bJJv*5kNyt}*t~)9czG1oCG~)n(PnW^vTQv9~rSWRsF(gC$Jq
zUinIrZU&j;(t~ie<mcvL-rykmG<Q@3JoaG)V{FdXN^=~pskk-b^hY|;+VVCcAB7b9
z^%8rvbDs?k!I7!KF|*7{an_9Y)9%=v1|O6^DK^E1k2QaDt8#0~W(Zh<#w=e<K<&oz
z>c~sv0m0saq|UC-Z@G@?Y2<O?0E$^GA6gjBTp$qbxa72UJTU}>t~cczy_VzsS|=oC
zYu<)jbye9uo{5jIh+3+U=2z=j1r;k7g=oJ`Z}{pDC1p{ZL*FJ$;kNBvaii1}g?DaP
zDjl0)XD>nZKOWwr*^^uzhNn$1CXrL|z?)_NW!*mE8jFWDHh6o9zu;*)f7tP^F2}Nb
zVo_{+D6Eo7!9tTbnTJe0xDK~Cvx+xeSM&X;1$xZIoz!Ky-KIVFXw)}MsHj6)?#5F}
z-9Gk!pFnD|t?a}8yU7IMR5Z?nAIiu~)z<v*$5D*o(`U|pODh}LJj{1OJgzrRwD7zV
z3_WtMG3<=jKJVX?_Y=+t$=y)tD~LbR4c__xzbxS_v>0Og37ebS5o5N647Ph}n?@en
zZfbFkVx0C%t`c-g%<7+AgP}Kfz{2%UAF<?Bwzxm1he|DNL%H$Dnagz@D-r=u2Xos+
zRdI2t?kcY2CZF-^9nTc;!ynSE8uK4}*B;6N(?($T^zZl3&i+``l!GQUp3GB%43~mb
z-~BpW_Aa}+$`0Mg%Sm7q;h-NaAkrD17mXGhJ=tNCWeIu%2BNjg!c#rQ1B&HSpW1OI
zw#ugLEy2g#ngH!Ir|_s)NC6?P-*-3Dvn&|HULcu#%#{7xI0@BM<r;=#{+)L~HhL;g
zqPq%#q!~}V<F+_Vg&%;NsgM5ujKN>^@_Qc7897_UDHL2R;8)gEOYt9$_}%f}aD#+@
zKMxIm<E53|GTuiL_18Qdb8prZF7%DnrT*2qjO7bFuArs^X$tmZ?jxq9Zy6L|)c(x#
zWwg{9>wZ15(p%(+)$U#&_|yHp?nFNp0k#i2BeD?(qx;>Cp1#VpSPRyQ<sT-dj_ILV
zsgkMcg$5jATR)%4PJ+gVLTR~eQ5DGQ{^?i(Oi;^_HkT=;OD0GZ;JV>?EaW0VR?78@
zQ-cjHuhzqeU)nrWDD%&=l+88|Pjvw5jSc+wEv>mIH1TCnZXA~kgO;;Ok(XcYR15k^
z!?A6dxU4jSlbylvztSANCxUvr1-8GX9?8}6QUpQ~Xg~QU0QbXvD5r*QR{9L1B}k_5
zOft1pG8LAV_CkCsm$2$kyxUt}FMPjtz8pnpY;8qgDI2DO0m!ZzXA-N<<JCR6(T(j?
zn7!LOtc|)xLq~=Ur`U6&y<ix9UWLzTd)LpONkQ<TI*FQt-7$DHe_~nJSGPDZ|Aj%*
zbQLC*Q;EWLXMas)e+Bg4EJ**1`M7aJ5Us6kk>B#XujutK$fjY_CQ6*^nVm>y;A0C5
z^-8FAIb#(+lFu2#Pt>SC1Y+BY@?4Jvb57l?c|Y+&8z2Ybzk>u5Vgk|wkjI{Gj{trE
zKKNo^5_~PNawYWWxi$Gs%a5#J`=bx_b{<ipN|z9}FuElrP`XA}IrH=Pz|_dnDhnR9
zoWu`mKdd6u<o$76-EgoQ;ajO4cY_)ANO-b?%~!U)SI9fW<mEy`(gUb~P%;Q;>?Bv!
zo7tG4dMw&<e2>zK-5;6M9+N!PQT2?EE%{qQ@c)0uVms@m;pyPfSV?0$7U%<8B~wjN
zYq8FED%aPR_9E0!Z$f7iQP0qh{e{J&(73_>(PgP%02M1VGb%2*d7ENn@^pgGwu~jf
ze4@$$L(64@08B^r9BtVYxNxjA@#g~Gl58tx36!(+bLFWMJ6s<2Jg|pLnh6Gk1=!H7
zIC!L^UbB_WtKDZ(ef!Y+^h2vLQ`SX5TF8BpK5(ADdrH3KNW!*LNvp1%HvC!K-{7jb
zXtKnSbh>W#rUJD3zirmg7%}%Cm9cn=^IT=K!<=O?@Z999KP2%a#5LsiEo2YiR&E|&
z?5`Y={~JV6YSIzoe4JGtT$TOiX|5-@f!kt5bZ?>CRc^TWfBUM*r2NbkMXHShBlCaW
zYNJ+A4o)NDPyWo~+B|)kf=X(#N+NE!vz;<kyXMsHUY<W1L8Zy<5o$t$PEJ9CQ|^cN
zh2Sh*qJDp}MoCQHR7lyB)kXya)6o`<3d3S)_0Rjd%W?8#682Y7HJ9<JKyz9J4ChzU
z*Dp{Og4Nef(@a=}Y|Nx`LQG3tSIbnr2@TviarvPIFWiQybV;$p$Akz`wO(uhDBfKk
zTY4t3MH89fGJCFb^LqHpx}UjQB0#-2ft@EagX<akrn<A8TbBeKof5&V&BH#l4HQ&Y
zB5O{|jg*F-*Wn7Ht-|noNq_wKs^YvvGa+)Lfvi6Qk-dM2=r~K=&|2mnf{6!$1iuKQ
zL{M5J8;+zbwEQ67JkZtn?z*Uh2`7zK74ir_NQn{ur^?fQzDX{!w!X2IvAT^nl}`DT
z$>6cqIoJXVOSblEY?ocy-co-euWZ@lnb_F<DI%(MP{+!79E<+Ij<B{?$Q0tIul>|N
z{U)n!A*HfoApNGX5^cJ_#A7G!eyHsAs@V}Lz6j9M_4bs`k{vyTyo~Rgr5FzG;;`}r
zmIkgOepfpLg=y2%uyc)7%%I+0WDisUjO$Pf3-JzG3dvdsdxIl<nR_<LRWtk{N3z7?
z#J23pT||W;iT^i%|3ZxYP?lQR#(vUa(UZ|V_GqBX65iRpeid_Z3@B8?y$c5z_4K6>
z0T0z|o)Q`RG15k5Uz+^FfMmS@TsFJ=6Kb-wwsfnF>rcIRrf+~Gb6rJuk?*xtk5*gN
z2-z2N<If#b#tPhNF6%Wr1L<7v#%kfjfTF3E_ql7$kLp(u1rTC(<Jyb5zvHbi$C2S8
z#>d*qqeR&}S9@&Q*O^m_^|`7rimtE?&bteaOf3O2rohE0k|&4VJek@L@|7D03z4gZ
z<tS1G>Er?XF|+k8^9Z6!UVdy%B7L_e2e#{z{g{=H^jtNiK}4t_8prEC<lKs(*2?w*
zJ;;yi2b3f>>~^c0PXq+b3#<I=^2?JIt+o(`JKsNj;v0j*OPgmGj4bkzw{`SDnnh|n
zX`$4!)cx{Jy<TeoL%X3Va|GGu+(J0TuSv=3zPY|@4zfNr`FwHVBnKRjZ&_3<F6P?9
z;VS__TS<N}IBUZbCCnCKiE;tBKfcknJG+JuD*#w`yD<tyjI=R8vAD81wmvo0afho@
zxx4S0<DnltRKSiUe#TCjugUF2wBrv$HyuS&5M8HQZMrQUMvjZdu(O0aJm!gzBy{<&
zdPg~-PHg`TbNO7Q+l`16rpn^w)#-t%QmY?LJBGkASh<;B&FC1pi(48sfkrEyk`=gv
zM~o9VBa+Kq9Dy@7HoEMV<!Z_?-LyySz0#weu1kEEuzlvb?#UK7yDNo%c9#eoj3^8I
z#-Un%O;O9n)?R-B&+EY7JGlN955rJHb6+05E@>Xu^@j0j%~P+6^u9iw0a?b-H=g0#
zLDR$F54TkaR5%gcRNh0lW@1(~!Lz@jymYaebH^zE5uWmm#q)^?OL0mzJmgHCpCveC
z$oE*jHny8&%4VdI-$0v27i`@^Bh)yig!7ue$E4d)NYs|!S~WX~J!#w7e15z%XF6E&
znrw7us@aH5s6ekNFboK^xHGpv&S)8*M**0bBut_U@ngnX3$zhKN*NPPggp3nM6gBw
z^?yTi_5TmeR8t_qU*|2z=dlrLT;m|e)QQWuY7EhBp3U->grjM5P;ng-o+UWxP5Lef
zRDGxO`x1~1s2Y0xTx#6MEuPJNr1ylMe+TLGbf@^6JDMMG7bXk%q&vampIzmTCfFRC
zh@?aODSsxmE?2Vs+A7NJtd}BGcS-<|Uhd~IzbT{wkR(8pYJZxPUE1VhYg~}C^UcZ8
z>6h489KPyEt?ZSH>J+#^(?eeRJxSGUrOdRx(5qN=u2Q#=Va$keEZ9Pz8@52H!n>&t
z)rc_{ePY=?jHUH}$7AKSm#z4Ef>IenV^^w3=tSBzZ$4#~>CZ8?ifXBt?w#8*dLRew
zrl#<?aH-~hWv!mf0)Xq%x!qK%k6y<H!QBIKd0!xmAeH>Pm*Qf*oi%^bb~!%Fc0eY@
z_|2nN`O7)-Gny7?;k2LSyPV&}?JAk>RyL|iAySUuJ3p(DAI=jyKZ5c|C?TMqWw*a{
zy88fdXt?{%3%I14hW<kv0i1+EH^MD@#NDeK4!OKV1FIDt{AGAC#mO=g8V@aej|IwF
z4u&VGHVXp*?{(!8xsw)FcB8il7iyDxPhaO-1ju~8K^ir>4o`Py$zlw{tteYBtGQ&|
z43cf(=X4%#@9hptJarDYgpw%fzo7vG1x~I~+>JlJ0h7mfSxnSN>&5zbP(YQEnUCz9
zmgrsx&><XZgDAf3iJV)gw)c?v%T(o#nR;qC#GzO@tuE)n6RL410s^gAe{}+AsBe@P
zE-)tX{FZ<w*B-P&2M8uYl>a<q@Ny7F0}HLFeD!Q~$QuP8=;kbwwwN^9L;Zdc9M6yG
zG}rnPbCi2mhD)AV?t>{AcVX6?h|}D{p=8c+bOxH?<<=A*<3jfokl=v)Z*&?1)73T5
zcTaq*`OB!#|LFzDsh%e?y{axhqF_>nJ%3Z4@%{S+l%xHb3c&hbcse}~QgeXbiN$$q
zJ%g0DVeG~5*?OkV(d3QzvwhW`SL8tN;H>>py|d<SK@8AVOKb1*16RJF`9Qxo?0Ga-
zo4rG;MDEfccFh#Vih2M;dNpvGvjkv9W6Qa}Qq;sQa~&=$c6QhMQY&GpAB#xaDC_p0
z@fV|f2kVioYP9uE$?Tk-)Qk$#T}KfrLD7OM_d{8rpB)G;M^(L@^UdhcUg=#><UF~E
zys2<1Qb@X>e4k~x;waidMH}*O)fqGq5`&>!$M|LNuJoTDL~*2D8r&YgYoz^Q{7;`I
z9NlSee7%`@!G3#0RPBbTE4%D5<NBqbtB0&1>&u2wO6NlEe+TsTKRXlp<t#{;IE#0_
zbWz87iz<w6u?~<4z?NYN5#C+_$d5YrZtlEN-q4|+jIfB$9D}-CZD9JYk{}Vm?>-9d
z<!wHc`PR$Nwsxh~Csy9=y`@qKwkKv?<t?yR*q&{v=j%AiyeHF~t{C`jQ!BNc9_Gj>
zOs~FWQ0X6R`X{J3KmdD*@A(v<N$k$Q!p<^Z7%TFYzp?SHcYEJUzvuV7_NP`6$ENG<
zF!V<A8iQ>05Qm*HrKf-Zw@XS^s9;D(f9G?M>-$ZCOAmK={74#h`D&3*8lQ?L=q}7u
zA&aq!AE8QsrbNarB0lo-wO0`?>*Y3&FVC*Y?X!OfIYh^ThD_ze>;BYudfmhnfjtIg
z>~RMlef=L%Ir=^|$u(TqAqfoG;TH5v1qtcEFF4)It{3Q3nRTBe^rXUOXdIHJZ>mPb
z#IZ!~65^^Uq-W04Ow?DpX>Mnd?&weWbCteMyM^N0#qIX8=9}-LX_brT5wn_%rJK*-
zLLq^XJQsQTZSzHKCI3n@f&a>E^JsIDCV%}Oz!J&SvotB7U)}7US9N>>hZB7fjMPw;
zbUWM2Y$ciyA;0SQ;uV%l&<?KfXNkb}zL6Lpjsho+-a%lkMz$I+M~SxgeZEA!C0PEK
zvDONh`oN*GwmoZ}>bvnHj)SzOz~Xp55aySQCsU}!S*=WS##QPpcwuG`9u+65<V+3p
z%C9qA%X{kZfQS1MMgR9~I{)`<%d6X&^df7K^<|rhwKsJGj>F4^Q7$Z7C=&s*GV;sm
zEwMLe9~hFjKU-R!^A)L2O^*DNmYu&}H3DSk#rBB|gy;XS%8c`dosBaZ>KW)G=V3$Q
z5mRBM^a<(ZwKJyzkq@fXL7Y;Sk$u;`uCai1*|R?r0c1x<sI1)Vy1eyTvYmhWcE4Vr
zQejw8aNO)OKGm4@FrIH<^w1*oHX6f6I#YG-6I0o}yDm04v7KyiBt@>cYs+Nu9o*TM
zF}|6AOeA*t2ZmRwUm-0N{WnC`KKaW(dZuX#xg*6v%>ExZ5l;;^sq^6aGxZJhL#zbv
z+_Wmno!Wu12tqI!ZTQMOmDFKI9PlrV1?S|xKirr<Yq?S0I2|GaBJdnxJeQ!Gk+C_(
zdDb_Bj8bSP<;B)&ufKK!d(J|ms4NY1lsKa|8$65SV1w+c`J)NzPg}<?maQa9cm6a~
z<Vr>=9!B`57E3f##(U51$H^%~N*vAx0$oe^1!Oz}5_d~Om~ngp00{PL;W<3Z(IT8G
zkJr#-AFcNYR^$4af{jx0${bh*PALT~&3I%(>rsjS&I+}V)sb<lLZ9GWwIf40$idlg
zFDEZ)R&H|Pnr=9l^#rAHq@a*TQKNE%`oaV}y(vbap+>Fq&jrsn12Z0XljXWcE$FI`
z6Ib>P-v87LiMIZsC;)D3msUMFLo>)z14}5-kIz`mX7W1abHJ#kTE0eP*uM|MVe~W_
zHN><ern+xu0wUXAg?htuy_xlm4W{$Vb`cF&s?_JPG3?~CU#7}VvT~UtbSPyHq6*f$
z-U92=tJ~H3?#yl;C3hSaHvl;*S|TuJPn9VGio{}?!muBYDhR|Zii&#M^Jfcn3n8N!
zoQJHwP5guHujtvFIHZP}$DL21PbD6GPlX3aHbxHEp~c;G3#GtjYam|gZu`P_N=ZJk
zl6<pVK<?1@z&1|Cgi+C&uKubm3$|{3cFWX0pyulp;TGnRH-`Un`b>A@T3c6<uuf-n
zb0)p(Wr~r#OSpfj>yeUY%8eg$m<Fm=p!U`OWeLRItRgy?A2ev6H3rdtef8<O5MkXN
zVv01wmHn{m^$^iD_ut?V?BqcQl@(Uv9;joNQT_T_>uXGzQux0^vDd(pcv9r4@vTgY
zvCg;J8$0H_myF_CFP9^IWX*23sKlEoMhRwrJID@B-u!FBac_*fKi9DES~yee(K7F@
za$OC$*uPv25WHQ3d;R3S#ctvzbl+=zY{$Y?avh;?fbobc!NLv(uxZ-fqhYdQXi4SD
zR!NqkdS(O-hy51=!v&?&6GEw|)rB-4xq)cl&mE2BXsv!Tc!Qn&;ZatUA|6o}-$y_E
zj)<1(by5$~WJWB`)&91`W>eN{v+`Midn(JH=kL7)lrZC<PeQ_w)|OZASW~k#+#?FR
zKBMc5jyk#A^F7+^Q#v$3yZjr?%zU?Sk@x(t>}LqPj_cdWxZ{gsl1IRZL1z9~C*B|b
zL#w|fvc-3jSxu+V-r4C~vr$ujD)7-Da(6)-{%5~+fdn)L>e5`VGGq{A5Q|Ca=;!hP
zy8j2royV20;~rmfcMM+XtVj%`==?1j%=f8&X}v%)L+dTfBezBy>OyQNL^Q>&9e?N^
z;xh}dxW6s-o1Esc1Cq7IG|RwX{gby8*Y9p{(2*9)<yk&3vC)-U2nq3z@Z;x@qo*AG
z8uHXJZB&;;R-`}~ZwxNZGy3{rrCy<p`b9R>*JIF)4By$+t%O<mkc)&5PbqK5<Yfhq
z8c`PjyD6SrpC8*9)L)#D@beLmk>ps4m%&)j@;Wr%X4}&l%sE0{YL8N+k7ZKe(uy#n
zTYK|jXb{~H>1UKrR6xU0YZHu)UGA|+&;GAQpEDJh)^`l5T)y~vi63rJK%ZGs8;mp<
z0{Jj-he3d>y%+<=XcP70W5)<s`x$NCr8c&x;Z{lSZNSEnI=^H76~MEhW3kf-0!{%*
zf%J@wzU8n1rTZg!IA`_9I>v|5Nn$NU3B2B$^*z0WX`uB0E5(?VlJMRGfoj6jM@Hqi
zB!Cwd1k3(M%E%LS4}Ij%FvvFYw}6?o=rs!u(30UV<hvf^v@gfz=;%ooa~NE-Q_R-c
z*1V=e{b-@e6%zYf&R7rQqq{_9X*E9O7y=!dx^h1m_pND3pKoxSIB=nsBq#&<fpSe@
zMMg$KspV7eS5<X$*c1tqAVhpfrN6t^w@1ACkWb4>y_R7DHAH_H$M{b1aPf*e>6I+$
z<4_Zh{GadiOrB<ntLm^VWabFWOkstka>YZj)p|cNzL&`byvujjCROp`dq@RwmvfZ8
zz3(MwgVhR_?z$c*ay$fvSBlE^d)gmSacl9*!392BP98QNP`l#o2z(8mV;-5pFPF<5
zdk~;BxV1q7P_jq^yDZ+^#Bj+!U*V%bs;c;~_}7x1OUqR9c@zKON&3EeS7)+ivwODH
z3=mlFbsnr5WY3SuWcN_5A8R@G6E#)<Bdbp$Yd|s`?{+^Dvqja3O7~9rON*%YeL_9T
zi(}!Ui89Zre4sJ-XHM)q*aW*v#(alCEmJR&Pk^7S(~|O_Vy>vjf(mkUKWe0>o6hK+
zS6F*7yoz_-s3ZgXyam#AI%znQe*(bk&SZsrF4I01$Y7=W$B4_}8-a7|M*)l#ZkKMn
zrxt!yp2-ou-RxU`9<hRmIIVej`OlwE&v&O<+JgvCP*6y@xz*wlQxQ~(a3BG0!T%Ns
zI))XL^`5_Z`G8-iqegp>G5CAit3}H~@)Ud5-vf%2zE7okd$(A=y?&&;iY#_smHN5{
zEme4Vsuwp>(GSOLfZ%N$H(Y(4tkGN94QL<znICBk-U4>v2w=AJx5mLjSi(f})e5Sc
zTZS5GXpK^6ni@+mDD3ub0^<>kO|FCB60^tAE6-%^pQuz9WmZc%qo!Eqb5D!^*^x2w
z1fBlcF)kZ{v$id01}^Iez=RY&n0StHoe43KVg5gmqT{PlEDMX@CN9D!P*Zn`rH5TX
zJv~7^RN{fPh~*ZM*TwDITS$`cI2)Y!tUNihCtp=KqC*T)@wG`X9QUT01*1QjcxZI#
zO<k`w`l1-0VxB}IFs3TxZH+onf9&XOo*r+yeA6>kQoj@FZCTiDY^R?i7d+=}AH1h7
zJ5o{K{ht5v?EZ%GLY;<G3N`+G^hFmD6H^hn_z&*?1b!yd^2pVxn+yHSIQ`<~JzDq5
zatZW6(6vJ#m$D8*TONzn8ncpD)Ylfk+y<fI3_IAwRD~;l$9ciVc6B<?OmHh~fI#h1
z%WuEW6ZySy9qA(JCDtE$*7y3Dd2xKV9N{}L!Y@p<Kw49KV4)|sMSZH)0FMg&R@$_e
zLO~zURVJoPT{&DedqTmDhL33YM=v1jwcS?+fAVki^`Dkod=^OgY=8O1;9mtbF0hRh
zaN?(93dpDw56j*yxgz-4R7%vrq(F$zrS+RDFV9blJZSy00d2O*RwC~5kR6<$_{yD+
z6ZcljE4W`PE@>OrSbU?rkE^tQfE;QR(>qX~<@@HA7L;et;NEg_#v-BlL&<Kyt?K4$
zN{Ek_h@h12?Cqt$RUOHaHktlMv%c;VlJV-Xe3#j@W2P+jy;444#%NixccBac8vsOD
zB-qp{$t%wLUl4L9gP|D_lkibP3SqjAFmdj$j?uw4rIaf~wPHCs1C;J&>8XL;UVASU
zpi8k@&Uv*LxTQ`&J6+wMDcHEs>CAX`2{nPzl8H`6!!MzFNbS3_Z;8G!A};eg8o`Q-
z_QKAdFJSeTu@_9x#xbuMaLaFlGuq)rePmpu@gS1|yaJ|5>XzTj$f<{u)iFntTya1A
z{8th1PT&1{xwf|47Lcz|9lkeHX}3SiRa#bNV{P5IumGh{O2-doNfW45n2V47{X!?L
zdEuWiu`^^^PQmdyq~2Ecus?~X@<a8`#zsez7xK~Jhf>F=$G&d$h>cTU)@mziu1sfQ
zYU?V04o6AZUT-nHWiRnwdt{hO;XpU5VJ_Pyf<0G>79_*EZ#nCf-P}(9nQ*yVGk-a!
zVPwqSg*^O(r|Wn&atO0#t32kg&s8NTvka?TucP^bkuwgj+38l!3Z6=h*;=5Iig10F
zmEE}7NHOY(cH4>XXl;Foh1CKcDvren9S*uZLiR#Ja~;%uel!^A_Vc0$c2BFJRc)I|
z?(`TSJrYsO8_QwCx8J%-9h*zCmK<*m`^E{yJ-2ZQjl*!vS~NlLBZ_>;zYlvOqhN}r
z{#8w2vA!xH{bcXLNqp^t-s(9&o<h&1E?|wzmkno$X!FwUQk-#HG@RUSi$F4^v~V5H
z_<}*7Z{A&(w3}})Lp&`ExJm15jf|QE?>IYpmGh+>OqB{8QN_RdO3uNUJxhdhP@#?H
z<c5C#(3F=t2Lo+Kf7T0<m{QcYwQy7^-7erux2C^iWDH47eaoO(b6Wf{*WTU311<hQ
zg7+RATDP_CsOGinvz-{H|GNBT2qA&G*sucTqA(KcKQE>Cx5Ow3dZZ5vR^4Sw^u_PB
z#P|6R5aIcnG|)C8hhJd0u|gS6XEP3p0|Z*M%MHh8)=cn0el*GaPYf{f2`Q*VPC%Ar
z_v=j$A6XWemDcjCAK$J9?3o#lSh9x&AYbkniYx&NR#c1}322ae2k?eEyFb@gHsNs=
z?=h~+y7?Bu&#!t{@^>_K)wg<j-0sk#Zj3BdGn2oubFgN^vXeA}G<g5wFAR{i&M<}o
zZ!}7O!|bc(_Dzik>{_YF-NDEE<+A+oh^zoMr4vZz#WCq4-Q8St^z@)&VPRz=d{M?J
zDJe<hvhxWd;Ly{RzlL=Q^LB&6?CkAH8yi!@;Y)Mz`u{I4@jGuJ9}C%u-ggc~BF%Bw
zz~D5`8{ZYG5<(TrF-2j=|J>(m7H3V3cTU;7{A}$Z!j$F#d0ODE`{P{7G@81;#m49L
zt=9$uiBds3o8M@^ttu9MQlT#}@8RfV<|HYNF(E-C=cAApFLgTmj)T5vfijLyAq25u
zp;Jaw3;0niK&?NBdhj0h)WBePQ<ns&IKaZ$Qjl<#f)JosD+o`Nb*!GpA-)Vl6AP!A
z3~K~9n6u6dd!<41BfuB|d$quQj#d8}=JpP8_@rsPv+_<XoK!_@MP{z2eB8lc{5F>#
zP(9pp^2QO%b|?EAPmA$FS*}l%fTU%qmKkC*q;T3SCpZ0~je1T%Kwxfeet2<V)Chfl
zg2639N?c=0-R)3ml;Tlg&Ba$An9ltO8o`wdux*9K7cKmJOE#i1jHz(H4!;pqRtfh_
zUYEe^pAi;64Z`^DjTZF^#!<SMRA$XUgoUM8gW-PjXFo(#UwC7_NazQ#C(J;D$~AJ8
zZ7_3TXAMVPHHp6_Y2be3llcBQWbc(3w3K8z*eeuQ+&eBJe(Grd1BFe;6Fx`3r5v@%
z{Z_5#L3;4Z&tH?z_&p0U!)FVOP*DjQv)NL;%7$SDex>#6=!aSkU#Fkd)P%>s?r2>i
zih7mR_JO4`Lo*~M@>LZiitqH{`RQSTt!M3N0q;gm+L+b&D~<tEYg#m!YzS{cT!el1
zb0cs>m6MWbCN9B_5D!mDk^TDh4P0M58{J66GL>v{Yj5uh78aIwWKpo77zRzf-N_QA
zJY_U9GczzayDk<WZO+O_($dn>uNCHa@W|+`BO_Rnl9Hwqg?+zYk#Tz7RyEyyGcqdE
zYkQ7_hA&cXI`)Bs1Ds<hAeqnYxCKMtb^ERC($bRmjs1FW-18tXt4a;~@WE$)rKFIo
zudh$e%#2jo=+C?Eyk|F`l>56v*{xDd<t!HXM!u@fVQ2h!U+WP^Er%&)s#%zhnW1pk
zD@5z$V7L>)1ycmNpH+OPaY#*|`?d-{xEHq+izoj)wGPew4#J`@*o$Pw@{N#OXc~T2
z74N6LNBM8-*!>FG_*bm@L@fV4*CSj`!w)=yuf-H?xK?Yf{4v-c*IK16b*BS&HC?Bt
zGPJ(-VeRf7m<)7&_gg;}6@_Ww+!QYxN#;)}v=cf`Ly{>Nwhc@2>JG;@B&L3UVd^&C
z8diIWzBDxonT$pQpONA2I)oZakwuI6sQfqtZ{lrcq(h*c!v~7j_O8h;XZq&5iQNL9
zfvhf^Gxi{<9+wcaCyrUK6{4K?3FNRxw!RkYJaO0_)dd*ym$!FVR8-qgx)>gt>1!=5
ztwaH+p_QM*a`ty#9$iu0@$)>DB1%e1pTtB$25qRJx6b9{+4JXVRtpW_1>w#NfN^SL
zW3vSI%*KGg135i?U|k(=T6+3asR8Oi{RQ#-{QL#E+(Wf2HlC5aJ?ob|<+V7Y<R$Rg
zRRC{S*Ve+KquX;8vmc+y<K{fm?P8yTm_?rHaV)K@v`m+qJ#pSCsQeug!gwGJJ-#$y
zURw>`x+w4Lxy3fxRA%c>SlEH*UPz8`oJ!$-g}a0r&PL5o4aoN+<v2}s!c{PCu@QUT
zl@ym6j^awn<@VgY&y|1lg!508oZGRkW0pA+{w|S0IR-6bCmkRzx1pC{b-jaJsQvth
zpgM(#nciYss&H=h%tTTkE0oD9Vv{&mhsVD!wie*cScIRvItDLk)wFG#zenhNQ1&TD
z$w#X=#5dcuPGpB+KuCkZ-4+GYUv?U|yVG}o@uM*%2!CZhEMm|;GEL`R;0oZgnIbfT
zwTyy+!5HHS_Oci_wFU3AWhEukZC2U=D8-#J97_9ix;5ep<>%+?RhH|vAaOhG2g+pE
z9_-j{^k-_rJSoog^Gn-vCw8`mGf|nCn6NQZs7XX&3Bkd^4976A5GiGor6eStKi1!u
z#B&h<9Fb3-uxqt!ID-GfC+W06azfwxnj6^c-ZOEyx=%z&`$m&!ly;mxNk%y<EF>tV
zj2+TUF*>YOe47o&U;9E}p5F7HzH>$R?KK4<;28Xu>x_S|C){+7aHvk3Z2-^)LRj~+
zondChmYi)kdhJVW;0@le_w;q?jTI?In(5d(v_#ZCu=e5-N<XHVIEoS(Dog;>bvZwi
zC1oLL*Zzs9D<M6frZMnSmZABe@&i=)Z+XQe?=W92{gjD+8C<wH@VKV~CG{SZdSioA
z(Mn@Lw=agVE2|)p`zvEvvmTv9R@)l{ZQz5{!F(3e_PjD!vK17BTWJp>BPY+0LKy-_
zFKx*9axF&tv$(iW_DEPlf~2*zwJzSX0_C2;KOYZjJ->Y;x2iirf<s4#(CJl6f92%l
zG!Hd_I}eGRHa~-jc%E8c9_lLRf3=Yn0DZGK=d|$t_Tu{!cutRNR`4m19Xmcb5tows
zHc_ZqcexypIX`gxn+5a~DlP+@9L&?>r<a$wtgKk}2==+n$-1=X%^n9ntFhenn@?9C
zT*KM4wHGjbzr(Mtu0n_NZU=a?MOxae36%fw7qL;0K8d>EyyJ@$uMJ$XK#}DNtLF$)
z^sS^#N2U8=q($JHm(6Gg^Ng4+XF7BuWUWvUCFDt<MJcJ`<qUheF(QNkGWPDg`fN{r
z^btE=dZU})NUFZs%$53O6bNT#yd*e1<P7eCWdXJ5f)#p<r{I~0cYiY-%MY9h_QpqY
zZcejC-~TF89*+i;p5xy|;3~<=M(#nX<}0)v{Xe;03O5r|um8D%==0+1s@NyrLA)Mp
zX)mkqN9E`w4mi={?nf)=kLZVJ$Mc8kL{!btd!nWR{9WnDZ!T5LuZWqd;aIIEe228}
zeWIYCyN~nmpmBH~y8#TX6Hx3wW2SNJ7WjaLz&DPJL=<Y)4wsv$a64@OiiwG7x=ZGD
zIWCNEFf=q|wU~~?qLLZRS7qr><Z2l&Q1|=&yRl<qpAn*xiiP{0o=!GVq}|cfv{X3p
zo}S)#I0J>paW~ayAlZDTA_z>ZL<m>Qd8yD5oE|!jWHvJ#{r14>9fu%MuZRdNA|6M2
zspe==m)5I+(wAlvg>3@^2g&>ZfF7NltwpI2?~63bmUbWUCjQrN`nw?K(k*P(QR*-%
zn!AQ)n+=m#&fWLAG_j!OOM#H>f-=F4G@&<AXkScY8T`K6ZE<qeKrEN_xwpmJ3Yil7
z%>o1RcsrRD4EKcIkrU;E%BljZi-unDPWw09C*xX~JKs5&S_si8nr=U2*0onoM0Slb
z7%r}Ex9#W~o!lMU$=ZWZ18Cd%xT;W^0CWYI3zbsWRd>uAf!Xg}DoZy6#q3CWwkXfc
z`&$nl3~zfK!e_BT{?3-&?c7`@kyEJBa&G_H1&nmb7zTsG#U`a(1#ic_=~wvpYI+c0
z)9-DY*%cKPJ+Ibcnjx#1ioL>#a<%v1!-zWVfp{0>(huSV$|qAsDNn4gj@N3#Gpnns
z>9rdOzl#RpGr68_Vy-X*9s7@TbRY+b;-y*x2w@up$w!i;iyad4Iv4Y<JeKj7vw557
z99((#<*NjzOPZaIPjl41Io&ZpxwP^)h5sl9^;9qN{nF$SvlV8gq|D;1aaJhPLIt6S
zzpQ7D58}Z3+UQh(CpJiE%*?P;y0n9KN-eVeDPsRdRiUi{Q}Va$1`CjzS9Xsyf!YqG
zCX%t(ftVV|(_`P4Cn$B>ofeN-4#qQX?15xc^oUpm7WOL%Q(Ine&&K)sa_DUiY9a9#
z635s_kIW1lDD_H93NkV;a15<)FKx~b76t%eM8yifxj4uMU)+4YE{>Sb_4U+5DCtLu
z*>NILQc`aNbU|_2g<Qp(6zJjBH;hbd{!8U~|8Wx4DC+a)&$T5#e{P&sbHR9tjs4~F
zgh{_W(yjgmRBm=Qal!K@C`QZGPsj_2$!=Xr%ViDwk-35a*x%p3hF|$1cnD>Fy)Plg
zi14wPFGDDRMWiP2bTU4&;d~!9mGCq)Uc-f|gpI8%)n<p76|&Sk_5Ir|nh^8RQbu_y
zq81LlEhYV5B!WRx|7PXVb8H<eVMF2Ph=^C<Ce($Ip`|#GT8yVFRHpQtpK@_lkKAvm
zbar*o-0GMcd}(L2d{htpgM!ci&&~$R-~F%&8C;zZ{`fJD&yD?Ly$?Rs>0rLKqhn>$
z`GLytuQK%hUXrU^O&WZGKY#vc)H;Ggb<YJwD(A7~xNi};xI85MV_|2nu&xFt5$QrV
z=hz=hXE0TW)pYdt)<|}l{gwilwW!*5j*f=Cu}m)r*k3p~U6dj-1S+&UMT0Dn4ao2d
zGa_K&I(UI^Wo}MnbCe?iF3ZHmN45A`A3>diJ7*#q0<z%N3-AU(R;GEX;5plNd1LqV
zzOJuPWPYGhVIq%&g8S=gE>%v>+7-<LbRbDb?o?h}P4faTS`@HgFhGGEzOw&?=%LXt
zQ{ow)pB|Z2u~QFoT61A|OhOPC*x<f!mVAI^jwDmFe5Qh~#fqS)VJY(~NKsYB%7>XN
zI!JSi<X~dQ9*8;q0C#?t_HP+(M>pcIG95SjQQejvmntDC{uXTE@a1&s&WYv4U7;@z
zuFls8k$*{mXerg#J7oA8#B9kcsXmmI)IV^QR@C1)e=~ei6=|c6hxk=(s=J;Z;A6im
zh1?wZbR>cMOR_J2l_1^bUns9%zb<yavh~3r(OTv9Idv0wAly>Jc}Vp7TPJoPeAVV6
zmc;FFc{ZV~T4hc8>eVZaD(fHb)hmjPhu%?2#Wwyz!TMU~%v98HZCY)q@+NC`0Zk^J
zHTUq{^>$tntvM06O-Dz{{_|0h2PwylY|8nQ8t*7HAU7V*LIhZcg<2b=BGkl>r5=Bu
z(6%2W`8x!?b_1AtG-&Zo%W)I!@<k@T$Rw$+9%H6}WOKT?<02wn^nW{8{}Loqubn>7
zqu(7_7yKO7?C-5@XMe!`=$E?8P4|b5Cm|h!P`~xrDVr-#2tkre2MFP*!pzy@wz@9r
z8t$PBD}b#sLFP2@ew$;LXzs=K#8^5lo=W&zZsa=pK#UK(1ag!#EJ|v&ym>ee0&W*X
z+9OxU*0*+6=gFl(zhCkEYNa+Eqy6$p{nL<XL4~2&Ep_={(Q=6|$eHK!d8P+1(@;<_
zc|M-^gNubQ0d&$R8<l6t7dsBxGW#C=b`8)*q8r79Tk?rV!`kNj7&%od?^%ssE*Dz7
zV`@jEpFtsmrS7OHn^B0l{>|SjjcWU+XIPc^{L89cF^+TmS&Buvv9Zzabq36M{e^3O
z4R%DBc1=Sf7jPs=WR=YW^7j_Em1U`E-up8yso!^b6Y0$EwO%3!z5ErU`EY-CLWKkI
zc-}gPq0G-zTCMEuX?_y`9vj$nPk7mKV*`3tU=P8BT*3i3@9q81x`qk?K|_Km&db@{
z$ApEwilY6J2KJWy_NX5sD$WP!-n~7TND};B698^YI+Ur2kbruY$i)IpR%UgnX^y`_
zSXxV0&YP;RWyp(>dZ!ZRB?Vdjs<omctlQSJ+zWii61TefA+3mb8I*!>1{Te55R$sE
zQP#GZ88f)`QT;2g$KVP%^o8!Q97#;Ev)^lz@R%XvrMoFfQ}+3*OIHs58`r6oAsYiJ
zSJQU_GFt=H={y-!`9Vz!D6@Tyz0WYQttpbKb~sdP?VkfLtM#x824M}R<tfj|=N^A8
zH-4wC(*G&58(yD_^<NC8ID32O-(NXW%);SZ#b+Godul>OAyh;9jlWbH_Scg-KTx8Z
z9$oSonY}>UZR{hIGEi7%WJNPjqQsRwFw${z;(x;)u}|U&u1%jkox#Mmc_l+7<m5Ex
z#hY|m40FV1dvKdMxP|zw5T@d`hC7$y+2#5qXvspukS|I*>qEQaTGYAhoaroT&=(_$
zaCA+x#Fk55Q?$<M-?v_%pz>4l=~_KSBWxOb5oi#a`hAdff3B|k7*T(+dL3C*M*CCW
zb)V#Jt)-wkO_LiX3b$!=6{8Z$Mh^G~bW3f@a<jsYUzQk^iV#O?!M?|{pKqTvON!>Y
z^L~HF!cx}hn5SHjX3!J;seu?1GaRM?^Span<l^G8##p3UCc4p@o1JZRe~ck;+lWl!
zt}_@`=i^^SZd;uwpDxPexce2Z!g8*=GmK2qKwbF%(R3A1QFh&01O%m&Zd9bZkq)I>
zx;v#iUjYG;?k*{jZt3psZV-^}hI{b)ugkSuz{t#;IkERsXXJy(hnHWO1+#nCx~T5n
zm)$S~2M_!lJEHkpK@ya3293eb+1roSTo1h!avIVHb9}zuDe&l|O3x}|^DCQBLh04e
zOo1k+%ymWx%P9>5nb3n|o`2i(l3P?0RO<~HL)=f;dkprZ3I+$FOpen83dh|O$j_n!
zVr6s{yGd*=%Jn+5M(+bR=SCJT4};#<PncAQ%*n4C9_>x<!==q5L9rsk`8@S|<kEy^
zI@=P{xFL3*So5LO_5q7G!HJD8NDX*J=o_2qlcxuQ<*hS%Kl$X=ID~53N0W=(6wj$0
zZW1YWp{v!{5LOpy>93zobeu0`NuyO2B_nAVq~|9^2ahZJJiQO>nr@HwCmWYh%AcUU
z@xtU*W(Yse2%<SXhu|p2atUY@OAcTP>ib&u`hL6KG3zkQ2Zd<OBr>iUx05wq)su=J
zjf}zvd1IxQ3a^kcyk*gwAzB@p*Bf1)L{7WqF66@DPmqXWKPa&tL!Tx@+3Umd+LuI6
zf$hRzDlcg*J|ggAHyf@X%`wKNM5SmB&|dx_-18<|{X2%~D0EG<H-+<WM;H&rV>yny
zySoeee6ds9#B4tPQ^;b8I!1!;{&MX$ZOJ8b`M|;6-rnu(ZwJ5#)XEH>H@IKqxSm8Z
zkUzvFB;fJ7pBuLMqrkz#AN4UbF}fV;7T!>O&seRuS$Ue5msjz5rmP$8G^*z`K0fyV
z7ehiP0rwYlVtQH>*pq*it?&{~6^x^((}N!n&!u#wrLS-=VJFVRGcuLu$aEENaZ%Lg
zRsX-3;OZ$5h$Fst5BK|Yjq_9QUH$LSeKj&N=Ye2d1-8py5-UAg;gN=xDSF~WriMdz
zbFcbwCzS9DRYeSqa+nN0;84{NY&~fj_k$(}{V1FTYds-Ntaw$d%DmtWq)r=;hg$S%
zuv4kAdWJ{OpOt+{`z!Xmp*6DZb;#>|9b&JaC`HD}x9+fgjZ?G{w`GAf&b|qmb>ck>
z*fNLqTH8xM&^@qdvo8Lqk}$^FIvn6n(q7e0TPo%giQSR!y&*@KTpZ&QV=Gz5UWa~e
zjoaf^p>6=y-E+c&s>Z?qd1>;?P(70D{?qv0Y`DY7TDi7F0zJY0t6vnZ5(5;gl$Je#
zrBWnEInnc$DyY~K-sB(OIm1wEv{w8*OyHnCzBKzQK%aJ#J8d~`IsR<PZD3g$`}4?3
zP$0gl2S&8{;$N29ZgiT-D%AR`#Z4$Y-zUvle8<effOWnvN2W~~KST!maX#Rv>FT})
zSj5J3sR4j2qEb>)9u{N^m7@T<Q)9pBax%bi1xkXGsU6UEmEKL}tAw#xOom59z=*L~
zFz7T&0XB&fy61UfIa`sT-;s+D=5ev$eAP#%UZK0*9q~0FAPW3}9z3+QiF<l_lDVBj
zXUa`jtQYHZuM^lTl<$v0!F;*wi}u@YL&7Un3OsXSVs9V@#pPCJ*s;Mx^qW;FY3XbL
ze>~QLDcsKJBqV9jt-YNc;`8SjmUCUsm=pObviZ31Z7%!I7qhau&h~yO7hV&Rf@8?F
z#^KKsu-kV)<E^2n2B@owrCT(GvSm~%384mPdV7iyQ3xD%x_!lD#qXlc(DbcN#wx-D
z$$fj2*qS_NfLvcWptfqh7P~c#-%G95+gxwvR(?JZImYTNO+gx8Ek7_;hOv~oTX%&6
zR{R?cAC-&Cg#?(O@Pj#w)#NT_s+*N%WF9IQpQWMcN~@JMboAozp0uji$OLJ^ECZ{e
zl{C@4-j=vgLu~@Dw9W>gzhOV%*Dq4M*6q;ZJa+IrdYk9nl;sF0ukE9ecKsyLkDAeZ
zu;g~}+elvclXgz55s@v6v6`q{?KvAu=EMe68mxN;=&SoeG$y!fq-5D_BOO?L7b4lc
z2GIW3i*DL1;xkg_gy1j{hUV#ZOs@vI81vz1{L4M;WbxLCa=XCH{kjVhiUl*9x36NC
zM(BIo!M_Wg)8oM~nSbJXvuMHb;UgED%<CRoYS1<M<0BFe4^K%+i2!{ZX;e>m-^TRp
zEc(lr&7hG#p2Fj_c?~8CZP^g$dwxK2lv@jR9Fvoi-G<D;At8Ah)%1dbf?pIZoB-bF
zSEN&_|1p+9C-vdP2OjD0?5tyco-jcwb}7x*7ZwHvCYXRD6f|1=6JcRtvff|+{+-Cr
zQ!6LW$;r98F)=aevM6roB!O--_A*#hlb>u3rR9QlHJV0+{^2qyDd|(OHpB}G6xc5z
zA<}_?ju{SHG0P2JcU%D0p*eK1i)}nSJj~UGP7%@2HZPCX;Ektl4oaK_Uut`q$9(69
zqW`L}Z{M&~6-`vAFU{u=(($XoHqT~%>h7Vh9gLJaFTC5Zu)~@e9wWjxQ<I!hefE(5
zBDQmPsXlsZZc2hejBY*zs%>4Gdk9A#UdzYkHLs=&Axf{~{=O?UdB;`RZCvx6Q2BvJ
z{CXvydRF%3hSjsotiYGb_pI8#7H+<Cg0qykef}8&#`d&X5{>)9bFP|n<@_tJ)te6X
z!sSm%3_15ycZ_TJ1@$x`tF$sA#QjY$Gza?T#)Ds7=Ok!TO6*oRA+y!euwG$D3(Ip6
z{3G5xeB7StvQ=({C}PuUFDk?ao9J`{?(AdoOlk`#l&e;N#svJRWUn|^bdV=cV!$6c
z;SlNPIOF@LlS^^EZ!js8p0gCU_?!0A6($_sougBTRg{&QwANe3tEsjox*sk$=6#m1
zjW*BQ<9L<sKSX{9g*vk}txmmq+;LQff7`C9);gG#T8e}c_J9xkyT?Ljr~k?3(5FU^
zk*EZ(%awrJyNe}GPR=btnr6_(sFoW`=PgVWStk!9ahUW(lBuezW6uqCboebVFQd;r
z&lWC*vj!|io^qkOcfjAkgkVok&osBYk+!L+IM61SLk3-I5klb98lxZtpR2&iLn=8n
z4{q17mkWoRN-m$KN1|MY7YAs%Krxo@O^TTs1N7ygyZ1;1!7IFI3tXif-o#cv6<v3i
z6mCfzhgus4n#2ODQulRp4Q6fvjIdGnZlenND}<8Zh?c}X`Yoxbv}M~y5+5eoUwIVP
zU_f{hTUVsDXGc%fA8y)uw3aT$Jsi%Uv$LldcpR<1lC)@}6_&D{gNv_2*0o%5o@x6?
zkM5mcWpbX3+mH(5Z_hE{p7S;@_Jmk_-6N#ocPy>Ew7+-;%3*Owt-9LnDt_%h!y|rx
zc1JzX&eI@yR$<ci)+{3oCw}`fcjwFTYbdVNd2NVHyV-Ohn?FtV4#U6z<8qJP!$bU6
z3Sr90B_Y+s%Lo`~K4Ner0xnsrk7aEtQ}L&!WM(+EQhwK|4(gifAS`O9a8uBz>EaIF
zsQZ(wy@jFJzk6N|V2fyJZx5DdQ6BrThEPj{BnUbWmzJ#DT>I-&Gi-WIv4h11C{L$}
z@BI9{d&91OU?5K^AN!!`fsaO|=p7YRAfWB^3>9ye9Kqhi(0J*yysYbwjOSBV$1Nfv
za(Q#Z;=Hf9zfhMepXbePIr~V4gRi!yib+6Y+tS^Q2nZ-@<pQzI!IZ20x&u!4bJ`^@
zjnf7BM-~IfaBh#fNFay3SuC(q1P2G(I6385&Q<j%afJQ+Nm)?fkmM|q#u*YEe0e%1
zD=sFcyEj`|e|J6yX5Z_$zC6tuW>EeZ&sAB@H@M@FkdUOTVw|dwldGKoaIkvGD^ntd
zdWaiS@MUVLXtl43*(gqe)P;qNQsU``6A0oWT964sZ9svrQV)@}IBVkSdB|9uVnyoz
z{*r#LY8zKBT6MC5V~>oJj(XfjRAD>WfI(n@SR(5B&O0dg0*5OjQm~2?b~e^Ge><7C
zHiauheP?kBNhn)UqB6ortonYZw+Hj@T+581Q=vvTH*6=)T)kNmv5V`J?v)Rd!oa$Y
zl{gwHo+=(Bn8?f@K^fKKayNjs&GI=X8j)7;-aR8b$_k@~ceE4FsnhPf+o{%zQR2eu
zG^=09W@TNQF;*A}MY0OBa{$?VCfw}`=G8-LOSRRQep2<{`LORwe@i@CaPyoFR+C0E
z!|mV25sXybP@EF;pm2-g6t@0SD1UWw&o!hIM1DL34Q~8YS@NB@$hyc1DP{HHJ~%>L
zb@Sfa;@78-`87FV^#Bj1%?iA+k<mv`k9d1~Ba`sWr+eO9U6r~V>0>i!7ub`4?cQN`
zIsyRqf3bo9_yqe5S-1HM41Vz^aC(Gn{utqq0d?!5=Y`gyr%JWAn$Y$rO^f@{YDexO
zBnG&{0g?`Pz_Pq8P6US#EghY$on32Zr$`zjn6KU+wHuy%NwTJCxVF8pyenzaZ7@Jj
z4f}Z*usJv+(-aCC%ZcEdYSQkW!S?>!tNFLl2@w*z4Ps`atZ{G21IH9Sxj7>R)+d&Q
zFzTOiJE2)riumb_7pLKTSg({EMTB;!m21OFyxL_Y4*lUtXqZ}%<TC8&$RB<OSO8_&
z-H<VT>U#m9d9Be;Q6={0%^eNsD2a|)W^<`7$^}KXxM533T%qe;S^jMCJ%Pwj<3n!e
zV%S??2R5I><kd^hx&6FiM=h`7D=BsORIx}qEidSS{Pvhi=Ak?xyc{7g462hfc6;G8
zm<kk;osa5_s=dIZxcy9}-IRG;zANX&ZT?UU$a=Z8hqrw(=pETMjB?c=XPt{YM+u#@
zx?`?-N+*}K=p&>6#>#d3%rd9MX9BFwtLKZgjvZT11z&!T!z4lZ65|`EeHX7jYw+1P
zW-0A(RovsSBp!PI3A*F9XZM>X1RLy`MEw?N&+rZG$HpcMlKD6hY_2V}>U_<kIoFx_
z5acNAWE7iFVxl^@y-=+Qbr{7$EP<sVWR{GC7CuosJkG>$tdwg_hG?Ee5?+Q#fqN$L
z@zp!JsGh^ZIadBR1&Y-2l$ff3)b_Arxp<+p&1CbG_n8bMWZLvtynyz%Z)lke-)f=O
zXp0_R&?OLW7pY90j`u5$^1XY@%iN>DxjDyuq3oxgi)ll45c2tZW+@Ja?`7wyEZ-ES
zDH>{BF5i!y{pR5l^+)`}?-^@r9`hnWvv`}CbupB0Q|x~ObdQ|6?C9*w;SlHZy1pF@
z@iL^2h!^P0SWT5{&;Px^F#5Zc=KL=_My?e#I~%Xd8&W>ehrv}W_?vM7V-aeKhsEX^
z(zzPjv!^}JnO;Zf0S}2R3iMRnDH}`(RaJ9k<*lQ8$_ms6Dzq?h`J1(^T^aLHSg~Gs
zQC^WFzTx}d4c$Bq7Gg+f$cL%!4!g;z2T-ZK$Fb*TVb@QM7<_d5(u%4v^HVDTcT=1f
zug#Yd&DUcKN$3-bban#YvV4P#0e<C{MAYO06BFCiPyRNq(BAW#8_<P!p%tvP2J*Jg
zHwXgk1v{3Vvo0prDw7=uy;9Us)c>Y@la>kIxNNM1Bbw_eBzTC7#0)4cr3!@x_Z5rh
zzJ7_de#CErrI8uSwEj0(<h||hEWkL`{u=+>-7$6(*m#X5gGI14KXgZf03(sCr%hRR
zJfvkqT`^_#<Hyr|6Jbr%)EEyYwsOkS8ue`bi|aJii#b+cZ6s;ZwLtzU{;f_DETlo=
zna=NiYc9^ihFs^kkosyjmn4nGe4k1L`O)_Q7!_VCrl$B80!B-%{G@zrw>7SGaP&C>
z7QEz0NHP)EH2T~ONxe!tZ{qsoKN<;1&`)ni5H?W~)v4dpp<n{rg88Mr>C_9VRFU&q
zXUtgC7g&6bZ0-z{_E?|Qj)*kwZ$k_ES+~XUbWMCD61sf1APt^xjU7jKJF45t-h9F)
z!{DeSBd1y5p)oMC6#X;7+TA-=jB1VN2~hy(r=zu)v&#8wt%fd`B(}S|dl`(p^V>Gi
zL%d<z;V@bLdEj`hmD}B;-*&(<-FWSeIg>CNg*GH#q#(X~N2z}x3Yt-KO<tO@gOeFn
z%ds{aZixDE^837wP8@AIkbDW3fg6(O46}TlhxK!UI_0LuYgIO{30n&pJ??P#ip?x6
zqV##PM5OW0sR?g{FUALc{lSXHA={nS_D!$p$cWY7w;}6Hzj{JIAQ46c!?=K>&J~e(
z`bDSdCU{1t@q{x_DOmKt`aYDQQO{N(+J+nt1Ny{r`*aR#HftqwZx>3xaT;%4LJQOC
zE2_OXc^W}Lu%?}nSaE=G#wmLAg36ULYNWJF&bK2s$UkiHA^=?=<K`hRj87DW?UFG2
zT_=;P7y+o40}FN?wzi(yo3|DTs2-pt5wuQ<FxKG1O=FPQC&5SEl(}5Lw!`WOOJ!$f
zhcX1mLemvnR^S#cHoe%t<``^l34z_E_QF6{MOcQpHjE#{?+IC2Ad|!fWIJXXR8Nvr
zAvFJb23l<wwU^i)!M3nT$3vkT>XsC5_K7i`RqC#hu*gxC19YdqSm}_^(h3n&{V2QN
z8gP1&i`|arl81Ww#?~GT9(PPiVF=K|U(^dl3YFJ3vX=-MedH!_J>Y;A!&9ER@<v#W
zJW9{RBp@sd4X8WbFfzV$cjpyJLG(Fn{i)6!PR&MO$4mRIdo2IZY*b~3PrsOGO{QBW
zS+gF|qd|k}0~HTata4R_*NHId=Q|YTC0|(@^CfY8`p=;)-P2!D50L4rpgd-kmi`DW
zQ0Zgx^k=t=VCCl$B=)>%)^<BR{j&Wk`F(wiTIS;FPTIM+`1ae11E%;C3+O}T3ZH`$
zj60FYx1w648P|&KqvwBOCjrcKscxb<PxH!og<Mn2`<c_M^``364~BB3=33?x)){x#
z3_lmsyxNkF8nQ0t2{xM5p#c7n`+8UjHYE#qOxMs<EEP2CTb-T$tF<~^4cgQ348oQ@
z+UbiYNHiMij{7Q~$H4L-M>A{NBY2QVOM9-=k{^DI6f1I=n0fj;{sl|%(M8qLW0KW)
zFk46H`rhrL3E@%`_I)jl)>oz}J&#rrXMsY1ny(CH6%~?$Y&K|UN_aNCyC-&E9alk~
zOGY;GjWp_12NqXN^B>&krUm1bahN9+`=d?mdw{rL+8&$Au6u5-+IIPCS-G1CezJ48
zRopFE+J9zh1}wrEH9X{`)1;pxH2+qWoIf5<OP$pMmz}-+*?z8Q81e1pTG)@A;@9=-
zhRk?(d`Sb)`Ws8mN^@Kpth>^SpLs{uYE(>w+Shzf4$$sJGBu9<%61lHb8HL`%rTQP
zGm7X~ZHGNJG_gD|VANb4T&!`6Qe%QF8$r26RiiPZfNyD<*;3qYu6+232p6MsBVD-Q
z@`&bHaH5W(eb|c$(~I-h=a%YE$BJ93*|Ic>*&KPGS5=+5R%RaovW+#<4^T=p|AFP3
z2hS%<3t!UAeao@WSPqQZd?=q>d!Ab+h=1WllAb$jKcleLKls>Eu)|o$)m_Ocia^}l
zhyZFQkpWFDH*fQdgF`g{t_GRU5dAblVPnfDKoKxbJ$<i`x6P!>!|o*0!Ttb1D)5qr
zAncqg^H`6wa=(v`x*sUJE{fZ_rqZnLI<aTwtF1mg)j8+{K)BwZ<E0K8)lps^QK=0p
zj5`HK?KyE?$M<XskW8ZX4|(&_KJ-@UxiNefg?^6Jzp{hqm3l>ml|N2ZrRl3)Xng<R
zCfTzf;BJOl8Tf3IvY8lr(mx`Da2~oxP%>xF5)OMayO;H^5b!zd$Gs)cEH@r7oh-mZ
zAz-IHtk27%I@_7b^}4%`Pr@ObSvfce1FAG~@-R_$%lY3C-?3=y&i*P12T&pk;eE=L
zw{SyF7O--{ci*gB5nMM5nUo9((H%;j>XE7y1w}s_e$TsTOBwo94<;k1#-Z;v*uqAv
zZSK!l^kI_?qGVgHKfh<d5OviRnB-;dce113Dy@H6#Lj!RpRaH!cLF^s?1dHfgZr<C
zZ#9a)l+lC~^z1mZEL`IHT3o1B5IvY2jRG-**{mP<0D17`1AOpvUk$ZHONzI!o&JKg
zk=571u~q?4hm?k$SRy3%SJUqP<I|AcP~WSFCmxIr&#Y4UvT?KjP%G!PHIFPw8ljN;
z3(J)m({$2gU9|4e-r_g4Xn1yE&fOl@K%A~Nq`%`^vdm{g3g}E;EGfbo;8{Q^)Y9gv
z!5J8FtE=+QL&;hTEW$Gdub#7&yvO&d4FpGs&fwzVyr?RF*%1_$@UDF!Mx@U?a^6!R
z9o9n=#;C}L|4*2QW_9(;%}^_f%XRHT);N*<#|N~x5jGY*W5$E%5(V7MBe4Zv{8zjU
zMYdqn2cQzIF1Pq{a!W?lvsz3Bmy|F9nRg8UpK<6lU$d}89Uj_fLwAl(cK7z)JO0Id
z|Ni|~f5`t2MF40H%qD(ez3_uE2H0nY)DEu;zU9*Ow}NO~ry;ABgi6wG394J9CcsYx
zaX0GSaNK9Le;rFV6SckBl6<VikqOUu%QP#amy^AuYmMp|YpAC&5YoO1*u2=FnbN<l
zF9cF??79$?rlIi2w++16>(N#7{mWWt<J1rq!F{~^=T9IWo4kK1)PzijYUErTaW^PR
zpP|2yHXa+MyqL7lS5SvO%QnAS5?P!)>S+6S+H=zprC%Iew>o(a0`h0UQT8iS8jQ@C
zt5_Ag;{jf2L-e-X44HQOSP7?XR(bQCJ?}1AKS-3&1c@atbbk?FDS3^4JYKQS^~D)_
z77SGh#-)QT(cyp}l{s1=(%CDEbw)?8OsRxQefGct@3B4`VVTkgow#3^W0Myt9Ww2M
zje-!}m1;{;5%9j1SSpQ>uE2aI>SeO^R{S2&_MDLuE|gqA`^}~p%GcbsOB#T5cXpBo
zUmC-+-D;2Vbt;6XCUzH{@q6b%gwIsQ>1%Gf=d9$kfAtW!t%yYo91o39_(em*AEXIf
zG70g_dPYAoJsy{!Jdujb3#!{K_63qed=4uqIk}h8iL5ICqQ=F;x<|;C$o?Un#7@ZP
z5nt=LV|Rb+3LwRvfq`t96fOWb!<M+77%zI&tus^)`KTA>0J;kD0BV-gzY{yI0h#56
zn`zE)Mgi)D<TNx$K<N@30=&u}hQULPFKYb?Zq0Z=e%kA(6@{ATp&L8;R$8IS{qG_(
zi(e9W!No`fMno_zT!&;|gqwA5xb+v@hYMwN7r(JGHdbDL&}uDz<?xw;U-~9Kx?OcU
zP$G8O0wt7O0TQC-()hJ;WmB<XVO`%B9w-T1&~#NxnjhIc=QyJ>dwoz)9P>Ayk2)>A
z56iAWfr!6FiG0B(W#RSvIK~$%!t+!WSI|?#h9yp4tly8u?26QSFJ=C6cZa`%SUz7l
zPxWR;k)1B-V=A*sZDCy%ieg2$2z4&?8)^)pF~P#lc4$0#(YI-(u+c#KD(Kyaa}m>h
z=AW=7du0*1B~;CUEJMv7Pld*ot=ON~k&<4A#Y@Iri1Kv(3~~m5wpY>#9-bN1D27sD
zRzO3bV(7|Z8>ls3BuI4<Ebwn+jvz5I(cQO-g8l(^wl2eo(Vy+UmW)bJg6jgMbD%Gv
zL!&4Lx+Ip82{vkZxVG8eT^hxh>>S<fo{+*%NlgC&jb#N9e#tj65|OzH`s<jqg{vzl
zk&W&r<n8C@H<%VjTKAyevfen{NEke6Fn%$pddc)%roo?OtItoifzbJZJk9AWR%Dg_
zWbBS`q1Q(+?y%@j*X=}0ZP!g?&)}d)b|n2cq^6Dy$;-Q385tSr>Fd93l&oK`#nO6(
z@TI?BikXE4rlhpArC%yQ5H8ygh-)7qZZN@k59C1cM)vFhrNV5C+{ee~$)nuxyL32!
zv0`b}e2R-1fXt^rt1gz$^IBJ5zZHlx&Bk(2o>)v4P_aDM7JMlYCPkik;+_VL6=*TA
z(>=suAWUL7uf4P7kij-JvD^_@-dS)lX@&!RiL1=Wi|j(Z6ArbqGcN9_`|gYs$nL=y
zZoZ#(xU+pc`I?Svtl@z(1cP%*Xec5o=FLRrI9m;pf=^W{O2X!+-?kebfpgUrRR+7e
zD4GopJ_FF;pVe*H^;D1RSQBVSqVcgn<Z8Nb4J3^~B8}>EeHhCF6;GOUpRw(A+CY|;
z#23jt_}Llm+D9m|_JSlJDkyhy&K0jFuDJL3j_I?`8E;bT&|9``Z$*SFe&A2aTgu|q
zk<tpycAQDYJU^51G`pNl`mFUnAk$stWO<hEcdUGR2ypYEXf!}Q0G#@iXRZ};eRpxI
zXNKvK?z<l-9FAVT7~B3~VOj8$E)Q9lX_rXr%5B7pK6SRMs_D<s!ByKjk^42FtIA2H
zR&|)!#d$h9LnDvHn@v;+ByaU`Tm~rvgiEHswq<oc_gE)S2$f2naIpM%OCOm?7*^O&
zPMq@mzk?$iYFU21_$Z-exVgTLoo<wf>FP-UUytO<l013(^yx7mcy52gkYd7R6qGyL
zt$|66%W22QQ5^6ROt8EjkPX1KysMtt+lV4j@2KkXK3+q8%wQSru>v^VvzTup5>yaE
z1|blfobGbj#=?bB=|`o`S?~<7Qr#+xPY0n~u}W69b&X;?j6c>C+%&!fDuYp&P+Ib?
z6nM`^KUEiWap*L8>+!g3snS9<q!e8kkC>drB&FLs2cP4FNw}yKUPwsi<b)KF1ZV9a
zkcxi~`!n5GIbVk*2yQ(nN?jaBF_kX7rb~7HrVfH-(uYbC9?y8^AFbaS2*f26AVSa~
z01e&h+K$fZw}l7ht<CN}!XM0_S=P|^!O05pK1*^+_ZmYr%*o}sCHxqbl;1ouLsZ<$
zMzyId`0!go{JxZ?c(c@=9o|0k?ZbnsM`%Bir<MV8ma2iB*ta?w^!oos+9<q*<y`<8
ziPwD1v+AEb?^`c=919BypyN0L0|ONFq}PZH*a@0|4dAcOzXI#)D|2&$kkUMj;zV6q
za76%yj2gkR%5oc@LxM;5eCv006bXKLQ$s_kn8c8u3gz;7);mUP%eyk07km^HqQA5(
zK7S6*Gn>Ij`Gt}oCF0jx`BSbYcF~jfo)9lRIwWL2jTD3x{$(|X3L)enx%<*`=H}9J
z_P5-8DfK)XpK$nC#z<l}7?&R$uS${r_XfU$8UyI&ulfL&Jv}pnHOEJ~c4X-zTXgeI
zV{Uba>R&CAvrE`3jzm>+?Ho%B?D1PaARQmMUIp~E-!|q9OwQKfk(43Jj&^510Mi;>
ztSJfQCzctl*k>thydt*yei;7gUFVW$UVleRPsj_t$~>H5Y+w?)GoXR5k|pLg;ibp@
z>AkKA%w#Vb^2wkq%d(Y0d9U67UZ7Bd{vn(-U}I-W^kAO=S>gt;RnS)5J<1UaZp4%)
z*YL(H#((N9Qvyi!t4a*mSj|=-H<2S0He2DpZ4+%84~=jc(VA%7UB5aW)<zRxG1;n%
z-2Hx?5^_ewrd*q{ni<`1ZiyHlp-BGlnhzk~z<UC8Rx_o>Fx->30_t`i8J1x2n=ama
zF5Vcv*qEpJi<9YH!|S^1QSjYbGnrd0U2#pJ{ed3@Tu`%pFn$|H5VH-9;{5mOi<y0K
zmI!!UZ0?WK1$_}wQ0U4mKYV}#Zl|FK*RA1<t3D3v#eN;DnX+s_158j0XlpG_++{`h
zWNcrDe3q1Jo^T-SbCoNVAZmlBM0^$*m%D!9N{Tzj@CD%Inc(b^p=bQTES2}nz%#v?
z;*vyqbM{e706pGvp!&-52LDknbm=_+=B$eTQstda961Ab9P)eW1>fNQFVlq7H|))e
zBbDucz$VLF5WoO5A|KSvpI*+iiel*sDa9+I_$&Fu%B<tlT$yy)`8R$_fz^P<cpDYu
zn#2*zR@%p++RQdquI~r9u&(VInw=C<_E^NzX(67{;)oIlv7T)IB@0#6*{#K}|GE%4
zb7VfBYOoVIYz=z@q9GEf*Z~2tJy*@(d3~B8<o5#fTl@1hZGiGrtNN^V@Ak;E1ODQQ
zAD{=k(A8e`61zh@7~EDG33`szN~eP|iv=N7camp2oVC__o^ov#dy805RLpau&_@hl
z#QK$oU9lp+qkt;_Pdfdg^|o|~L2J=N_x7H*ZQ3g&zVVci80#0-OcU6;Vd!e@j+aHI
ziD7wp9grL<qou6wJKf3Hc_Dp{rO>@lg$P(_qUE*kezLF6V+pco^zJ;PcQVy@JYU*g
zWM1lSAus5&mbiIaLZzk7<a-W0AB=^J_Q=1(y{189?|L#^pj9lC^eCyEx<uLLEulcI
zjZPL|7|*mHZ(O&3_t;$fM^{5rC#ZrJBGwLGi*FE%DSaW*m>xqQGRYPwa{|V|3hSj*
zV8vM8*^vaY(iiWXlQ^sg0Y?zde*|9xTDbz9rqnZNXKJiOPgppB3jEUg)<rx<;!pOH
zR7_a0Ii3g)IW2>#fXdungpyZaG2wXc*e{T%N;i#Q_$Oi`-Tla%N%Aa^kjw=MOaOkJ
z@V?9<5W~FTVq0;0wey#|Ld_G)6tS-cfj~7nRytLppR^c?Y_xx=tzhRY2Y*@Xw$ig8
z3KfLt>B3)DJyRtpKMG3#SNIfcoezcQCBX9{$VB&k@*sKsf=*N*-&@B@P=da=;YhpG
z`Q<6`ZAb-BK(()%lI*s)AWC#jv?Z}?A(%Hc{;!3VitY_31#Qiv9N8Nd0a)D_%2Ghf
z>VCev0)$xr2gh_N1p3@lXm=(GIIytAZvSmxjlj3NeKAu#YT5UR^tD6~ehde5mgDjG
z>4Sb~ysk8*J$>rpP@}YD^B|n@m5(S>;4no(Z+00<$Zm}@O>z^qG-Z|RRWIB2{fbI>
z0zy=4rQ~b>tkzwECmzQWZ~Oa$h&z6Pnv^q}e<lI%&^wn^9a0^;TE6|Pr=S$xr^@Ay
z8;G`V0rU!ghYb(`A|&9#)!$g;8mWR(P=ipnR^mw66`<px^7#ZHUn5rwL$9TuM->Ux
z3!S|7jnwmc$6S*v+x0*jv;H4~{IWMop)a}Dka@PsQWA6uVBgkkaEr+l4YNDlk}uM%
z5t5LAUbM8dfF&lFua@-4G_}0sI>VQ#t40u1p!az7!hJ2P=j->#T>k}I#<$1{cH7hT
zlzU5fnhx%QE~r44NrTNtsKi%~{yWiZ%3Jf9&h?k<S2xwS68#GVK(fJWhz~=z<a0B_
zc&o!1OG)-7C^FyY>|!OKCOk#-;LVun{4eNIK{FSGPPwVRR4n!J+Nhi(`Eu=aSVizX
znRfEwI^=vO?kVtvbq*}}L<Pk;^os<DZ{UBAO8=q;m^>o0yZ1@9PIOl;$hhyr++ZwM
zlw%!2Whk(qP&>uLDXn;1)Eeq%!SSL%byPVg_+JaPlECbOve-I3JuTPZ<_P^#DZZy=
zr+8daWRX3F$aqa~U3SC;8eCX^L|gq$P1k5cuI<@lB3~c9000{TN%O*I0SWUnT+XAb
z{V%UESN~K>e(*!^ttP9m87;)%#@PDfZ+^xL2=X9j{;S&V9J`uVpxbXgI1wo^Y15C#
zenY0O@m}1BBv~pu;@I{*WW)Edc7vZDp(cXKxARUcfLK@pd2_jnKg-WcJ27njuK0pi
zPn1^lyl6F(iC<U$m(gk+CTHx)avlE=qlE)U5xBYJ&70wNn8f!!VOY~i0{-axhLYkl
z!bjnmtr^MO7=J7?B?d(BExQ2X8HnUmW;m&PKT}Fb9so2C+k_KPS;U}x5x%~CmV6z;
z+8aUd(m*QVhYOlQt_zw4*f;-Ryw_WM3GZAUXGaEYWY0$Z1)x^pD~vkAbZptyvlR2o
z5J6X)33RGQg>$EPB2<WpNL*E))J?au+?h{DuJaFU_oqYkG3sBFz2v_#e}$6N)YF^R
zY0t8Xt~mrzmNh`-GN*WaK@v(kvA80jY6DfQPxdR;iYYga;~9KmPwujzcdy)55WPZ{
zh8q}Q?Pxb!095eE+kd|F<T9hcST<D1unduw+8%euJwQ(W0xDePIl#~uT(0vYnqFZ8
zxTJaEkVe7v+gnzvGj<Z>6z@1^l)@@fL?7sJ!{8x#+wzE5)*qW+Hgxtch1s9Xk^zQV
zaD4*lUAs6-7k7=z>Zi6ek%78LBv)QpjB;%MgTFWuo8{@wPawL$wdMHx!n{&F^TN|N
zhbV99H}6#IpzZ#ala*GJl!6B`X#T+__^Ht8b%?rCNBNIG&JG=mK_3$Gzd*B29kS@T
zf$How<bieaD!o#m=v=pmTO@VKyrKA-%-em1Q5V>NBMx{hkR4`gcl5(bSK~Ixag?j{
z<{c>QK%Znf{sZY%oUHx+7OIt~e`NAY-_k(5&_`lvD*6kymhH(g(=ze@kfLnq|2|0=
zs;d>P0ki|i-0R8ux)_E?LIUrDMh(2+C%~+Y`eX)r5%py8+3g#FJjYPKY~$w!37RsM
zv)(|%_jBo-M1pzzo@YeVE6pev1fGScJ7mZQ(P!5Z)yyLT!QEluRDohO7~Ow^%bgv&
zJkZy-cDT~EJ6{t8EMhN-h(gr0=7UiqEug8Q7-g}?2g>D8+LZxwJvjw8hP$W=aD0aM
zpzuHBz2%OYi^ypIxYXLwmIP`-Mm9E%OQ%#UV448TEl`GbhXmW$I8AZFT$Qo7I%s22
z-*LN72L$$<)OG8rzqmPh+jQ|bcOR!OV|L_n$hU9HK(|N1#Z}8%ivxl01CA3oL~hvW
zR{B>8HPMi$4--QgSWuE2e<;P>kDE<P6z>_D<ec>F>S=P8CXKIO=4w~lJ7nADTOkuR
zEc=POEYx}>+$F`??N6f;e^>(l-J}DQ)-y4HVd$JFKeqpVcx6w!`G&52T`VePTOz}!
zUGwu^3KKgMn-pZjMOngs{e6IHI`_Y`>vjD*z>-Rfi)R3ji=iP;s%G2L5(xxBwn_Qm
z5r@4B`HG)4*wW6Lq)BN=mE1;0;N}Nfq)-EEvKC~Phl{WXXDfYxf&iCUFHokZmrb(K
zc&sd8eS<2Sf;J?*9$u9|@E7{Naa1I3U!!yr1fqMHuaT2_G(1HJY^+gec-aD%HoaOR
zcp=_K0?-O2OAEsDbgwKH+Uys`w}nn1ZGj^Iz_YF6{i`*dcRZJq7M)9TUIFQbE(7ux
zq4YUIsn<12od>Jh&k;<VBa1|Z?dYd><bSPbGB?{VQT+Eo2r{zzQn=vIsYgR;^Myc3
zLLrmfJTNdYob*o!h4Z)zAdz2eobo3L()-HL4^N=f(`BpYhGo&Ld_XU#?aO9Cz#Xf8
zqcbogW%)40HX90VhWmTS0R*vfZmmnGZmt4Eb3Pb~8^2z2VofdzUs>b#;RL_AJvM7h
zZfMcH`Am6X<;MbKus@&!O;`;Bmnm3vZr?FTBt~Sjxsmgjy{p8201bczk<U-6SC)70
zJ)S}`_>GfzphLSheG5R^0>|4=)+)45VCd@b8M*d)*6p6EI~Xnf(b1J9tUt)p5LFp{
z!;HbfKvG|Cfy*xTzqjo&h^)P3y-@3T)knZ#h1sJe2mx_NOpbEirJfx>Rn3n1v!I>k
zZ@qF?{Ezxw+r7RP60hNb5O8r~j^{{V(yKVu*~#6~h3xgB1VVCUKQ*@WMy(cbmq{Su
z(_KAnI$SNhV54NN-w^;CWAB$C5VO;#{p4R26<OzfUNVp7=EjpU4{yWad%v*~=C*7x
za8?4@OWo9C0NzHn+B0zGeeLk|!2M{4PmHtU&bk71vP5f7N6??WB@R=M+O6Xjv2{CA
zRc8`(%vtB~T>oqZ@v!O}&p@g3CA8Fklb4~qy@EA%k6WPFto)FGhcL|fMy~LotDf(g
zhMYmC$IoIa&H0;8ItTdhN#DLbj@ucB@(i~2AD8EfUzz2Ehs<5MxWqnv0Ueo>q&}^0
zMofA!&i7qfK(T@;0UYGJa}{5KG{`#;*?8(ptPGj8=(7rEZ|h&anqU$_ap)sPfJ`*L
zZwD5u#fWnwbH@RZOs=HX`9!%n|DWH2Xj6C#PJQVN7D;%pKwmkH4hsdO7By)WOcTb_
zOS{aOT=p}||7k7PcbvVzMNV*KBztOC5pN@na_dGQH1-o{3fZkuXe&AY?I)BPWp<bz
zCsXFs7evkJSNtVcm(TA*YQ;SyUPO#A5k7E4Cs<6rV)TF`>Xd!^j`^=^04*XHi;K@x
zUmp!L1K8`~1*FIGrxl5m?R_d(&QZP0QiVmHcf)+zFHL3V?~-=aZww*^fG4;Q!lk)m
ztUw>H7_s6;*~=a&Hdke~C~?|}jRf+<8jpRFJf}lkl5Je`UPV+L6<BVJN7-HX;qoX_
z1+typk^X5g+o;%twRQ6!Q>qcQNSf`#kCa*Kqm4IEZpH3&?zq9k_?K7qdCJ@`TgUVq
zDBCR<BULhdt|9<qL8IaRSbasv=k+ttK5pqkc=nruJai%<8%Z(wJ%jSY)lMi?^<#c@
zZT#e({fy|MsBcUKtaGYAOOSDE+&8|IS{o86TWxOHB)Vpq08Uk)S`G{e@m$P6bYpgb
ze5wI0c-3cSX6C>D&zhr+d&BYYR{G~;MzS%f^Mm9w5A|2;%vA;P!l}&F!^ZCHv~Rgt
zh^W<sL~#~;JiAad`^CiF<j6&itnTfrDh}mc^VYvs?Mf}f=4tQ5@;9;Q8Z!9%4=k;h
zQ}r>ivwM?OF9$&@Qp}VOQQJze5l{l-!LwjQzYp-B;#pfO<&Vu%ct)P{M=HK4#oQ5J
z_`@^gdN`VWA<e<D5SC-zCD_X?^xdE~5O&zZzuVH_5o1E~``^?b()_HzOeYfh=JFaP
z(WTaYJCpBrHjj?TlX~T|nYSi++h}|I>_##!#+Gb%muLK98C7pR&$B`3E>*{A&vmix
zu%9wU!p<J)A@ohXDmVa$xExIHj~hL2kQW*rpP-&`kZ_Tj&S4^xxFHVyawZZm2r-fr
zFD&g$yX?#|%7dIpwbBgVf+f(L8WQBV(<T5w7euu8R1>q7t3TaE_xij`lTRp|GW#bo
zoOY0kls3KkRo%WqK&=L#gLnrlH#qUYd8${3M!2UnPke>JTLlrFm+O6LS?wj_ZyZC@
zSb%yh=fUn9eS@fOR&<+=bB1h5ba#pkRwHV_N?e9XcyvkVKn8%y`n6nTB#{5XMu(%H
z{%nk_CE?Axz8spH@R+_?oV(<l#;Oa=i+E-5^lk>4%|z<7d#^*Q-g*QHr^&bb=;8v`
z)T89NP$Qe6BF*FYIT|q!<0<*S1HAv!x}hyEEh3?Gvqr#3o$QMegx9o}su>JR4OsMa
ztYuTy+-Or_FPeVp@VnL8ZNq_@jG8E8=(Pe8=}!w`WfAJ?A{};B4~9ta*|I>6jHFO*
zaqC0nU_7PtYg!Z;^41PTm-J>pK!O@^Fw!!@VVA8yDrEkS0ppYXv(W3*k4Z(Nl^0?_
zLeiE(izwm$N&IDmcT6@f9ru~uEJHy2?HVx<iSi$m{M#X1M3f0RyYotDE^A`f!Z?<i
zrGiE?tX)(L1?OTZPmV7=M$jQeb7^zU=Rz~OheC!UBO`EdaEu0>$mBA~Vty~M;z6(x
zfHhhM2aDPnQ1<ut{ZYjqE1A@qtV>g~mI2MCcVu6Z@ScFK5$D>@N~Wp;MTO<AYRw8&
zz)9KP>u!r`d&H}Dr-ol{dHQPnaH^>ZCtTC&1K?oIMlv{nx*8o?I{=kHKz0mW=mt+o
z1-+M|q}uvlwFDq+AA{C;fY>=4lO#kTueoK02%P8Zc1&nr|4m2pV@2wmyGr~oB(n{n
zex>JB#bLMRCz~6LgC9jo&)UUnn<o2*^<pVOcoE3vDf>LbJebNWDI<f*$H(^w>A=7~
zlgktdf!<C5K{KQV{F9G48C+cWKuPL$zS{$$X#6wR*0|RId?yOX=;J;X7dN-Jas0%v
zwMKo|AT1*Kp&ab>K&t+jU=T$i8yOcT3ViH%tfns)>YN!$iui5-Xw<jT7BKVMmh_Pw
zU2JgY<mLvpdV<EwRU`l$<&Hs2o>;g%EgGfz9Z%1=0bhExTRz09twUE7{<zmMcmp8X
ztY!WQM_;^n4|PW---B?WvUjQG#WC{LT!g+--tp;<sH{J))3J|d<0^pvY|ah=8gb(q
zp#PFN0U0crT|&nKUb@F$)+)KC=>q3W+&(tm=PQ7=Ad%Jc1(Nr^uWxeXAP6Id{ybj<
zwBBQ$OdtllWVu$OM?zm51B=NZAv!wxi+2`+z1bmnEdEy~n>a3ftgO+)!*a%h$>6~h
zrqd;Q_ZG5$_K%N`yTS1?yhQrK&*yw^&i?Xn<*DbvQd1fL5+8vcU}yBnHfpKEcASX=
zX#?eu%}VRjJ%E@Elo|CsX4inB)}Q}vD@(+YfhocI__BSToXem+{FGu8*qi$Bv4!uU
zt$P~IPANQt_EISHg$#L%tUTj;LJ@xnM+uyzul7CZa#)|!-rYr*n41?Vxlk(+tI!jh
zY{;%%?e`H&5sv$qr8(=5RD?ZR;YL3}7?ppA278p$#Tc8vq6I3DoQGXMqMP8aw#3bl
z&XeT<>T9iTE0w&bIY3o;$190|O1Rt}h%sKE76lk?%op&$vAMhZxwW+wm)Ff9!?}||
z^LIEHqfbB@4`|Pi%F5W_5z*0@vGiKsg)-8Ctq(}BVTXUpla8A^LSC&G_jflxC;r}c
zC0REnEM1i7wQ+h}GFn+#CG&YE&d+N`OOW9(Xt&SJC4tZpw~KwtaL=Rjb7N4&wYRs2
z0~<vX8UUL+dV9Ygt#x6%dZpJbd_T%~iMaS+{O4a~qhx5X`4w~&X3prZwVqG9-S#%T
zy?gUVEzd1t4wCpD7CXyU7KdJ!lnFq>Zc2UD{Yl&s7XJ(0jG%nh9(7gxbXR7@LIvB~
zX0AtZS9~4E3(FHA^8x)s_8qtH_IsNjBBXSE?)>#iI}(W^wwOe(!@{&WTEkNQ9ZWul
zcE5=T5CrH<C>pLy+1WCD3Q?w_BE|c9TRU4@p_eOW>+?1CPrxF|61aK_GLC@ntp$81
zO`U?5HwAc9R@c`*0Q3l3wjHD$^#O1i%rJxQ@HYk84T-XRSMMIqM6<EJQ<uIZAn3Dh
zx{D=Aii=aZw>bMdK}t!9h=YSuWjWUg<Rf_s+26HqnC;fPb5qMJD>n+O7xjmrRNe~H
z5ua?yIPBW}{6RoQfy@BBTFJ*5;IaZBUVQeQ_6#kW<qzl;47nPjHPmK8w9QRwxk5?f
zNT0h_z)wt{y2MLnI#`>ClM!g!e<eqi8$mA!0nnIyf`XJulx`2*B{y-7a@-^WezjX&
zdkMICNcFIZ56H%~+<ri@D6Gz3|4uIJK2>wI;YS#Jm%$U+T~x_$_jw7w6SNm7O;g|Z
z=DMaxKu;Pey!KifW-v#cu!Q167LZ#4eWa}oHrxHb-TtMvaiH0#pQk9&-o6M;aHik;
zcY;FZe_5mt`4Z!(-@W?^{1z4!gCLg?Tx#lZHlYF^Stkp*ZKgQmG8-ZS12RMSGF=bI
z&4F<N0)2qjlGr*pte*_=Z{8d<nOy8Ic-&uYy6s8Y0$T3Lbe(gt*dex1wCUqg`KmDb
zOA<2Eo6wT`t7_n-Lip~_Xu4U~TS?aXn;ygPo@D_Z<R)4&XM*V#YezjhHd7VUUuxn#
zOMfS!;THHDrGT^@A!&l&sW8Gq%ZBIxMzp$Xq-D<nwh7?zr6P0Y?bmZcxopICaE?Q7
zApWy#Kt|@#b=jQebtv*491)CWHWogz`*c-Mg680k^*L-M{(Wu!oGt9-C-v&H%#;7-
z;p_SeLdyaevHsr4x?)jJ&Y9W23vm#8t@A->VBqt%0JO)zgwH@u#O=JFu8{X#b8|#&
z2IK|lnVYL`t;v_M(b7ugy9fyhJ=p?!hX_l6Y9cj*g?_deC(0BGHb!8~)U&iqX+l$0
zR(=AIPNjTL7o~4-#g9)P9SLqMmj?h$`+D=J#;mgq4buH@6oE8o1>9rw=g+n<@CyrO
z$4W3e*!tl$E-ST~`u4^$=02-%@>k6BvR;d|Hn2F*T&Fg#OeD$i5L0w1iSE}@CHOms
z)kGwUK{di%f$rL?`6u|Gqd{Y(qiFQ5t+=G3%8%F9hCJ``9;uuK+b7c2w)jJxm2ZWP
z&f{kW=68mo0~FO8e0TF&u-I!9^MzG*)~c@Mos{T-59?cYYU=Nx{40YVyuf)V4yu9U
zouST-4&F)`DJkT~`zhz==Y>eZ$mQJa-m*!*dneX4I#aWCfDlG59uyWf<*t(>lOj>5
zUK!6}O$^w0Lqo%@2DIYuCXveny}hg8Sp6U(f)3%~UD2{Xg8zs^DM0u^LGdXu)Y$?V
zc%~-&CfWz2;$5)HvT5dl2&IF)7;uOcFSnkwHMDGP_u#xRdK(v6?3g|Bd?SAx$h8uL
zH8_D_--gYXYNACX$Xs?|K0Oyf4bSZMdcmZFy|UC^zx%6B)*9OmSF-Gg2lqpJ+YcxS
zz&BSOf9=50$*QCBzkzwvFp|#qgu58z8qSh9CTYfe=fG!Y{Gh6Sa(Zgm@g2+1$mkUz
zVgHYhLBJBXF=yS>(bl$>SDy$>aZh%piWwN?)%C(Osw)T_jwxMlC`UBn*A2JbzekOY
zRfbj0#C`jQm?M?oQ&aO^1FHJ`hMt}t`oUJI9wMaor6*rUAK+EMr~Eh*K~|T6nc3sm
zA08dGbamwl)RA~ydV@*HWHrwjEn)Pk;gLvh4cY|A(R{5F-Xju2Pv%0Q<HxP`u{b*4
zhmHw!TJ0MGDBH^AN<rk<H4;Mey!VUe-0^rV-MtDzkQzZJGp-U|u;H%i8Si$UUsybN
z?t<$pXcVojAzOU5f-RlrcarUPKa3CqAyw5jP|0-N_%?a>^TgXI;SVHh_b<H&tO0_i
z#)eQScwvfrp~7c$m0|M#8_5;c<3btuIJ@LpoPnOY3&+30*~%;j!HVYTqCu;hA(W6M
z$d<djx?*zJk^?~^j{z6J=Ps4T*C4<RFsCO@`}1apxPQm<V4Ihq2Q9-S=>7`f3$Q=q
z%!ovJz6S^86Yzara2RH4*4W_^5RgWR>sX;<V4N+a#Hr`Kb15$`H=8LV1K#L&tgIE&
zRS5|R{i(c3!0Kub3@D=z;y?35YLJ1uiX;+=(iakQ$xQ8`cWqs9y7O4mxF{6pZz%Cl
zdA^YRNVjIZ_4t&SRZvagp+msaUxRNK`TK?gX<=|1V$bMT66ms1cn_xql89+2z#h*D
zd8lMa1I>dSz-9rr@Wg$g%7E6_nHPrhdg~H1bE+%q(&I0bhMN@MI(surf!o-CTDP|R
zu)E;wA${@Cd^FX-;f^K*$Mp8wY^I1ji{(`U+w5p8HMm$+p~4|UDPO&invQNFL&7Yt
zSSeHAS`eM(+Xw$$$Ekxbj5h;~AhvpOsEZVh;=*I_rvBht(#_yzJwF15-UHnw%QHUY
z0G!a4oFq=D<~P&XNr=NX+k6G#zkBZ6VMPlSjuwYEpSXDL(2rKkDs-vZ?P7{%T90Nq
z(YaMy8f;?@jOKl!5J=c|s|JQI*b)#bNU!-@?pHi9aqqz?pn2f;4@R>kt#-lP)*d%e
zk&%y4bWeEQPAQLEk!l&L?Mo`@E>9kV?<63?2R%h{_(gE+l$#6@3%HJgoEblLt;0O4
zCe>w`%EwSamtj1VXQ6vHX`@dwvI9mB&gVKA>uX0RmrWx_<1X54jGX=^Bb6{cvoec0
z(&6}ulTURpZW-NI6zYGgwOkWA-+0Hwa1uHA{=R!r8h)Rz_M+${+9~#MVb;y(1r&O#
zap-!V|0!%!?bIDN#jWuRxkh1yuBy$eoK)Y7R^1h{>7*}2=cgWj92>QrQ;jW=bP)d~
z`v^tL3fDRPDHX-Hp=J-&gDl^CQaSc2OrqB`ZVly~Hf5#tIKvvJw)a|L8|Y79QL8k^
z0=+_CL+^qv<4Zie9zX-aY)wLi8mR`rts4)rubzNL#Tb+=;Nti7{h|K;_1&5B{zSH*
z@bGY+E_Sm~xG!Mw!&t(=+v@1(03+U00T2_x>3VEv)%O*YqUzOFA<Fq=AXm7#S)hDh
z68!K<Jg4p3t}v1ztzbGnzSN#53Q)}_qw|NNkzy1T(#}SZLZYfoUc660x33GFbU#5#
zJ-)Oz&Bnsr)9D&}3Jh|o5(f$JcH4R5qH!ff+`cPJJ9b}@<~HCzQ}Kpre3|$(_tR@W
zjqUOsGo%T^n1s^QhpT)?BBPG!EerUEk9-&B$+qN3l>0oJC&4TU0d$HL6!(AFDrVHk
z{zhnwOr=dhe3z-OFq)Px8y{?`ZsDp|t$>C0s9;%AKu!{$e-?}T#=Y*P|G!|W-;n9J
zxw34M(A?Cbp>X@h+PnJL12+e<;XZB5T95w^CMX{^C_8ogL&DQ5>V9qjhaE;Dnv{Nh
zLBCxGf*F~BieKmg0S*pRmBamJuR1UU3Nc!&cSXi${W-h_^0yK<Z$KcK_oETs^V%Ut
zHZ4>nQ?O9E9Qcyv@oIo~ne6Rb9}utk=#=$fB49ObPGqy_Z)5<<fJ~6kwY%6**X)F*
z<9UW<&=pF@y4W_j3cm;{{tc_4H2$XHgu3dQ-@n(b()c`QN4FpIp&;dm=<sst0#TSa
zTr<VqAI9&IlCWqfMD&I3`+nFB8Rb3`4WY{W!!0{ncuO-&UETEu&t$RY^Rh8E2qAE}
zQKUQ4lNq$C*yH+ArG*fm=KF7?D4jicc&6`WG_O4wQ-oJuvEEr>!TLVkCcqVtT#`_9
zuzM$+zoxH`{!r%vB^=0nwPSk^@2;T{+jzw_<m5GPUc&%7T7do^fx@rcepBWxDXFca
zWA?`&Y@UbP6EIka+}w2@T_9HZ&tOX5qYb#$feNtuAUTwU^Zk4IYAdZ;hixH{L}Pz*
zwup_1`9Vw!@v*TRO68rebwmQenIkSHMy3b|2oR{J9%o<>4udhUqk^m|FpUR5d__Mf
z^oLE4Ayle8PY_uBd8sjlOfvQ{@(A?tK|tUKw*0jLj-_{$lwvRHKbE+7UpUzt$^E*l
z&U=wCYh_?h;Ws_4Dw#|An205z*{>9OTNhywH{>lLt@I#}$Uo`(BIZ{NH53{ZR96<E
zeB9UiA_k*vCVRQDdXRoUOO%2NsbVH^#FJ{1r>5p1Aj)1wkgG7<Z&1H{#f|)(d5b*C
zqLbX1IF)JKvE~J=r4u<$n~cp#CsxHwf{RjtPzV>w(cv${krP~$XGPjv!r%pfSY+Fi
zpI>BS5*3To>xP%oz$-(9-h5d3<>FpyznDv_T9^7hA6nZSX7oZup>eqVRH%O3BgrtD
z?*GsrSq8!qQBSu<G8OMWA1LR_5E&1p4uB0p(nBhV-O{eOogo+~YC$7?x;IzdZYy2G
zz|PJN5=Hz$d>Q8NhyB*_?mjRVK|{Z3H3#nf>DgESX|c3WatL@8@;tsR&A&b#lo{=z
zD?}k_r%X&NBrWHPp7j3=``*9>-#k|Le_VZaR90K}HHv^VDxIQqcZVX9lG5EB(#-)8
zP!LhNL%Nafl<w{>>F)aNNAEknF~0l99pei3@j3hKz4lsj%{3>RMLFUdB`W-{DO_pt
zE`soI`%(x4j;CiiQdW-XidqFrM(u8F4l6j$q;Q4CA|nYcR#6YQL|ZF(sv?r^pW_i_
z1RgZeE$+K+#7tTdC(eCvwC5CDV0#LATL?q>2|Ytz@j!^Yr?OGjPsMRlFrlzMAng~c
z7nj3M2z|aLYrGf#co;reqAj$RTW(HvgI1s9@eHrBg_-`ZV;t5c@8lG#+X0-|xhHVw
zBi0-7eIu$B`eJ>YQHvz;;PmY0n>6u1BNgZ&7Gq`3H6k%>%LqD{n3zA)ph!$w_j)0h
z02_U>0FWy<RNmeeTt`0zM-wx7(6Bc4(p@e?mx%^7h11b5h&LLw&7`%<NHJ_#{dn#j
z*KNsnbKM)6`gZH!ua<_Eus;7|!5X6e43X~y{=M*D;5e*?a+>Bs%l`bX4kz#CaX_g<
zvUxC=ZM^Z&h1+T>)`=c6X>Ha#fFc+5Ui3D?A@icLDcSkBGfANazN=Sc!I1GS_*Kk^
z>#p)~=<b66K?ypC1-P#ADM6R)ExUM$%eKWDq6cT$flv3D6zbOttH(0rFHW96Nkvj`
zv}alo){xX|9natk9n#Y~tsGG=y)ks0wvx|@m=Na_|1k8b1cM1mf8urcu+fX#PbLqK
zZ6{>l&#<K$!K~$9Da@n3*0xx3Z9ew&p4-#fEUYM_!-uwDx2aXykffApf!vqT{$`Hd
zd#2pdbkniQX&aZh{^Y4x+7nQ8s}yRPA3DHpZ<brzU2aeIZJe)&X^QwHDl+P;JB^G8
zCt?{F)i)5`rfQhm#@g9eHy2uM4@_n@?`_%Jh==>rW+SL|`Z-u}A1mdQA8u9Be2|Fm
zzf~)>qPoF>zom;G&m`Bzj5h?Kp=@iiwuwBH10pt6#T>?<YIgZi%!l%|$xw$SJ$Lxq
zQscA8k&|LtMDK4Og$)-)_$F!mNOzTvr0cphc0wR}&y(qJty)JFb4J!OI>($8Y`1-r
zNB!wdn5xxwHv;?Lb~-J|JmszzO{E-3!7+EpLR()fJmn4EyQCdZf9(~L#<mnqGkk~F
z6HSIreRM;s<7t>wX8kt~V~MQli={Pu1mFX+i^uiZZriu}^>|zRN+e8Dt0AqgVlPV}
z(*d!pkyIdfzs)#^qyX$Hsy}QU2zwq&%Gy<`QmT#0n^VfmH>!o8OuIKvgrta5eno^2
zHfiXcTDiw{`&Q^BMsus-(m`LUA}wwD>234_d=>+`T#>?n9OWW{zG)$3R`ml*rOg5O
z;$=b#bMyP(rLf@BL|);1cH(ts%>ofd9#7^jCGs8UJRhY?sX|@M_kEzOC;2jaB_CT{
zj_UnX5iW_88aAUP!6t1l{=$3OJua8PjF3gz<eN5a`8`8;)eDZtV%KRS=IRIY#|c6u
zrIW|_A!4|F_&L}vc|1+g==!eZEqDca%W$^iAJtEr+5*spy)t`KaQ|jeK>xB<t*!~6
z54yy}#Bi>=<r5N8Qi`%lO@YJTzhkWrBxCwyWl@9s<d+!}U{8?)+!1iP)i?@))dW)f
z>ny`{qWdh4&HeWTIt8)(d1Vq@bgl(dH_3{tw`Q~9_-Fwlf&4st$1NAU-Q4Ui7B$4W
z!Z{x5YJDut8O@)GD1_>Rk{EBmPba7No4PFzY@5A4p`JFz7oQ=@GT8{6cYl2mD0JmY
zW{jAF5`+6V)FXO_7-vV%M&5>Oo-b6uLgijVT7zpJ{RMMEK)Vf#Hf<)ZG{VWT|AeK%
z;C2kV+w;vF$+x5>4l5|mHX1p-SV=A$zi3q6yIhnBP3Bx4_CU0slBab`R$zAlEao5t
z<GBK*A21Z53zs1%0%RG5Zw}x!3b<U}fjbny%RM7uOVPUCw5LC<-b%{I7&m`J01^t3
zAVE=Q_`4<xB?z{3!KMg2t=Woakm`e?Q6CLiBa`%_Ry<og-kZa$fMLTy9*OZ5%hYNw
z8z?v;-S50r?^tG5fS%2Go~#Y`P2Pk>Q`LE+DoX3f%LsysD+86-F@mM*JeK_zz?cJ>
z>^?{?lJG6*(e;sBM9W1J-9fha;rW`)H^Pe)rpaucDi8QC$gCOs4QQe|XvWvcJ&`~6
zMw`NLuJ+F&FdYkD9TljJ7sR~(!vaoI9)v5S+~7w2iuNc<jKmawH+0C%sI5h^<YV!_
z1$hzOhFS*)_?P|G^;md#9l^{EF<DJBh1&cLH%C3Q`#=j_3P4kgajihEq`17?7)Top
z=gER|L^{LGpvt9_tMfrRV0kiI&(#U8qCit)=U~a!OKhLq_H`~srhhZET3XEgfNjr$
z#0ZQT(<4lPvgl6pLIqWCJxp6EU#ni!TyWeA%l6&<6CzFi4Dr3r@Z!z&#J7Vs!94-!
ztSmP?H;yDo)TUGe>e3?>a@YjIysY1}<5g+f2AjPpyJ^XUkzBh2^S^Rb6zQ61prKBm
zY)6BX0ACAXRz@pMPYvQ6%)Yp;EE?P{L%^7-G-N#FE^1Co>VDy!={k&*%ZKIe$Q}Js
zHU{?%Y<qy4dM`rK7Y`=j1ke*Uqt5Rl9Vgopp@7uFV(|NMaMCr%5Qqg_zXMQyy&@R4
zhHzkOQn3pvn-dK7&Yiqs6l@xaIw><M)J0~t&(?qnmf8*-+&s<*=`P1Urb!;?^PKRa
z`GwLT%;_NMpGaP8@@>6)yvH3hB}|-R(RY<0mDd_RaJYHo^@!6NQ<Yt>`Ry9<EoN(Q
zN9uyU>luxw8}iRmvuUarUSVQ7NB#0zlBrrr+5n;TjFWLr<zh@YBo|v2!2kphA~1IE
z-p4|ba|J&)ed(s&7;=LIiTAyP?)sa`eam8kjiKz6VEda??QyT2|4zp_=CiEah<8A|
zUai`ma>47IL_t9Th~tp}jnxdP$ZVZ@zHi?scLY^YK$Y)#wV4IeC&pz&(Xp=M0lh6^
zUi<VWuS(`GqnnNZ!HG@CD&K@0O*~ncv(OOX`4ao5%^%a()JEqv>wT9_>B;%N+el9&
zGl!~!ou97s>Y2+0^*67XiNTG_pPHO-WjllMnbt{0_um$H%dX}6lP;ovf5QKY$cXWc
z>0+!%`q){}mEnTGGY<hS=Hy=SbQ=9qZ^&6BbjN4`{2z9D^meSOLs>?Ufu*IK{}!kD
zG+zv0kjfV+%sfq|zBzed+?vqBLsP)2WmWZIon?AwM+7gF-V)SD-?2Qj)t~o|^m;e-
z%`TZeeUgQ#*eH4JS^H81la@%P&vPfPrWsR8z-^%B+@?{kv!#A{Z{f+Rse+GZlq<fG
zG0rV?KR0lHd!_-DSC616vCsB}?@;S1fvz@~$3h3_yTYQksp-ZnE(Z7`lSE4ZlNgzp
z9D$1~j(7zC^L6)k*Kfad_4ZOwQnsx+sA)P^gqOds+Ml(xZE305v!Q?49ZjS<Ku?4t
zJq2I>oR`%`t6o0$yhtVSY9#kP1b5<2;%?35jNx{#%XJRjVm&Wb;D1O&d%G7R_EJ68
ze3GAM!(3Y_?wDfst0Gsg*N&1D+HHp8^2$$;J?rTRJZlrvRpw&2(D+>ICb(b-3f5{I
z7x)B@^*&F;1BG#ViO<0$otC08Vx%5++-`!8zYjdc^oxi%c^cZ+Nz>L+bMB*($7rvw
zNB-MyKM5Oh4Ne=gU{rj^>#|1-w5j!gDsokAJntJZ*j1c9vEd2#UNc45L-y-z4Ar2S
z)u37T-L6`O;?m+D0~Itsa?|x71Ty}#A$Kya?&Z7H_Ffkz_(_ce0X9FSe+Tx*N(0S-
z#z&PrV^AFd&7AmpJxw5pM8IzH&iDGO81zyXy5)3scIM#~tSZ&Bw(hILE+=Ls3;`PF
z9E9i=^x9Mmdzc_YJ!hT)2|ecO!w|8w9|t4Xyk(Z~rlI@f<Y#=bo90y>my4S97=6=%
zZE2zB&3&z3g;Fqi7IMoYL)1`vtAbPwE>!ZAm+w`<8gIzn^fYS|@wgZ;BmaUXy83&Y
zT5F#w-)yf7i8I;gtZ$g=>)_<k5ZRdsIg3#SXcRQ(i5QPsY?q<u%ICzD=r>P%bZMhQ
zXuuVqSE4#=$LVDAsVCUy<PEJiCEs{r+@r^mGO<lIdHz_RmoJg&+}`w39b2OS7vP&>
zFRc%{OP7W27X<D4=I8Gk+yx=PgY7icChxAcf4_LUnUxG~e*-Y+WpdxLyT3mGID4qG
zs@^Dd<Nb?wb}qgX?U}PT5|5wZhqTW=ED}qrF7l?O|8i`KaeVs``4j0}q)T8yTv?~+
z&ljz{NBd$bo`+oZNnU>5%(CsxLsNWR8VbQ@g6bfNOpsDR3CV0QhVXr#b-|TcE#!%k
z9SkI*EnOo#d(2<G-ycHPBqlQ*DGK7`%HSS){>8HOm5}fcp*PfemM-@UxEgsn?~PZC
z*vh!m;7DR0NNU``$LrbJ)?aU|XrilzqNHnm4a6p?iaR9zeyowL-Kg$c<R9lN12Wq#
zBV%77_JQ|#HqU73Aa!5atT#myhL)D}2=jrnbGHm@rV8NY1A~;+Qc>QGRi%>wckqvm
z??it;hKI+s#}Ab+on%!0H22+o{DVO64Gty2&7)QisPjny*@6cbpn;6Ciwv}eG2<95
zqYwJB3|ap;$&Wn%MT82e{L}*uI*ExlYIHl_s4b-#AXw`E#^Y`^(QP~D&$jLE$$m}^
zu$wX6mvB}>Opx}!qN-qgEaMxm4$kEqttW1a8OPDa0P2Nz^COUnUhs>AN=c;_h1DMs
zXKXH7(d`6L8)^~@dkKcTsl7Kknsst9>DN#X?r^6g@>tdvbs91KX%h;cVjOs1DsPgh
zkOy&swGStOyfOJ#mhS0dacLA##lwnS95eNUWf3sUr<U${sx0g67i1U@2z?=Y(%Q5C
z-3Q%qnI4#yS<ipNr_Ex!+S1!78>*kGJyw{vx>-YQNW895q#)kKtf!}+ad8-b#qP4U
zO&}cO<=c}b%W|3F+^cQ(cmSmV8kd}x(s0>o-Rl!>Es=;JQMyvE6N@nP1#k*B|IPVY
za<RowtzKGy9p{o{@{O0SwpB!nt~4h84pMRwHK?`rznTps;2_yKoiiDSMQ!Y^i05I{
z!*Zbf&;Va5?4)O9#e0Z^M4B}xbQcz}9<)^pG#O3TBTj!*o)EvI(SV9-Fs2asFg<?_
z5=>avV(7Kug-cvg%eu{4#D2gT+;imttluW8M?yzto1u8H9MRRM4!YB{%#ybtL;PCo
zjIvCwY!g?^pjPYyX63OjKwnSy!Q#$Nf=7JsJdQ{NQWnO2tIP3%JW3$hgr9fAWpM4i
zau=yk0VYYkWR)9~sQ&Cawp?P;&unMZ_SdiS_e?JLYwnxyH?F{)M$gi6;ExL~9`8wK
zBT?jXVLjsdxT?t>Ze_NrGExlQqXNM%)!sjN!;$FNp?!Eiaj(a8IG&JlteqAi-^<>|
zfei2kd}r-hQXy<zzc?NeRb)D4HArF3x#2}DdztUZ_?!+wJUrDbAAXmK1oHCM9-c_=
z(;b?MzOtZ&)_<}s9ZSQZk#M6%o(w(vn;na+y3z=d%<gf%6f^~%Vqcn5OgDGzq?Yzi
z2?s~(SCMA2^+c>6f1tV<^>(&W(=l<I58WY7aNYIWkBqJDDQaO>prqd+P%v>XtV5*3
zSuwr&tPMR8w0D}eis~oL@?=lNeN-LwLXz_xG8`l>OETR&Z=EOYZAHN6!uIH^ScV$^
zsg{l0q(R2!(k%;5;=@;?k;T`UW`geB26Re=7`4m#dkCFO`Z~2M`r~eJI@gOH0(C()
zhl36WmK@gCkF<JxXg8!9)aKd%-3eFG>BRjEJr1fsW}345_jI2FTiz%v;)r}cqA{8w
zEYv|$4hop_k|Jx;K`$DITb#%s(`OOTVzin@5^f!7T~H3H>v^Hyt<ICIa`9lvoMrPX
zn}affpRRXkg?19t5%Im1*6VkpKx|S>{Fw#=kHumnA>2*f#qBJDS3Yp4KAdr3kFbUK
zRONB0;vI2uwF5M#KVq`s>x&bA6+#Yt8adTfo=|0fhmGWXXE$Ka98?>P`5x~kql|8f
z5Va^N!0(Gc-mPJYUO$2{l}(hdzHM)s%*I_?l6b-R!FX>nvel%<=9k-U+6#6u+9*5<
zZA|*Y=IvCh@VB)#@NRdfQ<)12AftuAql;E6?YSSIkpFIdVKigg-TrD}`22f4V<!%&
z9p&8#=L=b~bf9E}+q3`rV3wXedrLRepA20b;y`XsP{e&@xr~9c69Gy(Tw=L^PBb_W
zINiLiq)4~9vL`RDD~c4j-fb#-XtKM2jj`p)^8-(Ig=}!}M9qZH<;5pW$c{VZ7tuQm
z`8#b#g1gTkr5_x`*{bo$Do{+Q`+*!7_MA5BVy68W)p$UfB+JW>-*+v|=Bks&_6Fu|
z;0f7G+sM~t34T{U^8m|*vy<YFq*Q^V$w^^EAH40K@$re)Y#sjj9h-*Jj`s2~Ppv!4
zN<0L{Y=|}QFK}^nKJLGOB9E0E;8ckuMw{C>c%slW;4MPX{luW(K1=&jmWY3IesWPE
z6C79IJYXHnppiL-%eLJi0+dv{1JH@OPk_yU9Wz0Q*f+AtT<pX+j~KlC5WEE0rQf~<
znA*O0DJ$>am`(k{)Ac@i8t82M%7qfX&}fy&)X=gcjpX`Z3-VZ4T3l7M(yG_#$Z5Uz
z2}dNXj*T`VlH*Bk{{pXBps^@ur?;Kx`AVH(PeCpt^za)`hIT8Y$(5}vMSh{)$ZsvG
z;qDU>v~pcJTe~XVG}93p+BnyG2aO+f1o3`K7J`enT_@U5{)LiwZ;S7hocuXlnpPui
zwYc0tuXi-V5!&Yt8e8!G{mbkAO(nn30*^Sgss_@z;b70cLRV3g<hkU3b<bU_@}hRJ
z-s@OAQ_!Ybjs6k!PE3PM$w5hjF{X9liO=L82~?vqJ$rlSm0h_=f}xIc+ojIN_=s^J
zCDo`=X0{M6R!|c`XJX)q#%Ikn>ZsyQ8lQG`3mp=-%~$=j8vRh?kQUAOHRnOv?UELq
z*1#vxf?=JXA4xbqD4S%=_bI>&ou!Mmg-6DItM_|$jv6WT7Qj*FBh|tnCI4_H`uNw-
z53|zJgWN^wkS4KFG%JH24R;S-GrF4IA+im>=W+351_{5UOtn7{iv^`tx=va}w}Fqy
zKmVo23XjZ(o`V}ZC1UlF@k*mV`}@`35|Z7*eh2!a=}s=dq+S;Ah~R)wBUPJZ{Vj2I
zMDhkrM)q~O3N53sRLys-elIB**{ULxI+x@8#U79Yc6l*^)Jo6P^zvg~T6kW`#a~l*
zP)Nlz<;x`k*Se@~P*RFQv{f}543m*ozIinE232lKEW+dP{bnp5?soK<0!Ut-wJQVu
zFAkk*hI75iXyHVi(yiRSfH>drvz{;dyBnQ-VMVnG4RzI_;s5G&k;maN%0Od{*=vCp
zY~FPTNBg#^zVNa?qMd{E(RJ}#ihOW5Ubl=9bAHt9f|2Nuuvhy+BjXEJp_(SvBpgpS
zi^V)xuZ@^EF*sy%Z_TeD5q{b-HieVsHcZCpnm5;KZlklKD6_<l_KiAruTl=*=f3v2
z9wdGb@g555|Dh<f>OsIBob(kSDP$J-y*B;y=F2()YBj>E5iZOP%H>pXmt&~~J`fBA
zp{LIbB9eX$Ts3cVMi!agKJSdwJmgj@|D)Z0hPpcYrM|6w%*@K{#!R(ywDoY+paZY_
z@h6&sCvobRY<J+ZBSXB9*Ae()9=pHnBORXj>qG;-g<2T4E3giHaH@9xy1SiOLF&mM
zHgy=-Sdljt3bi`y!E0wJlgEw8ZqR-tQ+NK{nND#XtLHeJPClX_9jz|2iXQ^{VlCA{
z3?A|?g?A)lU+Ohg@rW9=D}B4$2PR~oZy2*5=?N~7IMRb4?Pq>!Vu|7!3aR!^Jc&O&
zV_Wn->88P%%{ciWl>T>dehv(0C;?V{nAs{M?h1K2>`Y1gDBEeNa?si^Lg!#>bTpi+
z$!K847PBU40{^e*fd%M+p91EJMa#W1-IY%+1`8t=muQ-<qd&sFu-sc+=pmU~+29-4
zO%%6pVt?z1Z&i)lER-S5ri>I^1b47QKLU{14F36+Q44fGaqlJoj;iYLqmn)dRBW&r
z7i-OM+~P2Md)xuJN67sA?{XhBwx}*XN84~odU0ouW^UIt$vd^8F$I4g<vY4N{U|iI
zg!GGRv7lD+x<A%WVuhTsuW=_hAhpVPjIS8fvtEjJVE0I?D@DJjzYJQAEs;JEO7H78
zxEtL7;0nuh<HhoxHC>8}F}^y9wlnVWf){QgS1mNSL>PjeM#U^rFy6Egy(FR772Z48
zD!zmjDG5pi^QlzTu29A|iS9yjekLS~27J9i@;cI#<<t+N#0(5g2RbEnHR|;k|8Bf)
zX21ESC1w80x)CF9l3Pk~@2R%6+l|kHWNoM6(L9xF`1kRb^tz65Ez{hMl!qD#<?nH!
z79$nrgDuAz0bQ^PQcT7AmO5@5L~AAIn63J2<;$ox(Zfr9@;u)qK;t8p$Ff|}$Mdt5
z&N1gN?!s5nm>GJS8n>RQvHYwP%q329!!H84KwAMIT!?(|bsn<jIjM~6ynM-sMTr3`
zT}pXlKHMhxHMT*i7Ucnz!rFdvkx-n}{W`MgqLkPWk}<&|0Qm%ozlycXYD>M6@9_x#
zhr+r3F`~#kS_w}6wdtpv-bUG)mc;Ul%X0?b+xCGlmp|`omL|MeYu?+92e{`KJ_jV!
z`HAk;Pl<`YCMJ>q|1Q^Thbp}M`Jc^QUC0G0@z?c3W)|ThAd}^B+JrdKJdmhyy)Rx&
z(HAu@a4#W~nXNr&z*x)<(A40GM!BbI+H++Qsc+wn;n;A|nyWAH!rQAZfL5`M1C`&K
ztVKo9ZA_*~T3s5_%3%;|t{#y3wS(HcWmyiWci|!v8Afbm`_~>3Pi+W@n=0zLi0oPI
zfQx{|ZrmavVeXv-RI*d$reZW+Qr{al8G7*kT7OmY&tk?2YvK08q>65|)ocbHSJb<2
z{-3ONh=~dAhoq(*_iT4-L9hnzt2?K6)iDb5DxxZQNWK-BYBb>0MxCFQbyEi^VQusG
z+=CU)5~#MBIj!;M{Y|v_;Crf<Y5)r8dw=Yq23r%IMF>aav3^5G4dzd=M6vuEdlV#>
zf<Uq&T3v(>cE`(`OZ$TWthKV*-LuOa$8)%Sp~{@(Yw-<hQS0V`KeD4Q9flT$@M|ka
z3LJV>W_#CYomNn4^3vnUuA6l(9JDI0<-zUg2%!<PhAiHt4Vmo?R9b+dr}x5ynh3i5
z$hiPPBGZMglQ_D6F2)nDS&eo&F4WypdHz9--UR4k;|;%*0Oaowcn;tED<%(zg!7<8
zc+@#SZlKir$iB`(rKR%qP~!rso$#-#tagjBTuE5cQtXq<xy=sHWUYGF0KeMtXYP)$
zKxUMlwYa1mOE<PP_|r=+uH56PI7F>#nPthT%9NnOMHqL9pY0>AMNSu<hxaC!@!;WT
z_Q+>*U8`mL-;x-68UwM09#uF*Z&ghANO9;^HguzLcRhy7`kwLd#3^_-h+FOMY#!f;
z0)>>eq5629n|(1GQL*naSbk~Y^g2S`0>k+)K{=8MP|R^z)xT*)gaecsk74J1Zl1lB
z&G21I%kJFf>N%0~|HuT1^i6$Mjy*3ful<zKbOj0a!nw>bbAd0J8oz&Fz!+*BP__ak
z7=UF9mGW>z%Qaq~ka4w5=gdaiw5^@!991;!fRT%8R_Sc;cEQ7R-W)GbgMH_kf2~Kl
zEi^fu^Y8xu92hy&*^jUtxvOgFQ;NL(745{I#7>Lf`^RVPtnA~9`s+kNMn-fwIXTcI
zkpS*g?U{>VbAZ?Cx|NfK&*w-3uu$!W+bp?cn6Dws@(PT1$=5ntBckCq0jk2m(GeYs
z$?erCOpOcN{S;JGY(Tr2%VrkpEVrDv0DXR#%b|NcuwWa9SrK873P^y21N?0L3w4qn
zpmzhwTu4Ox!v_|?0Kf$H-pvye+zWSr0^V@BPE=tv6~@2bMV)K}&_fQ3i2GmMgds2j
zTU#*Rv!_o#>J()Gncq5q*aNT{v%Ka+IwbaKDMjc+J!EqFum~I_JLaj$RcvRcw<n(5
z55H&_M-|puFL}8;7_7Gc>+p)GwS^2yJT0768vkUrZli%!jshqs>Kn*&t`|wTu%5{#
z(6I;lKIDMuib2AM&0_;vBWC3qK*~IJ!#QoJ6WAoh2t(dUGjHC!K}JVkTL~35*@x!q
z0sr@f?z=?UK!DK!2kNgR@h)CZ07U_OX+YxG0iQb1@IFh{ZflelKzMIFJv|#9+Uwui
zq9ekubcC8NwfMW)G&*iR1ui`XVD*7l!+StAp7jFg=_4d0z@O+B=m4q+{=l(m5J>aG
z_)X#A?;DPB+DBvBAVHmR`K#dyump#2$GK{6r#^Re*S=)Z+|(2}!xnIYLnQI_6_40e
zm21x3#!5~+kBOH;8$h|dC;WHke5!Kk-|shX8y_F3dsQItDy`<(KB+qZ0Ulfc^Cy(q
z<b;G?;0g+(Zy5K~p*A!ABd0&Z#{Lb+ot>RfXOkYw_lSt!VU*IH{r#EXiPJMPaoz6#
zuynK>%xu5Xjt;<afH?xDSE<vs%Gur=Y?dF;h{*Z*iTs*hnV3+4|5yTSY7yliXI#MI
zG672h1X{pI14>v}7z;j{k*Vo3Qqum_uJGf&1emKIVA!VHuXXP%G(uzrD#5Z@D?r^J
z#;6A@++%=WMe!g>|7O@ZEJ2KIrU#~3sfxAU4oViB?S_)w@3adZsOb1P4SlzkN35QH
z)s~=M{c8#<Dbw@Tp5*GPMP7I?dr|EvxYf(II7_E*|N5t6iM(1a=P^!;inUa)ntVm+
zF`TA2Xkw(l4Vl?`+f$B$eiArYYW41%JyDFUQ@s0Lm%xGr&h_?m;u!lCkRo$i3E@6C
zJ6i{UBo80o`X11BSOGmcv7(9NZu+7F;1InQf9s(M7+x?jkOqTYK(mGeu3UCWDrg{U
zA1%^N2BsZR7eZ8&l&r=<!?M8N&yNF~KOUFlf&xasOIiw-6Y6q!2e`GcO3#J~NRRZ*
zO#sVX3^JjwK!6nRN;hsUkHLTrc6=bE0q&Wrx?e9q3mzZJ17Q5sZTs$G6mCUF%Ww7r
z%ZL&OA7%!dN0u5i_w)6019)gOZ|A?5>T42HY(KV`CdK1sH%0y%0YUuf4fCcqHOZ|-
zSZ7=REg?<gSz)OTRJZhjof{koCA<rFybkMd2Q8?~tJSy3&|R}03yb-1x<K1p?dkpk
z*<aYn{SH&;BBjY-ZTGGkxD_)r*V6!c^b8#6Mu{ga+bC4nE?_i)CqBkDusuFHS}vc^
zf!70rFy|11iGFA_XUt8mu){i<nBkLKRGHC~*YLp|uw%bBzFp1D>tKb+P?g?0U9kOd
zWXnhH0hFlvpgcjZO!%Y#jVz!w7Mmv-@fGOk&CRU-WiWoiy5y-^JCT$bizQ=tyP3B2
zN9D-7FzRSnnwb@tWPw8*t{H*wIVh;xnglvnm`blSos>-yh;>=;s%)x4^s_Ht29EA1
zf-b;y4W+Igc)`MWG^)B@kpN*kIO<_O1`a688yl)y{(av_1yR0lG~jGF>tBNS$Kd!5
zYysE|4-c3#5DZqITmkxxc&_mr=8Af0Mh=m0ht$BjS83*btG7GOmpFn7)XY{9*|&YY
zkiA)Rj4+cqeP~UW98r?yBhAP6a!TZ2zYc<f0)@*$0z$(J_7wlXP2UnUtZ@ItQus3_
zj3~)M?5&BSP{4?SO{M^j|HZ`x+yU6V4`FY~vjfB<c&R7n&g7LWZy6t&n5rHu4*^$V
zts2KKwC^*aV&Luq!uLZTz+DF6Y$JqwzO%2C2V4?xZn*OFvYws-a5!{IG>*Ny2=mL!
z%X{j*Eu_qr-%0Dh`yu56-6xthZD!3FT(uIAU!PgHGGmD)uBv!);PsP|Kc~5uDg%A`
z3D$F4)H>~~i64oPA!*?+>Gnc<4nH{=O`1Y_9Kd&3s#K$rCYfz#7yyuM<L%?qIiJ}^
z1D}k`zQ#^7slD2{+hU%Cq|J*=a?>PJIOS{KW2r42c|=OJfk7NMV1xsV3ebUbXtNo|
zdW9*C*?L}9)-an5pi-jYvoz&9;I$77gyPVv6dQFtg^hE8PCB6S8IR|y_d>Dkfk2lM
zBreK3ymj^UzWo04VO82@_1Y1P5(WkaNf+lY)^CTC9`yccCc?#S_eaHJxT6GbB%++N
zM_7d`smMellnGN4|L2MM)vYtbKfvbGn35%~kMvt^o+Pv<=9vhr4;yK>9Y&!RcT`$1
zO^ELU2!`+<&UVyopxLU|fEa?BU^iXo=CIiR!pgX-x!IRWCXU6V7ayqQw72-94sD-;
z!<Z)8!EKo-PzWQKAMz4$Sv{WJT+sXu;wju<nfd#?wlb=4(*y8OBVlf^O>-`4B1Q;c
zkO$Q0jzwnfDT_hH9ARnX@m>AhP5ljvZD_E41%Y2QmMc&gxr-#dxg8of7&DF`dc>8H
zpkMtGLpDo3RmJ#Yl7%{zThu48!+^=h2H>rVii#e9sX2HSU|#sJ0~kt@_<+@gIq50@
zr5}$-I&g3vJ{J(E6J4ow+D2>wT97dF2N(_JEQU$D2Cn-yUeLq2;S56(l9U9{HJm)I
zE&Z30&*m~z1F(=rPAUpGOhYE7uiu{!-?zAM^K)|8y#(rPpwpwt^!mR=mW)H+d$pFb
z)%~**mFZ(D_O?KxtnUERN2#`F`+|NNhI3(4V1GP_;dfzzF)}}7@m%SWl9B>SEvgT=
z6HDDsEfgU|MSEbYX<`pvhDl%CTaaAkIK=WgPyzPrGN9l8aqNEKjOgdv*(m`M33VeI
zgte<7ak2-4CC1X;_qQ&?BO?zogbyO=Rbzy_+>>xI8X6kV2-qI^0K*fzqgBa&cjy1d
z&TV?Srhv;@APGQEu>P*uzZdE<t(1N>eMIJEeQHn?PbU5E(Cg!d6rlSxkoHvGntZRI
z#I-SBDwyt`Ch}*+*B_Blb3|}NtdVt()7bsEr#<XnC1~jg99RHk&WA1D(NaX{`Iq7P
z{|K&jU`&kBiAi5y|6L+q`QR<^@9~HFnV33j_Zy^vB{R5p80Ca-9eoXL@V?n<0!}CJ
znk44=09&jLq6H9@!RC6uOx#VES>B1J1Kt8AFk?|NZXJ4P>&wf^Hmn##K=ZdQ`c#EC
zo9D{96WyhhtxSMH*@D`kcZg#gP&l4~#`ciARL>Bt&K{+q>QgtQxBA80&-_}^C2t*y
z4-S@2THc6ZH4GvyQp|z9IN*5rE`jF<fI5Mv^oM^W)OT?lU;L0ThrxxaxKRSm^xf4q
zGx!ZL&{rKY<8XC#9Rzm8z&`r5nHdeRWu^R<1+2ut;{f?P6OhU(6=>juh`j;ffq)e<
zcmyUhyC>Ly43RG&9E}%hO9MWMowIX0_ygcik^t^uyXD`=T3T9CPL16!GRtR63F=^k
z96}d`pesw*^XfCXFq0XtM2+9*@v_d=)8nwS6!m`PkR_h%1~jdw&@msd7<5(2{;C_|
zUgnyQq@25O9}R~L+s4piyT+s3rB-RndHsnNVcwG!j1x=SrfH&XSv;c7QNKOeYs5w|
zAn4TpDwh+pSg{=eywn9LFYpGPV!#>NduUAw-jc$qHz3mjk6(o?ow7<iTKadu>lK!X
zd|3g8=vQaE(x5fS80AOH8thS*pW2I32QoS^N)ZTw$+r?PPa;q+9K;W3GoNVa8^*S1
zh?b}d^F1nAD8(XYSc8AWc?Op+3>d5wp@QH+^P%_X(pIrjfheHi3K_Uza0rY4JCUR!
zXNBJ!A=Bptfi5e(3xQuI-GV;#u%@?D8E!0pvZTbk(HzUrnV=fAdXK4Ol0QKHDH4j5
z^yQ|c^as{d_8=uHdCUwJC8k$6g?wH}`3q((BCEQ5*L>M4?vh0j!pPjw-$i2e27$@^
z1TFc<w-QpHA82wq9t1?~fxag3C{~fb18p(2Z*ll_4BwGm65$11>%tXUh?vGY>fdju
zk5bv4R7N@;GTtgx@i+zx^92-qzQ{=!0uuENTo=tUvu2<rEuXiaur;9z8883-V#YxT
z*A58uAcO(dV`9b#ES)?819SrOhS^7bbnLj|*9z+S1psE3$2p$3d+EW=YOO?_Q8iQj
z@MiE)=D4pOnb{7alXgl0HChqI0%yRN^5(GsnY_l;@XD){2?%dtS>oSm2mOyOz|aM=
z!=<(hLU17Xbb!fF0x*Yf!-}W3*yIbFTLDBL;H=aH^h{yKJwS=>mm?qeMxsJq0kXfR
z#^CQ<o?%c$2{kon0@|pmi>*7RTV7}C1XH5%Uc~D^Wm7EEWK;;R5Nca6<?~WB-(El^
z<VsG9pu2aubo?b_01y!W9^l_^n!#5AL4DO`Ad#P2PXD~|=Jxgy6a<~Uz3GzSR20<I
zntUYSmtT<Kam$Ix=%nW-UE4iMplQ1v_(c)pMEaG*y?cVEx(bhP6m*PFJJLGe$@w>X
z4-Vi-O;7hskeZCb;}PB6K;{pYvM?hA{@Xzyq@V)^+&NH1tDOT3@P{3<_1~v=mA8Nt
zta^iI^}qE;85O!9Ky0`nP2bi(QqRzSlbLz|&vsILMEH&uiNHOH;Pwe<rP3S`cbwmb
zvUi-l;uLpHwb%^LUxs^`3xBYpx*u`p`tLy@&__&aQ?;%Jm%&~Ih(CL0bGOJLVupL?
z^Yio7mG{7-s1C%JC|Ot|f@Y9CP3U*+Pcmtf#-O{vyka3j(&LX*HZIdNzfD>XjV`z!
zgq4|r29)uZLCp!Fj@3<Kyb2xBz@=~dZ)G<>`j%YF1imuAi};CyGuBjj(4&X-zdm98
zoGrOP(*dNLhqmXZr+ss>a&iN64YdcgVEFDIg&jM`qw$C5-8SvCvm4>l@&i@0-p@XV
zsofL3k0<f{=+wnQS53W&uS~Q}JFU6tM}2&g7(#)djGTPhM`s;4;<@X}O}oEu7W^wh
z8aKI_2n*{K$n79Fb#-8?%mW9TM+TB?4_^U7{@%_!@MD@a9J2_XhzBY4XM@Yj3FGCg
z#vk}(eM=Cx{o$;_vW*T>ibzH-=m~sX4?S&a%~Skhm`TwV9u@YOGqEY>!25(!QPpkW
zYYm~k4)MRm25bEd(D{H!;#$)KWAd1b3s>mo<_6An`3v>M@qqAy$#P3PFh+r}2<nFF
zjtYyh?tfDfl<NRCDy-N2IM4heG7`b3zP?`a7Yw;am(zrGB7WfGS$qF4_rh8WYRl?%
z-W;ZMxd!E0^$hGz+dBgv2@~c7V+vrbL#Vp45ug!KvXD4HT`52(qVqV5$H|OepZff%
z*X~2HH&0M)kQLC7Iue9V2;7hTGU!=}6SZad1XzXQ4n~l8a7$uAF7hMd!^lm-|JHRD
zk3q<<QoGi<dIr$!vp`-12Xb*1>uC<~H_8GI$KGxwr`eznn5<BTogJEpAw5WMzhca4
zp*Z1v;Qi~Ok@cy4@T4jv9z&q;IXo`1%wTBq>n{DsOc#k_sDz9E;Ms&4{Gu2YoDR;d
zl&6G-)C=hLhL?P10?Td5vEjCSiz@^42n09}l!+6xP(J@dDM11KTJP{-QD7hv;Kd>A
zT_3b!<^Zl7NcX8BMn=Z-GoVpW`e8|ggOJ?=8--jHS^FDL7amz%XGXDMH#!p+)*p;{
z=kLZof8ITP`E#no8F}XJ;u{zbV7IyfLQ=%}@l~#RlhaerCW|@2`n?t<e1u!~*!&Ce
zyGM&Z)=8u}Jh#afVD!`QC{O9?)FAlL4@x>4U3^s$!2hw>Pz#N}Js?@g@b{+yH-L+c
z9RJs|c&yKf%X}cRbc>=)hx1bGs~*r2n5}aQ2gCMs%DG1N&>QkkEN#tmL^1b!>_*1v
z^lvLS@?NvM6kiTt_fRle)qK%EMZeRz#lh<*UQfIv`fz^qOLlqL_xR6>7og!>nZ>Bh
zltIt&MwX1&NOd9~TTRw<rXu<!Wt4XOa_gH&QGEU3ExtTLF8?}-*(0WmR^|!GX}55<
zeSy}DgJ)<2*Q|7ryzfK)Rh-3NG2zA3_z}oS@r@N18NY#141jNp^OR#PljAwtlQ~eK
zzsl+h*vM;fP)2QZoS(x@I`rGaS@MHEw8)X^jelzJEOSMwa6Ss8Rj6C^PE)(S%}Tck
z_D_mT@ut-z$NyS<!G@o@5FD<M7u+yxJoYn3%5Gu~cRmjueGY{*`=JGIGrtpv=P_<c
z;G%9B>kAedkYmvP_BXnSMEDASQ{(p=-X58qCIDDt8=m?#F!qe$vL=K;5l%m12trU9
zXj5LC?A}S)xs(Z`(f+LR;I`!lwZMb^1aRAi>ejzsuMRqD?rKXk)r~Fejk#qZwlRsk
z(R*1;@$?%t42`8uBGq0!NF#?z&#u)M(-m}z{Bd0F^D>Y;H86KZ^VN91I8eWlN=gG6
z)8BZIS9*a4&4&u#;TNc=XPaG}Y#KX%Wf=gqVnJeySL^2=XtiiNJG;e0YQ~DSRpQRK
z#lD)LU0;uPa>_W@d}eV`%!DS~56W}dkS92Y*&3-<PjSxu^BFSmEp3QWyP2nkY3k6a
zCwBcNWLwK#nJsFZypBb@Sbkv@+&Vw@Lr6OCL(MW2>W{&_h0dUJk$TieRNl1kYw#CB
z?)Jl?3&We6r@;;GzWTzA1!b;Se@?GvCq9CF+_}P8rT>3d_7N;@gL<Bfz|kEee-eWb
zr%tKfh#frjPm+`Kwv-WvKOGQ5QW{N`65XBc`Z34m0>;y_12LN3uRh^jDctK22~jjU
zNccd4tVvubw1;-PBvWUJJ2(EPWuE=Fm?$mdb3WJjsEubr@^e4(#QPm?M~z1e{=20=
zRc-7AZw6JG?oUroYw$iJ7j?sNTNQ|v-0U{6P$qtddGZPK#xb$jx?_`T(>!*KMoQGs
z@=cMPPKrk9^iE9Uw=!+$u$Z2dm~+uEs~1adkPcy<__bWcbJwlsT3!M{A!H4=pGpot
zg_=WkaBHDtbM!5hKR#LWf74Biqe6W8-)ppER<CdJ0J_JsyL?VtYOJuY;e31X);P`W
zu&v~Uas`qw62NUKUecnAhbl6~QIhJ=;m9CWWW63xNq{cAnDuAAJp*66hMu7MjX<7q
zx@?vh$KDL$g7}Rz@zvFrgb5dRmt%XYr3VXQ^OtV_;dzhk7TEKeJnn$SKzm(u^cHwM
zyP(V(-TGMj3Lg#8z|b%jmL6M!bc&lA7HpNe`$Ub?!Vs=ZuO?eh-znNqN&2crKyiuT
zf{cOAnQMdndK_Ut625W0Br>&C<*3N4XUv*?r)aXL!f|}MYv@9q^xwYyjPy|*VLl}#
zW%djOAxE`HCucc?gnvB25ga0Uck6z#WGjY2_gS$>?o6yq#hee0F6~2eQ~4}cR*Ay~
z&9hPDyuxJe1^snAu=RsrAG#NNZSbQU$Ojw(^TnrFSbDok;Hr#}ZcC7!SSbxAReQ?I
zs^!&NN38Q0)$Y&^De*~IA46}qrRIP8fRx9iZ&A?u#<c+#J%o_6MUMsU11MQwqnsVN
zN^}Vc2>`lm9~&D>-~iuIO~~msB4mUXOdY+$gd%eNdI)&jY8H2PO=4F+v4qQ|ko|h1
z@PUtD+AWs!k0To#4*U&6MC|__zQn{iIN*NRjH4O>CY2HU3{2=UWIESA=cwTNgE{+=
zLy!mWrX$N>R~F9i2VWv#s$#qwZygry)5)ACsYIB~!iVcPuJ>P5mz~V5QhWbBKP)HV
z46ZZ-PQxgG(JlJ+tvX*BtPJR%@<kNX6saNgW<R{oVt~)DL3}mxV3#o_5o<kXrW{!|
ziK&>Mv_C5rcU6>)Sf-_QAYNm4OOxw=2mzLHdp<!O!r|#@xuXPVOMULp-Iw{Ji!^jk
zh#)hUYDcR9G2r_SFpF4m`6d!7)cc>H;hkghn<DYmO^5f5K)!PQt$3jkHc{?Bao4rr
z77q{WhbA>{CgA_&Gx;rMELh%GaN=h+FX-43eqZ2&Tle7XFzxjS28ZQ21(S`576SEe
zuf-(Kk3F58&zYDo3RF;E0ZoLxHgkWh7nqpBkZqFZzWQBU#ZxenZSE8yVrzFIrsj!I
z(YB;Glt{`R5am@O5}_Ep!T03XF*e@Wr<U4{#oJd0WsG-C@#+UU__Mt){&eATY|Yi|
zgu{>%@PfYrI?$Q5Zfow__g{k0m_PJJl}i`@Dt}pZZ-&^wfx|rzbj~A9&d|EdNN5AG
z6Ncogmj;NFx;>(%rUq)XvS5}q2P82`b^z<;1-eEiN2^_b98C)EK*yoYjH?QMS4=R6
zA&P5_*wX<aEcl|4ngd}T&Hr?Wr-q;1^)RpM%5@eD+tQGii^<#%h%H}#7EX}#Wd&qH
z2K=uGWGU<|x~wc`99r)>8jYVr^}=3Hl!i4m{%&P~NjCQ37<O|w2P#You%bcbq#0{I
zyaJbJOjmt8;zj0xZl0q>kIK9rC~?79p6%#XO`yWo;CBi~KteJ!HSGW+)ZJ~>1#S?j
zs4<%#H*lSS^({;Xk_<XBv$Bo>!T8|;uoE2t(hVgNAf<=woni+u_X&lpt`4Xj8y9?p
z4O4Q69#wygW;7xor+S7#*}UcmM>m$&-SpP*AfE~v!Z~L~a-r6_*|ffktNHkz2!qBN
zuOax8xM1_b6D{lKl1QLozsP7zWF}PgQm&3$O@saKqkTrw0dU-inA7r2?3*g3?AJ4?
z!H1|@P?GHnze>o#qJmf=_^y=a=Gv$~@TRG3C^438A2Y7fqg!e!UN&!RU58hfwBNe#
z`uI6rxSTzJKl&N>oVP~50NmXN$PDST;2(wTf}#7pGk|brG|_*mqK*@|#V_X<XxU@F
zUmm@>H6oMv1pno--EAUYwDyPGV79I1XdJ{=`{(uJ;9pK<Y5R3gIEZF_wRLa0*5LII
z&ig!9ln6OtAFxt%o+*_^crFBE?14ST&<jesHStznXdMg3Qc!Bb4q67cD+P)8d3PDv
zogmIleChqhM9g_Vd++5EoHyszTw1<AOESp)#lI2jgFYl2Oo9v40jZEm9)N5Ra9!cy
z;T<Yn_GY_4n~otd3`+#NbF9OUlnj08OO9W)JxxN)XG;okYJ+XP;MyW?P7Q=*?b4F+
z5$qO<J4Sdn?={WzwyY0s3%p@Azr%mT1cYay?kUsDm&ynd&m|TlX8sFbxDI9m$vbm3
zgr-GXjwF?-lgrTF+?ET0xVr&1^?g!X*i_Nc(obvSn4GoU$zN$&i8FEM8?TKHH_bp>
zsAn!o7!tB@8p`@$XV*0citjR5Ir`bae8(z#jOp|GfS~O4g!{0&5sOUH9dx$GU)KfH
zQ!ZjNCC`l_>(3eFu?g6)fBQe*L%_J9`*_8pK-TsJf`9wg)tl}kYGlp}CV8&UUrWw)
zG6shQN2o4upJ+7Wn#D()Fzrv+x#EXiHx$YZb6EH5Tx!idc2S^xq>QYvZE5H@1+SX3
z|DHQ{`}THhD_8bYEgT$Nwo-y@RzvK`M~cni8Q};>;#4v8iB~H<O2NEnyvlt}pCz7K
z&2NY9UZ^)>%tG1B><{mG_7&)e8kK7}fo5`|-EFn|IOq!~g0TT0=)mWCkOJ_}91!ul
zEP!H`@I8Y-ze(vPa5zu`1EJX1*y&)~S>9Y6cK7v-fkvNkxdtadCtj>1Tknl*gaikV
zfH85UMsF_!1cV$=8C?MstKVH%8XER<0jUGfiH`szy;L6`I8KXE36Sj%X3E3|5d$fw
zECf{CKg~#FN??E-_V)nok8OZ7j{#6}fG<$N)e~WV6B(9%pfbnzmzW0^?F&;<P*Awm
z5W^vw4D?L=IK7Urr%(V%5z1p!MYRHnvE50vC^+E(nN>$RZ|GXTZC@b2>>2IRX(`Q5
z73Hz*NzvEJZv?q9!7{?7Xc14>zC}8>QV~sb&YifsVhdR^tjL}lODn%QzAihNFt?bg
zt_;i*s0G?w=)^qo;7bm7N79dhkYf<)VS03K_&+Ye{q1=qcp6Y@CI4OA^^FaBW@bnZ
zB;X^TpS<j^>a(eo4r60u8?AOQ-JPv2J*WUZb8|4}eg)b|qm2->wzh_Iy?ImGr_Hht
zCcim7FAiF&DJHTmfIL_}_q;2ZaSmlVj0s!)z;iNb7;4_~3E5Pu5kLsQq6t7YGK!v_
zb+GXs*O_N>0t%}apwc#nfV4^+pHpu8H<COcc~>!c`L$`~!9LKeM`6CcjAZfw?oFp3
zIN`q1T*?D`g@ml##*Xf&*!Qc>y28P6TJvIpWYw|?jN4IeNeN{Px2IE{)-m-Wn;pSt
z3z1#vHA5kPx2M~}+1t1yc-V^41vt@r|DYJ#=jkjsw{e2_wV1=ZvOd(IZvGG_9RgPo
z2S8g5;OEQAIBsu_2T+ia6+llkr-O{u;8OqwNiOJ1=Ym~m%FteSKJQTu)CtPXhD5+}
zs{q7v&5;kJQ5M?J?WsG^u<G_~1isTkf+vG-2-!`(<|^l90zI|97}huEvksw29(%lk
z5cnAiCMNTtE07MWpn4bcJZ~%)=ww&|ybbnudwVxv5c8yfP8s$l%^(o4lSvlR2J@HI
zxk@=NN~iKqP8=CE$|PK7?E=L@Neh6CN%D}?xP$=dK|o+2y^xSj*me;d0=?tFcy6;v
zDUME|Zev4M&jY>ZvVre=s6dJ_9J=16?;?RqmLpv?N+XX_E<tcPoNBdJBt7=+a=YB{
z&fTe@KKiQhLvU}@aDBq=jiKMV&~<gmdv?n3^r&XV9;}?8x7|(>ht<P1N+jRr@-{JL
zl>DUQEDtV_u(cTTJ@*9uZsgwO(49yM<${H-0=pZ(Ji00=FG}jYXo!(pQ>773E<*JM
z*lFR7jg2O~F{$8jX=wiRR#<>F4W|&i`jfU{{Y4{zzOYK<edasbBiie#DxSUs-uJk{
zZ$YC(QEgIq--Ee!hpj;Emk3Z*_8O_4v8j~-dyej=5`5-f!?n%r-Sz$z7$PeJv;b{~
zK2AzPdQQ%A=tJ8+5JDUSWQ3L(2HfB{E^F8;#^v1yz#Fr@wwYNrA_j3b7?M&35<a8U
z$(}y}S{-?|r~H_7zZndkPqg$<npc0g3R(aoux_wJztmZ2d6q_Id5Gk&b`a^PN@8w8
zy;+VsTz{HQ&Y6+wlz|stT?keAVsGV@FF*|AsxFPHJ5md!mzGESad(5Fs#PAwoSQvn
zn|@rw_hZpExs-G2;_fBCPGrP4*VYy5;Ycn#vZye<wUIrRySK?<s|y#QU@?p=N%ww#
z4m!px@NSR63o0rqBK5w|m!zu@-sRt~dnsA7k>t_MaR=*eSf|#Q{m5H5`1}NV+5gxD
zmBV)a=4)M18Q7B521R5%A5tjHfmQ}3>2|2FR{R2NxlBLR?)@3g2u@!nb4WXIZQ^%5
zSZwLZ;<gZq2NiVH!4V)}Y;D@Z;s3t*#a^~s>7<A*yxGcC=)*mzb<?auwi!TUuBe$K
z`5mKRu3$eoh9RrfZQiW{x~mgIosXluyp*`NrV<>R<TSN=7gi%C@oYw_K97!3NPPkQ
zUSf)tz_>B(?~{h2qS@~zhee04pQ*V`6|z43$aIs$fDoUB^QjsE^oPszTM;>}CgnlQ
zp&gk7UR{i5SD<Yu9mWIzx6&k!F=f?Ab6H7n7FB*th>Jp^5@A4xLP|iz$||2?vz^G6
z!HHz1!g>U-A##9#Bs~w08n}W~^2$GCoPc(yl%UDmr9&T=!IalW=jYKBpxluH_fz4Q
zAa%Y50J*e~l7e=~)=Xs)?3BUP;@OUGG2=;*Hw@1dqzCv?^8czyJ~5Hb$!v7Kt5(=<
zfVA!VTiKr~a=M&s6fT}!W-5kuhV`v%ycC=ftzi)93Ve3b)3Xq<^Y;)nAt-QmmkUd8
z9ctTz7!C|5w|Ds+9b+oVOLSegxJ!W?K^cfX-h$RBpWSjY&hL>dxo8T(<6d@S(iw|#
zTn{fr#jtef;l#31k<MA<kcq=#FjN_&DdQ=GF!xY);Z*^Jj`yW;xy_u$e#3Pkm-Tdd
zZ*Ol)$SevM#P9WtC+*O+i3r?);D@qd5eW(TpiBG*3`Fx7Zq$MXqy-oXHZ3!{03lDR
zo|Xu1#epy3%3;=OLYUN&VvJ#J`&D+)Ny2tfv@}(3B;)o@JSS7^0Tg~zVqG}a7@c&B
zr26Y!Sdz^Pf6X$dc|wvI`(nq9zahI@CceL(Qc|4jc4wUDkUyGcAu6MS)vFu(RZWi(
z+h$HyQWynjy-A0yI`W~ao<te1ZEYIjWj&vik)$;_K721V9B?X<4b&IYhsR1SCzybU
zsMy`z-4zH3P7tV5!M)bC^r3UpG3l3ZjDnlMVx*QJ2Z{zWV5wOE=ONQabs#j2h?99K
zUF-uaR|({g^r4DyN*@9Z&$oW}rzw^`&+$N39lg^$*2@u3XOqjc?6{dU!0hU&Fkd;R
z*Uq!EEqY(!_ONewx13UPyMh1QLLau5oBT)kpqi|YZ(ce%_Uuc2nNeA{t6;)smaHi|
ze=ZlPVu8T;4#vfi90nWFX=jr4^v}>6uy0Q<JVELjNdbc0a()MBUpK8BN4@R%MnWkm
ze0O2^WdNwj{-Q7k$xzKw01@ama$a6t;Udl>(8q-2YdzGFBoy+j0u;4$5Eh{0{0=ly
z0W}~#I|lZRu8uKW_M-D`b+c2dE$0N?RLPcRALo3FOxt|jebH|BKv#IpCu#i#<K!$=
zoGwQT(v3@xyN9rc=}s<c%YA(=zzEbN8XG;7WW_gyZ6LE&Et0vT_}mGrlqA(T1q|h=
zci>Hp0rf5t0ay82mwjk_BN<|nbU*5%?lErAsdtwQ6S`Q0H#Wkab1(rtL!bt$Ufrv$
zp{nDSL9q#Vf)ijZq#b3Oy-L&=+UD=TW^>QvVWz9}Q4QM~FUTuuxRR=_uAXj$8i9s#
zA&~4b0}IpL-8~F4mNAe69x*(4g@D(1Tf8me!5u#VCG!b;U#B_@jP&=pwJ<8)zA~mr
zr#m*P@i2S<dn|$VI8Cu!m3ot)nX|z0Ue8>mVEeF8tKs3U!nZ-Kq>w6Zp3BT<+%j`~
zZkq(4T1>N<{9$ASBsFOuh3Pga%>kv~bfs-FNQa3r0s$&C2DAk1V=Eb2T3b&6a|@CP
z11t{?pe)fMGri+n3y6=|fa!Dk6&C|&PW)JPf5&Y#`8=?dT$0c2C_TyRZ0ZHkG{Fy|
zo=K^Gi}G9m<y#D<H2cS3`1l%8)NmOnMJ~tJN2`5TM`58~5w$oxrbwu1$c;aiP~^;W
zd@j9-oP(0<zT~2e0DCy>wGD_Yd!v0|Y~)<)X8~$EJ^yZ1%{;H<xKD9u@~Y$3s~uv-
zWHu$YY63!m6Lsxr(*u9{D~7`AbwbHua239gtl2KC#m2&tw>kC#oWHHXRKM{e-6D`0
zC);U)yPMOf6H6wB<+YH|T8&DXXcBbi29&-eKv6&bo%~IZpveVvu;d>NNE(^IgvbJL
zK!d)0W3t}?v<Xgw-=FRC&P5LXsIGz5i3ZrxK<p_G5IVO(QzeO*^cAJj<m%j<Gm@+%
zM%Fu_wsSX{gpQ?ojxdVa<6hRkFAqm#_p0x7>s~XPZfx6ZVPU~+{7?Jz#qqXO$bwtD
zog42aA5?5p;Y~6j9(>0w3Ldj&vjNX>qqWE*=ZGCt^vO?jgl-O6*Z6cLBt8Q^Sthvq
zbKH*C+2Z>wW-1EiU02YpCyQSL(rU!X0F~+d>dHzQEJ9dL6pjHMT_qrs6x(MqmZvHQ
zG@cs)dm#-J8rY=&N8DSuMb*7+peUlRAn=L;3Id9BN_QBDba$zAOE()-K)SnBngNEc
zAw{|*hM_}ZNXen&Eb#aH&Uc+Z;Cx)-HP^8BUVGK^-1ohn^(<S?#62+VEx`INWHD7+
z0YW-eL9i@7h2SNb8T>>jy4>n9|BzCv83DZ6l>1uYIKbqQ6X52|OPKgUfEz1?ty()F
zY?Ggwc2swRu!~8=pv#G&TK!N99Z5sG5a$(k{dN%r@5}u3d3e)hUt__+F)0~sUiIn*
zbr&&lr^glV8=>bV&`jfF8S&ifFEV1%ql?zd=j{dcn0+jemf7FJLsew^*XneS5<!+_
zt>1<q*4TAq!g+IA?>h%rHQW1nPhlAZ=Q7yqG}B#XyJg;<L?<5vF&9PbvIN)nIYGFN
ztHV+NuL*40MJ=qY!2&?~xPs?LW2?__t2DOQG>gRt2B3jfe?%1Q?k;V-?sxF_=BYd^
z)fW}xydU&54LmLVh=iQH3!U+U(sO2yG3I}Iqs^iTOqmOuiE&=vK=6@VR_v?pXUUOY
z8F0_MgB;tD(3aP*=R?k%0y{Czp7Etazllk-^rlXxzSB=1s#En%YR~<c3NivA7J%+W
z7N|p(k^A_$4<@ynlT%X*!E_=%7kc(gF<f}(%OxKKI+Qu+C20j0>lcpQtX0c^TH+GW
z{d;XIe9245Y%DW93+MJ=MAD#^HcJZ1Qv(~b<TXy};qZyA*BE~%-dZn+Ig{0reHy38
zEeT`JKJfi&Tr!orhib3&-L+lrgYMb_$HOGVtml31%bCpHQK{-584>aI_B}c?;Cp}`
zmgec@6&oF$aXIq$Cu{-apb%0wDH{W6P55mJXH$h-iooJDZH@C}Fkrblw)I)>TKvsH
z8UaD~y<7linWRd`tg^JV31)vQ!mPcJfi^4@J#2T(WvpK&mD52I?<!!qN5m9CB0sG#
zYUl~J_?^^2wDCG&S`>Znm*~#+%mg8ic<VctE$yR`AWMgH(~sy6Nz2@aWw6s%4P=Bd
zF3~jNOaxZoE68ciARwRx>{kV)*PfDXd|=rJ=Gj!BgU>DIspK%RtoL}wfK;redp|(F
z!8oCzYq?N}ES=u=UT9JGJ;FC9rP40m-{stgkf@ygdA<%SkrTpIn+rbiYww;HJ+3K#
zXWyXWZe0~)uCdoEqFSM(y3%*+Y6N`DSIY1Q(d4-xU2{G-Xmw|MTPc=Fjoa+k<1^4^
zNgx9(%!T<`)%9sWX^Hj!fY^qTc96aLzCNUna4x)jk<3*mn|u&-&Aup@X<8|hEHg@f
zARf1uUZ|{`Hcr?eYYlH)NC~+08OMC<gs@}5Flu^&{>)LANs0=y=egoXF__JbG*gm1
zOGnfgVXk`l?Y}_D90G$r7mT7#n<{1w4kd@B9wm^L&iK2;JHI9MCXWTEBl1VLudi<_
z<vbcqPep~|21mhsFcrj#)eH}SJ5`9h>5WbsXY(pH5KPlKZO(HWjCncO!K&<8q^<j~
z(9)nD_DaN#+{L!g*OtJco=t9)U8G8HJW<K>SMpkfx4;vTVyPfGwF){Or?vK{#<#Z|
zHv6vJHJ0la9XZLAJ_sU;bf7&lDbpedXh{Wg&z_Dd`evOY!h#!~fn(dN`lr7=D?MCB
z?ju02%GK}~n|K0anF%hM92FB7++El?)m_*z)Kbmnb_Myc+ME+?B_KysrK|Rpxn90%
zi-H9)wS*>3atkq=t+az*U6i%~c?NiC8Smz6i<o~w-$CFvhD<N(nkoZ@?M5ESnb0b{
zv(a*`AH)!&Gt23BLkvIW`{x@shOO?;V)-G1ScZf#f_wvCj^IrDh^+BvrdeUu0-8px
z#k5Tda>o@T#w~E=jeY!X&-BZY0dNW4HFh6d`e)RdUg@<;Ug?htqHl#UFLp>7(`I#x
zu|RL$@jO*4G%r*x92FriYt`Uz!6^0UdBu2BOKc;;7ytnwfnYx@)szZHFpF!Z+f5>W
z+mC^Y63>+iI^F)EkGO~f8Na{J)2vLn%wSZdutNHblw<vs@&}*gPv+p}QL;6D5nn_v
z%GhHFO#>9KUQ2Pnjb+yB;tV<vv43=p{pvf=5<O$_gMXh7lnU-D^({J}B;^zQ=g;Dd
zHN;!~etvHM{YE^>@g=>j2G)aHtYrVWrs?ZHu2;hk;LHF0adNd8@A49HH#a*2IotRT
z`dVXSrJ4{+>EX_Pt+M6sjq6ym>zaQr0@vM?Xu7)o|NAleF;;a%io7wMF~id*5<o?0
zM*g_3N~;QnO%?MMT*tck3fxL6{v(aQFtt)s=wdRr+)&DNl0=tPRnDVI1K1diUS3SS
z>J#v6(Erb<uj^wmNLKbdIy*mhyp(U`1BRT8&#>}p)7R|3%RRg`Snl1%S(siTTL=#v
zKpNU;h`Saa1l<I;V`2F}2Zb%tV3wu4ggigWuM0BaoJjMbqK?&~saY=spDAcNDjzT9
zRsx=BmrSr+sUVqeF`qQe_~`2H2hh#rf@|Ue9di96<sLIGeu#~iS$PzhI+n!vF0_n2
z)KmY`dQCt74->-_Z+FScBDVbg*7N0)SsWkxt)X25COd<=VF6vT31wX7=38_C>C5zH
zO{>K^vuKr3U8vd%b~kEXmG@TjMjh7dJ3t&PCO~jSzB;s0Ll*_lTcA&-Im?Mu<rBZ9
zzVWL_BDt$j{_hQ*S2sX%=TlB|MS?}sh^$5Q(<u#&sPdhm8DY(bS6Ag)flhDwwBxW-
zZaPMy@n)jRV(iu`8KZ=@Bfm4eG-^;p<MH1MRj)4mT}7qrEuUs-$OFO1*-Pl8QhzKu
z@;e!|h=wDtEGiaLJg2AggD9i2v6$G{#SP9b7Iyaj30l<$@-@6ArRa%8$Kr3u)}FZ!
z_x1a#(COQoQPP1zE47M4GiM*apVCO>wMQMMVXg-ZVKC^D#%f|=LK>9ijsAdOo#uW&
zaLPlXYS8&=uBSXSaU&Mxw__aBj(S(c9NufkTrs&@cOiCGV^KC{R5j6!$Rz6S?~jj0
z(<|oP-HN*u#O&?u>-1hHE~3fjNwe@e=Zd@Ma{0wR&*>2$^ax_lk0O%2R-KwxefiFF
z)UnVe#=9bG62kiovyS{R2L58FUw2B-Xb;2<)yJnP43L4_>qBn`L1PzwBedzOtT>mn
zgNvM7xxC48bSkY!FTYON(9hy930$A%#JzJz)yOEbs+*zACsW)uUhy6OqN=lcC5t-D
zv)5Bbx5vtjcM)AP{YtaY8vw*C%BSDec~E}Ql&)p*M$C`69{X&L;j!}ZX+$lc(db3#
zSvhA!J+xKxb{`Xqe2=l0*vj_wXt1i`J-T(H_~QX}d}_2_bx?fzz#{Xr$4wKoDH`g|
zl|Z6mX{d|O?dd_gU8KjxsvR=UfPg$)j8JBJdb+3Q@~_*fCY}M<KDX4Y?=RLCv+CI&
zgfC2R>KgKf**p9l1d3XBx32r#@_Q}8kPoSeFp6r7$13gnhAm19aoyMEQszlQZR`48
zjhC;=V2ooCu6;1%L;10X$9P4s6C&EkJQG!Yk6o_Jj8BlC4=I*#!BC+Hu|6WGoKE6+
z#dYPDroHkpm(sPl=d11pV*RB5Ts@5mAI$WvXJtqYz7T|hs0o$%Z|Iu&^B0{8`iV55
z6U>RSWQQZqj{H<lLnW?m@HzN*>$<o=E~+N4$YY)oMY*=Q?&Mb)SXq`Hkwh;i_&9Wo
zc2RKHc2CY?B<Bn#{|XOcJ&1)@;rY1KRl~EZWOARS*aLW_qb??Pl+-)Tc1z!=yNpqc
z-Qcnp=*u8Pe*giWMYCn7<H=}MjIh{q?F0s=I}T?Wz5Px-Qpx2Ua@xt%i-Px;H*@k7
z&L#Qwm+3Yfb_tqU0n3ci?Z?rrTX*)Db7)5WW`?z{E(XoY9XTVcYHD!*xf3C^C)t=_
zj9%zi`S2#{IyQE_E<P4k^(bh4ae>-Wm2zkKGe~u_Cn4+NY%iEoJ?9NfCFl|9tSPd1
z?PTR>!d5Nx%LjH<+^ePt`2w1`OUvlZm$qp>S;IdKDl&Z51GQ5gj_V6j0x9yy0;0VA
zP_lf500P`CL*vI+H!)Cv2Co)BWg*cUW7O7g4%&BSy!yk}sJGEPm!B^E?=uE))!(20
z-#<eCXC?nF_|pIUA68%gvi&cA{;ysF`adiAZ^8em|Nm9>v7rC6lK&R`xBmZkF9H4U
z!EazoIxEG<$T(hYmoZYPlLIn2GXp*cHf&s~9mfZo@Hkc-cf_l|Ll$c*LHgcSc2?F3
z5aHPDy<WWdNW!L@3w9{wgWXMJBJSlqCS{~dH~w>Ve?Ip$I7nR#vYZbiTx@x4CN;t4
z3J6fO$X=5nz{tQi{Du1U+Lz0}f2!vRY~%k6Ky6^$BCaoAyx8mE>dytw&Iej9c(yEE
zG|(Z23w8MF!S2AQOpw#A!=}okM*BY{elW<o?b@FMmKmdHa)$E2QW?9(U*v{`^#@^g
zT|b~j3346l@9z(e{m(5vrm9Jlb?7p~b^<A7u*Cte+7Eq(*1XAJT{Nh^cg2z9{9^Cd
zs9B1=JK{5L%iJN_vTNSn&LM4o(IsdzI#_RRO{D)*@c4>>S57p#rL}0*sd?^07y7#v
zo$v0P)3qp?X&F^_w8zD!@J8<|iEcbd9*!@*=%{p~4j>LdrlT)c6jtln;kq@-;BcV@
zfpC!MP!;I*lB>Hn8w;!e?3h~^v+62yL*T03ubkAijrYN&PDaV3yzlSVaQ1Gf>K4a|
zNRFQ0AX4<*a=0E91J-;2-m;DNwU5yWI=C>^dA4-vyVfOcsM^PRLmR#=bHB)K>%J{4
z80oo8Iav&R_p6fmF9!+F9cFe-mCpf+UJgiS4%o9CP_m%FK0G`;6(FqkyL*6j6VMj`
zm*hy-&w3u}&OHP|r>v+bryJs5?2f953X2nY)~QknyFmKbxYWYcxj746F6lENC)JE3
zrD=$*r00>m>AQ6Q9aqszQ{%O84EcS8Dg1YI^x_Bp`@|2wIpdS(8|0Wt+_>f8`ntNs
zZnb9C@OSLodyT`pGSPwP7KuK_&Xo^a9{2kWQ+?6cuIzSp{4*yIVoLu|=+l^0ks=%)
z9J$81Ep$gQPi@=>wc@y1e=t3?v)GNyk|BjeFvBXe-p_->!bX6iX+L6T`2PK+mH}e<
z6yiLekDZ;>>%H)si%}WV0(6OGhbOsSPu?%quxteqJNsUh)35r%!lI;I7_)NU$$UT;
zPKeN{7b<UTBChR^Gga|UTboCz2TPa`icp2or43sSH6YQuqzvqhhvBaiy*wQu8F(6*
z!$GP)K__$cq>8oRXu|Gqld1{4bi9TJ6D?MCc8wdhT~!mt-Mz8tjp?$l1V}5^Q&N7U
z#hst(dku7V^~HgOK9+{q{S6+P#QxoLF=l2N5K<qN+0@i@i3*Vxg^IphSpBdVF<;yt
zQ$P^U1v|pQ?Br%Q@gaAHC<OAG4fT;>Es{o2!w7FZe6vY$IqDz878T2`>Xjqy>AwgO
z`xcn6)LORuP=Wsj6YM0Sc?vFKkIfcZcDS~rT@(_^JpKIZx%3kS1aM2VoeW`Zt9p$^
ztN#8SG>mLdRb&V#1{4qnE6IMrJeDi=a*udkv>Y3(Yz#twLF6^nTAzcx4}5}x+90;0
zu3dfZ8fgZ`V}G?PCtXF_B-<Sc&#?1y&zGxUk_bbTxZSHCEQ6W+87<5qWg8lvAvN19
zF+5(xXe`*)^J~<}B?&2H(4C7(4R-WVaMxOfP-6!b?bKvA;bmbe%ui=Z+&z%mB3sk9
z1f!sY+>2?u2|u>KMCyTTmiEZ1qaX&DGz+kq&0?mZUf*+D;)h-netWbE_i~m-Y*b=i
zWUA0II@y+icanA1kaHz2uD5*WkIrzByBh0MxdfrtmeQQ$19rMyRNHEa6jgV>)yUp+
zjkQ!!ksy2*ywBQ&?s@H5NiMtMewt9BJ2?C^sljyxr{KYr5qh8mtQOvnDX=jrA1w2L
zz=1n+Y_r+)Vrd0W0|RTbjJfKkasqm2*A4ZAXf(y{fbfUr=f|tze#cI|<Mu`6?>oDC
zk!PFr>;Xf07kWrZ>||cH?V9>E6p|E$iI*B|q4Lhm+}qn>zJFO9g!Mo_%Y5Ji*tx~3
zx2nBS3C<S8f5^E|(C&RW7m%-0^~aA1|5+yM3E}dDM6ep|WAS_7Fbj-2XT`@)m~-Oz
zwGk@sfnAk`Up@(aq`iHG7bh#Gt*Nqu7&-El{Sj7auXAK5xtSSaB&pUMHl4^JW4KAK
zPM~&WUwpd04nhU9BB?_wF;g~TYY^#td#mr@@o~RIRAxKk6?1urh~q6@%K~R4%wqQ(
zd%WDXk-i)Dt3h`whi0X5ik)%yLBGoyb#x~DLX+keT<fa+W?7heai>8fu-Tu<nrSxd
zTyF0pk5C%(54&{6lr1#oio0~<j?_xK=lX^Fq2!QTZf(%oY~a0d_t~#Wul-@x75T=L
zIr49cYlnjdXY97et6kE+C&$$F{`DNyuBu(cby$=09|SAbmX#!K9>SDd6Lbvql$oWY
z%1_>;&`ZO%oT&tuq=NDuF79jMgijTz82&Y<r{qA2qK+B@Y456-nHYX;lvt!dWKTr}
z#x_bqW?k)rXxx@d@S<oaWTfpqSgtCgC;-R;1NE<dg{YXs_|1LI#7bc)<%q`g3B!la
zze3H1@o4a!8?Th|2XUW?ze`p=U4cHN;&-oks{@<4rujO%bax8R?I0^tD#}7F?|0xJ
zQ{8yZA9O>EYE}C<WN<htQ7iJ7IOV1Qb>Kcsa)P&XY(v|A`HRl{X~U)uO`a)&mohj1
zx8+QA<VWn&iuDgEIxyD_`JAOqF`BaQ@Y+g?m1f+lnJjMnq6zP-CN?`lHEQqTJzu<A
z$YF@=S+OM?kPyt;@PJB|nwd9A@Xen;8R*3f3?Q3n{Utk-&C$Z>v9Z;G%24W$pF}jD
zTva|-_VRVs;`Bw-;O;zUm76}GJk4wlea++e(%oImFDDek`mp~y(oZ?I52>Zx4HqqF
zO^oA@Pq~xj*jaUGrQG*Nl7+R&|2|Cj@?qUG7Rcm-8D<a+kuy^v#Fz!CqA%E;qZ_76
z^^iY)?5!8_#Y+Yft2;)w)t+=UNOyFNu!^02Px15!?C*8pcigNy8c&bPharA02ZaqR
zYW`(Y&jbO1GIiQr6ydh`g^}*-L@o}QSPV3`phf6s=6=b@Tnx@bERn6K0xfn?c>cR@
zq+iE_jDI{O%=3cgUu)aZ$ZL2D=|~Q7^?As@SFbEeKFV0y7=zPQSgh{mTsT}~x>~Kn
zdi}ZyZ#HGxdsZ~G`dpjV7;V<HN9DLi*l2b%m(k_I-^b5J%cn6^`A2at#@yW!sqJ^;
zhS|Ohd#+Bhx_<s7;^XNVW_OJpCKJ4*?XX@w?s*)04j~+~oM~xE1)L%2{Ataa*M*{I
z%E5$`N^3VUn6**V{f9)q{kr}d>BvKO5Zgzb3?C~Xe0`e!UYmPpxl;wxQdEEZ4yEsR
zq=kIUBKyQ!oz}?M+}7UW-SR|+HMtxa&kNQyB*q>yO-y|Eo8M*#GB_Q|3im#h+APq{
zfn3xJRSIX{B5&izNh#~T&L+sol8NdO6(#fT#j3|N)#Kj&XUk^$stcR`1zi}^k?M6~
zrdZXzcXTSYH{aD*ZnaBdEgY%n>Jj_vU%e^X*Hs5Ah0KD^5uHXm?<<G{8j<-`0W&#w
z`E+wS8&9)qptHilTLesO^XYN;y%rE^q$1vVz)RymV^`3rPLuVjq-SKHLupACHsN<s
z1~Bl;jA;4xc~~{e*d*`utLhin*tebfzj<R;-JO=FE`PgvyN60LX#^h0BSS(|ebSK=
zi&t%i^@(~GN0H8iib_J_6$2CXkBO`YVd<^kZ}M2UA}-fd0RC6J4MdJ6oLLY%=H0J<
zwuhS0MtD?((b9I$eaJO2$`s_b$+TqO_%5Se89@m98vLfaYw(tX_wc%Lx6|H&l=0$1
z7s^WAO!4dcB%Yy;=O~!ViSsmD*$Zh|xuJdtK0;MO?zr{yc2=e-11I+@kYHP}v5$i1
ztBmO-q=FE#mo@ew2Tl7mrEs)Vu3t9oyV``y{Tl0`hO=X@4Z(UIuk3L?o6M2}0yS`F
zy`pieVRt|3)4bR)A0BD9w*_|N^%^&~uAa_;$wqbaKL#+>DiD0RzhJmJ^sPQVzfd!Z
zi*s+OifeBmT#S*A9p16gU1fztL2RlKV40q>ar>r--B762`jzr{r8_VF1SqY-wqdB-
zKt7tLB9|y{LWfUd^Vms^o%<D?lq45t#on*GUI^<P-u0;*SURK>j^FUyc3Mof!KBP(
z(}$Eo*!x9%h^K5|o_SQJH~!jM8;RU~6)tI|=e}J%-9zUw+-D6JJ`Lk_(93{(M8g+j
zq<iV)w^j;o{ka^d6?U7V_+-2J1gUQ#8sTvNO3N(EWNu|xekAU}fr_$oSS3x$@Z{79
zx5EirJ@1B4uWqUCOyj8m4@_lZV)4pRNCO*;)zz61Ejp4}vwPoc8|mT2j?%N%={kjz
z+(CZ2POO)a^t$yGWH}NxMJ516PN(mF0)4)+W}05hz*i&`7n3=1aFK3r#mf^M@+(Kf
z;G2wvM5>*vkxwH)Rk^11<7876!PEK`>b7ugn2TwzlKpFcQ8&++`22&S(J7DNlTOpv
z#OMt2B%ToujXl*usN(IHy#tNtFCV6U6_n_&dk4PmGsuK+P6+cRG8D|lPUeTd+fcBp
zUB#U;;8EF<VqwJ&<bWw7<$%tU7WW~Uv5N*xp+++)k}5SQ$C_kuq{jyXo-&(VKcBBL
z*=-KD=rO?lwIQyu0=&t!$h@p9mnl`6$+N1rFbVkHdcN&Vf`~Dr!Qq)D%Px4M!Pg+>
zFz6m5Bbo3LkMYvN@`{Ym+5~wK8_}P0olnJzXU`Vg@xyIk3n6##-B*aSOic8@23^gn
zG;yH$Q^qAEW#@<;$(lq<4cR{U3P3{K;UwMNxg@Xo3fwIW5@rBpV7G_I8)gyr%NWRC
z&QG3y@Q^*vQ4Ev)@SkV`o9S1!5N72`PNw9=LEz2>vZ|_-e0PzXLP~bHmo<51F-=cx
zuwn*0e;!_~dq$2#{mHO%H*G7hFT8g-X=b&qwE4~6q>R0FtEq{Gidqz_m%`9K61Fms
zU^ce1i#aae?E3a22!t1zovv@vO3feds~Y)0dF`&%hC)su`9;wBq;6=MFmi}iUGr8L
z70R^nyhu?D)@CqJ_H)kWT3Kgi_%S99X*B|7Jj=S*)iP}EA(*U9pFl_nXn7_n_v67n
zV@h-h+dyejZ&I0O*3QnZh?>6N{_ZMISIO9Ltp%~(Q0BYDwk?iCT&jbIGI8#ub-31j
zJF%rgkD%H{FLJfp|1b<aaC-UD(PM3?t8$Eca}!Bfbh$i-rKo>Zmh&Ug@1N-98ERki
zAo1}TclqbMy;rrWDwVmoD%{)I2-P+1<c^#gXjYE1L#6AlV;nVj3xIns`z=qW$8AmK
zZBQm3wN~QL-jp3mE7e%2wdO2$`E(yV=+mWoyewV^GqpjiA}1+fUtKoO=eKRbjc?$b
z`A@`sH?_0-$U?$%n=pQ8Ww@c*FpZ(`2ea#R&B-XSDEc>h(RBailinI#_Qmr+^P=f2
zHEfv~vHh9+o3)}A=@IsJdoSgBdtzwRSPg@Hhlqup5#en$B5`+Vva>p8AXLHQ;MPJr
zT*|8GJs@?1pO!6*^EE7C^egG0_b!`uUffeQ8oeEFD$Ga4@iQ!39n(I?sLJV7zd?jO
zQnGjVUsZ)=CoJ`HS4;989Y5IIeC>?TQ}H(VYM=ohnEU(+N4wBMIVX;iT=<POIWGfJ
z^KNr<RA$U0osz0{QzHxLZ2Ly5sY%zIBNG$Dn~i)8*2>hE((*-?JrT3Y$hWG^IUlED
zCUvXYnOxUb-QLnx-2HqDXXUL)w*Ah=&H+!nTgu0F)>CS?zRQUrZVV>IgA~wC>&DD`
z=P6Ws2Vtr9f~gW==4y>;{svs)E%1krVYXw98;uDU0cq*Seoz=aEo_-3YOGyyK#pv!
z+cZtJOt7L`xZkgIQ?u@-@PWZbu#oz~dBJF<o<Yxr3b3z#)+Z#X#P+}REbZPhb%NE1
z{%)8zQl4`n$Wa-p+8E3n4b^riI1F#7g6XTZ2$ds6HncnKnP_I*IFCX&m=WV@FpG)i
zm*aMvorU_P5HCW{Lq(ZJs&{Q<Th)zNdukPV>8&<orIkJSUTl0sZ?x{DT$)s$E*~G#
zXT@L?E?1&v23IMBXr%!2Sycfv)6c&H)t*@$BFie|sX9SGIcNy%Zote%u6ahfy6mn=
zRubAsJE=D8IDDVN0rL^RRCBoMxgJm8KN`@NbTxX;vis!F3)uYc3TKo}MZ?8~<EGl#
z-nQXW5$|*n<58h`42B9KxrqSXgQTtaexLYJX3D82#T%)i^jn0?j6#y>{RUrcN!SLz
zyUYyIvL>ZFO!DHB=W1&voC~_kN{tCfcXbY=IzU7I^S7fi$DcG`abv%m-89iqSf;7j
z%3Q`JB?<2_$RtGB{6e7&vP>fH9`DXRInpUy9nUvD5`94Yv#{RXqPbS{tn|Pqt)bi@
zv0clXM*d-?uo^y|(;6Y!>p}Y3`Z(LMe1m;T5Os0p*6%H)JhM!Y#sn4}z(nq23v+1@
zr!5Cn)I0d`fHpHbUk{THxo_-rJ4$?V-{D9MdPEMIft~G4271JwsjncM!Jg**;-$MY
z`+%?z0hN0E-Ec-`J`Lc!Ff%ss%YQ0a7d3h;Mlx2w26Z^kvDLh~*Z$fj@TrY);Z*eL
zxFeO{cE3ngyzAGO8jK*3!@NI`8D`=9-eKpQqvGKFXW$YSLx6@jLF-a-=lzDYIR-Oj
zVb=TG17H8duSBOq$RWIV^XarR>CRp?bP~^=MSsz%T4T}q8L_>zHU%l1G_GK{v_%qW
zIW9XjlJCc9<wiEV!HIKx<hC`fq3fbMAW7ID_8b;QIh3E#?r6S|WK3pyx~eHP=#x+y
z@bk5U;eie0z6i73+dl3kL)N@Zbl;|1Tv!h7omd2{{<tVU?&Xy*=(D3(qD+cKYtR2g
z3ERL^>vmpVZR1OF*2XmxjHPDd=m(7`!Pjai1wO)Ioq7Z_yO@DXBab|+a+7~k_@RJ6
zt<~_ouC87+)HqLt&FF^v*O$rfTl6|B1f5<qLes}9*}k~DyF;rK2*>U<+gc|B5$g_S
zxMfRP{iQv3muviiEN$gTmaWZ`P${HzPrk`(%>v}ilNk?#<O3>x^dp5C&R1YwW?N6>
z0aFpWI{7?!^TGG&*3XD2qbNtILWbT(>tg*?yS&n!9UkK$b*Vu-m&w&>3f_MWs35AT
z>700)m2I&&2KpKORgQ+43Jtt{s;E}0k<XEY9+geF1x}xi$tLTQe{*qS#)FVqqBD#u
zHLc7sNy)Hx8W@C`JL%>0{A+tcfbA)EnI76+xQ}m*lD-FgSH+F`cXCmbp@Ab4UhW-n
z>2Mtb50$Z+R@)ouv&W=R8v`q|%mz$js4n?dzV=)3K#TQ>uSHAKj`b`iR9=mqS>C!w
z8i9SxwwIVOoLRTR(@qbX8Ued)NpNrIupTe@Ut68YZ2->DUHE)hARKN`Uz;Bg(7EDh
zVT2}dM<|37S&N!RU8+ErO#rU@@$utuzIQzlQd3Ko4B~mweC=Et5`NwSoxOA;kPPhC
z1KjA(i5qZcoyh?Vqce?lT?G--)wa0+_Vai<6*P?26{@1Wt4-=fXUgR+S>vFdON@f=
z<t+2D2jK`0dNwUl=G$#@7Gw{L2#eO$)`NaIJRHcYFESrglq9;xWA{hFRKx}-8rqeR
zF^ol2^myO@TGn~T0WdCm_TfcQnE_%cnq5rY?GYOK@hXF#2CR8<wdR#mEqN37jw;TG
zAI`!<)~)J=9Qw4vQ|qn#e#_VLBFH9SI(cg#dtL||TthN1G4Xi4>In2!$7!KDE7?ID
zAXP~vUhze#+ZkcmrYXCRfItKX=Rc!}XL{7t^&8l1XVl>Y^=sGFbnUqF)Km<@y`@Fz
zqA5=?3iZWCS%7i(yUF{rDulCX7MPN-#sfN=$S#_rtwznR*m_xcEQQCjICo42C(~)A
z*qs{p3JQt(T4`1)y4i(zMuY`Y`@oK*{j$8GxEmDXU>A^PqJ>vAWaxpsLom3#5JOb)
znLz<)Y%vK*bdZ682&5Iy8G3haQ=|F_l{{pcf~l4g1DiqWCwB*l3IV2spinm6OP>{9
z)!sE@F%b1qa(FeG9g{8=XVDtTzc{XEO`a0<o0F5<X-M7a!or{OaFK9vd}EbrND6P-
z6YR@w3NHEs5oJ*6yD2@nc8%=dK(OLykGcNDrNB8&HtuZrpmu6yIPP;a*$IlQ;&1`a
ztIoSD|Fi#cRS7)LBY8x>;fk>M48D)%X%a8Bcc}Q#@)}2nwYJDSgEM@nR5}wF1r4jy
z+t$;N_y!pdi+xX5;RLNzRh5m>lG~|QFWKPtJczDf>gxge^y_9CRM}%#vzDhI8chSN
ziUKgR-NJp=g-L5;65X*;X-^PSY`cvw_A~b!FKL<@HnyX2%dv49eYOccPv{*~<)?sh
z;Ihe(AxyiPuZJ_NEjP~W^KXW_Nmx3sUwY-F`v|2dX0!gTZuUDKw2VMgg5Rl!WB_w&
z)}hsSpy3iYFhf-jH>FoV0v@+=M9aK)8b!>REQ`B)paLMtqnxvMXduR5+Pq$>HM%=(
z5??tM2CnP+P2$d!qTe4S%bk&pun(<H4Po*%BulpG%f0)jsjtCX<`mu4<31LWI0f0W
z@2#l8!!Bzrce0P*fx-~#)-E_B^{Xg5%M`Nbz7}dc$`eG<d#HU133XwcF-Xxl;AK3j
zkabs!x1KpqAG2!aq8tQhpT_`BU6krmYF%bASB15Mb>!NNENW0LIW{gfZy`lDQI_sV
zi^W12mTj{!C8IFKnJEj)VuqPGSy=WZkb)P}@6~E}u?wLJQFHg`V`9?2ft87TDz%qj
zWnwr~q&Cm#Q|ZN%E8%X|>qecA@6S|WMyY%XqIDqZ89tJcX0x9dP}v>EC5Q`8eq!7#
z-1!jwCqsK$nUj;&?N7->Tta4J)g}wk7Nv~z&a!Y3N3eG&8yy(0mLtLaZiqWCgKP#h
z2jn2KHRLxJ55$!iD=zS^=H7>E0x5lE`*p0hhw_S%d8Oe>U*CTIy`w`F9o~+%(`aa!
zo5srdlS57jeRBxi?niZC@7@ntX?p<T_s-gjG9BE0H?#VB2D}Wzb4+JS#>4Pj)j^lP
zU>2uq0NCYT%~I7}^}-P_UL1u5AT6*u!SnQ~qj`5+?RlvT_f~09>2ch;PKimTrOQdi
z${)Bt>w}E9JzVI(2q4h7XFrn90pcItQs6&qQ$ZsF-5Seh0%B$@X<mRhTyDS+i<>Fc
z*P9ai+DKWqze*HNN_jssv)qwld)ClJs(VSoCOGIz1Pz6h%)zKI4-C40MwTkq(DIEh
z+bd^9CeGrb#C35I-Obq9Zgi<>vRNZv^iO#Q*H$@0T!nzwayzVW_8L^_2ZW_70>=B}
z(<fE9yOZ7+hc*Bz8UX9r^p{zj5Y&rq2o+9N>aC1v+4HAxO6_1UYM;n>xHC;%&JOw}
zq))vZIgUI|2%uj5$`^R=QER^fZvckLM9p7aEYTt-f`Gk@ZkuNN%kyqd<6|#AZZ{c9
zp6UJNQP!ZLZc{{qV(Q8Fz1CGGR1UU3?YC2*s)9*iO1%vAcAJX$D`RJM^Sp2!$#Zp>
zI;B8%iYMaKt9v@YWxKg0Dl@G0vk+L5!o7V*71CCZ8)>D|2q8Z+AWcomc{v|tjay|u
z;XHNTrXv^?q3N26T+it65zks$V45%)Iu`NZdH;jv>|hsu2~c@*PmIwNRhF>0aTgYS
zpxcc+O&yQl=z9~?-!Z%&EyXFdI$pYb)6troCjvlx83L-<sTCFnD*c>uxApHb51OFR
z^s{>X(^F>S2`fL8pJpL$`5_YrvyQozR#GFcI#|KTmutCs?A-$)pXI;Kvz)l+BC7!H
zBrcs-<ZlVW#tL$0AJdwm&45;PDbG)*@<?LWI*tgnY-$A8W^H<T@;`b0N#9-VBE23F
z61KBS_RO>Y=Ts=o&jGA!M3Yl`kj__IzrnaTGPR(QP;k0@xvS3Pl!zJ8n_tz4HKT<S
zj@wiboivFz%mq-|XD#bb!{s#gQ%uu~bC-o+iCILq<xR5h$9FSEbXKRv{6s@=G@@50
z;bc#xj?dgPwf%^_mV~NbsPU}%JUIT(X>v)!F`03E%)ew0CCrTO8x+lSK)FS!M;6Ds
z3P!1>&;Pw~_3!h=^~hP`&o|>zVtYsZ(m%{CvEIp4*0-7ny4`!ayuAAHc`Kcwn=vOr
zYum1=CVzn8y@b6L(mFEH6kC%zpN6>JA=%K7WfYr_UNd^kTPP^7*GY%sVPvAj8||BE
z@?l7u;^E##9FIN&Be_EeKlNT6t-V09?W^lsZ3zQBZ5PE(LS>Lnm~z&Sfu0!D_^oF@
z`jhFLSBYDh9*eT4PbqP*)IWIbXhV5V^e99tLvm*$WJUYr9Mqve$TQ|s$#y0Q$()r<
z_7F#5aP@7UP(J~8vZC0~8i6UDBL|js+>`I`AFi&c(Ps1Z(dQ6BvC)z;3WUO`<@E1j
zrD<ExtVS0&PMqw&S_9+9E8l;uq9@3v!(vcndNIT~>dqBqz-I77m@bZ_zyIw}oxq)q
z2dZms8;!1kIie>~b>H8UB_{7`za{rSaho?Z^g@%hE-aNWbNW-U`<EV^O33TuoA!Ea
zKC5$R*A91Bwr6;XhFCKT>K@(m^3u0JC2`1knWNWT>CMyBlz*Hr^LyjhbR8|IE){6D
zHb?wR;CX&~klwYW!aApar$=K(SyekIAeI(hY1{J5fL*w-@Q-dr$GWcbWSGfHmSSb9
zXm`mk=F{DQlRZn+zLJ8$e|q=E^V|&j1NOJiU9s$O@t~EHz7ySQx8TR5rbY7UU4&;I
z1HDrid3(~Pu8xQjIj=F1clRgSLz_lM6S5{;b_S!UEDHvN9DaVlB1y%~RxQfieWccu
zVrV+-75d07$7KDi!F5i6hUhlrFY!T0sH$%s>$1Bnl<u-sRyvwxi#$A=C=#hVTjPvj
zB4)3!gnBWgIot9%``m?Z{(N*NF2Af%x#i2{T?MX24|>Js1D(p<ecSHx{Y@jo(2NRV
zV|yH`bW!-wwo!b{4yGm;t0#3wwZe`|k^uYdd1qP7nj*7$o-zlz`k1U<(B3t48(S=)
z`y4O++2D}jQG|C&il8s!dG98T%e6fkqH!TeXh*7_Z8Bl|&Pz{W#`x}sBGBC&YHAA}
zc*~P-Aqj{G(5chwzrwzd=IodRI3KFhxPUW<>k11Kh#xc|Nwarsa>Q|Eq#u@T-P=!T
zRD{{FP1d&l64+&6X3uHVFtq7$_3!s^qU~s1Nh2O`rlS)^mOgCH2{BzP_Dxt`b6|n?
zaO+JPHDGq<!Oki8Ca3M`N~+pj0eT;+cYfU0$Gg)6qprbQwOlZ@?46*PxXa;12FJM>
zp`-e<IGp4I%X(KvO2gxSKK}jp75i%1eg~O5k6ZaB<)ux&R^7TqAKpCFRodo^T`$7V
zV|_i$B)dTP!xsaWjRCS5xE_S%HF5Lb*y#42Gs)GuL{!vn{1_=ka6raWc0zcAzcpTX
z+n&`BZjEy1!iEP!Of1SgMdh{>i*S~=AILCaY7tsbcCXt~>?{cPuuelGZf`9x(bso`
z*WcaU^OwLEo~JjjZMDR0tP6X{TPFA7p{cK(|AS|GhS8}IB~e6Grj|m6R5e+~2@lA2
z#&ho?+rsj5VUN-xTZJU$N6a7jC@7?(RbLRfD>L>JtiySn2a+23$;kWz8$<)r$Xf1S
zEb-WMExon)(Z<o@TInPMhSaC&bQy2Ek`dnRm4~v&3wZaXX~hVZd(ri^@0DNn>KJ;7
zkYJX3Hb0$w1Ff`kgo0jWWtv&5Fhvh+rV=VK#9sD~PCG$+XOF5S`R!+U68FM{MF<>T
zV)wn$MV~D)msbLt8k_L5oW1dRmU7AaWvnHqwa<}peAlsL@%N5u6<kOT)_irpz5UUi
z9-WYwR}uO@-LvI-_qP6Xv&V>!{benLP}Ef{0rcv*>gEp#Ppbk=5l_X`)p>Uvof5k}
z#Z02x4jKh?ImO4JO<*=LPrKof>gbF}OulWKei_4Pq2CF+-hkWC5zww+X!N$}xQ6Pj
zm04~@Z<yb12`?YL@YCStwW=s>O;y-D=+ExuUPJ_Qtn51S(_CtKW75SQ+TfciQ{w6j
zl)@)i4nsbOZR6RU&nTi$x;E5!qHk$sCB2&Z`EVCOSh~6jPL7OZkLGx{W)r45{F8@=
zGqX6eeEhI6j8c*={%Pc!L(C5GG{0k=I<s3W6EU8`&E1v6-8ZyEpzRIL%1Xu$e-Cns
zUYKHIb^GtZR2cC^0zXsT-1{D^^Hkax;!L}64jehGVMZP-c5i5Ni8M8i%wx$~B2)N;
z%AYn?CkQ}AgJeR6o#FRm<A@f1|78Dk{M8^p8P7WPk*0vvV=S3UD;S6QnZfS)4)OiC
zm>pnCWNoyZu*I$Q%~Ad<*UpVd?-##Cj^E;b`CkWjWA6FCS6x8`hfc_S*xj94y54yF
zR*en|O-UTd7pRetIIXr07#V%-T)BSgD%!`(ThQR1D$jiJ4!vP^pR7wL*T<nV@6M?y
z-mZK~>u6I%qo`+i6P-6tU+3`XmszTYmY&fBJVx_!gKbh7c0VOuHq-;E<oNzz4?+(2
zA$eRQ&f23Yqp4e)77)c#8qOd~yX^#WvMq_tBAy~q<U*?a#&&Tj4HtVl@)n_G2sp{d
zvsB$HG-+(?u?ACO$-%Svsy}qXPh||r6JInN+~xb{beDbqRV#I5E?h<Z`IGMbb(phg
zS={;cCa%IHL(BI)@ObzyCye{PWc#;*C9+i6H{TCx<vhLjuP;IVH(jD~)75A`@VKp2
z8OL-l?(Kgt)h$bcqtAfTCpp--*0j%M%2;pdP>GPB%h^Ij8jdC=6c<aP1;eqolu!ix
ziHW5vKN*W!+bu&PnPX2K_V}@aPcXT5m^```&-w=o8-cptW-Ef$wrHw>tq3b?=H}!>
zx3Ez-t{N(ZLu{Jl<R!<m@7@m$lDYc-$C5q%=f@{%UafndZ~hyf*0(qj>8hi{(bKPk
z_MZO7qPvS3HR`^4XSaOV73ZHV%oEoY7ulmzOL;%ZXWnZz>$0-T8*~1MES#Ilj8Zs$
zDY9><MH{hv=!%L8PPX*MkB^7l<r|9FeirITRPjO6ORry0ml3AF?b{yOvsn)d4<(NC
z_s=Aa{~*8<8c}`+3(Q3L9jdMaE`$7%i`L|->(7vl0>F~<ys|XO%xMQ>cK3Wh<+Q<r
zedze0kjO#giEnbCRJn6G`t*WsclG?~>0OfMzKMbLItn=zA#y<;`{uEp7^0e#>bmP;
zKOu0seB>VY;L!ZS@a-7J`;G=>zJ)pcTT7!#YHf^>ogK_fFBGWf%JlWz8Xf*saao^#
zAx-v_Z=(rdqv_D84`tkvA3taQcp?9jK2r6oe#O+_bQrL@bB2bk<6GqSJB@y3WzEkm
zXc`yj5z{j26XO0_sk?TAU*Ok7OKOJQ?Ng}4MU0DQ_;kv9i3Ob2g&r?&D)m>JL2__C
zd+RRFEGpm?2H?gZNczCUVM##w`kU(D7*YZqdM1AMXd^hx*q}ArLE&Rz&M;51fz_IU
z;Y!Hr=mYZYMs-$A`zN0%O#3)29Wlml{sK$z|L69AH(R(bu=&aIrxVPI!=W5`W<_S6
zin7-3@gNvSiP6I1+*{);p6TA{w2T?g=AzkMma)=MmASiCwRQ+;T^$1|{YQcc7x>Eo
zz670AmJVD$;xD@V=jq*C#;9nFO-k4WE2UE!(kLmCd5ke0Wju5=JOXx$Ojs{H>sE!~
z2z<=;@3V}YY9bXUeXIcX0BD-4Z_w8(VW<N86UuhpM7=Nd;*V)=PCK6Usn(=I{h(T&
z@*MCmW0ojJCf3%T(ZjSus!FR{U(0l!T7`y|@bE7E9KLN;^RQDP{>MM^(gT_od>yTU
zq!X3D^Wu8{Ska8kz&IvsB(_&-p~JI6t%IM;vC@O~_Rj`TIU0rsf>26czW<@CJvW=L
z9KC|M>=xzi+psUN86L9V^PTiL{dR};3{>)Inv1)bc6sg&e!VI3ckj&JsC$cY);xMi
zaWCzHF5u;6eWMf4Fe;dq^acMTA||4`|CvS?L>Fg`?l^N#-gGF!b*(KdRp8z<;`ekc
z-`nrL>FOxpfSU1Elj>T8T#Q`NVZc%XY9MFtOk(dB^}4dMd~dG_IbEc`o6r5t)P-(u
zUEb#1-L-HEw7N3U*bxr5vm;G$9@R^s8&3*f_Nw~SQ9J3)WIJPoT&8)|f&G^@HYO8O
z=uxzN<aEGe&%sB(cf^QQ(C8fP#bNhzY01#U%m)rt@RWVLzE#<9KnCtOwX*JbBTFa1
zOfH6cec(w+M671~==QV9Ja)t6<0v)d7n<&LjD8XfYD~I4y<1gPEn|}fT`9?`dv?r>
zd0nP<D-UzA2${PDksc$N_9?^b`L}J%s!tI1DVY|$+tPi?@m;x_^=}`?{<gN3&~(4a
zcaI1%UtH9JpP%Cs4I$CxQrLP=O9qnu>#~0|{vE6xuu;#RuF>3SuQAoVZ`K(VtE1YQ
zzjCaT-TS}MjVytK0(di@A4hc}F`k7Xt1f>4vMdO>)5(ZTnJtv1(QpZ+t2FESNm4P9
zkiSEww(nL|aO#Ow5XuwJSqyP;d<&P{P@mM+n8w4wQ4@v9YOX179(l*Mt4y`o#kQMk
z^_TSA!T%L64Xdxhlaa<ME<)F1T|j?+4{@b^xnXOgbhqu;9V5T5kycfwZIjFJUo`kw
z*qyqWSy&w$ABeg~#82Yx_cKIxCn)GcapjMRR=;b$8y_n3W}wybTdfW6v#d?*q+<IA
zgMyCc-v1M?!qvgR)+Wx>+BzB1EUGU&REy((Q0OYhu6!_OW_>nGUxb~d+_}%yeCoB%
zANcSg@Vmd?8cT4wxagCMq?r71VkUiS_4Z4FuKeX=jf|C_KM8n0AK2SVpOJY($Ix(+
zmG#m_!+Yx?v}Cqogo?JmJ`eMkjY5`dy(k8_Y}?$ZtBuiOgu3cyfc4mxOdRbTzSWoA
zQ|_$y<70v=8K=8g8s+G-rha>uQk%ctBGuxxxwd*;+(hcET?+o7UAl$oz2eM_RHBwq
zY|L*Xll)zNfHFH`yDT=XEgc*{({-5n7<MtR-Gsj8CF`V)$HvAx2(L3eQBtDGHhmoV
z=7UplPAZ&TMEeAwE5*$kk$(u(<6hB`Dlss(PbwY=s82g2{jvUk@x(R_QQZ&2=XuH{
z=c9I_dIyAjL`1OAwV_IUDlx9uu-;@5kQ}@R@&09Q@m`3~g)<#TBI*7kYC%2-ewjx2
zrpNd9ZfCO`$F8+y7Ng?pvLQ<Os=B=$ub3id|159)S~4K#tVu}}JN-HSQj$S~A>Y&c
zv`fu<qW+c#vrL1B+s$Z?g95e2ef78nQqp^IX};m1f=b@hqJcO$9|fWlbgdEU1pA)D
zh1gF|JtMF*xQ5{e$r@!!$??0;6e?E$OG&_uMVPo}BsnsMIhJffAUS#4L{Z12+&Z+R
zfs)hjBe~YtgNF}h-&^X?Hh5S<BNU=tv3tjjzbVyA%=L6;^2rZsM#rgR-+bw2eq2>h
zpbyh$-kfMdPC88Y^?jC*0B_Wc#ZH*oUZczrmQxnp)?c5JT0=z95?c=4Y2(Ep-BMm?
z2}BCGy#dXr#~tp}*i+}(aA=C%eeeh7Ok3s`czY<|&hL7ud&OntzudeDKgI`hMRqdk
zO7(vY*2zdBSIyW-gL&%zFE5i0yiB;RwmjRVN-#8s!rF2vHvAG!DrZ$#!p%utR{c4m
z+|gO04f;Ss6Y})QyE<V{hG2uFq9OlxgfC_)uK{CpiKtW<T<tdBjTj&Tai8&Vzb&q!
z*PA2*>Vm)Qx5sSk=`TyRQ>5Q4Dar~{ik&s@JUIif?Qur+XPa}@5^khdmeEVZhc%M3
z>7$76%1*zTn>u})r_&C{wGDv@cUR(wXD))2-1S9O8V{2;tE}dKx)`9Un~@O&&snA#
zqA98}9{iqf1;k8~)DGLHEso^QE+#uRg93n$h1gg-5w2trr4leojDC-Z>>XJ2tj>KJ
zwUg01sWf!xH72=vv*OlU09PB<hf>qo{kG>yhB{)U3qnI4tj|lezwYBtPFA9hhwcGb
z+MM9pM@P3i4PXqVJ2)UFAW~CN+*h`3mcW-x!$EYSk%}EUc>2x!;m^gz`ET>m9VvsH
z@^my$u8DeDl(s!kpr`+?qB2JQuzMk{f|=b;jCj@I`qpIAH8Ml+5~gb_Tvm<5dhi>Y
zVL^5`ad0e7PyL=?j{(3TWFrZceU&s#<zo^4_g;YPx2CD)Z|XeXT+)U50l|6#&wcPz
zL<G3U!LkzaB9bK^ciN+p!&e0Dbj7Z_>gSU#J{44TDjRmfxY*be4qM9qOxQ}In41x2
zmB#{Ps@rw{V+#KXZ)|@mL?>MGUU_Jl>qLZzlXKgpcz#WE3R)>F5jB@-SVKi4HQ>7g
zfn2Ah4t#v~?u{L7e{BC9Yc3>%cU)!L4)im#3Qtx$x1(j5te!ukX@H#!Y0_h+<b5Z5
zy5>F^=8WDW?k+>@&oBx-E71rK@cIWi?)B%oA&F<mi@ePGvZn?IEBe;g&85U}fHd-K
zb3u!3x;7LA#a!5VEK(&{<jI}xA99?Q8<LX4XO^^0cr^yO?0EU+chmF|e|^909+hw2
zpy1(TY@o`rNh#6^x~#T9U%{l6`KYWt9LXP+TrZ+`rt?AYl<X<575a_t_1A!cj~3o5
zyN=D>FmnDJ{rCr0sq5yl6llF)h8s79S&+9UR*Rs>v^2`l^^G@OJ-@qWgHMHv6)M_`
zdVlZ6GiWfC@N%`}Czbv)baT(RdDY3&>(BcNbUxilhT(0w?xtVj=^RtFW+4v;YM$Bx
z1=X?zzWY4>@gx&bx>KaK=FcGJyAM{^xbkI8LpiV~s#<>jidfScTd$^QeIu4G7@nb;
zcet|-Nnp_--xD9%A`W&|!^JX`^IVJMWR{#Za_w3Ws88cjx|$t7qS2a87_z5JYzy%y
zgT<p}>^0%`TkmA6is~9YaiGtBTnIytN3uL{BNQE&l#Neq-dh;`ES9IBYGT3puZ{kE
z>zmEQ6y6JLCI-fL!>1B7Mn;?fS3vaHvbl6Mux$3&9o^)(lvLhGlKsj@w{OqKQhC3~
zQWp8CrC~e3rGn)Y^>cjmOZ`2`lCK{0g~@LLWPP_gI>)j;@R0c5#OUDY3Z8!a!MdoO
zcfnF}hLhhFfS4e=ZQB=_({8RuVFnPT%!amu`|2xC>jJ6n&a5?ZTT6<TDX7_3|5~9^
zt}3u)m!qV4h>o+yc*x>j9MRk2aQd7aIirXwdX5^m`yE}|)(bDveOCDrPYx5!u^t7=
zG&K)@eyI>T*TVK(+cNH0Xtc!P^v(3<qgUSwQY;g-xSj2dl;Nw<0Xn3864wjr>_ZLh
zknxZ5`m%-{8yf!&x^ZZ@4<#epy!hRym+0tpyL*<_W`mt5=@I2IRRpP6vlWp~^Yu1X
z&8&M1l%6uGFeKDm{1YT7>L48(gd(8KsMhW4ZVC8pVVlDt&lJ1U_t_4sN7Pf<uzn%Q
z6P&A4ymv=7H1z0j%v{@$P>*1)f(inV!WgsV3r1OZ+k#Ej>!l^=l((9l*`GihkQ~tB
zzs<}vQ_+Wy0knL39P#{qS68CbEQ8aTI`B(%r5FD!O4HhgQJjs&e*GUc^WS`W;Cr8z
zeZUA(zn@=ToNMcv!{*@0o%(1`$SmU=Mim&H$Ot2N^j>4@=$7!%IHLwb1Qt+~$Y0}U
zArO#d;kcB~fglbi7(=S!-gX=}-Q5Z8^DB7QGmZcozO%i(aM0|<k-u1XYE?KU;C+L)
z?X_-b;3Cl3E7}laW7w;f2*cBV3Z*I$T4ZE92f58R)Q7by^4=Mno7db?r2k$OLh(96
zW@9x(nwE&far0+LWufA=8XUi6O94{3NoZ1>8o&k)^Ak$8oNScEBOl;vLuN|x5?@>+
zy6r1>nx7u8Ds9QmPSf-;y1C$XqyCzmeRo$vwNrm8*+2%uLSId1rCw^@{P$89P>iSf
zUGKQ@@C0w>DCW7MJEWYa!`iyX0tFNl2FWf4hN3?>6w^Ox^eNGCj?~Mqp)8_<*5EVm
z`?k5>_+-HwLK@<B?nq>137=rz11(LAaJ~2QlAMJ|-$I<+(K~T=An5dapHa{nmugpe
zMJy?JCAEbSSk)7^-jY=KR$xr6U!=M-?rZocOzx|jp^J&auR2`)|LG@J;lLa4tv5e&
zOt*H!^b*LH%j}o2?m!kw^t(2|wzE+DBKHwsR!0g(Za)?|=yl2U;rn=fd7#?4a|*$_
zK9Cr<?;O)%)T2{TIY+N4d5cYN$>arf+-rXpszEAq3WJM-(#rRC1{fw34h;3JfbNN=
zrgqniWcamWs}Ex9`C~MNf^^5n1!TPcdga9gS5eo*Ca7Akhd`U&H)cPvJ+F}eGra-)
z!FFjSH4vR_`E2rdZH}vCz)3R?R#zf*v|n#JB_FabSX7YL_`1ig0o7%>!Q^SDp9Js+
zxvEUrUPp2j)<3>~7jW1hBYEu)@I&^dz)zh3`3N2LYz%)NUDhwd-q*lu4&+iNidOl1
z^mX-w6yibX_l7oorN2JBwi$n}Dv!j!u}K0vazd}r(Tef|b(3r>9ezR`ybb1;S;~y0
zao*v1k$2HiDl3)l*!1`E8?&*FI3I3OiL`PF9RP!xz~RXNR0!m_$;y6erV;vCejuUB
zX|iP~a$#(Q_2JT*#CsDuvCT;}^giLo`dC^WQmCi($Is7zI6i1};@dvL(Vm$6f0+6T
zu&TDM>w|~^Vj&<cU6Rr%g217ryF*&K*@%R+(jC$*CEeXfN_TfR{OfqXd*9zjAGy~P
zd#{*t%rVDaqp#24Puq50#q-Yz=_qP66KMadCJ7N0#dC;=*GGchNO*FzOf%o9UdMf=
z)WLp9Rh<Xha>is1XH?Q&fVdnZ43xM=_9D}4nuB4Y(UX!Ir`F3iU(h*u6I`oyh+u}`
zi0#o9Whj#m6i4p>G2$;KKJIqo*%nPyzVo&~1`re)nJpY1fR6!p0NC0y$LU#Iy@kO+
zct=MA2*sc?w8j#gl1OM%br)%`yI^{xZJRHXZxmTNDBUj~-2fG(kcdaPfynfH>4>wB
zi~=J`>+L&tBrtgAG-o6Nv^Oi2)Ay;NZ=IjY7+0H0HiAZ^M1<ty@U@)gp2BRZY_%sV
z=9O`94i1+YV{d=?gMc{qRs#P!XC>$_J&TQOJQ(=oWAO*EEGi-rn%`PJq35}L5kyA^
z2x?*Vs*{K|$>}MM!)D#s%4#bPw$E%(-f~RB<+$S_+p`#zG)znzx;JsX>=3Dw{V9?v
zomw;4y;tgZRu`A8rAK+w_sk!UDo&|aJ~G0?`(bmqc6*9-@1`VTR<PiCXrzFc&NFZ!
z_<?>#C-WFAm*$w_^j@H9i9Q@Ai>$jS9=%bZ!K6JQfN8s*N|nWiQAC(4ZS$wu5TWM>
zg|KJV-(p!}gQ;not2cwt`?vWzBICZR_jgg`XGMSQc|EFC-#Sci`QS{Cu6^OT@%-=T
z5`I_Tq}(06(ue;og%F=V`jU=hG#J{8u?n{rPLgB|e+=E##Uz2hp%WL2>BQTT-0GxX
zyFw_B_6{0^yM1(7aR3vT{Ta!alXD3RODA9Npqf~{3L<#eaOSK|nl02+k>T#@yuwqz
zHP!nG9gjVqzn}n#JEd}GF>8`tXF0+YzYafV|Bzhm7ZT%eySQoZbnK(k!^XGHUKORb
z`qrF27pLw)npFgFD;1nX#YkIQtA#E!Ic3KUe69OhWHgi^(o#vQVMDRq{)oMK|LoB1
zRufhVft4$3!Zq0!ZvgOw!h%DR-07Qw7$Ojn)SY*XL66JO{0!Z_4!^BGignD_u8h%0
zvpg6i_TWUJrYs7+t4~bf(~YZ61}%*F?YpZJfc%a6so%|3H(ud)JZ4#RQ?){*+!Gk&
zWVuB(gUfzTT&@?dRiV9gH1bcqB9rMl)PGgE`>K5rgOgCpjaSdo$mquPRI;kfOhs?D
zJS@~tLaK4LF?hxPmtE}Gv?}MHq2WFs)BSy+%(R94=A6CzDc&M_^jAl`vldfLp>&gH
zv<vs3_frEeO^r>uwjSy^xgCl21lnB^i~KJRmkN0*Ig%B$x9VkOia=_VzRt%(fYY1S
za(yQ#A#Q|inT?^IpuQP3$22;YnovTT`8@PNoz)kC&)ayq&UP8{G}IBDKbhEm_$5fz
zUTnx)1(Ae>vbrZme|M6O?^|S%E$^ovxDs<#mS@?!Jq;`XOo;Kiw#Y}-drEq`#=EO7
z{i_ji7IyY_DaBc>SC9q4d6t@pd)3RXLSiLRdMcp|nZhpCZ(>%=Z<b?5=^X3D@9z|B
zO?HlWi)XLL*Wi92Z&afRdzKLD4;bD8%rf8U*FEQMpZHfCHd_+2ltF;~uC6SUBH;tF
zejVE=$lpJzT}U78<`7ST&mf}ru_jHi$7_m-dwB@*VKl_*eD{(;b73z1DF`i{J^k%{
zgPv`VA0JuH27ms%y?IRPT_@kwU!u7jhn;!Y-8X$@0~n8(7{GYwCKvifWaA{Lu`C8L
zrLGU;ov_-{)-!Cl^nbpntB#3i4)XG9rvUs~C@wf%FQU9HNN3hN0KW$RovZV`aTJt4
zYN`wC&{m7b$@a^qf26|o|C^$-D2#`i^C`5n43wEapE`&KaE;m3CY8iED`=ZYNeN0w
z8fA}H!nmU1D&oXjnWm*Aob?~hM3;g?@p^l*W-;#=*&OdLVPYCz)ujE<%nMVQR&<={
zAAU<DE+}|l5X)ZE)zO_BEqz`1y(9+0`PSHb=0)>x=1-|@(Bc{rs28F!l`jvTZdkoa
zA$jOtUOF}9qPw$VqSJQ-gTTH47$wTTGQ)6eo3ogO)xb4mCvf+7t<?5L7ifKyw};Gc
z@9W{F)pzZpNv;+Y*z!6&2`H<?33EQxm)%Vaai;k%gft<OJg<S81dWmlLlWi1veh-s
zB>->*5l9v_V6Utw!!H5d2bWEkR)Gq}$i^~kOplr#dQ_wSTyChSyr^wtrml<jv`j~R
zzYB^{PZ{FbQFM}6;VQ_XeN9bH&W%_3?sGss9mt@f8_}ZW1){DBed&B+w0(X8tg8xz
z3ss}1Q`s{8P<fReD>GwOwDj?*hIs_du$%HnQL#829M(S^2AGS{FBNL*gU}!C`elc7
zRpO4N+O{?)%Vu|UF}$Lqyf3Gxr*39$>wFSKyMtoBO0Pvhnx6c?PbeX;JkK!xmfxYu
z?f(Bl<~uh802+XkKaV#z7wzl$A*1M?2@#qo!l?@hIhuY~2w97F?--K`)XG}O24uY6
z_>FVzpQf#@$p;JCkU^MQ7ab4d+;BR_BS+E8$*VfAzTwRd2UG*jP%SO!kA`G_*xp#L
znwUg-s>YObk;Rh{Nmw@d4fPd0Z^Eqoj*ih@CCxI;6u}nGL=iH>BSKM+Lg{?yqNJo$
zh~&F`_;Z=5LoP@|fiNWB+q<p5S5ktCo|crJ;kM62R3i-&i&&0ohs=6+Ro7Jfqr5F3
zO6_nSK+^p*N#~W>?Ea=-t#oPI_U#KUHtN5WVO}^px2$3&%V%$t;np2E;*O86Pqlxs
z1zZd0XqBv9dDD<{c*e5n2tlh-fIY34aM=QkCr+HpE^8IoIR`&K0dD4LSe!$@(afkK
z;5~FfGXenc-K%Q)S%LqSkmta+9D2VOUcSlV0a#fc%~t!c8mm2ypeTBy&0Z=noFGn7
z06M-dai`SEwn0WKWT}Y+E2%zk@wC3EK({#=8T!3(qM`aLYor3(V&+G4&sfY^=c*a6
z+dc1^c(&ljN5{659jy<=1(AcbD!VrMuf<A?GD~j$_^BSNsSq7&OQ;j{x7335S`nJ)
z2Ntn5#w_k)cUQL)bdDO!e@UWc{-@>m>@KdF>cB|;GhV#v-z}p)UV#$s(`CL5F@<lA
zp#aN{t}*r06L&z~f&Ndt#DF??`LD<F@nJhRvE3+l&zzAzF;dhef-I<{bl*-n^D#QJ
zl)e-_C<zgi-13(E^5Jg{Zc?P+cFJj$+B~#$F}qdw*s08o&0|?VFR9+R?`n!|t>G*Q
z?VoS!g*8;!3YGZL>6kgak#_ra8lX!XWtX!D=kMQ_mX@c6_;M5AaV~x|?DCE)kFK_6
zb?@it2WbVMVAd17Kf_PlczGYUlPAk$Q+?}nG`M{>aZ?!##4XRm;(W`?FsG+A)^~RN
z!B4*{oe-JPgMSnh1gW5mqgJBfd>-YIQpdL)H*6X9_TRD`x+^dcp^QM4q*b>1i+*8O
zgt&dHM)Xf(Bas@`DEN9%O~ieL`_`t*NQywc`8k@Sjw3hNP3KfVijtz_T9M_spH~XO
z7ZZG&>G{-}v*{<{5m7^Z<KIuY?hnomA<KQag&w`2!EA(6UwtAD-!4yeyf`eZu-#<S
zl7OT(Q9B_(VI&u3wu6!w)L-D8A2JUQRwEw3VqI8-b5h7!rw46paH{v@`ZTM90T;b%
zkE*YlyK*+6Jc|(pNDd#oas-?NkXU%{c+>3Gk&h0SeRSd$1a)PHGpcUE!ojg2jm*VP
z8fu;HUQyT#KDVk~4=9_h=N1eUHEN+Y?u0u`qlK3yI+#CSv*3F?<aPNN=wv?t(^R`a
zqr?cU_#PqohCoUE=2=Fauz9Wuy19|DUsyhR_BiL2%{IlYsYbOJc`$f>`4tQfv;;&%
zJLQx#k)=O$UrR_DuN0hLg{%$wf%e&JH1teC(rLCcG5Xi~((<n<e0+|~)h?lnh9?uh
z2N)?_uQ)lqy+NOKUnlC7&$*MH+gOe>=F`Z?MoVHP6+piLO;%s!Ci}U~NgQsss(djM
zd!~pf6zt`-@6w?V>Fa~>J;poYJF%smS-%`z?1F{UuF2fl)tynFS?$jAZt(S#9MloF
zWfpMF7zs0ZivQ#0AkSAkvbrdsFbrHD#RN>^<MtMe*vQXiMbD}!0+)6_2(E2ySW8=D
z5956dGFfPjF~&GA&2%o?O44}-xP8V-2kEJ;?XuwV!!6F6(_`JPW$U`N=^!~dO2)zd
z)x;%wk3^w&O69_NM7&-Eiz3<v`Yj8+YAQcX@<(q#MU@Iw+25P1P0f@U*2SE^4n;*(
zaou5tA(&v+UuhC;Un|i#UADQl;|VH(9tTKA)X9My=OP7U$%>N&7NGOO2#ak?KIJFj
zq*dE|Ia_mf2T%)YmhY{&K9noe1V|;M$|y=5w(Xo%A(*n6Y*GO0$*^7=G`BuYXd8FD
zhOn?@WtrGm=0FY4+6MTtp+Yw`P<s>Mb4yMaJ{4MC+8;`1Bx^vlpV=&ShQg4}&W<<_
zx>zhPG9>-f`$XFNVT?8{KTQ@B@i<_3Ha0*a@+xgXB(y5l<?401H7COpBLA48A}KeP
z(r+D9!aW+IQaZ#yL-h3GY6!qs$Q?1Z>1uc4@y`roh@_<BzQ%cT;z`FZom<xDT9NN=
zS?M7;p02@3m`m2i{{($qA95^|(~el8f2${s#<%rcs3l-%ZD*%zBv4|8-~PFWLtBV>
zG&1&oRQz3nuZXfA#HxXeg%MXYLBU`?-eC8{38iq)i?SoAocq^ee+n|p&eKTZv_N&T
zB|Y=aad#omJmkG(0~El-XpZJOXdXP~XT<_<=DT#^t?o}4!>zyCW@lt(oYQ~z+$JO`
zTkcL#R!#iAnuDCYrj$nryW5?Kk3rC7xq*6(UEJfKcup3lyf&}l>(}V$D*Pr7!qnKv
zIcekINsxOA>%$do^VT6Z7v#O+lUgu2qw`<q3cuoAVOe8>vSj1b&@|}!yevl4!<+I8
zCw^Qq>xuh3j!I3<?gJQT&8FFf)fiaT1q+erSjxXnN~)B}v^wb051d4BI0AYiA`_Pi
zTJQAQeiBzqOn#S_=>WAA(BhO4ie$^~A3$}@QypDxfD3rvFOkhQIx!VkDVO*TFhW3X
zGZ^5%X)96ht8Sptg1$vZ=VUqP#Z~4e^H*>-rzfV*yyCFiGUiq_mUGWmjq<*fWP%4`
z_JWg^+ReGk+D<KJu|!O1yD-J>Ik^;@Hi>RYnreD)f_C=KL(=u#k!Wjg=(&*^p7YpV
z4eEbMwdo6@nML|RU9~OIX5;Q;lHlb~O>kkiw8eP1(qy>+itA5T@6kW!v}3Z$vR!xr
z<}rV2;0bLALj9EZ8s>O<%Kwcvr!d>Z!mKCb+~%7o*=s&1X6!Nk`wa3rSx|zz-A?o@
z58HwO#c=@jVyiBa>Mkn(G9m)Zn%PU%v@RNbb0E5?f3zh|d*21<uJpRI->5D|)gN=!
z1>t^Il$Qu$#j6O9)m(f@quuHKRP&F`0}m&;q?}*z0_ne}s~FAS@`I3{%D%f)=#SgY
z%J~9MdbXYou`o^*>CX4E{Ztp(6FL0E(t)x9()pdOJ;Q5Vtk^lWHpSyyIN5QWX9g5k
zI8IJ!%S$HaHoE&`U|ZU}q6nVHbNHqqo5M<&{FXxR;8!h(zgQQt)a`ZG#ci?#g+aJ|
z6@=5G9Sq5r6G5OYlV^pWF4k&fM9;W6$8>aj^oL_taLZ($w6S?_Ix-)@KTq|?L8shf
z>dwm+eJh4$_-`J8_4b2xvx7SCM#OJXsUS2l&KC$Gb44DtB>Nn?vJ3L^jt0K;0&@()
zfBfTm#Z0IF&D%S{=b&o%9$jt1x@^r^%QN+IWi?-ph{$g9D~P6`n0y!zJo*xd(mJb1
zMT%INR=xlY4uoqYk3m#bS#AA>$&bM3OBtwrQ1#by8Av0<lnHUA;`m34*si$Fi*pLs
zDVMN+?&&j{-zbZNY7Q$SBmP>4DKaBNwOzT!i593w9Jo*x%q$dF7eVE?SfaV1(64n>
zB4C;)PqTa_Exz~3nxyjQS3*;^y@NVZA$>yyl@xU-L*^&QS3BT;0u>D)(liN~!b51f
zFo%~y%4fBOR;4;#pe9bbeIwp1x8mU|61xM`omNcbL7C;d27i3EK_l$<On=lDEw_R)
z2n;tn9vdl=JN_R$Q&Lj*ikD*5{I|M5ZU_)aZmC~q?_#rFrt#@2SAl`y5KzFZPC}(=
z6?uV44qJm<iN0Z^_v<cRkRl@ow~-H6kDCCbm}eDXZV#uf*R1orEoL>VC-DB9NQk%E
zn-URaB$<isC)AWwGOKSpMh{wJLrx>kKnE8er^2aF@BswplgSUZnq`UaGd-DDnHQ2c
zcZX^tR?znd*<#fDLVMMIM~R(7Ka|)_G>x<r!1m3!V5cn4`fA{EqJVb_JgQ-@*G3<T
zHv=H2*gnzm^?aHeX>a)=h^t5bze@2b?ms*XQ2U}a1R&+xiY<6>H8O((g51$6yBy-m
ziw+XXb6U1g`9j3fb*%$3l#R!2;dl@CbUpY%UH``lTEW+=GG=(lr1$5ks*?+o$qfmR
zhB)f<6F)j3NGVI@1<iK=OIud4KfR${m5MuYcgMe3vFS0*`kW&}Gmh!{7|^LkYURW0
zj<5?S*0!OpP>^Oof6dAvtInq=J)62*iC%9ZQ|ssUk&2?apLT1+1JM27D=KOl9Qby$
z06j0rFp`xokP3|j8qlG<7>Ui<a;d|K;yKc~D<e|0(Y!^_%Mw|54v>6CWi8!f(EALJ
zuzo}9Tk21iohaXRl=4O8Vxyb&#BIJ!Qr5<LZDaK7s%m+fFL(}dm%FLci!5=xk_faB
zRk<e?^$o@R)TPulk-lT*+6EmRIo3lqPl@cd-^xo7b_+I-yR$`mt9DVsY)!{QJ4T=Z
zqmxS@qg?@uCdOuqg^pnBQ_0V_B%hZ&NeT@0f8<d`&97+pe`erI_lq;l?~eoA?B=R_
z13Mo3;kefA-sUpxmGe0&Afo`R(y{pX*5m`zJWdSI>mz-fZ3!{3?E6s$l`G}9I}J#w
zSBhzVVER(SyI)i^SeXR$L%Jg)rdu2sJ=GX(3g<k4lTJ<10n#d_g36@6MG3PRF9aj`
zNt6ok^&yhzdjA6IyYro(tZpLxuDp-y#9tV@xnuICq5I41y8`QsSfw;ttpvik<q!2%
zH6z>7-Xg;gO&vH6CC&0G4p83ns|nS9($a;pGYxO0LliX=T`va%Y%LY4^|yPygo+Q)
z))qz|;OQY6?t-2d(e6(wyVmy}P#o`Mvue^r6kF~10yQlokZQ(PyN7vQ-Fbe1wzEF;
zdm{EBZ}nlX_|WgdX4+i`7t29M!(iOLT>hJ$9mr<2lbUL&$1Sc#B|3VbM3UCsMGs2~
zmo0K=Nx!u$w<Z#hIQY>#K3|xu5E{(u-&X3P@ajxS3^hETb7!U!H#~K^o{?KaBSC;i
zn)Gx#X=$3eBra)8fLylV_MCgNKi*)ci*-sulnjVs_X6{{nPmO$0(p@uqmynz7_{JU
zy1(OPb!GXM-+iViz}(35VTd0&8^f(sAec_l>bnQFvonD^6DV*xlMnn74vF9x4mXBU
zJL`tN7B3ri3zD$z++l;dPls5`)+la3WF{Xw?+)b8Kt`$8zWot2dBM@j`5Ua3|4op<
zfHx2)JV{W`7D|zabS+JO4CVfII4o8yJm{t#R5yJ?O<O|JSFE|HjEAG6XC`E=h(!~T
zGwKH(nEy+zP(^fBJSXTrK63=T;{NGDvR|%g(6Pd9RCu-&s<X<-uH()QDJOUJsJW}U
zI>*tn5e#SmZOAEn0^`-MrlvkFE-L6b`Asmhv(=T5K-|^I-v+X%e#psM+5F55D#xb(
zWtTht=v!zypOoe1triZAK*AN#+_FHWEZjurS@!$)(Q9~fx)%Nk@Da&!vaH=|0P;g=
z+1}*_eb?Z6Q!H*2E=K;$YTv89Ek%av<NCS)ttCSO#=@yh=`X{^QtH>yY~Jxr?_`UM
zse#0(6B_jqwzpShPjMLzEvDJEoL1WP<Qm^LFo!FOJv8R5R8Y*CbqdbgWbd1HBQr7>
zc}o4irv8(*#mlI=y(9(ooEfvOQl#gS5qN=}%dzZlXzA&4zrM*jRi?POw-g>ptVG3_
zwKx@ojt*G5hXk=B=3(WRm{AIZ(t=Pt_c3QSkBHs!{L)w7FumH<4;X4_HWznGQ>5R<
zIUn1sPAS-I10^g7TAvDA4fP^c>(lafRzbsau&cX7@z`#&yKM>|&*tzxpOm0+$K_<&
z0vC3a!t_*uHGY_nV=zto?|sha|G{Ce3q+DmDPduKPx2k;o1j2c+k!^4sq9u03st*x
za*vXHbQqAhRj8~buF2-er(t93SsJUXE?Ed#7f^|O%h5?Ks1l>4kv&zh3clHl--)mO
z;e4t9j2B0XdAww}``9A-O$DcBTThR&jpK)#E@sgp`WM0u@D)T-<e$54-*aJ5pHtLj
zUazc-j*Ff%fh<Kr6mK5wJpG#gzL!7oe}ilHx2vjSXeOZpf2d=O12m&+Q_`PgQDKC^
zu}YzAHVz1;AMpsy(n8uHN0i(U5dX4RISf^X_{?o?_zRKDyIRwTMW7n<nq=LhOusWw
z`Z&(%I*!%G;r^VuHBEo{{#a$w9VjMnFYxtiNl8|UCn(db`HpqqBbl?kcXFpo#KgY#
zbhdW3sp^e?yCtH`?=rHpqgS~m7;f4A?as~vKHML@U5XJJQT-t+%hGlQq^v#{7l2Ok
zdO{)H@LTnYk>ua|8Tf{qv6lL3_=7>VCmlPf-HsrNS;0sCGA)QGh)Iy^Zli`(RZjfs
zK(N5^MrPgCYr3`J8XV*It?`nB@e<iVfeX(8v8<zKA+?Iti}%SVq!7sG2ExR~32?;y
z)C+9*t4Z~~Q@?nGY7$bHsEOPs^Ky##sRv*ifQZ6w$7nL23Y2t$-1Y;?4An8!c$$4u
zUg<_aHIRD<aj+K`S4PH2Vkjwr_RsoMf4zbL{NW;&dsjF^9R&@!cV7X$>qsWINFuKI
z*-`dg{x^Phr1#$d4f4hFGpotR6Ez%%eTL5eq4#&qK-vyaAZP`_+^b#)3tRa~x#52e
zUIeoNMu&&{qeojbJzd=5;zCnQi#`P8AZsjAsvnr@ebgJ@YK6dST{ebA1f31v^24wa
zQH9g>h(JyA7CZm~DkOd;xR@FmnvcVds3QDfy?WgbIU?)y(Eh&o``W6vb)dP2^r@sg
z({d7qsHC4^rN|M1KAcv3?E`rz2!QG#n#9x)>4YKk;)WmozvPk{T%eSnY2h=L_3c0x
z?x><xWVAqqs1H>>BiUM+dKCt-B1t_CG~ehhCcdt?S$G2~Rw-dk!t0q!k85lO-VMS3
zkkDEG-}_!0ufn|1Onea$^~SBA_(e`tI?Ygke+&>Vz`!Fy?LT9yDgZa>dbPy@)0}aL
z?t!={!hooNVT)7Lsxm&x2R;ASi<agiiQBrfO11?d*XR6NsJS+mK<IUaYmM?oPmeE}
zXCJ9&Z@+8}lKU=7=IXkEud|M0XTg8kqGDmN`gCrwx1r(WiMCHuW<M`-g9v3G08W{G
z+eSou0-*N!!hLtJx89)4$}$kg`%47o-~Rhz(4-yV>aHpKxb>Uopy6j2MEGI#<kk>C
zYNL8hiI4Iu2^tz4ymoioc&5sJfzF7*_&0ag$@2Rk<%IEHhsOE&Juz<kZwcI*Bigjg
zlh}RDGFnM)zl(bQxI{C1|4`fpMCHp$<NJOn-WA*&d&_A3CQuO3Y$~2!wF5a{y~p=B
z`(bCjbRhU!cX?vPNReo`WAQrr#5R9|_v1a#Ja=5Bdvtv41at+XURs5AlNkGOo?15|
zun&o}+|K*j^rZc3nn>hnFu?<TsqEVeV$O#76WJbxo(u=m=iQ<n3&qR{O44W(C~W(S
z(+?Lrz(XaF?$BckeS@)_1le@m#!l@>ZyylvItRxzf8$!}=r$FXYl@J3-&;r&-0~01
zKE^Mf4PHpE^#vpU8oIOt3D9)<q+WY6rUb#+0=)-Y5RQL7d^=d{PLJ8FPdV4$M@Yt?
zi|t0K|4{S%Lt+F%0<!pyNXXmfcbJ$8o_m5jWWM5RZBn)mZU-uu{&AQCJ%D#NG!1(N
z@IxW;IoE8FqGVT`my%p;rd@z8<SNwc_t`9$F;QW4rRR*`+hp5!XpxLJH~UldiBQqd
zkc-Gva1%t4?iptSk9C*FOnSRIvZZ7NhX%}uu40FeIjCLsp~j_$()xa%opML-41fmJ
zQqw!<{K<|}>~~ZVO1F#e05HWU-!Yu=D|xxOId5dzoZmvmM#z8uGvm4iA*}(dSwuyR
zt8?CTM`2XqtDytrfuKAyVe^SjY$FD33tdD7Zc;Yk4tQf%)@vuRCF?fJ@z!b)sZb4B
zRdZKG2>#HLh_A6F--Sw!O8IKg76|k<Yy)o<!P;}89fSQ$(~<^HRBTMOd+sFc75`co
zSbi#{{?Kxb`oF@H$BK+bfnYWWYRZ<U@$~d`Je_ukVbp0#O$_dPao?VQgL=y9rj72R
z_7bVpzyM}zw$$@fi{Z*FuZWF}1gg#Q8URX6VPTAju=^h60dH2L+(XKow!;d@4VX1A
zZbX3C6x9O+g68WQ<Ha?P)R>1r$~)E7rz0)Dl^|rJjls{>fBQhmrTQM<hL6>n^#jC?
z00<-S_!wtyB$|(Kr$S9tJDHNaGWjdUL-<yZKXGR1BV;#Tch_S~P3R@h-UsRH6!E^U
zjD5KwgchJJ(CBPsiX1CJevHYU<&8l_A$x*G0@!t_LiChCUxv(Ng&!m?R3N_qdSY$A
z1>tlwhCM3@k2Z%to1?ArZZu!?%Th)3s8#wfj9K1o021nF?lZO@mjXdY^8NC1%hE~-
zKUeVLz8{dW*FYNa<IK0DSG7EwETe50lukEs<S(KkK()}IZ7J>SR55R&6D)=To@2oe
zkk$tT6ZnK^9K4c+1a_REeA8y4Gd+41W`EF2z?fTV7&oH$V24&casKADr_u?L4v<h?
z9bDw1bG7?q$WT$`p52DkedhDI=M&;6+>}ZNlu%GagpRI>r_d_p+cO3#YP$CdMZ{Uz
z+3Tc<L3>%4QQ_L!+8zR|28&*|_iB*@%iJ~9)g`Q~m|$2~STAvKcms_`x_}#N=lS94
zc#YFxYNG$E2T#6=?u~vI-Q#<i&WytV)rZ%I5Q-ESq$gJTJv_|hYmL(r+2k3W|0cty
zS`-VRv0EzOEzYlX#Ai}6<wLXW?ij^+S9j`E>%VK+@6?5n)hB@}uSJtrnb90|hQ9BK
zlTV0m#Qn$nP37XzzkgxQp6;T&bri-rluk5|)|V(F#|9Et$67%(O;aL|$`mA!e}deV
zePl_9O*ef^Hu=%e7Z0{Y`BW=ok+B(+@E|=S<HoK1MMt}(PZEK);&tQ(hWnjW^(uCA
zPnj##^#Z?tM!Llt2norMXlor+%(S}M=DT%4woxiD1lkJljHBqPJF~y|#~NGCf|HY$
z0=04NLzkzZ!(I^wKz_vk{lJF<egMzD#GLVe+fxS3O#N1F%c8Aq_A_X;gXxCPK7Og^
zHDL9Ami=s)Y>kQXd6I6lsd!A73o_}&FZ5Rc23%v)KTC)e6Enp66&VTSacP#5Y^BG?
z0wm`SSM!G8CZdLt^Z>rbM@L(13Q{98G9p!?`lxRdP@hQ&)Ujj6WrxrQd243acKq$_
zh#PdDn=J%KSQxT;c=<!GGA2S0Xi75fT>w?^?~$@+h{sGmDd9Xt)Xje;t%=El!nd~C
zbr)-Ii@_P&Tj#dVsK>{*c%ko;aCn;jG6*z*0SR=xs^>$M?+=}CpDvnMwlfsgcPTOO
z^`~?bdrbwCFbUQ^OR7#KHFtAc8!Mn&Da&KO)KY-X@yqP2vy}M3W7ZWas^7lFz*U?$
zh40-%XX5?0V)CL4MX0$k0n<QzeZiQnm2{skG9&=4Fz}@hDkga=`Ab|J-jiIrGA1Tw
zeNo?6KDGmH3e%8^a6wg#?-w5hHt^XJ$CjF#b@&pJJ4e*dYo7Rm)^I;M?p_1bUc|_x
z1Na+I;w|^(^tKxxx28sZk9zXaJ$R$*_plczW0?;Q0IDf;0KJO;!;{iLmyc<ZpC%#f
zX!g2OWxQi)x#G{n;P5p1ecD_RURHsu?*{MR(bP*Q?$#X`ET0}xKalqCa;ZxeNF=o*
z3K_Z=`R_+T4{n7)wjN%Scf1?GZoWL1-W}etbDgS5@_nF4MNac2^H*v&m)l&W2Gm>g
zg2x+}p5-Mr8zUj802-qtf1h5gJWqVZLuK)J!{tWN{xM+vY<Du{3tkb|KNXjgdo3Y>
zYw|yAC{1SEOWYJjQ=RNlgvOsTHo02_Et)Ytj|lxdLn!fN&&cqSEzo!)00gSWG0Q!$
z!#-XP=6VSW^KKZUc?ttg9rlx*>zC%WSBAkiRb?SxL&>Bw=e(w{Oouohr(nRE$<Ra%
zaVOg$_lW%q%;q~O6!%bZ!J~tWW_>&)VRD~JO6P7Tst@-vy1cFzp;P#zZMuvqt--VV
zQ~_I_uRT$O-@j*L$9g6-zna^(_F%Una5F0vc_+$~Yp;olMCv^b0hWM3PDEL_<O|gI
zU@s&LvS|!1UEMSoW@&g>WE$5cZEo1LE^$il#J7E$;&tjPxpp>{_x*)>z~JOnFv(Iz
zvHdeqT)nXD{*(1sYAD3yMvHLhYwo3TI6jqHX17|7okSp;k^V=?37uopnNr!qnp)!k
zGGqPvx)Gcwd#K1&_KK+kGWloA9<oyLveg}5#d;4<GV(*yzF!v<^f@BV0h$Zr@Wf7!
z2Zx^R8QtUzH{(=Q_U@LR&v3J**!apW7m$rV5*#c2n4#b>17)l(UeL^{WN~XH76etx
zz7~BP-p}}m*J<@5d@yyEf~LO1?zc^zBzGQk9sZ+Z?_L$O_H85h!LZY3B&3VOtjT3I
z$?M&m-DT$9fW*bw>Qj^B&w?fd7^zIiizctg5m$4Dg{m2T(C#oZvQuh1J$zy9=G@(j
zT0Yrd9Ynki?ep=nx(%?>eB!ceM`NI6=dWLBzJonqu<2fyJ5e^W(v#|}-I^0Xlj+Wx
zkp5fCm0WIccBe89Bd^hj*R}goE1PLf1aPU#_F}Dy_IK$1u6F@-*L^3eJ>d_-oa@K-
zLtH^}4wd7FcLzJX{?ccfQ9eFvUsBjR`Wfj>E`1?q`=OtI6vg<h-tD>;Rrtu>iC8o$
zSuZh4l9gC&yI9J6F?}$uOW?4S;7|>|C+AiqX(PHhrIUR+8T^q!GyYsfO~xn_aB~sZ
zB>3(~aS&^|MBlp@6_s|XExh#87@<b_oXzqJuH&N$Cfw0-hwZxv+bj>~U{7q2XNw_d
z-6aC|UgB-X^ei*kHQkO0PN<=KrSguLs~dI}y!Qu5*~UnHuEycYh4U;@S<-Pz;-!1S
zFCt#e1yLd25>av4Q{xt++usAy^~0StuGczeGPQ^8_x076i1<7%GMWwON)IBUMnJ6!
zG;-x!Jr^=VqvXb?opR)D3DmELSE-@wD*MJ&+}ot(E<3K{DK9H$iz|+R(7z{&T{k<B
z&VN|4HXd&yJD)Veoh))oUp2<YNtx-@E#>1C?5sXr%dt1lWg?fcGmm2*nkd&Adqd=!
zrXpTHcyt`c;&x5pefMm0a6o1m6LVd;pt;0LuzT%eEsyaJ;Sx=%IV%G*JSabDKjY>A
zx*fh!$UPowIleGmaAy5EiAjK)xAiVFc$td_;#mKrf>!H1OE&GOAT4h5_uAY98_`0p
zMS?f<?SDD5n#*RVI`rF_cVnq0C8kKesL4vj_3_!*=ASAQu^`AsnWgyHd4Vp*84eD)
zDbauFLvQ>?264{NUM}Ra8C+rd;^k9qJN@K7<LA#(AU-aIkl-qfq&%sCL3Fs7B!6CQ
zX0iU8N322q^6KE|ek(hopsg{b8~1eS(T9lTm4%C{>8KyiK@6{YT_X^&YC=dz0q?r%
z9kcsfm7ySk!OLqAm`KPj?A`Y+p|dZ7w?Qj~nD`(y@D6<kCPc?D6o5gw*6N3$ADdx)
zi1q^iV^mU5L^9%olQTZ-H8n5Wix0j}QqF(v?z26{siBaFmL*9n@L5p=Ruil95*6)n
zG1p%!wzEgzGFpV#BFWMXR^^=vj^ayfSd6DCx@??E-aPijDx6Fv-(=s&M80pkG)z5y
zjr;;*Tp4oXE*;y;MV#ndP;50wkN8>`O#TiE1CRSynqQOWm+*Ke`Nu}z-`%0PJvzJ5
zt|H4e&7D<H;wh8`E2Awc;P_N}!EQm~*Xji<PV&7)IkQya@Ct8s?0FLT#za+w<~ek_
zF|jlKTH^27sW6zOP$Sxt>iT+%wQ<2DN~qv>6bKq-?e-C|lyux8e#vZhGXIJRd<RD=
z1dxwwe+aT>=3Pn|8kug150#m8E8>Lid?d={<|<XFUtu!A$A29)nP0kvpFiawV^e;c
z*6RZOMwjNvH`=X|lb&UbOm+HuY%T0lBgf@Ku?lGIDL8`9lr5`4p%OZ_(K^$q^`Ekk
ztTO)DpTNXR$)GjHdy=MTKAuDS1J|^<rRd`0Sr9Jaq}cbyMrOoDsH)Mapm5ksW5p2O
zOcK0L%kt_iLyuN+yJ2t91@S)v*bpGI_rptIx@zNsaJp>QHu9!Ix2z{%>PCxyceIAc
zfrjL@L`YCwsxZVnY7jQ-C_V{|XS(;s(aaZc50i7{H0}GO6)Ba<HEER7l2FN5(p|he
zRB}_}*7@f~=8i!?>$;V66Otq$Ir<%(;C!|@VqG<jT{`WVWWQclsMy8$sMqDH-~Rzx
z?tGeEcMhG}@%Sl!mXjkgh`vv~C%3U{%vZ&K96|eE<H+D4H+sRvhX*D!6#rcQ#xoGI
z2Vc|3YSr!K{b*`tb=iE}>UNZ^NW>w(^Sh=#zHe|kKxu6zeKxCOdo<9Oa9Fa|dDLX3
zYVJG)OEgpy_fr|kmoE&?m$T(&DZdFOHk1`3XVhv+s}}#c{$#=hGE*wKkY7cRD*vKn
z(9I+Z`R+_=+(Nxfx6gdl4mc81rTHG}(V?8MRKaK(!DvPxX=s`_F59qg$aUep)0)AB
zqcVo;vxoWD>qcb{ER(q>+7+D>Z?~rf_qiQ5OLn0Fy;O_BHm`qfCo2RDS3x&Jt^DSU
zyI>N9YvCBbS{hxp3#$<{87#XPEArlW@EhJyM?UFD@^g`b(fhL&(-d&r!3*p0N2!B{
zw!y)&0ekl*YA6*b`}Ap<YXt1x|6{B3mRG1z4Ne~=E|od7@iM}&N*$Z$B)H<HinO49
zFqmMhEquC*$4M4NDpQw)bY#95YvZ80+u%yoq>5g5rs{{|I`nX8ThU;MsDsDl#UbX0
z?+RVIq8m@gm3{m4H|M^Sn!Lc3O+Wu+C2<O2lUT)NEP3}E$rBCnT)OVHjj_rTm7mX!
zp$wTHxb9i46t0qgs+`KXO0xMt6unx*wpRs%GfT4fdpq~JCHCL0-rQ`eYR%bKi`Q5H
z-&ZO}T51LcX$W|gD7I&nHh6jl%&dfh{C=K8m|no}`gLBgXk=~|e?xn*lz1pE3W_49
z$l_dyjea6ewta7`q7$Eda7iLh;RidE2*+BX>m}>TvprziLtKc)asO88ZK!|zdm7pA
zBTO=1Q0W&A>zn7Z8p*H=fBux&LE~}ag6$#(j+huCHW|(WucGNCc%g6SRn^R*AS09P
z;Q<@Jx}sryZ!6?Qwyv+~9<hk;5T8IwK_mA0D$zvY`_MYI<()12akF^e!6?kAD<u2s
zR=75K`Q)|}{n`0>C;zL41)-mzmt&;G!1pbk=5^=YT_qWrSJ%!g_`72=;3o*;KR@5!
zhpPC{`1)6GudZB|SCg(^V?TdR;zHp3#JGHNS*f$X3RZ)Bdq*L(*D$QU{&a^AyEVf1
ze3Jy-r<bB$Y$s}#)tCf6VYMZMSt80$z)#%FID9a;CblZDTw&(pKeql8=icRPJLa|O
z%pT94)v9a4@1OfsbDRG4GbV>6o!Z~;B|(NSeOAvKHt&#iGSb|t+?i?I<w?5j=jCYW
z?=O@(M!~PRGYnm6J(mAsTI~>~^HTMfLz23J@+1}K;#X#CYdeMNUx+>a|F87VCZCLj
zNYU2@Rd6^fFC5zo#QG06F7)ApSx95@N-Y0uiDA7LHE&s++@U6o&@`hut=z!)sel=G
z_y(U_rBPw5Fx#jb`>g!*KL`KAWS)H3yEBrel}T%RI=5*{HkLApO;?$PQDDv!pY12E
zC>lEBya&B_{!&@)ALnnFOo$A4$Hx`HC%_T!M68b|CwH~gRKr==dof_j;##jer0OQ@
zeHS%kyAflhXJ>lrUKak32?lJYU?+a`S=N2eEqk})Q*74#h_U14iDTTOeIwkf>PXxX
zVnro(%be(+GME+WT+tEvmL~+mzWg_N)0EPkhA*ABA&O1u^1z?}e=NQ;k|*%#Cl=pj
z_=;PJSbu+9Vc{vkz!*4xE}Z)&wl9Ij?SOY<q)vk_<Nx-x+uKI8VOX5b$A!5~7DMfL
zMT?o7)=l=cRpk42AbX<RyPE}HY}=SjsrAvC7U9*M>|lJ9ONxLd!1ommj20_ZhsD?c
zUoA_iG(&FcKZBl3_+#$)gZmfDUtI4J=IyH$J{?%5ZK30KFfa~maQys6ODTeKejb{Z
z?1R3H{Z!kEL^*q{2L~=I>R9o2bq{5G=Dmi8K6^n{|C2EQqfzWV3JOg}pW?^Gw&HOP
zr6LBKL@x5EZgRNEQ%nOZT;+Ulkn_2(R|f;fpA9M)jHs1cE{qBJAgl{n1^mH=pj>3f
zTwwYwKq;MA%5^X2vvF)!cek9jb~5nw$_qwD!^FbM$;uK9eevQ2SzJ*4=ksDu&pU8g
z@PdDKTRzw}@X-4D`Yo{k{ItP$r-K&&<MwVa8jB6@&d<-StEWTLzI}TdFw|h;<m80<
z;K8?gexgFRn4xtj92n1}ah;A^11Bp0XJnesu(O_%kdTzxtV(T9RSs=XWl>}p9|)j>
z4`G&+3*8vTV^&T*t86g+V);gLNN(Y!Pa~0ggUdb91zuz^@Fy*_5REM^zmJy)n1}1w
z-G#AsFa~IShBdVF-d-IdJnYGgfh*0%V0u8SA52Ku{BWhWYPar^3{EN%JUk@LZ*D~b
zvuphNj5Itkk)EHQAGpgmZg9XhJUDGu)_`plIR6w_o#8U;e+BP9d;VAuDNGaj?Z-z$
zERwl`u(ulDA|o^8GNs_5s^u2>h`Yu>`l)JRQQ&^Qj189s{$e9}DsRC5%tDium6cak
zrUDToTaGe!5tfA<Ovsx9E<-dF*ku=d{rUjBY>)g!%|A;AHkYpa11YQ*nFhMb2Y$qg
za0!WFU>o*fJ{tIKX9I`pvhW9QHEa+*LTbMx;BYv?ew@B?>G|>InDm$yRQn?}H9z=%
zG8RXI6vE=nvuy_W+T-%OyP{Lxkx`1y$WBWmb35#1pQv>yx3;mtBp}Eq<aPeo@r~Hu
z&##D!1V$v#h}K{f17(0?%Is#CV6Gv?u@_f|{j8QVHP6Q~)6=Q>_%!$T_YubKSJzIo
zC}*CBwOUl8zsx6brS2pbeBw%QzS?h(h>cZUxF{`UNe9k#-3;}^*Qc|E#DYG`;ZHpn
z=H(v3l1>|tpR8rXSpZz?!KFXn<Q>Zo)m>d*A9s+3tmQzt1+p?S-~9dkZO`^}=v0bZ
zHon44;BVi0+#Dv|hPzz;P8DHwI?%`QN!gjM9@AxpDPAmvNr%VA=70&|Jzomj7)bd7
zKF)5X2M4@e2wbxB?e(eMa)eS2Jd8rI9Yv<U=pF+s$*5#s_Q70zFcE(N;1|V1XWULT
zT3ex3Y7P!%A04V!uik>89wxfkyB|D*@(L~{D*6qSD2qYGliQ;{NHV6ttd9a7%Vn2S
zWxM_Zq%?3vKInKRQx#Sle~MDcfYD!PS2WD^Za#qT4V=Kck7|8OH5T6rIhiEG82G@w
z4{a&`3?tHw%>b?rWZA&{%^-Mv8mcZQqo$<H9JNN;)A+kQ>cQT~Rm=}1(}1pzu`-cl
z#?sT$I_5u7F*D18D=g)D?4hdr(#QE^iyA&rZmAWc7aA5u4hQynIc4fFSL|83yK{$%
zdCDn;br)jB#>R+@%w@lowOLRt1zr@Y^HQ-LStg*kxEQfwA-Kt4nxFmdOg?ZL1ipmk
zsOablV+9(lW`BNKhm0c4snuN_h{FfB%V+ys?lrVa#InFUJ3F1h(mDqIfmCd4itg_2
z)*Sd38@Yw0=96q?#zSo-^2o4ZvgV~1`!LsG;BZ$srs<MDHOcL;lee1a5`26FBPxb&
z-FdI}njtXrgV9WSEP8E$%<a=)5yx^^>2CWV%VcukKMulqBOH3P-%gY(@Z}yBSlN@{
z7SbziRyV45>u9aqK7P#FsX3Ga7F$B#DzMU#=fnY5;?a-#@ZoQw+cS~lt#LW9J-|JK
zD-;tG8w|vGbF>)3zi~XKHKuiQfSn7a-pL|IK|u*k3!{`aX$!=`<+M&i_$7LKduJyn
zzeU(58urC!zYoC5HXSS2n5hM!5lPwL`r<e=4XjAHojZIqbaaySV3e{qZES47hRBJJ
z#|Ps#;HeLDMMnWkPoc(QwDO&xAnT{Ow_tpPpYWX}BOD=ndVp-a^zPZ$g;cXCi*o1F
zoqU(w+A$k7up=xg=I<@9tZaOF!Z$EgX>;7iy{n9g>Cw8dxJaJB`-8}1SEWeT8*yoC
zpa?(cqBWVWDg*lf`22-``<4t{EfRsKm0;K#n=TxLt3rVNmbV=~3QK;}8^?b7qM_P;
zyO;yrV*+|kNVrqU<9fEo-2MocMKsiQt^o-gF0BgqKnu9Ew34u}&z3zmrOX{y8DJI-
zzK<Fn%jZ@FR$k#hn_yr-X4>_zw+hT`cv4a!m>ku3EDwg{IpF5~B1vd{sWS|V5=lm_
z?;6(tgv@Sy3|!oeN_E$FY_K|mR>&VSG)<@3r>Kr%?#p9*j#qG#(YzQEQqpexp?U~{
z8}(l{W{1FrSx|wid3&APC7EBLJQ5!nBcn`jENd890=SkQrpLaiEsn6d)@5sBU^5q*
z*7)VVTKqi^O#k{uy_dq3dj6I+)zf964?QeX!DTV)8ycE`O9@Ro*@h;;3*Va4XiAq<
zG@R$Rv<Sgv<JioXcKg>K!L&-ig2u$e%vLEe{2MkO!_ZzNC6Ut6<xlZHC{!sC4jnEw
z=td3DhGjhlo7YuVQ899*FCq8(@-#O&xxvN_3>jjg`LEZdL(^yrYh)$k*vdd%hW6+s
z<6GbastT1D2Ceib6#`R9xKYTcO09F5VSgeWoLs8mUW0?{&G~9*+J{tOmH4|J3O|1Q
z7|v0kg%5%iTu|$B3e@Rvu1lwdILo@<@tihf`>IIE_Ae;{8ew|^nbJw4UJo&7@Q3Q$
zTx!o20$fq>_5^&N5RHNiCH}hoQ$h7kRlpN7$m2;@SJ&X#{vtUXgih4QkAHDIgXPBa
zxv4(kJ^BWsL>Ivu?FN`+0#^j7BhoMjm<K-EH*e%NNArh(k7b$VEZ_3#YAy)18^hUQ
z_t0_EXbUPl+4Vf0ba!_T4W^49Pdm)652OhCfBh;GLdeUa+w!37<L{;oN7(D1j{EO)
zdlPxpz<7=6h432M;lr?OxDv!bHryD=1zU;6w>br@Z@Zl-74{j&)QMl_rglOwK2J|z
zr}Ot$%?s1856Cu<4E+8RLg0HBHEl6=vOSp}ym<J_Y<%I|EiE%sjR0}9corriye7W`
zQD``Tp)RYUf&<~2+Su3#ruM__&GqqK^AlQ-^zJ{#|7IMkD3WiNVcfND3w|p)2XAee
zNmFwgbcLHptX_j`&vQvZPQF~T*njPIws80iVLV&cIlBS=O19|g>l@AbkXq_`ejr<J
z0FFTh!(}=O&N<z!A%&~{)c)KSOrSQ&wCTOz`DfW$B|go9n%*pc0RcCuf)IBWE~~7Z
zrCIB23<9yD$loGc-sA5-9Nb*)x^zv0od+J4%tgeiST=JS2>go4pax?F5XZ;MlWo;v
zeLv7UeJ%C(Oz}G3f8N`MXX!{!NlBRtEGx+^*dGak1&9iqN5NFub<>p(+Y5m!fO&9q
z;sTmC=*JHd0*+5_Lx)R@`em0b6hJt30|~!s``6?P42;ydIc+fanut&q8X5}GOxbrw
z(j{V8q-SP|huUvXq-V=zZX9$oD1jJ#akQRVKI^Uqj8a#7Ln9*64Z0(Z!E&hr_W-t6
zI^wg;Cl$jeBxQl^?O6V$>hvW$FRz9)?{PSo@?7xmcH0vQ47Eq0Qb~Mn0!|V`^5BKS
zA%yu^SyUj@%0SJ&Pn29;Ym92=@hmC?fN4ysNXUtq!&1ky@ojs%tv|nQGyp8+J@<za
zoj0-%4YM}%5I4KQ6+8!QYRLbX-Ckc>g7#!tZ?%5s^B_WARYXiV23cJU4*w0JawZ5;
z2uK9hOZqSVQ!h1(ay2kAqi(0uO1eD|fC8mD&G&c@d)}y4+nf7&?x}+e0&?$%nDXuM
z5)qe^t=}Mwj{vV;x9j6E)F)30djty<Mk*o6p<%E~SZ&t^p6mUxoGk!0!JErbTB8|~
zal<KszSf5TuCM)?v_R}7arJt~eO)@u8f9?c{upriUAE-{zZK*md3jxRt4LjNQ5NOZ
zEq1Hi)EWvzh!r)54l*KGQracs2yb8bU@WVt?cQ7?O+Tl@jygIXTb}&+nOXog!)V!<
zrdm$NWxm6#Te*CtOTinXzATs1ovx0KEcqE-JT|j|hR?{GSKn(!Y9W4*x@B9}Dgl(a
z_ZVL}2MVV%jLi@ZeEA_~igye^B^I+kj36#3MW+|%T>Q1-v;=??1`~6d_r~Ulnk+R=
zrA%~u7sqbNXZcw<61WLV0%Q>i#*5SqSY>j#p8W*o5+Era92|tl$LE3k4-Z9zQUFsF
zIaW)Jhk)UHCIZJFtq)mOL(?==R0Ckhc*Mm3B%ZeE=Uwh&0@rYTIeR+_h7lO|u5!L(
z;pSFD+z9KKI_bLEvr6mbt^oI`vUG}*T~IFM@9lkRZEtruDFnV7;JYyK@M!-6@4D-g
zN#=5CQPBZ^Oj@!8UifTfeP&bByX&*Xkd3jzB*g+q9RcVfCT<4B86p6XUe8S88M{6j
zl4h1HY3T=7YqG_JSZ|umcflraCJo`y6{9)Ez`)Rxie_xQT1yS-IXXHz?0aLiAS4%x
z^X!>UBEWM~R`gNb_&kn3sXNU;D9D}+3kwrp`U4_*%;5_xtXPdI+x<Xp$Gtgkh<|N%
zX??Q1kh-(xcvP($zXaf`{%bQ}G%b5-1E6s~J_zRn?=7`3F2^j&<6>fB4Ek)q?&z<$
zPvrWuhpF{4|Ht)$b!86-i~SYFC;NQ~+`Q$xQ<WuC?TQ#7h#0hog^!rQuA`MWOB!Hg
z{pJ1>dMh`84_12Ppv?Wn_FP1o3<!`jF)@j;0kZ&*hIHr=z=e_3JNdTu#$Sz++!P1L
z?eu85K1}eQPRZ?b&iaSEq&op<EQ(wDy%}2^q4OFsd=R(^)3S<4=4M}Wn@_N4IIpK0
z%wT~E2Bd$n&|enQsv9R$Q|kR6D)rif6{6@hx_Y5EV1rq|kylV4{JZESsT`4DZz~IG
z4#e`6i{6coj{`hEkXu-nGvDk>%fd4JhJuOdLs3zYL1Z!x4o>2+CXwSDlDCf!4F|`c
zLyVU%vl<%Sg2(}Wx3;!A{o-}Ma;&VRkcj>?Gu6swIitQ)0ls@36gy*}yd2Jf6lg)6
z>{9$^+kV<^Oe3_fU`$n6S$W<~H6`O@*HZj?7J(dBZ=lV+`R2s##rBZGPl<p2D0_Q(
z$#Y~B=N$aAMHVdWz(C8Kg3w{y;DLn8sO@ocwaVYkBAK3%vC+#uI}9QbB^;N_HY=Eb
zI|G!J{r0PoMpBuOoK)ta=WfZ5cgMxu^tqqZ;(9aK%qOT=Srxz*rnBh*8E=jA$MHKb
zE#uoKXoOmLCMLH&zP?Ho<^XJxbuL>+RBcZ!RDj5g*{Ao?hs$mwEeMZYk)we5--~N{
z2J+|J4^a1j3$MN0slJ$;y`fk)`*5<J2_;3oRFYec(RrzWdpPe)q$yvirle8{b?e0%
z=1)gq9wKZL>=0lWmKExULFr6?3$iV_1u%*}QNIE9!1}Q(D7P(bZILd(1d1?^F$`2)
ziM%dk@O;%W8auW9i7EjBfs45>PfEeUHZU^kUbgPsws&EEty;R~dGZ-j0Jo#~US2+Y
z_2ArHI<yqb4sIu6)t;dO!~&BcfB*h{VmBhDn<D_Q`VA-|bGG^TZ_XA`EobZ8s#mS|
zRdZVr)gegr@WHF|Lpz{6$_3*AQe;@KA+6Ptk>qOvnS{SzG?XF98-@K^S!VtZ0Ah-c
zP;t2pW+5x0DUwX~&le)H)v7f>JYrK=FJ2%(96WRex^*L`pit~~>jS6<U~rnlUm(v)
zCi09Q&erpek5+CDV-+^5Mgo`hb+9W6bq8Y=<Sxr;nwpwG9+rj=iV(S_B|$fpJysyp
zBh16SZ4zAW{DMHbQ`8bgd_7?bBBP}r?M$l`)}F*Z$H0jA{yhayJN*+m8OAgq0s=Y#
zv4Q~3+#a`B>iC|%;tZ-~o~Xrz1r~#@7ognCLV(xW?&ZPtq0ETz@HFt3mP`@IH%e{?
zos@~BZ#C$n+^_aUj<+Y3Ks6Ye2ALMt{u=hHLr@g`%b++@^Yilq?9(C7FAPelkA3l+
z!29|7eL`s-4*{3$x4$^8aIeKM;_uyQx}O{LFN12zw9ykKaTYQ6;6GR$_q6lXD~p?)
zKsmr01rTiksL@6xw833A1OQSF<{`^wF_i^|L>{uTvND+{&G!8Ag|{2~j~Sb{uWznW
zp{8n;ZMG434!l-LN!{vwcBSesvupra8V+YIYmXxlMND$?D;ys{^)Bu17J~hByfGp!
z&9m?In21kZr;b`KBMWfc4l_<5+&A&q*sc%02h~9+8KA<}lWwn!vdy>ct5-XIL*Oo{
z=;@`1iHQ^2u>hp$45Q2iT$@~mM0U>(Z3_4iu|c~Qz(=*q-AmfdRF>SD&8ga{+GP1S
z@Cq)gqmu&A^y)b;_+$kCqgrMXv9Pc(4DbRxv{&)M+!){}E{C1haMe23N-!V>FMW)N
zc!1qH>jlSO<5Nq(al&9ZTYi*@19>VXUA5O>Ta9)_&|*G&_JQt{9n5SkNDa#~GsyzT
z<NyVNlGb461_28Fn~U$Fpcn%jcyYZ8X)Whr=T<!8o9(a6ak*8DK+x1?D%s0SeIrGB
zB6M1HQ#SgJ;8)CIAMd*w<$b<SmFmz41|ov(>E0+Ju(qAFlvJ?&)}Iu>LglNKOCc~f
z2;R0J5rK$W1|m2}UBAB(3$oa*$*`IXuhrOSx@?zq_4I%OC-<ljx)uk;_6T^(3lNIM
z=_`c*b1~?Fgv<geXpm_0n<^hQ4<y|jQKEx7s>HAt8pPIeO#zz#cx+qXa@*&Dd*@^Q
zIFLNG{pHJ-GGN0^r(PigUN{JVn!LPx8lM|*TkrTD%VHeb)ZG04wzd5Iz3zMEVJpx5
zjg|#AoPkaEB4EMe1sqIV7q&KPDX^{wSp@7eUcRw08Q89QZmAfTlan*^Gboz@OW69W
zz>$S5Nk_SWZN24D)6N0IfZ-MQVI8|1V8OS5@AjU`&CgZx+WBOIfMb-pel4qjd1<;;
z$qNNfP*Xi`-_JDQ_#jZvy>?Y#<C`JiHE>wO1lSfhU;q8(<>R{yfMN~ez`iT+X0WZ%
zJByb7`}ckS>*w?9ryYI>?5i00)(2Z!S^}%MfamAtZUweOJud;PB@s!<%)%s~)&tg{
z5inrlhy_N=yqZs*pm8(NzR$pW($35@&ejwKS^2+!1*Aqt5LnRy+n}Jz9MbpLBjygw
zYYYo+@dERNw0YhYP*7i89i9!e%c}TU4{#{*sw=Rs4jQZhmRq5~@#L$4AkVB%o3Si2
zN)}Y~gFwMY;Fu?b$*@NRR26{0%ET@Ifm9!8yZPjM%&x$pY-h1w|91xV7sqrdn}L*g
My85}Sb4q9e08uq_cK`qY

delta 92979
zcmXtfWmJ^i_cn-%NGXamh?KN+gM!i_-QC?CH-dCYgOtS3NOy;HNevCs-JS2@`TgH7
zELg*unfsi5_P*+5v?I6vK`ya!gZ4>(e1GGH+UUHgC-mftY(p+9!Fw7S(deU~z(?<X
zxx{M0&M`<&h{dr++!>?$Js`qvY|@^rPu{2e5fK6X{MUt}QHgBVg+*Rth|fsBzsT4*
z;U==DZ0KEO39vaam0V!-$s7MC;76Iq+G#Yl$sWi5N?K=KD@k6Y>To%$`Y*J7>Sxv7
z(UFYrh=j;lcB6S@UoK9#zB+d-`i#d<yvS&WqYu^cd`VsVms}oaxha;ZEnK1@r6#GX
zw(Oira)5qQU?o#<#&K5w&7I1bGx{mDnpuT=$oz7|qE`JeQgl7So3qpk;xg~S{i62n
zF)XJpqz@(+$`s+pwocAaF$ts|F0TK1h9hp|u4vVd(iu_XM(wD->pB;{OlDUuSL10#
z`$nni#LIN7{dxUY#IR{Off5$*9k{N~W3Y;g``~sf=Px%-xt})i{~mt1g*g*BR}j1M
zI}e|~7?I=xJ9YL+t~FBm_D5~y5jm_x-fTvc(rOZ|mM<n_Wyow>m2V;HC0-n-LF&V+
zOII#J)jI|RMzm|fut;0ifcM{D2tGR$74=aD3+wX8Lc893`I<v#cSq!$X$Ji_soC<9
z%y}tKNr|KJ{z9{@NmuZ@8JXnfSCwi)f+ioXahm-Z7~zY!?L$^eliK<S*tc4@2w`IX
z#Ik9;95-1^SZh_Ocaou*_`(<y-|A?5`>;8I6Os7h)!2K<w!zlQ;VR1PHO@L0M7JBY
z2EpxC0RmpTa`ARA2i!ZVT!maVVa((Ydtz&w^5r@m&Pi~H=_1lKR!wMhJzKth)ZLp{
z-F@8$rGmJULm<I7FY!qZyX4B%-vx9nnc}|n?tQYrx&~zH1#alX&D7wDuxU+9hsv5X
zVg8MNvOon}iB(bL^ZBBubID@TfkyKj?KBMCSYPuL8czS7<KiW~s5sIW^o8R${ASSZ
zYt>$OyVuEd^XTV@Wy=gUZXQ@Q(vEft6fIDWOHA=8v8`#SF@Ixr7w(Yc3!(<ltabOr
zk~rcJ4!UkoFrnm_T?Tu5+NK*_+FXU%=W0C@i78@X=+tHRHw}b?7ODNeYj20QkMdNT
zpA?5BYguRr2}|XCI{r;AqGhn0W%OBlO8w-y8((^>`6MUTkN%kV+-LZL`=7o<pf!4|
zy&%Pr(J|fsQb3COoTySAcmJvpWrG(>(~Fq)ji-EuBZ)I)9zXCgDJcv*$3>+}Wydyh
z6Dt8v1VToe5T~t<{*%utc8WJ!%v;-rxe7FAjQEnnTed5!2Cr+Sh4najn)V0AkwQwU
z99PPx2tKftyNJHTu^ca6h>A^o>Zy_Lh(bjnkC>VNEr~iO++tZ{IQ4l@Y1R$WPbR%K
z!4o&4`gKfb<;1_Z8*YSCE3I}TRW-2H8Kw{E?5_-7>`&+!XdGb?g4O)~A~u%X#qWh+
zV3)@5w=i<5&+;%?D7oNCTt4QI_wSl<3mFBMXT~x+3}u$LEexS)wHUE#QvH>_g^6F2
ze48;SZgwckM)m$~JDS*!h@lCzPc`S1td;+KT$}=VYZd2w(!t4Q;J5fHi68bdZBR|7
zo|rK@(jkeSg}cnt$ypGCk&KK9O|vNu?8wN{Z+pXqA&Z|Q!E*u_<Qy!CUd3K5?%E6t
z6=!>z&QzJ^!|V2;lH&0(=ELrfgel+bOjX3pCP&2=xe#tTT<)k;zlar=TK3Y8&&RBs
z4(x<(W3w*K#s%VyT?|LcMn85AXapqIVy{<8>YwI}b2a{>U9x_XJ|O<2F?Hv}<sE|s
zsNH*A@KO+GGqz0HPs!WU55H6W;&WXRjw)(*)S7Wa#}9&a$*%VG3;kPr>t^5zsrF-~
z@E80SXEKtK{hXD$?+OG8S`!DN#Xy5&hj`E%wM3lM=&FZq2hw5fD#xl)8O1w)a@;(@
zl6V-8HT_r4(L$7}ufR?a+f$v<W|G{hHOO5_E708hY!*C;^-P{@PM@4wMVdJ^Z0%H+
zbSdCtima}2GC_f3(ASQ};x}qJ%ea&2ztSRL^PQ*gaG2{DOa8lXBfYG|2dVNyk6z+b
zy|UGH|I~j|U{q4wPdn{|fbbSU?DHoDYa%TIt}seVH|&{e>3Esp)`8q{VFmLm!Jwrp
zDi&HQ-Yo~RoiXmglu66sDiUW>9O)lXxFPCcPAXiA^o{~rr|!BIMgvo?>bnPfb6FkD
zAHL$mk&f53F0Y7F@!FU_6qG#u;bqPJ6+1gq=7M?Jhl4Pg(vk|-&o3AD+wyLEH|9R_
z*t{1inqtTJa7{^tX-O&|qi(u2QMGyYj{olTVmH7RJbmP1<*R2}Jcu>-go!`ZNA|}L
z(QuD!VYQtPEtIr3`u~b|w#eKV(Wc<<Z^>IwvzQ-Ey15|VX`WiQ<57Qt?;5<V#pV2O
zSL9wy&su=fK@<DDQq9*lglaC0Xs{)HS%N5D9F}p*QaqC1cbv>!7W?X&!)|*TFKQ|C
zYXRFNPwg_!a7>GnXCBEUUZtA;*XAJG$SfH7;>PIXS0QEN-i`%-Q%mHxpP_S|E0DYF
zYuW!cY*>R>g|lleD#@4HoHE*la@8_Fxqeq>m@QCPU*0Tu)@&Q8?+&*+>#$nVZ|f12
z=$!wHt}LUCU%e-GjBy)MczlyPTjwUMnycKIRbKnlOeM;ZDYj<g!)(J%FR=#K;Tihk
zs478?<0)F#+3)3>o<{PHFF#^agHtkTt>*7oO>fblZ@bsJ`s*jT?br_oJ@-S@B0a3I
zkS;1Zn#2(o7MEgqmMfI&P@zWe>gpQ7jdu8K^PUOoYr%NN_#3SN1>Va&EzGH@FL0+q
zG^}MZS@Te)aytfjfi)*$amY53V%nN<!Oi2Y{p{UL4BWPGp@HW#35UOqlS3I5*J;1=
zvp<9g`lISv4c#7L{p7|$*V0ewf|?B*1iGEx{mCyP^KC`q1I*{j^T~W3o3}^ECtjCG
zCvJRhjl`8#^EAe>hsj-Cvr2E0d($ThGnvgeD4AdL<dmfyAFV0DC~y5e1e;0IUQcH?
z4E+^YU{UBiW+Y~4)M8AdNPX1P^JQV1dk(r!CG76+C#Nkqim~kN>8YjSC+&z9!D@hI
z$>j~3pb?NN;N4#P_KfcA5jYrDBofBwBueMhuOlUCV9HeA+<5$2Dn_!jV&|GlR8gD3
zQD?0B0GgPa=3Po;IqZHWYiNjeW8bvqE-W;r`cvIa8T1yuef0c_Znv~=2q33OC>U0m
z(wg~$B(^?cb75}3tj7H&EL4)_*3tQV|GPMh=JfWe#Dvv(er<g_#LDE^Dx6ffvorRN
z(6o}vz&Kmq&@v+bk$&Sf$D@^7pI?jTJ{qVpRZIN$_t*BH4eN+$<mm(hgGo{5<(OGm
z+NY<lWCLW)D4A_-5F52*l$72GL&CyONY4I@Qgd)rFD6;)>o0d*Nf639adB}WB*Yvu
zN`hy*uD*+3c8A*@o_rdSe}xDWF3ORS`dP90CBXRU@-BgnO&W$YaTa-JR3c#z%-1<j
zzfV*_*pQpkJ?7d=S(sr@5usZYjn-b<IU*&xpt`k7x_H!fsYN(^M{QsVdQ~qP98t%@
zt>9Z;j~S+Z`b(-h{+Y>GRsUfuhcSm?RK6XTlUQ%5{b8=$W^sK(1FxOEJrfgC-U*79
zaWzjfs-AuCylV1DK#2M(kI3bvE8%=?RHplc9jOZmbx^5y)S}bzQKs0ou?hSb`^Joe
z8op~}Y>Y#0Pcy^x5;M&MigV#sQOEHtHa0SfN=!s~fB8sWel*}^+^0|yU&f8^FAT|`
zZ=j0=1{mB=2&kOhQjvGiL`WB~b%GA}r@O<af91WP(BMmayYNbhoWJSt7g{82NCekQ
zeGCf41l<D#IYA8-(OX3UaokPaAv_2Nm?lbF{KQmF;Z~frZm&on+f;xV=q%<W=9D{9
z^pzXKb-VVK@{s6$GDRgR%OLe;eFbMJAz>ymnpj4%1HG#7CX${O$^r=F0Svh)6)A3A
z^s1k7e?$1JHLIHKRWW7Sp(Wts7fVk(Ki)^iYTGqu|Jmkw!<OPzxLn1HC{>$g1<*&W
zj1CJ+Kl^J!ieA5Dzm=n?m~_ibV7+wn3mm4kxX@>(=jQ86BI(a_xPJT4LuHA{TWP^B
z3MDr;yl;jG+ZSlP5L)WJtj!z=zquEMn}psgzv5$Cqv7yXveUhmJhM>i^;`7x;5J7B
zT{?J2phQ!<s#vw9(zjWzaeVu0aHOAGW4L)hJoNt7aYK3OlthFkHrQzQy3FtouOJ3}
z`W*`_=@|q4-cu&!b6vybk3~<5LM)rXi*GzRo~JO7<m5}`jat%aIl9lY{8cv+f32G{
zVehr=Fi1GlDkAg<kLMEeXVtD|K5k!9wKb6jU;fKiPQe`qqns`rbczBT;rBS2S`1<8
z=uk_yitwknoGr&krS>)ymx6S&gO3=GM;t52h)>*pR_<!1Q(%e<7Q>I>ZN(Snt;)Ta
zTQe;P93~b8lFyx$B{2vkzsJfF&Ncc=<Y0cV>QwXHlLvSSASJO?mv~C>8*e&jSg1LT
zU0}&t#(E?7^s7e=SA-k3P8+CdQ6ZgWHQ%R={L8h}9J$j5Z%ZSq6Aqk^_1;;%t_}02
zuV3Djk4`X0e#TRtu3+m=s3knIYp+oJN_4=DUShm!ITwz)pDtIxRc(krWm-zH|FM|L
z<doCl()Gk`^Y(t6kG`X7T%c6v%KewzsM&$c$>k1b(>LEevAstslXdb1?K}`K&nEL}
z8lH@j5xzYC#!&66dI@6vCje`oxVo|;P$_wR*U~-;Qm;_WOP>ElzHx<yulf5B<!jzJ
z0Pf@)rzP*v#N~rI*f8%xKf~mxJr%JahJ;*l=K<Ca>D>L#yeBZuA4rvy?k^rX(`Q39
z03LM3?>o`saxp&C)ghGi<8kprfqiqbzKOLo_v-AQPUKG`YA5?;Ub>y2GWqvk3ijC6
z0mf!tkmThr0oUnS!5MQ_VKh80bZ_5~fGB(dxeVM_?L#&Wm-Pq@>}GDTVtTby-hzn?
z_514X0dd{)h@td8+8a)RZW!Nj4R!?<53=%A6LVE}mGTgYKJ@S8989eZb~jYE8B9i}
zipuHk@03Gc5`D_+nWL*Lx6$ijfno1uL$P5J_tjv|gHzU%7`6~&qS?rh=839J$-;$2
zz(Zv=W-j|lmVeF-6W<R&2102cHdzXFE*Z?|--7vZ;f>m~XJS496dNs+N_01BVrzrX
zy`93p&K5&-ENr1ieyv(7M#hJ`lU@edIlM6sP#a*qr(y@sK`mjE)3Ssa+IJxi9|{zd
zIGwS$XaWd({ApoXtr0z)bbYeaN;YyAY34lh{Fm&C0cEM78qbmyF^36{Z2Wo2X}$Ey
zsu45bykn-b^NYA?uZtrR#VYa9gKC|;Ez6<$=oDh@8*w*20C|@;Adw6v<~0C4N=hoI
zGip<{jl(De1R`2+jF+|Q?Su3R$N#1He81O5rHnI~3obM2QE;g9P@rLZV0SB9hZ~NU
ze}3l-h#SHi5WlC*g^I%{*|@&Zu7~q~4@~1XT8r~w_GA;L%s#MxfAdmzQgpro;!Uu@
zlg^R*Q_RrLf7R5c&-U3XHSSOOzDMLyI?07PepA_V*c)3~<fY7ZmHX@d=RvX$d}K-X
z|H|f*Q=z7)nH(Xg>IY!gu?;#iD`bj+Q1C`ZxR~~e=>jeSiFC~uM<!zZS&&z^cer#n
zAou4hu{5)Ih8*7j^;9KlcK3*8!9(_sZ<1b#6D~)$^v^~Otyy$*H530;J}WyrMqt?e
z8N$x9+f)UbR&JmBBySP@r-CkBJ-trkg}eA^^ciqikEzb>E_ZZyc(=Y5v_s<9#><G&
zOia~=>;8XY^<MA$ua+a2Oag8)V98l;NJBt{Y<E4oumL%^o>va_4XkW^(^k3uLew^7
z(b4%Q4c~I%(a8<#YDxg%?P#X(+?ogLDx+k7zpfZ(D1WhJL$>?JYz|ry_Rbb{<F}ko
ztQQa{`=`Fw`kl68-XLP^YKSRJvYF^0SrUrRe|zrIg&e9-sLda;Qy6L&@&DIPY~rC!
zLC@_RhSU%q)x|V9!7^oDX%tjbtW6{S*5uS$$=m@@W*@qwZDJ!R^|*V?u$$Ng&XCQ4
zm3N9eLUtU6K~7inYuoUZ?hph3eh#lT_BTfM2jsdg$ZZ@B^|qGU=@QNu+}Rl)>dMYz
zH2?tJw&^??e0>(?=*+-IudH86P;s?2LS~bizxFQWpEJne93R2V*$p#RO<r;vnWfU9
zrFi@WSbYgVI~_<aHli{(7}DPH&=N@m(6H_tQz#}icHe#R6??|l6Mnz#mj^>)xGO3x
zIsO=5OhLvJ3OZ*G+z4rL^g>mwx$1|9CA;$3z2xNS@rDEUSgZk7F%=}GbL}Om?(v;V
z8;$a0gbUXh#UhNf<vF3$#{%CC!kQl#uDv*7yKW8r2ZzTYgo$@0MD@o!v4~cG;tk)2
zCQ(;ZP_2J_XB@FlUUECil$Z5za8oU!79Y?wjW;_-K%X(DzR~J;`N|gN(^bWLJiXsU
zi-3shnaAcU(70hj_+MuBd_iSPV(e>(NnYS#wGYjX4vywf7S%U>OpEW*G1S-2zDRv-
z^hPQ4#*DfVIZW?%p9O`xCUJw-J*erZnVqKcr?;@>=Il8-_X&RJbc@06Wwblu!S96v
zNhUYAyT3&JnjM455Si4$-gJ#hP|h98FF9Qx!STNex2J_nD3s3_$=7=MLFF|V6Y0y-
zHYA~CNmrMG&7!%e)sGb>|DLiWc)-8KwtMmQeA!g4QW(kT);9Q7{a$CiPP@k8>TMmF
z*m(t0bw#>ZPU?yKsSr>Y_%lRNof%9par1taY95A#V3}YdV{zeB0A&5ZGhE*NPu9wX
z5PDGlZa8umYTkU!>t(0;+X>@FK(X}~h_l{aZ2J*8ihFaX0H)y!93o*ivv=%H(wy~|
zi%YVVl1zL(9_DOD{aLFk>!v7;Uf&YBzgE;v0mW)@Ss9r&DM|o8^@Te@SR{4anU-Pj
zK<DXT<w2kXYscr=$9K+_K5`s&YUMak-o(xqAoPg_H3=$-zcveFTC3Wdbbb7biMg?u
zkpR90p8nm=m8;l+(gz@{0mM7{FR?W%hw>$!6HpG`he}D>+8nhMs9A=FN)bCIV8$dx
ze^C>ERP&Q2Iy=@c2M+;-sk)Dp`9Dk`BJ<zVe}2;4FZ!fqvT!GY14>Gu^)!Ix#62{1
zzod8)D=mMS1IpL<WcnUA+yB-ykr2qp$Rs7jQ*TDLPYsN+^vpJ&<&W?y|E_W8Gc=63
zQF||wxJiwGn<Q{?#m-dyXR0M~+xeWDt;f4(_j7G&S!oXF9)P&{8X9NQqQ=2!ZS62o
zVG)J>*V`7Tq~9R`P{RKY+aZ9-h)^b{-#e?UlrMKYKKQq?-;-|-2$#?JEBw^qgFkR_
zg4BvyTc?nky|Y=(rW$7{eR~IbmS%(IG<ZD&FA{7jYPkRrmK|><mYP0kYn^IeY*L`X
zcEb2gPSx>ibTwTm{%#jOch&4%TpSgMRxf@4rPF`*3-}TBI>Zl#9?qaOW;q?HhOU<W
z7neA`F`DNqKm2^o)9(0)-d;Pyu4hnk{p9biM!kzqR*mS<@o|Z>T^CSFepYxr1vA+9
zx~$CcBTOk%W+aZU!~lk=jvQZhPVs;<WKsUF-Vi_qC8$sT!1Bbil;-ZthLm7?w4|)h
z7W0Vx2(XFUNR9>8hX=7Y9P5y)n3&nq?bFg5F3#egpmay|ib@y_jF?*7nKN*y;`vpX
zkL4p5H(k?%?ua*5Twg69b|_2kpVFCUwrQT!xj-{j`3t_yWp9Dpe1CxlXMD<6^VcV6
zlDp<$XM1_$a`suM)E;9dbYQJm0AD{Cj3xh`%F1~r_=x;Yngl4rbn4m}Bcjrwx_>gP
zCu^-<PR}MISIOp$SM0F<P!^VK%M=gOyH+S*t8u-;!X|!nSi7j;FNdWZ>pnBr*L>pn
zFeC*P-u?eZBwmzlTR==WUFtJq%pqSi6$0B5Dej}%8H3yq@T<2}Q%U*R!6fY-8l9&Q
zkU^ag9_~fiQui&V-Z*&?kJOsCLL{X(^D}L#`qa78qr9Obno5)3KG2i|ot%s5X=Nme
zBK}!>xs`KaN`5*MI~fZ_@T7UbGrr0=`G9J>tpg=3H}`iBx^`^c#?N1X<gEaP^F38V
zGy5@QZE}DQYi1vMJ)@MAG;q_?K2`3Dhfrn3c4*3q(3LOsq1@?HQi{0QiDKK373Eeh
zu~!OPT*~$vCa*QO*~SoQk5dn?$58*hPax{dU!xpd3#GcqQCp6P=)~9GfRLBU|A*@G
z^Q!aFqLGo;S(%)0ha}Q2QF<wTO@fawjNBL+f@KIKcl=AC<*GBLZ|6@CidYx<MU(wq
zY<Xg+<Z@{LpU%4zXIf2fx5cmT5Twv*A|A?#lo<K{R#x;nwb;(2>-nI?&(GtOL+Nty
z4NeB#kqe?B&0-%jt2pbg6+Xtq&K!|Ym6#nGP%yDhqNds-A0rcX?J@j_LhggCqZ=81
zQj?slryL${@*fcSR@r;pKh1`22rpgLJdlw8p48^4a4Xanr&uq({n^MFO?&ONw(?Wv
z(!TX`hPAfwin2tAq;w^a!|rc4QNf57N!cUOF*3Z!+;pb+{`8fT8xs53AnUUoIfI$m
zWV$B8zp;JB7`nRK(u)%1dDGCz<u%n!0j6VJ7zNV}TGt)nmy1|W4_5LYVgmc=W?@hL
z&a4fWy$g&LK@$edjt>9qwyfhC$;NG2*_-*D8t<KxrAmK;wSPK%J3XSnZfF{BVWA}C
z)r|<UdB*UqSJIQw1u}N-7`U)BW7+)~pSCPk1MR!?9Mpt3y|T(*22f2@xJWGi#kiT=
z(dmD%mnNNA`qKl~3~XbyoJqnzRr~SUHkAc})R#_2UkRe9oiZU5H>n?;k~q8H-${w>
zso}SdO^cUpeCnC2w+Peuta#Dz6gN-x!!F+-@q*%x6QJiz+vV{+L!%=e!}ouI&x0ws
zHxFZCrucO%3b{oH_`n)*z*~ilH^1@p(%|#l9D#7BcUUnP3BtAR7nxuA^#FV}Jfe%x
z@L|VZa)9o+WdVHq^4~T`El~Kw`NKAkU0;m<Y;=L_&PfOx2!?NcQ^xjAi1AFtB#CF@
z$FgeP>ARgDYa=}1K9uxUifIz>AMJg>T0+GSgI?}?9!dnkabo)bU%ugV<#c2FFi$dH
ze73l<r)gy7GW9AO**o^-cHDr@`c+1;FCeP1k(Y<HC2uQMK4q`9&L$x=9PO>>iOhp;
z4w^Z`ie}hD!}|pK?)~M7G5Efmm3yea_q0t8>hOa8?c6kROqtp`Ku(;M1XX!Zizsc{
zSQh6SNlzRFnu~rC{7Kv<J9T7SMeG!Yv|4TSFG)$7Ci0D1Ltu|1qhmj{H)&%HyJnS1
zAx*lEXZCf6;U=mabjfJH@LJZ3+||At;;3;^pkQ@D2c;CI6>mN4RKc~mbG!Kkt4&(A
zUYDRwfTyARtn0jJy`}12&lKAK0*_28En%?`<OF*B8{eiJv(fdS2n_wra}u1XCauwM
zR*S!30J-t3pHXlf`Fxr|d?rvHDI~G*NN1QwTVGez=a`lzh1Zh7zS#{4I4cdx*xYeC
z-`zCLFEXCyD$*r0MP*&&zDsJpY$YtTB+}7GyLp)&qw(Rh8L@}!6DY&p<gYCPJr6o7
zuke3ET3q`7hIGDKBY$S|B~hhXd%B^uy~(D!LC&8yO+H#fMbQP5oOyU+FW2+ng|7f_
zx%>A-L#$AXIkvE$bz}FD)UNpaCubZJD-jP#PIcv51(WjS>k0eaHrt)Xt<jPp9E$l+
zUuPlwv&%<NqjVREfTxy$Iv8O5DV2y4<~#uJ_MLI@i3K&ut#FsZG0R=W%|Y{P8|aIW
z$u92ft~uy|0hhIT==!yHobS56m3kXDj0U?XfUae%EtM}R9y_Vx?&g}l#_bA7a;=>`
z1GQxkBDbbi`feql@fv)oDFP=#@o5r$gF_IL&BjI4pZ7E=DccY!<&wXgd3C&b508ln
z(yd0%z@WkEQqLE5$HUxSDwYWy3%#cjZMq|hfu0ASt-PZUFvR`OhuZyE^T#bZa*N}z
z)%PE~$`?zv@7Vsci=IzpEF(}3&J3Q&q^@s$`!dUWN5wDobV;!qS!u8dRL&157E5X;
zj0_lWUq*Jz5ghiepSE}1su&3Fi)D?jmY?r0d@S-&ZV+Gi`@)(tu77@@Z{zrA1DSoQ
z+>{z2?8-~k0L3?BZ10ADy-Iq$qY2$0i$dR2(?xI^khjrz;^i48&!Czb-%9PVR-jfj
z5S+9QHol>Ja|X6D(&ZX09cncXYAVC=<_?=VK0HyJoW$m1W(F2OwY64To|$~Z_9@Ny
z9grbpb7oY0;GTnbMSZS@pZ}q7RH_X`6cm5>mlCeQ?Fo*vzr9tTLq3~3Vm1D5$^Qrl
zA8BZJ$V{!P9t9U3$}hRU*D%?J%YVsoV724x&tD4l+B7#oJ17MDlg+1?^EWye_)`B!
z<hNL@hN?p_KCHIikrM}cll2UD5<l8)ZC9Y)j>_^Y*zEmo5kHD(GM_Gx*?fo2f?6=r
zMg1_7WKqQa=M+ENV%zSRRm4#}?K7zQeHZkjT>^!x;pCfT!MI8H;(7LPiQ##UU5sk?
zpSQ3($lQ~_=6#00vF#0g>v#--vUKDNf^V52>wJvsruJN(4m`j^a~a=N-5LPfwW!#P
z1CO}r+{~OYojdh=H{-65*V@0@YH@e$uEdxqKDl}NJw@xjr3xT9G`DqHKuF{a+O~iH
z@=Zd_onGPcm5|~U9nT(m*|2BtgJH$M+zO%oVrO2O`GWj_aU|9}KC8_-4&L$kl|wSF
zI_}CtW(@+iD317*w2mGbN1{@}%hP_#iw2J>izF5zi<ri$Ny#{az&w-BvHc4VfIiAM
z-EZ@5dnL7Y3atGo=RiE8<Si)z^QD1V4~w2h;9TKO`hl5sME?(ocL+gx>x;c55l!o`
zf7&lSTu^ajh~82&L?!P??j+A1oaBj(RyF3SHMYv9d=GrqC|~tU2~T7y&wE<wMYNE@
zMddYC1B&g-6$9g?B{)`CkZU6Q>e7tQwul!Q^G^P!U8w;&m$Osb6Jpk2dnzaUwF_w=
zV~EK}v`x(8;Y66RT8%@|0PW(%Og7RoD@-+&TOZ5qc4=1qW&_Fr1@j4wTjTt%?NW(*
zU58D9&_y!!sWLrcP@3<^=<zCrePb*6#<$sX?iVo+4vV$3CqJ)Z;duc+v3R!F@U~L1
z@41gW0NH95J~%BS)8PO$u1&$aY_B$)29b9um>7KeLlvMMLJiyFvmBt1=MoP~QYt*S
zj2*9}{_BAqLPm!A4{q$}@2<WG?ig`u3Q}R!_kA04^@K4;Zvb=iv&!jkG$UEGBKafr
z;?x=>23So;Lntl9m8JhJ?ClvRo{PcSW+%d8HfRkDv-ETg%@6X!)X#p|_7NM-G`U!i
zLko@CyR>Is^ue`$Y@iFr1dTov^Wj+3f3hGc4Ys_!gdwTA^4Vb75#k4yW{CA*ZWlP<
zF|I7rJ12tG`hgpP+p{&el6r!#8TR;rHX7I%?H+VZ4fpI76M@O;ni<J1IUXNZbbJYj
zOnwojrvKfy92wN7%xU*YFC`?Cq;3-qxSNg5Sxf6wdB=M94(VKRteBB7O{6X<a5Z#I
z^&{eC91AUWNE}uE<F3K~J7_OrgMG^b%3dUDo(*P{?e88V&WMFIYOM`q{aJYLz~dJb
z^4xGfOIL&6GcY;1@LH?7qzopuo!W4`Ii#X8jXSnJ$NV}EDuvsgnCjWk-N(g8e^L@>
z1z^o26pbX&_3VP7s#x>m!M`G~a`VSO!M7fmK1+v7kO5ooy`x3O3DA{W1FB2xZyA9b
z8J1k=-u%qQ?lfEVy6ME#^!1)%KEarAD;tSo>G^hM_P~HqSKSxF**l#0A!6a;dXDs6
z<$9DNh+9AeAYNRL<_D=!Fb0QO_<=A5io=|ICwkwv{W2JiPtWZ4;7s!emIBc7_uQwZ
zCtd=<2{`f5`1&flvu&S~XC8g@MsyTHyxR{UMcY>oW9f3D>Cse{H!(v7U^|k25<faI
zwe6cF#GDmPMhhhxd~HTVpP@NSUVzl<h-%4>Mx;;bJbbAHLqah0m+I#(>^|dQ?5ccH
zvf_TMBBYLNVu<M#N8Tw?W;rkv>Xt_=Anld3Q&<+k{Kh||$ungrz*%P`SL;EAC)9j1
zR%;)wHtk#d<dEr~g(_3Td1}Ex1|HepwKr)Mr};_*<~&q22^gEMFwC!*$&l#Nx4&d(
zM$Y9CeK0KaI=u1elJEHArAa<pXUgu4o`ny0yh)a-@!BMBzWtEue05dkz8ob&RQ5mu
zQkj|m>$sp0!Wvlsn)(N5jm-yHv5&Ex6Dr-KqZLxcf+Fec`@?Am1UR)7>mom!S#uPU
zf|?zDK!C~CsX(w2!bcCWNQ#f|n~W!T>9w4}Osc4C>@rA{uQto`DOS3ey#J2i(_N6k
zUBW#Ve2`>)bc!H&r&%n?co-MYp2g&eR9aGSVx>gh+2+WeE5hTdePxT8g7xsDUF2I)
zvCt>Z^0+(bMZ2vU_SWqN^cEpP_sZlv6Aa`lZ7BRn<u65WpkgqN(46m9w6eWey+g+g
z4wwT-wZ!G%C))aaSa!PRgmtlz50n?qKxlkZcsj{d^JcnQt%$Tllf=pL0-L?7Ed)mB
z=}w2TWGA7Snw3$g${f3}?wK!14xc4{+V#}9<o{<0Tt8ZFyYDOW*|5A)5{|0kNd@R~
z6}o~&qg@}|8`yFD2VY<Rk5_$$9Rrpe^LT&hgNd0tR93<nq;^4u6)XP78_rK%p<0Zy
zV7yFXU(?{s4`lnU!A7Sbb|BZ=-V~QUVZu>vcDNay5VwO*;5HQ^Iu?=4el3#IO42}e
z>T2&NNaiewtkrIAz1F5<sDL0^0%U+mlbe-qT6);hMy@KqJOK0q7J{IL8YE!0t2uk7
zcUQ^k!Y|wuWGHMWN1x0WJn!@)uh<EVUD0+nzPO^{7)o19l5I@5>Lcx^qj4S>?@|RC
zqa^WY(oR1b?&4GHYwzqI+hczpYaJ>K{#914b-Tlr?>y5V@o9u`lM=A&!tQ(p^v<(k
z8~4HA2c;Jc<jUUKI{<6N`<qZ#yv5w}3deB|keO(RdUijojaPd8mZk;%i2pEvnBic6
z?M~|u>5URQ7;#5JVXA>t<J2=z_k6V0w|Kz12S-`3N4fW*Z{<1wrZ3k1X724ff^6-8
z8>}(2?!QngT$xl+BHqhduF~yRFlm~r4j@7o2iMx2pEL0w^3q<N3AYE!+M&TCirrl2
zu&aH7+`M|sPXR!ZikXg6N?UaO{X`E6<@n>u;5h=X)ya9sFYJ)+<hsfXx7}$AqsBbX
z!f+$vTdpmvYaH~ai>I<IAD9_<W#iM*x+)?dWSX8i<2l_5>wi6SP#KH=CgRrx|6y}u
zRnFyGEKr&`0|UCf0Z1_`d$g$sY59Is&i6e%kDE}ve=2G4L7^1({4`~?%9Yb6xEJAa
z8M;fcQBpP8qwUUcQo4{jDQ;X=vm4zS*Hczh<4}cg_=?)Y8x(y*??*#GMK5>*L47?!
zkg~<EhXs1^KaJ=`><@969?%s5k?|D@I+!vPW&71+Y=IHlh=4PCfNJu97Z{yl!QPBk
zSX#^J3r(`ta*0(BVb@D*Nk6@Y@Y-OS_N5h7#y9<rU!;ZOfUZ%uU`%s=F-g+!oCN*y
zW0=K1M_5ejjr9UGRF+rdi%KG+?I*0?5xF2@V>XAg;qZr~9>^J&?&9V^n%odrAyad3
ztLi?h=U^aZT}hTRKQE#==P%PDB*XU5Lb3>Z0K?V4Q>j~f+rFwR4^VOQKE|JEd$Au?
zWRdS5#MUC%KB0IV^t@il;5c31NG?+gwdN;c=8%ws^dU4OI^;W4+|u+oS)=jNrdRt<
zxDTWf=XsA(cf*UT^kggEW`vc*9(NnHW#s-EH-(}3(Yx0Vi*-TrKX)oauB_C4ser_J
zQL%{qaZby2&cBGj>sKOfZ|NjlJFdf6Z0LQ}PSp=DlLrzKe&W<W&?)5TdkG-BXZ@Me
zfo|bpoDCM7fEWL2c3!ao6ENOszz_tiYQ0XJ36uFQx4MXGTraU}!vvsI(NuP2yg7JV
zQ&xcBVIdD^60j464k9zCWD91pkVuA9!t!b<ayUHOuNb%OZ#VW=j9{~)(m8!(SH1jB
z0|YH+Q*%8%k{LsNnRqwzF?#K)5X7N->xA7d?01(`^b+mpV==s3_p&gmm@zmDzI|^S
z?jauM-Q4&0aq9WUoIc~_w0mU(dHq)`{7H?GfHJg(NxB@P(LM7UF1)6a2<`bMK&IPm
z${867Ag2c#L{)_WE5Os72MWXk>at<+^Rj|%XBCk>R(a*M{3r(_-Vz!SdgyzBOA0FJ
z^B_)gwl~+8EtT{P1A~%=Mm#b37cTraHXrxLhw_xc<X{xYrswL7;vy*nQ-34=dwZ0P
z4Vz+=3^!lLDowNyBJkEu@R)-??R$ik2IaNRFqN0+&4#iS-g{EgdL?$m-lEn3Drht@
zsKRyi>p`i5xPmKcwx4W2$cokcspiAqzlt~&$-Qx_i(&ZtR42oqVenw#+1=eXGJf@r
ztidtJHzhW$ycdumYW*XRJlSmW!L0J-1jAjzOwEs^SaG!*bl?;taB{jIP3LY_Y#-W{
zrhopfMY=R$2;iUhHy9;xRuT&x8rjC85zudkO&|lez$3fkUTQ2oYax1C7{7(-slzwe
zzvRj4Z`J%i)mNT95Mw|gTeVS}f%8#5{5#l3QFc)WZ7{nR2N9)K+y@7k(GFWYF`%?{
zO)ZtLaQmhXJ-1{1tOk2cjz_+x6V`{A5-v1<11Yvxb&vlJ`n?kAenzoYqj&&>K@=Dq
zoFBONpgkSF2GKL!4rA1f^IjDIzPf{u2x7bb=-+Wf!2zEfm_CJFxBTO*jXRJ%;ZZPS
zyAAO8yzXUP;yLgre1D;(^0=#8I)v6_d1L8Ixs&I{tn>Rj!~wKD9)4~ORTH^RawU27
z3+u-m49Emy4Z+%)l=PEb&?ec;WVIs~73Q6k_=%xh1|t%%9<o6q@~1Q5$Mq&5bo@g!
zo6}iY!}P*yeZ$9<g`)L8J_2{ou%hby1C&#G^K&w;hzwc(pz`gVFeiu?hcq?=JnLfE
zVc^Eg;j?S*wbpSLHrwE)mUwZ10V?Qc<+c#~&idaj!Plw_h^8gBTqMTts7Cu6;8F2R
z??9m~wYIgj9h{#dzkUBc@ahyb&0Uaps+X@dDJn{ExihS%yZilxYBZ&k$w(GTcD8>+
z*317o0@YA{B`&uSrBGKtIR{k;a7V_>&_U3J0>4O|&|F>QDF)FH$kFu-^iZ;M6M|q-
zI72ElImsQ9Kh)qspO6<@_So=W#*VR)@Xyu)5O>9|-1?NM#5W9lkL;ZPvj_k@&K>`W
z6%dJwlSfw*+ivU3=7(xZ{zpX2r<ySKnyW5SOwJ>wRr?q4ftWW^it$bc;VY*0V!;vb
z1OzZaL|k&9_d~`7>3dG6--f(S7!ajS);nsgr;X1@KfBgCY+@BC73p<_6D5Og17dJH
zJ39x*$C<XvofTGdKi|-+eT9j~ip#?r-))|{Rj64WqW^hSJo~EQcx`}Ibnoiw%5Gzr
zbaqx;wO;(;#4aI9N)Fhbo)jJInuLbh+MGRZ>CSbqPIBN*k?VuktOSig*9p1%C#rQ~
zNkFfQn{YDpBGf8qWO19WA>`-8qrHv+^4nR{XGF`rIeP^Rd8^slJ&+zF1!2zQInQ3=
zE+IJ^&!m{cD<#BEv)U@Z7a4?EFxX)mDr1S`_O0VUW@ctkc=+0Ix=?=#501CM-G%Qb
z-=};#AN*Vlz=mGKA#Dhou2`*7JgWX-4_)%}#?8j?(d#OBhqv~mYA8(qnP4T0f1A|k
z?nWZex8d+Ja_WWN+>4@7(Ny{m^wX0ZY0dV{9}ZsEci7=F*%_yy<r&_*C-3}Xr+5E!
z@X0PI19xa>g~1NdoXzP8%yv7K)oKa}h!)J&R3OzL9{^Qj0oMJu+WN;&!T?w-k+jwP
znE7;&(e72u1}=7T=65?z5oa?EoRiD%<<1VvX9E(wrR6m3U9dur?%uA}dONKdaBge5
z2L>oUeE889&nU3<m4=9fgv58P9!sXlPUb+X(XCRa7W`bDYkWZM9r|6p)2U-CE`S7}
zTR?BWpf-!Ntl@TFpC=vBaUxqp$Vb&Py0m<B{-D_I3S({0Bu3v<+?AdiL-mgjcKOB2
zry(w)ZxC8v4NW1@iU7~&rdH=25B%0{A#gKLJ)R4)W2|SV>m2D!UDqO!DUP4-n(Qd-
zm;qI<2N0`DxdT4%>&`hiQIV_f|M)<r`y-Am>JJ{%1GJtguGS_p{3nRdtL^_-KOEz?
zp^lkVP_%mN$oVnu;SoBpj)_+%FL9OPufFZg7SE5<Xs76cZA-p68I^g`qp7d2Z`7ag
z>HhBa@Zg}OyIYj{Q*5G`q$Ha0P^#f*wp2o5?2UEVi->ZtFmA4{FR>u3N8ZBn7!mLO
zqkp_>z~${oPodS3X^6^AaruIGYnJU#%rHDloS)zu`&8~|(d{Y7ws@kHkZWAtcl_+V
z|8_Ria3y-4qvWfkGCdh>OQ|}AiM%}G=s@#i<T!Vxfp+c9#43RSp2o9z0QZoi;*x)n
zPu16bvgbKqMwSSJqJA~UAqGTsbN=Wa*%w$_9K;U2w;}?a;i)whs2ytkYwGA9c?!ex
zFV%Rwo1aYL3kmaoH*Wn)e&pO)4q$Qkj@CfU{)AkCSHZjM-|z~uY@;_T$&&sRyd?~)
zGk2L{KQc3AbM{u%w(FS9C#o>jmHK@f4-Y~mHhJpoo7o{tcuw-WSAu~#qlK#F&*NlL
zd0iKD1c4VD8;h-^q;zq4$)wZb9YB=vB0_`f^=cj5euz)GBsos0Q<mtzSt8>NrY(&8
zpsKTV5>;HTBKj4b<zun}>K-9yc?y(u!))@a&g=G!r@es-Nr?x0T!2YOMl#JIHyw+f
zi)2Df8TR{dQf<uU1qMwjx0n^SVb}t}>Kk>{6s{^?#=FrkVtZ!}-weYR=kkZgq!UZ}
zr$xn1oE&^m#&th==34$a4f*78<Q8%-HLMs43|M~GM!zVKb4Cx@YBwHF_stw>u8b^e
zG;%udVpExTQO6w}h1<8@&O<{c8KR?SCJg~~hnCFdKJ@)uYd5^zy7zK-r1~g1GgztE
zR77L8plwruGg_@qx~*Ktn0GCXy<+FYqPCL?fRSjB#S|10LNuAKFilK-E#SEInv#+-
zAtB+#%k&$|29HZ-{jPAB*?7K+n%YRl7u2TvyKA3X7+PdvV$acPA0DU8+r7O#C?P3H
zzCejyI25-v=oJ;(9u+NZNGz=~v&rz=e5In`ygX`RV&apXsb~_fi}bf^&~sE&+R2=9
z;~^!JWVNEWN^|(bzj}CjGO@A!nV5(Z4#Ph@o3SnjEgOr7L$(QYIUc=^K%Z61nW|Xl
z^Z&Lif(T5iT-E9)8z+#ss!#g-L6E>C4$IMaA6oX9zt!N6ROFA<#sT_r!Ey+zmWu>w
zdRC^HF^QZ}3P%`me(jZLAb61N7?0B{&>|NI^oPymRFU{JWU~A=T<sESqm_aPE)N`p
zT|%+I#(^<R<R3W$^mf=_-hqIf446()8)?J{^M7I<m9Y6kn750diB{A~-d}L(ky&Hc
z_J&wPc`;!sL`9)-40(RKFXF!J`dgaJUL{5q-zP9Lv;T{I865-l*I@Ke6zV5u`cjkr
z8a>`IPt!46{(AT6IcbNyV9}lSbnjS&g-q1O!<eF1D|(Y6-PPuYnx|RM4%Uv6l{Kcn
zUoxDC2kOj~%{(|bSOhQrIjc1trA$sqaXlSZRxLL||NHl^fw8guYA^QMP@0U6&Rfsx
zGYVSTf^IK%i%IY3=vRyl2alk;ORDtr^p&+WENtvQ6GiH-_qXSVM@Myx->!x{^Se1S
zGct1JV9$s6ujwBCm&1CVoP#5NY)nBqjeh_FpSw9z%{W)@%y^a%6fC=kqGfGu4c-R1
zWyZQ$N<%}#)%w3pTPj^%S%cB>E<a*)_33TF4paCrjk7g_prb5CS@*y}=1@>$BWvUZ
z7ZbjTY4IH>lhZ}C^tVOHYSVv#F`QRt!H%{IOAWGNE#-T|&5ygLRM1JTKS5HoAnP<(
z#U_9=<#DVnV6$xZ;js#>m^v8}<5O&LzPYB~SvfQ?(%_|yr9TI@0yNrm9-adS<i*4)
zP6QL5fjHO&bOd1-^NEiqbH(l<b5R{HW+~UfrePo>x2g79KFp?uN2GXTR+IjK;}=eT
z^n0;{lA}~$I$1)B5P(4hy`!Tm)VR7m>>_c!-mOOU<YQxt1vUH<8(YyN=}!pGXpRgC
z$Zm(sPIQJ5Y|hrQaHs5jD;`SXi150-!k&>ajb)PtS?+iS&A{;R-#;!sU%o%B-^|U;
z6}(49Lo+y9=^>}0l81?7dtLlPX3(trY+zszxsC0({D4mcd7<BJ)*34ZgNnWH`>Q?C
zWcKC_vwZ%Je=O45E!#9Oia9Dhw412gMn&_oer~Ydj-$NTrT!(t39U~qO1t^Hu7zH2
zz{1@ZAPivA&u2Vbp*XxHXqozqr>iMPX@;wa?u{MN^eHLnw#5h}<O}Z{gYD2me;c{j
zou~O<dN(o!K+eBcCdOI5u<ru2Jk)rIXRy+G(O2?8A+JY0fgtf<lj7o*J$vnq2eTy`
zP&DItJG+v_VN17+5X`Ba{Y4Za*<kC03e4h<@5VQzAaUy0#n%n_<7?5I1t7iQnrn*+
zwEm6l;${Ywv-5M~*&5r&$jI^3q}+~M3at3(%{R~yYEMrOh=bYL**$yyoaQ}J`nVY$
zx5EY!JR&e~ae!mtam(OfI7}iQ3lC5A12-pVPtjB|Bc=Kx2;qcWZN0Iy``hbl14#|%
z^X^aXi9hHd7!M|s($b=)NQ-D`5u>1>?94YN#cR3cbzG|w+X{<_*dH!^LD(8E5c&G`
zg(+ley5Ugg!*4ZL_s6>VR`~YnRDv+`Jqaw;V4$ljXOeUgVQOls0rbwN=LK+_qinJN
za3zzY$@=qEK52-<aOC$0(Cn%(40a_JPa-XQH`>G(fz1DJXRp0M%sR510sk)CPq?Z#
zP+vr!g7i%iDpfSV7aOfT<4H0`g;1-&ugy%1=9bo7h#D1XE;i^uynO-d=S-v<UAX-x
zcaz!tbx$dfs{Z6=mcg8;rWi6}*(HgNG^=Zi8{qW&7f1*Iwa2i7d=_+k*R6UJW#LwT
zTob*Vu{%fBr$xkic5mO~F~E=(*H*j0m|j}ezETT_;9;?sXu7hcLd}L6`vy;&TM7xp
z=QiG)Et;cL35%EUN+v#w*dRmzOr?FJ7;10pJ$t$T4OA=M)yZc2#^20D7Gn(1{7px*
znJlMS5oYJ+Hm56THS3*V@;TBjE-tvU`+4#?dwHAYdU}F@sBpR-kxM19ukK8hTg+62
z9URy+H16LXIk$a>P!J|(X6%<c(E-%obDit&4+ZC}+<sl!?QEy#Y<K4JsE^V0@DI+u
zDTmc;jn69Ri%bAwxog2qSUEX4hfDRle8?@tVnmYf=}=3p=D5Kq+TU*P>Ouw}RLk@9
zuk_;^)&jzzc9j3lTdOqDd_G-88Kb&LQLI{OE987#KtAn6(jl`gxN+H5pJ3J>Yln_1
zq-z!^*Y04-nJm*jI3L@(FS8mj9#5-ihjZr|-mAh`srYyOeNDq}Kla|M)_^M9u?G7E
zj)dktd`;vAKa^}6&X;e#)2y;J)&G{vcA_RiHpZOC<w@FCPKNTM>z(nE5{~xPLlVj^
zUO=Zm9R*`!ay)_#+ocjb(3_f3wlhe%k?*~YJnn86z*Ro22WB8UXf%{M&=XCyIZ-sj
zClC=FtT$C=h(^qdD<dN_U1=^NDk@rfi%!H{F^n=38WwiB8n6BQt`W=&uux_o%v6Qu
zjmKg%8_#-61)NhHZ4bm2Fy;5ept!Y_l@h%_&k^w1ExN(hhJ@9V_tHv(8NiBf=-qa>
z+||b>EbArC=W`BN%qm{Me;EE1E#(nqt|E-2)HS}i%*$eBz%DS~V~VTp@T^PVgTsk)
zGUl|z#_2I2hPnm@p&&`dw#A;~meL|FZ+)J2;?`&v-p{tERD}X*1oMbI5jZcWn{nbD
zXjLiduzV|m!zk=g<*iF{N3t%xLwVErWtA{7VQF#a6G+`37L3=(rDlp0Dcvf0@Ig2|
zK}^Y9b-vvPFXW<@O3JFb=4Jpk*R*jeILd8*Ef`3p1{m{Hl_kOE=4LXlYt$<$X`96s
zM5o<p#UDR-ZJ<Sj@d9PzzBu|5@bJS;{%Ck|J5_tnF)$e3+arj}U0S6~-`}hYgc5Qc
zrPftC$M5r}r>3Q8AE<eF@W#dUx_ftbcl)G8;IU-SZqLOlCue78-vIp6aC6j$hR+%Z
zPRh|%LGc3syScr!b#S1A*L>NG;clG)aQ7%w2>P#`1Oc&2Oe!4NJTnSDf6Jk+k02LC
zvrk^tNomAtHts82^!)y-;*^Pfc~8$Eox2@Q%kK^Qn4g5jbdxbxmNkL^l}Elzll1Z;
z#t-{<o0#x(G)syRTo6*M!c7$f;lB!KO|h)J9F-qeT_xr)gaDib3Oz%^FV|<gkm*7b
zAAokGZjGrwe!O_7vYX3eMBp%T*xK4o&InutCDJGc-^ocADJ}2s%ZQ3TgBNQzuUA5Y
zB(QNabQ*0pWvjh=%H7bQAZnH0iUo14p1Ao9OOJ6`O<tfAup5Am0je+b*RNkba6%#?
zP=sjh{^;%F<CQz#Lgc8SWgZD2jdOBq@{{8D28K0=nc49wO`LJoDyF-od<>Q9{U=o8
zVgX<jn3HecojmVBVM_R^y120<c3L;^a8q&YU#z&1+jDT4fQ;#X#xbGhVZItcu~}u`
z>C80<rU2{Yfy$hWeFpgx+tXZs?T@2a6yH;YnO<^0Q$YNQOow+5^t4UQ1w0+#87*<;
z7^A@bmn)Z7Qc*_6M<$jSm`9hxp_xe7*PFc(alY-yZKcI0b+Wq++(|tvL*L9098ld|
zV>^eEUtyG8uj~#VgR6%i6<ccWkO_3A*~VLmv53+jyiUq&lhNZ4YmH={R|%@B0vCj;
zklKPPFEbfQa&Jx1iAO(u;i}dfK&Tbu^Ar?{wNgl+o0s5(Q%WX$^z!1TRVviEIQZxH
zTVJ#p^JpmYx8dOsS8lKS2}6u_vh(<F%_MDbpY}`0o0r7Ig8;is)j86J;4m7seMgZ@
z`0$NFBJN<?yf}8-$xpZUc5bmF=oKo&kYU|)ncX(=QjoV6Tyj*cyik8gLaT|lc@`(B
zs=S>1%{bJ|g$P!Y<kR;h-5yC9-P<G8hg-chxx?35a9lk3W>u<GPftR}BnOuZ#JDza
zg+&#x?1!T(aDSJ$qsf~;l6Ss+;%X?F7zK~iRaUJv!Rk6IqM{``ZGzzRmOG~7rf(lU
z#Mn7a&U#%eb97Us(uJ8IZiteMezOVAadSr`EZq4JO$E1R{finyKHYNP3%H*X5q%kd
zYI9LNCAIJO8cQ5YeB;|=LF;tdKvBA9&z_0PXHf^f4fGCt7kosT@wO55ndpAw6Q9RX
z!=zc~O;bm=OfLenpgpU%EuMotT+wqp&hv>=QcerC$6D~-tN<!O1$G72QN9))Rc}@?
zRWTUlpHkaR4sgFMc;)*BRVoL#Fe`h^0I~&nv*Um01C<I^PtxfvG`ZqwaU#+^ca+)#
zW(%%{OV@Sk8o3_nM=wQGOH#J?z_4VUi$CcxaN=$gFSgr5D{2Q^z4c-D(r*-);GR6T
z;@wKiD~d;OpSRC^-ZpGJMZ^8UQ9AiIQv_A&V7}HKSw?0^@cOR_>*vp(5qO#M63UM=
zhadoCGaR=kxc^*Tn7~U)KKL$BTjz4#OEARCSDGmT9w%OHHJ89)O>DnDXn1o0Zs&gR
z1`r>|)dKP>(>-{+-=D~G?C%-Rs#?&;4L;Tda2EAAs4L$ux<_&CElbq$+9-y5#8rYS
zS9-rBB^OOhZUHi9gFQC5M4W!jlQw+6Jf(Lsg1WKtty$&Z^oNMp-B#S9{u2z!A4UGT
zgk$US&-hD27GKprGnt8wXwo<`anRt3PW{;}1IZ)XtQ{<jASpdRuz$=U{~=Z=1M7Fz
zEBD<u=5M47uj|tAxK^EiR_w{<@0zfhDcvy#4fHh&Wy<=tJ-gY>`W#jHNKVm)B|7wn
z!I_0t|F_*f7yAPPH_V$GSaH4Ux$ygF{|c{@pXIV2$<%GwkX|LmT+}D~Zy8bINbeba
zHRgckiI}3b_C1v>7SJDY+pyd`N|yA|vZ_ONW#0S*5lM`B?8hx!-d@Xe4@$-hW5BEY
z@#Ez1zM-*wHi~rTkCjK}u-{rVdgP9_;ydKk+My89p^qmX7-q1S<@^kt{~e!>F<Ka5
z%b(Y;nMJ?0t+Kwe)->E~u(eRXCChBq65I8Mih_6Ry`nMXe9jfH9OzY-nm+zX&b*T1
zF0tOm%G0FfqjYdDNdbRv%m@5e5jqO_B1rE6BSE#Swp$ek<X?W{n3R;%wGVXL{`2I3
z0Fn74fNmY7lk4j1*Xi#smLnRj_v)V#^_b^!hUgm@bR;n8;|h53A|!EG8$n>~8J7J&
zn!W<6s_pw4=|<^BM39z{77!GW?(XjHJV=TX(p}Qh-HnuVcc-Lu^KIVm|Bb;LI-Zw%
z?mcJkwdR^@uDO{q?Osv6LKd#i7|p*n{|f_p|6jE9C7Mgp<x5Z9q>9rRXVQ|RhbHe^
zpH~>-YpM=ys&t9a__QLhJMzCNdeI*y4Go!G-=U1X{US5c8Wn?l|27sMlcsH~F$D}t
zG;#?i-uyXq;CjWe^&p)aw)2`gLQ6_tkJ@Y<89IR#Xj>c(GQRvgPch78=y|e~&(k#J
zY86tGHPE+~{I3~Vpaq9<<Jg4SZBaRi`)})*av3G+WH+tKy)9eqD`CRZbBPU9eo<zD
zo*|0)_E6T+O!>mLMJZk}CbI01(T6=5{5yAo!N^Z>!|81YU%!Vwcqm8SbD`}IZf!2N
zZ$V}|tMt@Vk_<v9x9jqR$t${22TH{%f1)V5k1v0@j|n)rei1laF*`9b%t3Vd-G1V^
zKi?3bInDfU^Db3wKCTpn-6l{MFb<7TxD?ddWrRvZNpl2RCu^cKjP;Ut1&vbVb7_%^
zK6#4_BC@$-jK>%*)>_EPKQkNy{f52me?kr`X0LL$;;uLkl=PxS2OA9xEW%IpV4Jp|
z4e`d1DwyYRe=#{&roF#;3JAx3NvwvVOBDcsWFGk39&mMs5-I8(Xo9}{O1uB5&(Q?k
z<D(NhJNuH?sWjN&Pn5&%cpg@<1|GzE-6|z5je?450BW2Re$Tk4T#*TV4329^?;xZf
zI^7<QPD+wU=CnqY{venp6%p=p63&{rANw8tsB^TjyZfz%HwWea7ZYs9O9(yJ1CJbN
z+F>O<v0%k)y2~Y2A49{Wi}R<?*?-+2mAm?23CkQAK1@6dy*1HJoM6UIE9rbJiiv8>
zb-0>n+<9SxN(4FC`Q^H>>=9oM2e2DTVYOV}6M|72*|$5djEr+x*P{;=9Jt!ZX;h-4
zN&^eb5OID9v3pi6Yny^y#LeqDQZcS6Vk1^KdGcraWA;W#Fprlq&iik59E%dq;OK=c
z#je;w!RWN%(N6!$Kf&F)yy4QF-n{|Rc}0oC8kYe3B}h-D%WSKEi=`fMKg6HH#?Ecq
z0!*u(4vm-xmMCP|z+fah(_D2itKpz{^P^g8zGNyp=G-vbnKFgvyukZr)#ALVMOSZG
z{KMd!>E}XlNMsC*%__`zM;*CG98WsJzs=#se(4TD9@d~zL3d?|Lzg0$toJ!}XzRw$
z#w|ORSs^q0g|%&Se>;bZ44w50aLrcxmjW0XcIRuXR}92Uw40wTnaWo1^Q)RT?W{7X
zRe%o4{rn(e_H5hgYS}tLA?wUJk?WI?5$E~ce5{#MMMkpT{O>@4c;lo%6{<-Y+~}g%
zVmhv~y1iApzJdOZnHfSuk@_njHTat*hpB@E<uDvE6Uu$~03#$M1P>1{SFFLYe{cZE
zJQ12`P>PGo%5H;bB9+V;b}(PNT|5ogt5>(yBN@*m<C$~BBFF)|_*1121p~u)Bts~L
z&lS_Rt?k1vC3lD8-r8lKKMl@%FlLJl9O+&sC?A6{pZLgxBnblBNed2hbMsH1pigDc
zqYo}Bf?>}VMIEc<6ysE>E>>HaekZrX2C}Z#KZMC&%AMd-7pq8WXb=OMWi3MRVI?o!
zr(>*Q11=f;g{g%F{lQ#y=kRa@s5!?#JOnT^vd?H^W8*nKem4?@*N>w*RM}<rRswdq
zU3u?D@O1<`Kf?>TsW7L_L7gBB=o{Eb#@pY1lR6X_YVae_?xASwFGuaUwALHeZH=|h
z7FPfMy`^(_9LZk_QSn#&-Gwm&f>a^Oxpf`ci;i?bZ;hHSaUD_hhIO9n6;6eZ6T~8K
zTq&lzU(hHc5e+<AJVz;y6yX%CGvo<=R7a7aIo3Y-8QmChTH$MJ3HgVy&QA|PO2TW@
zY@3}I`O5Vp13J80BGlj=%3;0ez2`L_Z=GJ*1_OSq?ai`y4p!Wv|3}#=g10gFYLvz6
zEBi7>mv8W9?@X=`cKP+J;Mu*u$$;<CPS{d_kcO|t>**UN*-GagYP=u6>kk-Fq=S@f
zcrW?B#U5wZFA)&`sM&<q<@LM~zFfo&WujE~t>Jbi3+I>R!_PUy5)x>PEf{%GcJ`O_
zE)(9R&v3H&Fv6)j#u~?NsGS*K#4QNv8}zraI{prn7J3f98t8g+X5Jk^A-gwKT%cC|
zYHMqYkcfyfx{rm8t)sg;h(<bTL-G(4)J33XC79dZ8HKXs>0{4!$B`Hq7>wA6Py_vL
z&i5QI_SHZ9;%GGLi-Mgl1>7ZAYlFAmOHJ-qS66x_CP7j(fA)r#meRocvFLZeA1=2(
zeiB?-TJi$ar+X`Cr+owOZ)ji;RSVVB)Kpo`MuE;-R#otjo;@sBqv3~(jf$DVflceS
zkco+?KOzP<V9#Nurlx+=Oa)eCs`_1m-v<b5<JR3PuBEK3>`fYUg#%U-z+tC@C6~+k
zM*t;VRS{!A53eUnt)`BknYszu$s3a?JhRhw3z#X!BncUpRp+>X6WhU3-KfHWpqE@`
zTKyUgs6<}!WQ@4`OI2U<`Cb@9I7V~}jqZWXnvXxQM{o)Ww4yR;r4KUWN&5GtrXEzT
zc1E;}F$*8K)NeK`EppOd_7CEAf5xF?e02Y*)Obk^Y^PAyejfQYmpkoCBr{m<Aupzk
zT2$Y$?54g-R=FtolWz`5v<`gr-MoBx*@x0`(kIusuvd0RPLumYFvLLyl5ZH(1%;B_
z_U|R+Dr2Z-2RFyQ>eh$0#(7IxC}{0Yy@iV(rr6q#v{rDBX4T0N6w^-4OvNb>x3^ON
z=N&0sIC)_@v)$?^il*41zvx{+DlD&{%)$;kC8lqp|9e}|ipj<A6A;x4Jo=yk9<huU
zyQP^XR+{dc1Vi;bJt<UmJi4*`uV_{C3;<7%;jqCIh5e4tL|-xZL^N^P>SPXUYHB(j
zF7VJR(|fUv1D-FU+4IuC*0$8K2*$8GRR8?%q_K-?V~c%_#pJg@Gy;6jo^O?v?CtIC
zYeR9SpGx6vL3}8X&k}8+7&K%jy1lx3|L?97&_fg&)sV&SWk`^i?wvswvUkS|_*EX<
ziz9KHAcS?fK7!$Hk!3#}LenD?C7YRO&`*~kTd-O|4$Pxw=MVhIr{bFfyHt{|`NpMt
zWF9X}bnvkC+k@QYyZyPd!Wc6&2RKbK+xh)<%q<5@tJ~<JfhySK6#j;=s~tWCg(e{s
zG0O*4b_nUXeC;sokCuSzXQ2Q1xaR}0)bH^fQd4N(>)yX_Q|mr2(=%tQeD&_=dDqqK
zUQT4wpQHAS?`850ivqT|c^YWb?Aehm_x}2-q%PqlUgxd%ZqxKPn87=q(>ZL<hjJB)
zb+z3Rgj7u{*vEP54C`&Y2}cC?ph)~158r}t3P>AC1(u44kL#{<295Bb5o=}L9Qr9o
z*ty2Xa`jV*kB5GHlt;mw@;%%`9Ybo?M-=XQC4+9;;DUYSyib(M#iPEZ5dYVYf#1sB
zlHch55IxoT^Fs&jrS*|djgsx<JzCiYb9Z-cXkz)dDedLS12s_8Aa^*@@bjkv7MBfr
zfV==f@DVTq<&~A4y}if)yZoIf=;S+U>I6+HaR~{7sUr2>O%bRhy3Z(-nB&{Zie3-~
zNl<b!i9<c9prByO?IGVQDk@1Oi<h7Qy|ZKHvOn|Ge=SfcCI+uSLqq$eRM5dt+SKus
zjsO_C-b(H`3ikkHzxe0RD-g&g86b^u^YBo{|FL&)_zAMqQy5uja<5d{Btpl_(_-H4
z#1ec>N%@7ta_abIuhh`iHX<_eg^rF62wcHTnl*MuOZ-jE&CcIsM|A6Hu{YM&JJ)`H
zFHn)jt$q;@KiE0nUdXW*ER}o9k!LXoXjeBtpW5a)#|s3ZLb7h(hA5W7Sf^jE%-B~|
zZ3Gox<4bODGlrwEB*Quw=O3zvFGs{LWs2}dx-cg*;h9qwv{wQo5NE^DcjTUx<T~CQ
zPrb5?ZJ~WfAFD*79^|ywT0$zFk;#-IV{$up-Nb5V5ILBPbkW^ei7&Yw7HvWE=%INN
z`KL_*)BIP{C+MdXP1Zr-(Wq;e6x|E8(57i&60t=Z2T$4&l$zgmHzO(@O?;x4K4;ne
z$_@`#GIE4Md)RSSbYBbUCdz5f6-ZYf(>n5ePG2S0dgcf!|C%%ez)Xv`QukMJ<BH0c
zZO{Uj3D%()G2L;pB1@6PQ>mT{yY~6bqjJX<az7vU6Vk>*DSVYPon0+V++OVS&aY1B
zgBYUM2^pBme7pF7_TO?D7#oZ7^@V|jg(c*1$QHW2I0r^seh7?eWgoT$q@<*F{zpDf
zGT>WUS|VbS_O^Y2lWdZSrB@!YkW|?dU*;Yh7;xJwu65gET0}HKF_tWaKLUXw&5aV8
zEY%gX#XSOS30kwotLQ!u?Au01Uw}9A?f@B$$7TJmJB*h1-0oyVvg|F=J;EMY^s&_f
zx~JWtm-YEx?;x(0R6X=21xzeR4W<n;2`3U2bbVU-K_l(RIayj6RAWc`g+$dKP3o9<
ze}+GQ(t~~9&*%RA8=FsBF`{a!Q00Vo!1>1(1wXELSD(vKbbR}DoD)vQ1wUyC4EEh8
z*V$QMp_E^KmX+?9DtI+yqijC$lVOk#epM&OB&5^rOk>3U_2tLvqPB^zB>KY_wiW~T
z$w03ztZiA~6(Hbcy`n(RRf<TEyn=B2YxajZ+6A`DF8rHI+gBTB0&9jxkx*iqUvV^F
z3AxTy^^WK%rcj;7j)2DRq0v}Sv}}hx!{M%R<=>8Y{3x}7@kFpz*?08K`rkYwGn*;_
ztQW_mJKs~-2Alf3WT`DX0$x$vpG|j^rJM8LJFr5}zs2EFAo%8Z)0xIRV&S0i;pWD<
z-=hqtchw>NzGFz(S%wm$QMxvM;8NtMu8B~H2Q5+1o}H%>y6O_Eo0Xa_mY|B*Zi_e`
ziRM?W82LJ$9nVn59L^}0jtBg8P8vC)>Xb4wDZ7k}L?;dH`zGO?OCUjh5Rpb8{_-f{
zb==xHQ*_^P!SegJltl^!a`dsF*G=N(z+T%wxy!%f=z(>!kOAGJ{_pd@74VNYtMk6C
z$!x@hUqvt}d)uo~d`9)Y_CyCQh&wD|ajC?SfEZ)v^u{ac)(~7%zrkH6rgl0-D5hOt
zck;LYWn@Xu<|q%_Czi+E%9L@7cfFx!Nz7SY^P}6ODF3_0a9TS~lZ}l{(4oTf0D=VI
z0`>_C_z~$wz`lpddzm#LA(}JLHCr9@EDI-%#xoD;QE<)i?1)h(v2_?5jI3{7EuE+E
zRK@P84o9av$6zCJ!DM!u2JTA9<RCi))Qj}>qw>#D7#DOx9sY(2GY{O{^^r-qNaFP?
z89lU1bz>B@+lh^)Oy-nBFs1duJwyLy$sxYr9$nWokI;rmGfGwhG=_PFuqA2f@a>ty
z>J*+6E>K|u9lJ=0oZj}1#Gr??`9cF2gShMfnwn2Q;Owo`qPy&B8}MXeF+_3uly$^q
zrMnh7h~G?0OxTis9=~YLAI}e43YppHA`Cy&=$xgRuk*gwRfsZ>r<Jm^gB-F<?oHmA
zpAtW__Wa4h+1Bd<faDXFETAQbE<*7;KKZ3GpN{s%7020GFkG2<cVYOR;TGmws=f|w
z>&cL|9laCt;wO>oh2fwXQb4!?QH73`W3gywxU*oT%Ri^FsC5c{{VbTO*Ii3glo#ZA
zSl0TpcPLF9MyT4JSZ^U0qf{{YZWH@UJlgH|hVOUs+5}x-4)7;$<?^d5lc0@AKwBWL
zS!s`vemk846lk?Avaa0xOjK0V%*I4mf_dg|Bx&nQvPNDXYjoUHn<mR^d%asWy0mVW
z23Oo9ObLvobIA)0xlQ_O%h}qy^X*zt)rw>F!)O}5qvx=fuZ2RsHw=z0TB;vvB(`*#
zjx_8w{-$BjUTiKatDru$AT}yEFYMHm(Om{L1Y`0ek(f(~$Gh6VHKCBuQ1rtk<jun{
zxth{c*@K{B!gpIpsSne1jupVp0==f>=9#fe?%9gJV@#pK3&%$q)=OIlw{3c@h31{d
ze^yCI7n}>>sL=mbrg856(JeN-5UudK%Hp#+7Xu=VV$p4%60%#xzopKu{$14Xy;}x(
zsZ<F&LtdK)d1xPmz)4(afJJFMRhP#2V1sK;_itnf#pJ4OSz6D(_3YFY=V}gVjyoy|
zx5swbr)yt!*)67vzN2%R%X9s^jam#YUW!EbOc#W^>>G~2`46f~U9@ko6+_$iwgdQS
zq&a>2YozcO%}wI5mZR}?(%%d@8<|p~(@UKg`031d6fK8XuYDCRTMMw3#U7~LC~{tR
zvl*+s1VA~lXy0k)FS?^e=LL+~Hd6APzN2A+0$kF}3kvUGcg36Xl|9N0`tRiYOsa(A
z`H;J+ocW5L-{GyJAPIAx;ogeWIj^+SQBPY*5UHt^t{~=PC3bNb>{228_rH=oHenQ2
z*02mD?P6&w`Z8w6FKEy$h8|ku<1F~V1M(8c<{)`U_@WStv8b(QlX|yp7A*+iKJhFz
zj=Mjg1IpcodO!-e<HrX}JisAaJ_`~xtJVN)D>?bi;4>=Z&9HW{<BPp%tk_>p6!6Au
zM4)_e^Ym0K)!_%(udH(1a3bU-@^IK+XSjT@Ei5xl>+Kz=mj-WQ8zK_EQY~Q224z(u
zfvprL6w@)Tr*Ho8R(08Mc<Xl=J+?p*@T|mk)K`4x5Xh>%M5tk;s9XX?C1y1`Uf=2G
zo6%5n{!$m19hpE*nh_Ti!du<519*3!h_&>I#mLTfqU60K?*~eaOm&Dt#~zW7O!R9>
z6Vj{&tS97~7^wV3`cAXOuToRlV0hW@ly0QA=4_rP3TBl^sX!;47eh1N=YCr$9jnz~
zsEaFSS}PA#FQ-!s$`U;ClaK`W9F<>8_PeBr?43qSM}5}3z9c&Qu3Y$v4F3MN&V{ic
zm(sJh73NPaSwRR6+c&Lax7(+GSL8ojBcSnAIsNcq&h8jMzOcDY-7j+L9#GTX_`IaL
z^zb#Sak}lB=a>i~J6ZReJ9~mE4Zy~q6iiezh3u&g{o91fKE&|Mj%lNH{`%arYtNS5
zx7yJ7fy;CTH_YN6*_JE`@`u|)xNnOx9v0Zqf>sQR(sF-LX;hG=VGn7`6h2sr?QT1<
z4YYJ$RG;<@_3L$Q{QfCg`d>m6<}m@pI;bQ21_vh_)0dh(Pse2iLy}qmiwCgm*`paJ
z_$rWt!6q_k;g!B>xl73Outdn#%1>7qH`5(e>IQ0E+Sa8TJ$lgR7p|v6{X74G{WhvJ
zrSMB>2c@KRe685Z+M%-1Y0hc4c%SxLH1jH8IQu|7RThxG*Ip;boXX!;kuG4kl29MW
zS*li#NTy5CQ4)r>x`%v~8|Sp=EKVjJFbIt%sH>f9t@H(QDiS6iAATjeJO15xeBgXl
z&&(bB)5pY!bD$CB!(&7X9T!SjUZ#7`OD&&^*P#15(118@=_gM&G@8aegBL7CV>mFN
zY)oKAcSgeVePx|*l(8PoxhTV>@2@D_kAk=*F=D1>B$=+csq)?}L0OX5)UG5mZB9^n
z6t%9m^-8?Qai4Mcq0aZ(2nzn&h__YsviDo?RR?_q{>l1799KVsVy+kqS`An1U_qI4
zw!_|l%Mh0eT6!j72Xxe~x1Mc{<iC%iwP>jK8Is%MSMO}Mf|UwXN_!nJfKZ`c_bf81
z@tMmQt;ITaN*0X^ODl6|%Royp1j!ZG2x@y^F)Ygb9}tiYYj16@HXHQ=YU7ooqx2(t
z92^|b9RCc86jt^>f)#*i!qU~%4Gp6RvzjZ@hg1Hgv<f(-AWV6%rE>v^cSe53Q$K6I
ze<$#xj^vGGm5zlJ_v3)Mq=~}o;>I(&(a$uQUXD&KrV|r6V-V9R`>T}?D7MM6itPmY
z&y#{c(zyvs0ne3h`q8+jAwB1XrS2r7y2B-9GSZ$J{gvln#HS_to5u-X*_JFI|9BpS
zs@U<eIl#4k|9d=kcw4^tpYN%Jzm&IKk@s1jW5=>E<Jy?UTJfzCu?XOfDFbf3$cgVx
z(&VNCZY)3nGL1TA#*XO?S4;4O^K}g#Zkt?eM9*TZr=R@GYPOH1pzbfOZ^v10#L&%<
zGu@HO=#}iUR_(L?&B|wYz$g4@9%d+)rCxMbbV`NgOR0sgaCg#ju6Q&tzW!CV2)Akx
zsu;Gu3j9V)#$5pZ3@(D;vTzB@*7)pR`Ze-iFy|dB6df?*Ic7E+_1cw!Ko85GA*)~c
zq-S9?;_n6i%D`$dJjF7~p~+b9;?oO$nIY;<!6b&_wR(a5s6gk3`-BvPo~P0Hzj4DN
z<!)>7zO{3RqH#AL{fPiHaZlp8hdcM7bU|4Y6B>Pe{VMBud>B4HKA9A5AsHFW__w;y
zkR;~z*49{wHeZYc+&~H=Xo5_EVuiQqLO`8vbU$G+Nzqb(k|Ag1?Op$X5%b+Wo@VO_
z6*X@t;Of84i|M`uV4T%#0IS!bTEXvh4|zB4pWlYXO1j$BOL#xU{0-&ckY(HMiGiQJ
z)?f-GC<b?TotZjM;?E@R(yy;7S&;rc1^YKE3lp2ZZO7U*ok=l?D(~m8jRrY0Xbya6
zxxt<Lmc0`!8{^mStS>#vDf^7r1X7L#K|_pkl7#HKREamE%%i)n{&?P+(3@y2wrY1O
z{()kc*@f@zPgEP_gu8Xjf`#sXudi4>Nt6|$il=A~`}XOx93kdcR~@nb;DQikyc&cE
zvgVVib~*H5jq<Z!A1#oK89@krQ*GU}=B-q0`Rc7ihNdM#%x`Zxdghi}qVN%wJy_`U
z>D)Ct3Yj;IjLVFTSE*K?S?x2hn5QXy3-(80c?XaA#<#)OF&m*{On(he&5b0mBSE%B
zLzHPbE+?WK9T+%OcrOm?89?wK3uBn;Z%6h@<ivmVd0yXT3MX%-W?`YyKFVQN7iTIW
zu#H9Kd^h+XP7$tuJ&x}@tBYh&odm{o8~{<6#9`STNhS8<58A_rf>ASMbo33-3z;a@
zO$W`+OgUO0P0s=9^MJs>XJtTMsRK1%y{(n?S1HT^5>+qgW;d@1Py)leq^Z^dmqk>x
zf*>2KtE&ScZH63SlSPOeDhMECmfMwg+aX}k+)pZ~Qk9*QezIra7c5Vr_-iq3N}B!W
z#9^rQT0+AQ>$~h$yrQBW14R$M3Izj&lr|O?8avC(tGnEuMkIo=6jBlfp=TeI(T>Yu
zkqqf5A-q_o$rCS@xeDXFT0&0q7x;0UA{WVp1G%~xI6O&=YDcA-{IWl4Zo7FH2a2v8
zWsBayRv`lg`yHEi*}JtDZCaQZ)@_qg-vD?o$=(cWs&;hR%zFDsd;S$5xK6Mc%kBFZ
zCw^(|)*LoW$^F(ME=c+1mz-$FLsq2`Dm?)0u%9vL>~S6Dw0_2R=JE?Us`UWv(wL7Q
zHSJhqqwGT@>QWFdE1t1L{UJK;b9I8YqfH5&M==l8hSqo}A|vBo@yjhL+?2D&^|O8w
zaHH6?_|a$N^!J?dC?YkuTzr&3Kmps3Xl53#xEWUEb~i2?o^E&HEukuwE&d&z{Jtf`
z0$L;Px@VhfyUkBc;9c(x1Ntx-X}OM8^Dz<3w%?22N5@s=Dt*|=;09!K6f%|QS-H5T
zukW8sf3~D|ky07j)c8fcZNGm>w;4Maj6}b7dccwGS!(?q%>O<ovQE;=Sx~IDQm<F!
zz2y~~t8&@lKY?IBYXX8BEfp7h4Jho4qrc<l&i8MAKDi6DkJPtpl@%<UoZV_(@ot%Y
ze9a0Qr-_M)MkXe(de+uq1qGP?$o;)Q$HwD*^Nxhe#%n4Q*-uUs{ei*^=!`&H4c`~^
z)v9~pLjXgBM!?+F(-RDIHGstWxZL8+%E@U0hzoc`MA$Nq^S=(!D-a%h=6<gcn~wXf
zp@x%JG~nu2XVRR3g{;i}`TF&o`s@tONX;8t^gGg8iODqiDq;0g+jDv}E?q<52gO}>
zAbYrs8myUkJqHF^N6XF`&WKquyHHZT{xR+C>;Lgr{%_I(0-ex&Thf3domQ(u$;^SK
z+AN#Tg-40b`|eyp$kcIlgWMrP(EU%Pr!vJ{>kWBn)4hE;RV3OgcDA)tth~{dTNqB}
z#NU97_}rKPL!fwCUFm3-=^p&j10%sVtJ1Te_VkrQp;Bs2W7cfL2@)rBs)&U`iW1Vb
zAXr~!GezuKxlF%1etF5F&ev!P0*u(dwdwdI9<YB=O9{QA(j{R>W4N1d?k9nJ^E3pD
zeSUjX?>|O+1>?Dh6a13egrDhai+gHa{=5QF&?_*dalHAqvEI&rKvTtQxOE%zr@+Yu
z&q>j$WKWv^?$yl?gSRd3ttm|}#L=GR<g@Dh$h|lo9~A35?5G)68g^8gCDoFyC-V6M
zCQz)v`3q+|QvcPPLbPazvd1M_^=!7YPr}cO$zS{uLq%HkQ6Q<nj%15RH}yo2^INnb
zBM8wR?UQP|L;LR$!_xi~TT*r{wTe8+mN<cINvY#co_u*NCfZv>@U;NPkh0`2@aC@!
zJg}3Fj(HxISWkT+gIAz3-#Q^aa+Lrzj`;;YquD4l!rWHK)<~k>7|%dRA6}GjQ9+z8
zO`)QD5CFnm@TKL-C8YhetLzGP`i{lFU)goQ0+Jes!aKg|f3l_$HN4@!7&^cDCM2?Y
z8nbr)9hw1SbnXgEjL6zfARFuYy%>EXJ*Swbzh_EjP@pzDGY%-fC5TMGzme->NWH6n
zOtu;$qqUu#OV4~y>CB79N8^#p$Y7#{f!CB5I)It_!HmdhwG<dM*!Uwc?PSFpSHz|S
z8|e+#`z8y;_UeiU``E}PLsju%!)+s2lM~^`hxfc<CI|w$25i}CS&pB-How7c`d<{^
z{L#+-8VQTsYr-8uv5|N83OHKd)6me^V(RJYB7DsWGWuIKDk`vH*VH>N)q`Kfz}dtW
zuB+5HEZrMr^C#H#_rpBGQk-S}+4u-dO()EiE<%Js_yyczfl8x&qV1lD;n7AngDoT_
z6Ry;#k}0QWpwJ@f9|syoSrw{e^q!NWjiq8oRB$m+Mfuddw~@ZFWOMhydBiGRKc7n+
zBDaf*L&f^Ie4(<e*42O%Bp27|3+<g45d05s^y2}%5f&D98mFkJSYBPtRN*b!oiz_n
zp?hi1$tg0e_=HOWqAIFOUovHH{9RVBm<TTRuYB-E5hjup58b>AN&U3kC?oOnogXwk
z`rF^*;Q#=IJS7g}Kb^Oqv*dB`9sJ6E;Jof$7R%}Xy)+uzzvOH_k_`tM<2OAjh*so+
zE;1fkEF-Xe_24-YxT#7meV#Y!OwDSVLHG|41vva(L0JOo*-D>#4;W6{r2)^QW>uqG
zFoUtFd6`*MB*vaQ{Y&{zn>_S{dfze-xV|-C)es%UkX&n+2nJQy{VjF?n*-AsIs<_h
zm^AgIwX)F$ycr7dyAaOBn>e$v{-(>(^MRiv_+q#D5m7OwFhC%)eKdJ`<#`^FP2EE<
z1B&|?mDJ_c1FT$y_Tq0oti%a>>pesB6+7J`^~n=yIml010_f?K0|a`ep}@O)))}5L
zvsF<VQPC3z?Hl3$(Llx~$7`N|lONP1XEQ<uh>f|*+Rcp{4iUZg)*Fc7-2VV+Q@^Io
zcy5@G;nRKB(MjX*&WZGkA3rCVjmroq+~rov{~T*^j!uyAk`QFYwaX(xeXmCvE$(`f
zVyGFmfv5A04ofv+=w%R4dRU83q$2wy@%o1NYy5alqHeY9d_N}<9iJ*5L>hJ|wAAaH
zK-%O*tiQdYV0HTSPae_ee6aFkvXq7e?A2V`ZS3mlU!Wxj%}2E+?NiR5dAw8zYL$=X
zPbSzS*Chhc@OzcreqZ0J#Af@A4^smkbmipl8?KOgpeUX}&t0zjDs%Tolq5h&37Tqq
zOA-H7l74a~=mLJVKwj;%I~JJA=NgrqEV<a=ygOY=0TjMD4x9ZTgB&h3|M{g{s8(Sl
zzU1-c%NHjw33+r`tXaFo)cSC<7lf$@Zm63$5mBG{>^~;J8DUu-cFI9yVwc7BSQ=*F
z|FNnk${prA#14?&61VrbN}MYqsrZ!)Wzc4zclt_zIs?rX=i!L1`_gt``a7mVxkH%n
zEmvjPPw_o+(&gG`HKHHoL>6JSsi~tiojn?7a<{_|R)By<j?OWoo*%tL;TYbav|hLY
zb!t(pn)w@oE*9m*10=mfCh)PA5Ti)w15(jJ%4a?zEH1Y8LMI#%*dIAT(xWg{J69%O
zq(P`Yww@FBz_`OfTl_AYL9nyR^H=MR2XI!;;Xf)IEj<){q~&nWQ%!S5?vU(XLD2!N
zY><gXE)N%H!E6f*7)><2KG_T`8ULls@bGYJ8%lobw@?P@e|(yC@ojBw&FuVd`54%u
zr}U;VwPr04V-%BRxX_8^+$ST>8?b<HI_-;MjFWw=fLg;bU^afBsPf$F?zvk+=QKyA
z!LyK&p2^xT?#?U0B(D#?8}!d`Hnr-%GT{t-LqQT+*aK#nM@vEjaxdiinw*5lB?G8J
z)RfDhoHad3zd^L+%et?(cLMggF$-i0L>{UEKP3<~>Gb4`&5V$A11nb}5@{_LdM?B8
zsZp)_F=}BcR0}3)9UWL|S;SO~U_=@(FZM#UuSo9f;95#CXmaX4XCwGDW6;$%P1bNw
z9UOh=NQF&qXdX|@?QsR&`;i$;Eon)DofbEUJ6TFum0+vJsq^*f_3h8%h(i>FR!o?R
zipo)o?9n}ZeBRf){|hSU`BFio01K6SUfNvj&-(u-LIP_0>+N)i+iB<d*4A|^uG8T{
zVn)VGKR>?{YM=jpPxnH`=In8YWVdG1eATWJmxI34^z&YRH98@Om{b`=6T>$hp|iE{
z_oJRAdZpJl*+T&%gV(XvW>#VygJu`dqiW<JgHqw=?^q4Oxx_TVLQXP${^;U?v(wLZ
zIM2012hq~4qzXqXHY=%V*&SE3UCpcP#2pJ;I9cj>3HxSPovF>s`=~+5f#F?E5@tE&
zTjrGDjS)l5=$MJ1BGv(gtEE_3Uzr?dyh^^E&L8TXZupplN~0U-X58ohJ~`Ct5F^xx
z4d3e>EKXp1PvBFh1N<N=m?SDDN{Is9CAa4VV?jZIR<lQJXE2uI)v-Rn5W*4?sG$7}
zwrh@3CFfHU%a78@)A2Q+U|_}TS^lyC?-Cju`cMrth|gd(TJCdSj^&6QZ!2rn@rB<p
zqgnNSa;3NixjLSf1dqY^XT)?5|G<<}ORQ+8HzmDW2LkS4Nn~EDKAfRV@M73GuZ{E_
zILZzMyj&0}CaXvmanfZlzAtpgE^gODyZB0#0S~bfwSq7h)~n8{O0)H(X|2WJFq|(1
z<*X#F&>gJn8?oJlFpSuK4qR#>l*(6nyV_FHp|jkZX!Oc<^e&o?0&eMTK`nNClDuMX
zqQ41<@er%oW6=a=4o<RcceXp7*eW%U{r^h*SDpji(%f7UsLB*7O$PJId7csyZ5*J7
zX1W4FB<`IP8P_|+P@_QjA~TjR2))6fL%}W$Kr1MFi+N}aH)1;D<bjoRX65?fIqGxo
zbMnzl*S2w34ZgoqA;qj`P9l}`Q9qd8n4ijD3Mcn(h{6w2^beIz+qrBCC6F^y8_e@a
z7*mi+XGflt^Vl(qG;9iTB2m@9(9S(?Z<U=l=&UbW&953+&f1`IecU5+y@L3HBsFEn
zj8Eo!rPJ=sCccGZrd;hJc}S-Y&6Xh(;=>N^X9mK#Y5JwI8s)Fp)WpQ_KgZ%BULYmA
zvixgwXQ}jd)Fl)TwMOPKYE|&QBfasmBZej4oH{j&cWR8Me^OWqGetUXNZt?IFwjpV
zOAr{pTo`-3eu`S)$D1Fzm92eR9><}*p`mWK{sr&M#=aN}dUsF$SXU_S7oIq=7JrN|
zT~}<qRt!~mF2TI^;cL}MxBR8iPUpF*2zJY(K_rR}3{_x?jpA~SXd~kOtqm2c>QQ@j
zsPMe1t!Bi9q37z_y1W%(y}E=Ac`GWu!*MOcxy?qo{D+U4F?mo_oii``QJk~W(H~!W
zUS2!1!&=6L8h@g~sUL3dQNrKg%}jmD)N?B(mvXz623tj<noQBq5I=%QB*H{Gv1&uM
zBxo_V|0C{C)9)ry6eH;xvCf37O#j;_@10+@Gl0J~ompJh7;PPi)#KSmvIjqIbx)qY
z?vh~d;p2rG*R}sdbQd=$=6HC^t9+)NUVC*YiI$Vfecbv>!?!-=YEM36QW!<&a&m6s
zccU`dp>Bi7Obc39kVq7Hx}O$h&o4iw!xo(hbSvMiJj;;I&kZEOys8M!)<T)vo~7X&
zDH5+9Je=yqkNsl2$j)yp(RZf;@7oEkx6g=dK8fK#@loGWbEwH>WZ*eZax4uW72dc|
zyov5}D|T1Zv7QT1R4u7E*f3{aZu#h>fSudjUB@#sUgW~mz=hT}Uwqv6`+rX|;6&`T
z+f%rEMt8lc-bpY_+yb+)@fk-nF)6MhsNm)MkGua0#{B%teaQN<1std<<xRHq4Ad(^
zpgAK^wgnQmwWT)2)&oJkvU~TUh_gD}`&)WA{M_88xhj-WPHTnS`A8C?(Iy$|r0D|f
z-1!&(yJ}h*o8<AYNefK!_q{dN^W4zGdwF?zvucsSotGhBztR@|czRu?mi(4cBrb)B
zj0+9u7y?@$P}$wx<pF`w;Q~}AWi?@)!r-AW+lWa+LoGQID+c(EfiqlCJ%FOW+|{9T
zL3sE@;alXuUH=FdiLi%M{+)I8sY2xh!tL)OZ?SIV5{O@_GyY#>-29yAWq^f+l?R~a
zClQfYk<0&9_kx6^xZZSk;FBB}G%Zf~O|7>Bpzk(S(4K%VfWK#RGA<^(twZGP*M@wb
zrNf~$cTFUMq?*0HjPld8WTT@4%lMgQ^B1P!chXuD2VTSVss-Zfocuc{-EyFzl_noB
zoK}76&s#mpGiGGkpkVP5^fEmeHoAwI+bs_a7??h7?<}i_PKYhB{)ZR<JSPZf7c9?%
zva?6><*C@%$fZkGx3`1S)6?q)m7kE3xj6@MO;O69hTr3T{cig;j$G_49DZF|V~4Np
zK#LF4N>>yhH}qStOqss^Nwij@^{B-BN&4MqXgo@6NDMsLzgBcO=|Wid=dKuTYnU^S
zz9v3#ug@;0%ElC<ukIV=NpDPlLs^Yq-Ve0(!N2>Ug$XLPuUeeJt*V3I7&SY<0tbc^
zT(gJWs^@R0hOL~7ty-~FIgK=Dp%r9IDfE>lt@@3BNo~YY!Lt%?jRVDiUS7}6JbNFE
z(#a{OoYT{d2S0KGgzwvL|GaJ8pmuF;YT|a;CoCzUPM7-l6oGW0cc8o+NFIxbzUBH7
ziQd5T5li*B>W4+DKun3Bp?KK|If3kfzLp>a?8-tKe5@s_$ihTBt0Gg8d75QC+J^C$
zijdE(PGsd`3%{%=so{kDU49ogL?Skm6Ya(8+1!3HD+jcTPdeUDi(xP%2u;c_=J>v3
zi9ZC%^OP~*pF_E3HN_8+g#;0&*9%NDe9c<XBFeU!6p4}O2dke5wTOzG?(%E-5q*=s
zV4xFf@AicnO`xA`E+l<CmSt>OFVPfJ9Rnr9MJ$6{`|YNO^#3hG(9I{f(QI_FUOEA4
zp-&E~PfxS}JW-#NF<ptnhJ+43kK+&mw|fYj0i(cH;)rn7N#M~Cm6p_XCcjvR=TrV^
zK8e32K4yV~!%uutUxmrcELpG~4d^SS_g8LtS*Fl%*<|+T+un{Nk(lH?*>DE|)sjTt
zA`G9}Uo;`Sbg4twy2K7&iDNN1pzzYzA<Bgo!oUF#Z{}Ot7|DvwM+yU-%Jr@x{zcyU
z1k^WCW%kv1hlS6AFyxS=k(@u!aQRhVQVOZxU`}YLy^cLIW6ehEeYow$dD_vdB5?_E
z-BH?}ZjLvMUcuqnYnt8#ta>cgJwDwrIe&O4><`p~X^}nH-Tb)TzkhH0<h~JfsPpu|
z(-<^WhBkpAje>)#k2YO7q)8fx9;6!RLrz;+nZQx<1}q1Wk_yXpVHt`!*lxB2ZL~RV
z{8loeky)aaUnrt%7dH3W5})GXK#-`9zIjgJs5%#K&LPQ3JPBOE3Nf6EpSTqfU}@uB
zwkb6)&(Bi+&cF5NYFqSq{CC`FI9*Acr5cJ`UerocLCVCeJc!}psOpOphZII|ih@*K
zE`m=ZT>Q9Q$!fT*+#j~@Pc_T)fD>v4u|8D>mDNexmmjwtdULaSt}eHj))W=$Uf>Om
zLT}j}&oz=l?*f={-rzW_Z&hNdKX)=YhA({@hN#eD=5qSKC<G#l>Ehb1f_f16$hv}6
zrAQF`%nQ#){Q2oSB{O!gvd1WTBwNu-QGvl}Eg8gTepZN&KSZq2;QXieNx9%;(yGMg
zmQw44F4^CO49x9nx=Cj)-f(sJ{b3$)=<S|N(-B0E6#eDnNX9859x}A1B_1pvPf1PC
zT01ZISe<K3ZYl1Z@;ty3P8L}*xICT`YJsP6k0l1p7fum<6U+8wv>?ORTvuYc;4uq5
z@6T+s|5ruMo>m%H`if6XytSeFYA@_?frKaF%%18qDSv1s-0iV*@|IF{wf@nyUXstd
zBxBDD{=Jd(U!Z-Z2c~bqnTM5_A|ui5f}6nGiqDpZiD__T@qfB1!^>PqHwR_1L5X&@
zckwam8HLI-Dk6v}*<T}8Q?C-5dK3mxS*hB$WT*k*Hi&^67!cvAGK>DRxpAziI`*M1
z*!W~VOWfJ0&^*=pa5v=}$xmTXRVX_=3m3@{s<&!Qc>Xp;0&x;zYt~p1rSSq}Q$2c!
z#NXDXP-ev?X4M>YTqw7@>3&bl?aD*Q4H3No`Ol${)lhU8`$QQ7=29ziqd~b-Mdt9D
zJh#&5zh=s(-QD3xO+%MBmDTgY>wtG~VX^MYN7rX>>PyLzvV8?h&u5y>4mF`g)bLQW
zV=5y%-4>g1(pZhbj3b|w>4V4jd)jUm!0D&H9s%kBEE6J>Ze$Hu1h3OUE|~qf>hGX-
z`Dpte^^k^^hiMas@3A8vvD8WGW&K$&y*KuI4{D0tLEExqPM~UvB-hxveYdE>E00<J
zhuJdU+yl>D(ux*9c%z{4Fi*}|Z)JFj0e<=?UkYChoN@_vp#8IIPhGupUE3i4Jhb^1
zHPL-&@^gY09^=MXAPk>juM*JlcZp}|4_YyraW%XNTej9LiE6}B=hYa8esf=V*M_h&
zU}mV>W2r1ei@G-^NNkE@$0&5%2)$|q{whs3&Vp78XKrNw$&HQ|o(_q6ut0QOUvqB`
z;*xB1?XF5_{3E^L(2Z~<zdReFLsH8M{f!TtC{Rd@bKZNOlaC+-AyTw!7AR12+Qzkv
z_#C$*b^ZQL<bm^FA~^)ok~(}3eA3W%#o*D#2ulN`4`#=au<0$Xd{P=_R(Z=V<VDbb
z`nH|j{@;c`*-+pVAmA?Q$TLw(!$1`~Kg~?U*s>iGTuMM;lPFwT?_U5d)d-*hm1J)F
zzfW~PQ0*la{;Fgy?xcA&LHJ>}H4O26i@~_L-A-v|_*5>kdAn`lGpfH}Tq{?s*it_G
z9itC04i;Q$DBhG3ilr}$g=~~OCB3FkaR;+KAtMY|4N5B!s(g6K3*K69hUo|PMi9YT
z=xJsAl*J?-#vnvX0Rd)(i=O`5lak!98y)=dX|q4}X}%p$C^K_&uNbx@{gacEGe17z
zhJZX~Ispj{JC5<U;wUQ?s2>a#?qSo12K(Y_Rt<hFyDSU<8mxb$l!Bj{e?gXu{}Snh
zjPFRp=jQT~G^#Hwf1cmBVG&M_K{@+3O?2dspn_P4XllqMK_gjk@44h)ivj|#_XABT
z4Ps<u0$}*qEOLk$Of`!bW?tb`yF9_m)tsy;vxF7{JQzV2tMBzc&*(}oz4uPGbL55=
z>iV_}dv+%)0y8@q<1~s*K);$%y&?b@SVTuh``6ZTX*W3Oue5!c04&jnGmyCI;xnp2
zZ9}PiA>Y1r%+Du-wJ&h9Lq$qTu6-upxP{K7QPs;ZymELL9yyvdEDPPA6O=ZcNGu(J
z*bjz5D(zN<!zl#AHHveA5db<7Tfmdalcs9AH&8rmf*zdJOxYJWB+TbTL?OVd1l-Rj
z{VpN~@s5*}E+Qb?RlqufbnnCW_pQKnv%qoAc&cD713<G6GA_AQdM3V~q6^-ge7G(f
zejKsj$Y2?3e*baFC0mVyxZ3A_r?0GMh!#qJ%Tpf$%ZK%5|9wiDGi?xMwEnJHSdJmx
z8~p!j($zpm8pae5M&DiLfBqn{D3RUg%lgcxsnQ*Z1e3xW`ir7dtpU6VtN}1mYw3Bb
z-OvThCKA8`0M-%LQPnRu$ABS0&qBT9DZkJ4i7~To%K(^`<N+}1;o-r3m_Myiv2Zef
z@~_uDf3_0+lS2q#wg;vG@6<Tk?8yaQM9S-gjQ0J`mFaKATj0rK{BU=DJ-1x%7z_q-
zp)#XBY?HyH-vKB%Y1}r#L@{Phfgk?aJ}HG)zLXpwkAJd3>5r7Egda-~*_5u+dq>+q
zJNKGK3dMgke)SKFnjPBEYk!64R_~}Y%HgiABBfRHcNQgD-2Er%Z%z*KniJn;bBqff
zb_1y52%2W#2uaDw^x~}VRX_B6Q6LKJ9D#lE3%dUSr=}k+>3mpig_G~k|K=G^_9ygi
zy5I9ldtU6X&XgNYfC=M~5xE!%YT~dU@YI0)5D}k?)fd<9;dH_DddF=AfViG|>)>Qd
z@QRoiT#$qU6biDz{mstHgBz){Hw30_Ivtk}_qS%%b4}^0kghjJ>Cp=+QM=<6AuVlf
z9^e#FUtbU0=a{wXeDm_&0OPD2*>nNKSpqNzJyoJD>E!edyukgo%>C4ak%=jh+y2$+
z>T1yYjkoCj5i9q$Cja3a9nq<paJ6}!NEfbWW6F9tJ*^82V@Rx*zc|nqRe%T3L`ivS
z%*-M#gs^w`QIs;vo=eQX7<S1%_>g(*wPmLHhrZ}cjj$kal4ZqTC_1FYwYiPcv-U2$
zV<uq87M<E`<Ta2pe?O>=+V<D1F3FCi4ex;47%eYUI4=JWq5)`)?rLE0EPs6?UsX_r
zrdz3|d`m-t&0JkwEm^SHK?S&!Jkwk4fC8`k2@;Z$6o3a4kg_+vnF146reG+Oo0s>g
zyFEbheHskTP<{-`P$?whv?2ulcx(Px0==FO`sRTh5{zA4U1e_n)|zGj&^rPCkr~gx
z9s4~yoA{(lKbYeHn>}~BC6t{1CZxy8%F6Aadin|wc26@|Pg53<-C}$yP^E^EaB(`t
zKZW_@1u#5yRlCy-W@%T{u$@XWG{x$P&4n4;N=0g6liwll^tOY4v4d557cIKG)E6be
zS>@zwav}E{wrJx=F6b>QNaaO?<Ud1#HEtoX2PZw>tYf97o@hRB`}1ucmTfF~z9B|2
zX+|#I49Q0*j`(o9t>Cp%F!pTA)|;m@NdtJ2^Ec-!dRv;k@5&tQD6btDyBOO0vT>32
zSyy+TC;J11%zo;pFz7#s6hNSEkmRJl+}`BmnuXWZzJMO}l9J=RfX0ntND>blTNvCh
ziK-zGh(^rbd0ot!UPm=Xc&>+W5rUhkSqs?=#;-69fgT@>8yE>V1!GqZc6KZ45rSL6
z6h3;uiO}orUysLuq<qplbJ(X~DdJZP^YfGcQ=W;5bXi3DjK*;EMWToJbl#AXH6Ex!
z@HKv}7Gvf5N+Ei#>2FZe)zA0h<4*_Eaby@G<m=N>LzWss1f~f(I1j267LX>evjQ|z
zPxKEWqS4C7JdvQ7)Wo4|O6%+m8xeK@zkTxInisM^*^00q*Ti7AtT(O7>h36g0cVEL
zU|Gu^|4S9Qc+1ZA^Puovs4x)k(uvOo29nsTf!xTS;>c4pjq|U@g9%Ze^|(AQ!xAGQ
zX}+Zo1EE^q%4(n&illl9StOiRD<EK*y-O0FcIQ@b0geGc0D*gT$}&q#Osw(CO;}hM
z2FUC!Gp&I)4@?`_>t{>x!K~pVT;U+1>lquz)?&rQ#lZj`v|8!*xZ*jA&ePRtslZsG
zJQ$<=wB7b|KIdq~1^><-$+v~d6n}1Daxr-h!qEI8$`EU`a|p}s)s~eoil~Fnl80aA
z>ZerTV>`}SYG|l>1+N7%4n<x|&tRDzT_RsA^Sdf04?fIpg`ci&%0Sz&CFK>}JFhQN
zGPc(|41}{^AcIvUh0%KgBTe)a>F(KLdNNN;y8n&PJ^{#_Z-q)G`^$Ck%{P#^B7J>o
z)VJo?Nky_|w6U)<U@CeH%mla`ElC4|AYhU*R47P7LUP;13+%XH%-#`$18~vq+_}HZ
zt>YBLeiHU`VoXgzee^lDi>kW1a%;Xc1~b_c1&YB(t&dNZ8f<KAXG_8Hf$fs-a9;!W
zA~3?%M0C^C)Rfk30`S)JfbCCZ2WzHe9XP-O38p|aW~fR-!GS7g*A0ZEYS<f(&#eU8
zDl}d%5NF5m4iO{y%-0Fk%}%xPijn!T;Ac8N4@X(UWzRLPH{~cGQx?Fhyvy|LxDZZ$
zbl&F%LVs8S#RLt-xxA>qy8vBPOf8$9hPux1Sb9VbZc+gM??v_HSIO=e2QkDes#346
zGl6m8A0}J}9?Nd@9Ot>EsGchdi)!AZ>I@bYJ%b{{`mfl3(_Tu5|1Vwa{8Y>elG3b+
zvYxhK;mn8}y5Dbc57r6`l6h#j%*1kab&X@v?EWe8z0TvDz9)iW8_eLWC)>8@52f$~
z1P5>bDTHR{=i$nLZ4c-epcDV;+&Jj`WGe(tzkX)t<W+HDZ+=1j{(TA<>@+Yj=>{sh
zCuP^g#YOL?1DF;Bc_1Mn=D_sdwt8aN%A`g06XF09rw01^LIB16ySV6{pXbwU1TGqp
zC^$4vqk>BG%9PDd__*VA`)?Xr7Zk7v_aCvdd8ya0T$ikTB~ecd25&{Rft);_deP$I
zca(8p2{uuB+Yz6u<UHYl^2x^GM^vKFf3WqT?#Z0dXL>G+k7E|qQaTM<H^$j{@5D9D
zF^b@Aa`*Wu23CJS->q&`=BK0%9-)gvzNUf`w3RC^=}Ynv1=_}<ic3i02gVu<{A%7$
zUq56Up%1@NP>sL7`uZ}1_M7j^k4hhmKF@y42#nLtU|RGvURb86RiaN7m+(y+K0RCI
zbU&)!&ZU!<S@K-UldC{Ko+5JQbo5{?#lv=HIf5PF4(>+QZ8>?Roy0-s8U8;Lx03N1
zeK?CN6_e#^GvN}^L<qMtpLT#JqlTDpA?M1Osy{g^iwXcgJz&*E%I6Y#eY$PPLYTm5
zt?|Vk^b!&EfYU1ea3%nJFjmVtX8|E(d@e8}6~=#F^~7vxxg0K_w1L^&e&1(s`9Mkm
z)XQe?{w61Pb$xP_-A+?sMpy!{p~2;#{|DtqWDE=@7sv`I^Vm&?jJ5_6Nd<kH-j6+b
z*%qx$<ih>!uH;y$r1@Jk4cB+PAjoq4t|%<1ImC{esOERODS`eK>(xnA!{=F?$(3;$
zxyeXx{k(dodymW>^~*~gClCI!Q{{QegrC2toK7kE8NM?iVdC*_q@NOS<G3ThYG<V8
znfH`WLTw`!Fn3qeL)gKsryDGBvR-7Knwn~O2(o1UMB93jjHFUQ_u^%{yk5_r4U_sa
zmDe?LSMFwA63*tuoAdl<+WECH91sVMo$2r^vDS^fy^Y_R81X6i`bHDfG$x8Wvr>6A
zctt9m_NFl8OdFPpT#uGuiqy*ez`T7tT`rWDcatR0{1t_q-{Y(|!n^s$<tsWmx`AXa
zEMEYU;&18#Q~!jK?(d^Qg5Wl#DD;$m?ayiBvzlQ6Jvg|0!79)e);CnV%Cvc}zOrob
zWU*>rs~C#9eG%}B>G_<;uX~$T8_X<=d~1yG+<b3$!LM?rc|*KdtvbQ4Q+XetKe;SF
z8*btovOH!dkAlN-J&Nw;k4`z??l#>+d`@*EjZot?w{a@9SEY<j#+PMGe4ms@;-yNy
zym;^5<INs=@!#kC^p_Caw$YnNFi+ohEPYBT+uZyT@w~+g#e2Snl}hNzcMI8`>*<#+
zIm`$8>fA(amLzIg&h9m8Q=rza05Z?ANsGkCHMAMC0rz&xBrv2`39KzjbefZ(*v3(_
z*KBN2tcKk%BZaE%;^=8<X#)wYaK6Cvt<vM%Vr*>eWZEbj|Nk2k91RZ*ZGpKw@EZRg
zKYqa8)|idL`T}b=&er0;gTRBy{ck}YjM;Pv*&CUZHZ$nmPg+!fA3?YL8E}kC7x2#4
zZEaECeiDSj(8#3p0t2j50z=8qpY?#1-#`*a;Y5U?h|xdtCb4i*eNbY-JoV&3H3Xbr
zfu@kjDW@qp9fr^K&<Lc^IGHE6xVi&2oblelUKO;)aCZs^-kG+tUo2e|!s9DS#IK=^
zU*0aw`xy<<yzEnZg6SI5imIW_&+)W(%9%fj5fGYId~kj|6(QaBQd0V2Pjg!3-;b#k
z7WPIguQT(j?5?fK_;kE6j*j6{-S(`<73~{lt=GJ)FF?M!Yi>4EY!%?wsA=mHx^ei=
z{Y*(ogM>syHdQz;Ue|Pe7VYfD9$1{*US_w{sz%;!5TXz8yTxE#?u$LcZ@$4M^Cw<x
z$+pD?UP53TQ*q-SRB(Y)gHP5HYinyy6%vqTFgZK|rXTO>A<;CAlAAjrLg3~r_z$^Q
z<$G8d#@=-4aC#(|<HjCJ<NpINnAuX(H4`h)%?$u7F(?o}1f}wg=SUKOY^!b77>t01
z&*F1;^=e(&3IFXevt|t=<axMIFEX^y+&($U0o_g%ta<)>`Yx{DLj#&*2l>(HbSyYV
z2DI<#!<q=EPw`o<N@q53%W9v!DW)q}a!9G$h!H*tmlS7W8EP_|6}G~(r=x?%t)xGK
z>iPKC>;6FADn}tt$AlwO9!j;OrR_gBar!Ske*O5u)V$t_Is$3ZRPX-7+|8~0>}&{n
zY8qM}-JSg^w&Wwqri2ACri%-s`#oOT&V%oygKm=>E4U+@09~7G52bz*6<yof5(Sl|
z8_<{nOGC3vv@r8`#7`Dmr^AB7QPh%G_m|Kz3L07`fa?H7jbw>ok$di!-}y9i!6P8N
zz{c(fC1RhcG{q7Z7biqV#1-kISl`(JR&U4vuj!5c6xm;FB=B$l0NT)m{2uSX#YGC6
z&%HmZ4~t}t9)r(O;r7n}pINIH?5Ka2mwBqCI!!<d0^VEvOD!R|!%JpR3AzQ|Bck*R
z#d&M%M`=oC=Ap6OCr>itDB{%K+1CHZ)K`aPm3`p?N=O=%qzVYqNVg!MG)T8}cjo~q
zX#we!Zlt?KP+Gb^y1N_h{v3bz-g)Ma86So@XP>>-Uh7@&dS97`&d-V1h-k-4gqy7T
zY|fZ8uI{07Qae&)9>@9Py<!T;m0Vyo=GROX59*U6FsWe8y!rsMozY6hXquko-a8F_
zfzK1t<8MCsrRW|t@qSt0RKJ-}!)JY0Tw59EPunw)<}Ekm)4etRtAA9!svJk!Z1m?q
z&yU*<QD$27T&8z+RE*)v$C$LPwF-q2c$=3huMSV5;a|&-MqP>A+NnR`SED0we3cuP
zU_ii;zKKiZ61l}VkP#*oT2g1X)`{}Bad60j?rse?jom@K*nC7n0>;j4^3OIvL3<km
zor{*0mBC>VOcJi|Bev4q1Yp`6Uh&u3E}=Kwt+rRtbT1;qfC9maH7^&}SJS!VWD*Qx
z3b19qej4gdvxgSe3uX<n4QAZ#-05r&9lo7B!5O@Xc?XzB3J98X4Vgnz=^}={@`%K&
z=#Jy%2zknN`w1nZv`Q~tbI342C1bRWTpIOD2KR8eGda@BBkFD0iPhcmrn$u?2=7fV
zE$FQl^VUrj@cWUMqO+&m&QdcHI2#+!BqvJ^)FhK|c?NKsj~`tvP#72)&)GrKC?`)R
za|tM}j@$dkj^jSW>FXynT^f(wkhY&>Wohh|F6|B6)jO_?cA?BW4&%u}f75~tgYeT+
zDKZYfG93Qy)qAKDcJmL7OGRz@Up?TYHoo?3-1t-U;2w`-mRicxn2ZWqg=%B#*2V^0
z*ct)F_c;_`B@d>p?d^|h%*QHssj0v(rA_UAYkZ-5yk^#-DM#?ed=oq7h>;+;$Nx#b
z0hRIoS-TE*<)UmVq<M*sSd0`BR#;V$Emru3kv!O&b>rc(ky{k%$Qb5qA}8JPu2o-B
z8Pj$pPk%Df9%5DZL~&<vQ`N{4oAKCY+QP@jeA}p~(Ym{??|U`Ad@G<}b5PRYG9VuL
zWNSiX5#qaJyZ?newD>`n!t?R>H5~qF5mi|R2Fy+{f}+V{MhG{=7pwBXj~3euBh*h5
zO?pYr<$hp0MSO)5DILti_cV+sc7sbyRCnq(`NeYA==!nSpvlPEJ4b4pQQ-ozeq#WF
zu?<72odm?3ENd4-q3g31B>n1tRsCSA8=LJ4{PjnQ5xZkDqMz;b50I=uMXA=}#h+fS
z2?YL`oS^GIRxfZ}SZp}FmylSTG)@zh#|1dZ#csnMy%#1t(&bVVI#sMR?a7A*s}qu{
zX=yEQw=f`M2CH9ddif{!sL|i)FyqKV0r<WlcMYT8W$CI~8yJ5cBU!uf_~pnKf1pwy
zFI8Fp;&bM>xI_+p>KoUU!yBGrGa#Bj<6;B;M-dKYI_qCb`jQqGHidt;;i?inD9n6=
zxO)_ki!|bdd6$xxtn1A%zR*Q8;XvAN{Hd77<jtCl;Xp;QW_Vck-SnRA%fsL)bv+k^
z<Gd_<SL+-EUDN#?yngbCRi6jZY61)F0ghYh1xp5NtA{_j+AL;j$isZGFiUc7d(pQb
zbCMXW!i@J96ya;Dy-OpVlD$MnMU|F+kN87j+>c_J&Fxr3Y&6i-fZ!dy?X`-IXG{w}
zNPj|B!OaawHb5E>6XhO?FY@IxkdD<-Yfh2($2uPcH$<Gbwda`8`Lthlf-B+|akB-=
z&5c|rdJ^n5wwq!tm|v{0YY*zkj^f>xkuxrijnVe;j{_(`(6<R(`3du<s&UPh)2$~Y
z=dTL3yAV}v@jf_QmujeN;vXy&`KFLQ3B*JZm5<OjzGpq^>%8?gML4(tfl;0C10Fu7
zuN$M8?Xd`L)eV@DeM2y8SM|NJh@LFr(no#9atBc|vCX%+2H#nTzYF?l#mtKKb4<jT
znOSd4(<y~ZZP6B2RCl|PX`<FzP|;E3BI8wPf%b8RqY~-D2@iATb0QUu)#59EHFGbH
zJ%hiNt$q*BNJ0o)P_aoqgX|rFspb0ht5=&6GyYX3Xb&Fff*TaU#nlyPm*9k~mKzU{
z!Ern0kKUq7=+7a)+J>hmVqPRibyHc;8$6dXr`w`GdXr}JD^vU5)t7H{NBEeDLP)IE
zmfw_$`;J=Yh_;Rv{h`iBNYn8f-l#+?>gX0|3MM7vbHfeCKHZ;HI_E7r?m|4Lz#kJ`
z+Fk4oqzEA3b$GLey@~k&i@8dGZ(gBZU$H!JQs~0dWttPpvsS)Smvqw~dMwPO6|lOy
zFaNg6+D4qM^%>-n&Ee+j^KIDZOO*WkyVWF*8c)=@ySle2&-vEox-vu<_~nc^g!QdE
zrhA%KS~pm;r)A;i4ydI*hsUSI#dZLGAyn}T1Lf>FuPX+g6h=r2T%Jr##~*xRF##|{
zdjw@zbixrgpe+LzIoECqh7{=i9%krGUj8z3P^){mjx-QgcGrO7@(q8)_9Z84U~yaL
zK%NChI>k=vZpo>VCpTNmfMwZ-<9%(eYcfeH9Otnn&Z-}JCUSj5^zKXSRm|5zYK{TL
z-~5z}PS8E<kykZa#WEMVYF^6s43thlT1MBkVho{z+IUm$u^x<^_J7*1u-+8vF5TW}
z%1SU!->y8{wxgW8E|X^1Y*uJ-07g0d>lm|ikJ6&rli9a|Nn%rydPb}dr$n0Y^6*`I
zGEy;qPbAU=w=ZYUb)&cc4e!FAJ-}s;#0Nn3i(-PsrKP7VdLlkZ)*wrPE*3+W4qJ4}
z4IThdm1+}gP~UX`!DWzdzc1whtBFi<cVX?gG)d%NV?YyBj3PFQFh3#NAV~sSEB@x-
zQO%jPrEA6|C4lAr<oQ<4m=L<na|MQ40ej3-YWbQ1L8vcP9Bb}(w#3#!$ytTopZnXE
zmZM~K`ZBGt6A*z=1N2;gfI~Y_zH1@Syq1f*{IscNV>rjL>|jaZ!-Z~_jv0%CYSo7o
zhY+*y!MV8JZIal#vc5_}{K}6VFGGWoIto4o2C)XeQ${TgKu+Q)49UUMyMa9{>Fj#z
z&gA$=P~rH<AK~3;GGAv927z?Q-`xR%#M;;Csz@Lm^Gye8@oGF7(|{{}_=i7bGkFL|
zYZ)>rJox;l;Jk3V>_6LUzOwRafNw0MXN-$Pbx*m?dc2j%a!}2-9%wr)QyNCw9PkKK
zg`{2-3{4H1A3j7<A2Yn_t)e6i^UZv<Y4MSXw`$xx1KV!I2?wO#sXKEKrW9#Er;9B@
zD)@(>8^lj$iff}x1V?B2a`DbWO&xv8Hxi?T{;PaV)0KMtN13LInQlR4pcqqS2mp(I
zI>xR<t~241B7KF!>^kvqcqv|79+!iZEAN)XEvDqNHjN;v$*#=rA>7k|fP*(jMm3m|
zm3!XxLtUXuChUWmuRYxT!>(`!8%qpz1Rp^EEy=FT%G7lUNRK?uW}WG^>c8gY(UfR4
z1hjo@1-e>9>}K=Jioap@hF?-r`at)?bh4bva;lQ$l^DnXKq2(%xS2RkWb?!gKv+`!
z!&Ap@tg!gG%uq3k%NIu4LBVY<3QS6wLo{)3=EU4OcCKbIZYcKd&SJ^@AtrQXb)s~C
z)N4H13G%=Vu6o&^hLNqwP#!H`uA?rN7q{1ebb$krkza!K6m1-V`bQJOWtDl)cPlNl
zB6U23rUw{)9Qj%ymaMrITR+6%dCqBOSM6hhKt={)R9;>)#X^Zaj$%TOJ5L8pGsVp%
z622!TR`Cjis8#F%g$(bY7yVH{*SpR|(=|j~>IJQUkk+&Q!q5Kw5`{6SB+?7>323|K
zf6ikYaNZ>=ajGX?OjJ1pS})kR_=yi->5^s2d-k5V`D#8_ZkxR(67nKIUT%u}vewQo
zt<!#Wmoz5{Gu@-ZLh)^^d9+C8tf9<VR#TI!T&(`L2Y02VsR<owZ~^K5M8_f^+ZqX0
z7igth*@c2qe;?+)hsXPpmp7@hk}Wzy*Rrmze@tc}+&=f_5&oE?y~b1dTRZ#StCkk}
z$BbGkoV0OesAjr{dX$>M48z<bTcgQtO^NqDvJtYZAA8^`eJ8LfK^73%9{G-#Ld1a}
z1}@J8Zf-7!mkXChDrSXC{@`FUpe`asEbLoLyttGHU(AfOh1$m)x^ad(&02hN=N|n6
z8g%&wLGwlI)~l@ErS;lEiozM9sTA#OJghpU!R{f~jOYly+7Uu9*_mOx-}|?MbcL80
zTQC%!l#*5h{O3QK*1}br)pLi{vf}Mld!GWKp0;!7VCi6=**+i?cyPH#>2<r?n(O;a
zI=iwQ>L)%HHunu1Lqh)f7@)42y%rh<%<8vgtVG3}Nm+dTU2;F{+wy4ey=IoB6aS?P
zj_+_?7=&_%Sv)Tt%*KjLPd`}#g9;;nW%<K$Y0!1NE~o%_$i+L?RPk|VySEYd1R?_q
z;nqY3uUDn(ig;;uAA^Lp#YE5C++9W6tB?l$QSPa|39wHmpL?q}!(PtHy1#$g*&@cT
z?&$U96bG|)b~scZeP!P1Hf>p}LuK}JMJs0Gef9tq>A62?z8N+v-L5Zg`TAoh*2F`k
z#nNJWUgB79qTU}YwKPX@*!&E|B^6#=sHrNElRHwtdb?!g^-U}-@)noE#w6~}P!#*G
z_`lEFe^bwv@E0HeHGQoIzw%DS_Q#(WtNf?8FxX+u3_3VwD87gqyc%pqCoYeic~i7;
zT;q6q_A&-H9400v(nlqT8{HGC{m$+<Wi{<>Y-%%8aR4%RsIBtZ9h%`+n7>&FB!4e1
z4(j)b?F^~DME0lbW{Tg=W_gjd&5do=YU^6TY7AUUpgo}pj?#ext<EDM<$t&coZ_ul
zrJo(Imwa-bnEdi6K}yl!{=3`T7`(KWUlg*9KhqyY+d9~l$&}yO^3IWr?~Nt5w9Ft;
zD~t+f=0mS>=hGA+1R@7(?E?E+yJNm-qoRe)vZFg#9-@UMnrwsE`KY<Y6yz-~*`nM2
zMW~U*<ze{gieN)crTYCbB~F>F_8V9ZYH;r-TvF-Hv9_VMw%5NB((BWh1ju$NHQ)QP
zHPxe~<R7MDFl>!79M=st%8*vIy1!7;N+0N50+|biC>0$vNQ?<eYal+K)vD39VZCYm
zswG4?cRaJIs#69gIWsIlyKiJ<v`XNzv$JFSebGGmu3m$>6z#dhUg`yg!}bJ=x3Dn`
zMl%8Wo2)kTZx94-j)z~kBLUxoS);A`P7ii_J~3Z1B$9>af+lMZlt_dMeWdRB<#z1e
zx2yhHW&~}pVed@-(6Hx_n`|-qg<2bUzp}`;*q8J&&w?u&J4`D16>%a;8TSPD<Py1J
z=6P9g`>l0#h2>8o+)O0)l=PM~_4OlrIY{iO8&?W;!{hhUWCQMVUIVK)s23Ma&Uu^q
zVl1PdE^Dk&>Am?M>}z5T8)w|+pB{SFzUA$qGDd=m!?19wth;J7YC8JwyG#sJzWK$&
z74@kilLCj$T6z6TAWx<KQuEDbU6T5|l6TAE(P9-RLje4|{m>GgZfL1ec(_<%JAmtA
z&HVMdovHX4{LNR#mC}w*wpB)6EKDTso8UQK88NDxtbFsjDJ8m}w)W>26n|}DY*fnS
z2dPj&Ta?yRSnf5;BRRg`=^uac{VfopA~Z{YwmF{jV*z5i3?sE4_btY1RK&QvfK5e=
zjw&-;SDpCjj(%L_#`b?VcoN1rd3TX}UfU4EBHmM|{3|e!x7v8P`QG>EC}gGTr_X|T
zRLR3qien#nx~U2LdQEinQKzc)s)K!9=_y1aO`~4gh*Bpe#`mY9`{}rE!%Ww5c8~Zq
zown!9X1opF;O>2{IH99@N1&`MG<ik<&!}oc8l88|&2#09$nOgn@8F6cI}Whv0<k(W
z0BzA9n}ZoogxcEL<++y?&i)V2K1B!K>21u6kF+mzVd;LkEAE`VJGA#s2um-7aDvLl
z<5_GUqP*=?Ke>pENE0&8Rbq11aV*|OY;Hl%{WCZ<{pT)op%a_G;m8lTbAR`l3>!&k
zpkk=Tn@6+%l<rGFkf~3|;Tcf>ew=i^X+bt)2w_fB-k2oLU)#s@(HH&Y$Z*oU730NF
zH4L-4yyy3LoE6>I`?*W;_j>fKp%)gP+d6+m)}#?CFEayR0d_ej$9`K`@kRURTwR_8
zTbkDW`Dl5H?~y(9|7s1_-MXiZ{&hp<Vs8DPlSq%`1hEw5y2}UJVQpj@mj354wV93X
z3;hY)iG@9)r3xX8&QtAZ{w#!HjW@qSXv8QQl~L8*zIqvJU`gd+K;z{>Ty<&Z<T`70
zUY@_1CLi-9+gzoghRwO8P#gDuT2XnE>GMrVGWAy>f}pnVUdbmEuVl7JA(cV`Za$S#
zYLiW~lcWcZ&+{-VY<4U^N<3@y2yO|#>qyjcC%k+U_h(N>=F~q$c^vN#-G2X5ml^>!
zXC~J!%zGXl_%hju551rhA~DfA3iKEZ9kPMQinZEuIvLLyo&4fkji}iNCPpdm5MyOs
zumnpNHJBRN<{wp)a{nX`jA;GJy07h-rx;AuL-2Z`$>ysAe<?Q68<1&8IP3k(xzQI=
z6WZwf9^&;4xC7ZO?hgz^r>6|=2&lv^W%vl>o_7=hjb6|SUgw~dR>EtGJoOPR{09m$
z9d)4@&Y!!z#dSA(m9QGGmY-nOm)e(gom)+(X$IGoRR3JxR5+zfsoWQ4K!Dv@8Qy*p
z7|Y-!jKA%mJxSV)8S(%1G9|!2?wPpCxr`XH(~pjcCpI){DJhy=2?_*EWzP-fEPI-`
z0M==7xz7(3xVX1fbGaPgl1MSSaqOLt#5<DZq=^1vd2QZnPbEO`X6K^8?srn9^@!X~
z&`rDego74CEWyiR1iF$8S1R0+HRi0In_d^ItPN_~zc4B&!GAf3f-@pnJbP19e^3ie
zx_84KE~7}#)H>0ozFm0aU@|hOcQK14K*W*^iH+VPWH%<e%zVYXn^ljXj)CK7l#TRb
zt@S|hb1!JPkHuQgdH8lUz;yJk)9zq}R`GudPXm<69<H}f+K;0(IV8NSGEX4uk-BP^
zc@iC;)jyF!l6Tv%|M>dxJ>t^03Y;eM=-N(Zwy$}Jv9sSuXdl#M5=a$cyi@o%rRD{z
zfZ}{oc-3}1tj4mO7_fWOy6yy0#owF6^YGS1@&CnzJ{sQ(s@q(2Ti8hVj!UAGoO6kJ
z$KynEV#}wcb<TROudQ~DCkb8MTh;N)j@+~@;Qf@uK)JObj6x!bobm`rN%P7Ic529C
zHc=SyVf@Rs>5cGQ<^YZ!LJz6u4CAqEbqND`p9w&ZsUv%f_S}kERDQziI9}bsOzYZk
zfogLdEBDWY)3K^Q9jM}lGKp~<Ee7oE1@(`b?=5Thf9g@I39YE9SQ!c@RXzM!Sj)o`
zOQ>Sa{Qc<sfz6n3#fM5}8ku)YTybsHnO()|^Rr&Z=PeiB^{SfgOi)$CLAXm*?tx%F
zvUB-Rp8q^`!OgT-Rx$O2*@V;b`i^8XZh_LF?s#%S(HwF6Hv1Nr%S#fkh$h42mIzJ?
z_p`;Y=f3+kwy%zCNp^NFHb>(KL39;Rzo5{-A*XV7Zf981Oi3m@enSm&J%mOMO7;G5
z|LsKpyxkM50<_kbq_W$o1{S79qOX^do%KlpZ+<a;ul(=>Z@MyD{O;}D<=ySxP3Ihh
z?<d{zUc4^xk}NC;Bl(-Mcc(k+ungA(J#ATumnx-!Jd9=o`t~E;S5G@<)h!xR7<C4C
zy%^RY>*H!)+p3Qi(%?cJFMqC<TOF5YfD<-5qL$AI+5h=edhda}sln><{Pw)fbpJO>
zKwT0#b2J+CObL>aGSFE!eW9=a>sMcAOB7C@*?5lYED!$1)?B?8;DgO;{V6X;dw{hh
zp~)w1x}kV_obH4t$b-V(4pU~Ve*9k~z?Fru!>>yk76z@^bL)j(TvwLI9~2}BRauT9
ztJz>!u|{Nc^u0Wc-Gmpfq_w0lRU~gi|MW%}2K-p;|0R}LeND3?Z0EZ6H9pqq9iEHz
z>fTNt=qfG?5?Jh5l-1C6W|$Z*uP%$EaTGI<sr-XLSmIAgSQ$$e`%l4Bm>=2>&}C$+
z&(j7s62|cH^+q_xvC^Wyd(jWsZO^!t{P>vxlr!=)6O>PoOqH0vOWnve%^pp?&f4sg
zrDqbOa%R=wkN=f1(9qc2C}1_lPyOFM_N$#vP?|saqUZfkSk*r%`iXW)31z3jn^>Wy
z<yl1nR>ZsojY;9jS<gQe<q0*r1kE$YbmP%c=%h>F4Qk(!Ew1=L5%V8{$r_te$+)Jb
zBR$!P9mJ}1%JZ$RL$5ugBXkj~3YSlOn&Pt=(TxfmfXVLS)`m~C?~3t{zpYnb8Y-%1
zv?K-|XDoflWPLUVZyk^8tA+OWhgIp5HJ`L!G_NhcA_^txyP|7ebF(X-Vm%Eq+m`^s
zjX`D)m&F`DKreuOg;0mQKKO`qEG0l?IwF?g`7}aTp9%C0;du~(a!Y87bCUAr&t{P$
zBvL69U%iu~zjmweg2e~d1^9_Ep16!NZvHHwsX{XehCVDv@l@Ru4wh@DciJ^*6kVDV
zG`(yp?w@L!BaY<^lqK%?yg4{0dqX5ox(fW~vKi(DtTUcG-nd=G9ovQ}=v_b8+Ed{4
zYBahWSby_avp++8N1Y8V&q(iKkI8hpz^Er(M?Tw053U%pS1{5?o470b_>_O;kD^q0
zHz<W>vdMNQQe%A_xt(_TEI#`nD*eWP)`Os<c^_o@va$Is+fC0X*+L&xtYxS`q5B^!
zT_#G;Ns&l&%{(UveZ3Y8Qy-*CO6%xMv0KP3Gj@=z|LzjJ7Wt8H{+%8F@@rYAa6K8m
z7Rr2wajLNXN?PRD&w#T4!C&_FtN8;duF|b#p=o%fAwD4qm6md8sMA#pX(In+4&9jp
zW42z_glkTzlnZ|xUoR?@IYUmZ#lh6<(_^_p1Nbf|isep;!ttU;`@lv|>o>s3LI`wP
z$CL~45#Ro{ne+&MBp`qhSFt{p!QD~s$j~CP_lBE?RJJcZOxxPKj9rp(7nY7lR*a-?
z_N8$6U^!H6c{4;?q))lfuuXDlaB8B}IBx#E#71=Q>(;<W4`Zb7PN5x_+P7DK_B6r+
z&bIq+q?20xrQ&UsJsD%hQbklB2Q{UT9_e_rZ6cfh-SW13ufcB8u8SV1mtM^+)3Gra
zWn|nDd^=M|Q<75Kolx9$u74@(0kcHk(!NVvE<6~MQA3MXYicbV-@@<!fsdvMbQH?h
zW8y}7V1l2$_n?wkO`C@J2c?1SJ_eiTR~4H(zswNJk<=<(F(S*#2FDZ_CSpgQK_)A0
zo$G|nutWE<w+&S${mEsSLe_pY19=SW4pi5n{f2PGLKj*gcJuw&<p<7EEQ;p(9h><I
zc}ktKJ9B)`F&T~VS`@+`HP<1Li+X2#Kd*`d?P1ssB(C~5U?_5_{o3#lTvVuTY>5gz
zS2Wo}ezMg*-nX#EvF|Avk9Ys=bz4WcLK61(yt^xr?engw{5b_6-Rs-b#1D8;cszmA
zZJP{4z`Ss1!{K6%E;wHr9k}7@5e+p^_hVy+CLX<-oYV53)B9G+RC#nCdS_GaLrIz}
z4h6P6(RD_m8O*nRXr9yTAJLFhCA(0qQNM!YsBhrtsBEQXdU(p8*goEJ(*{BYt<!l(
zgg|H^@ILf7MXooD+~0CMF&_E(_U0+~@=a*_pO4OZw-(#G3`y{ntOe-F7lVuZ?S-~f
zS!0&-jtFj>OUzJiM6JeAfuC5KZ9R}oiVsQ5J=Ey=8m{-foxj?AmVA)W|3@YOsYeR9
z3aSA;7XJA^yDK`Gz3d6>r@R%$9I*>{;P|asvQ5aUdVUL`Z+0Oc;ry1i&OAen-3c+G
zk5qaoFe@gW3fRjbv9|;#cI<pSK`I3&^@tyK;*~g7(g|Q|LkQGX`!`A_g%7o@N~A94
z<g`7Gb+;|)U}-)WLo9xxV#1EpXb!Jj12aaP{R_gAYi%OCC%6EChtcW$GmPI<i|NZN
zAQ^M^HgI8@Zd2tst-ITPfArnb#qA9sRY(t%kzBaDSLlyX_ud&&uqqcncU3{!mY}h_
z22{p)j!jdD=>GkgpZ}%yBBB|Dcw<e~A`H9Lhu&wKz6uXYb5kmr>tBvI`XsDPFuQWY
zs&4kJHJ45)8T^n{Z&Eyg^maD8FNpqcGKogj&Te_Z6PL3rBQ(gSQqn$;Jnp5xZfwd^
zsiM8%t<1Kb=BVZ7R^!zwe9)A|s4k|!YRm|kd%v^-Cvjq>hYH^X5xy*|rhE2>i&Pwb
z+iQCY-$bY=zr;lQro|Cp6`&5bcxqf$nudjBX1ZR?qJZ|h6yYJE{^i+Hjc?$j-9Ef&
z`6%gYXko>ZQ4f!G(6o$j*oHq^aZ1VG7+KrpZ^iwiZZ`K>tGL#)x~xTLx36C1(}d=6
z&x?zlo0w2_{-?WNI82=j0k27w*B`xO6B!1WIvTvlAfwH_Kbj{b7e)j+r9Y<fXCgD6
z9N}jkm2nEHYKN}>$a=jqR_Vk5UcvD8VDI5>e?O*YI>uY_b-f+MznQU!Xg1Aua*fce
zR~de7ybaaN((QMMBDJuK$;N41e3O5$H{%sPH3e|KWMTpy2B5q8qxy3pLyh%ZKb-Nq
zztEvp@36semd4`<mpMPY+G}}2N-7PMAf5tg4A9kh41>Y^e_D*>Jz5(`qh<u1^JN1#
z-JrDP_BU`jyV$9j1DfinD84=b6)X3+u)`*U_}&8{G~gYSZ9Uf<O2Qolyf0@T*8s-#
z^~t!7X0yjr2)fhQ#lgZ_DJ*T)1G@ef`aq$-3osaDD~xIcPJ=V{0c2TjGV}saM41UO
z0g=WBAm^;W6xKpYRpaN^_-xlkE6zd!a8n9@p8f^vmzf<T7(UWMjB?zh*Me{wGOmtS
zJM%slGyW?dqD2D5qFlI@wVq_e1eyw2urU|W_Hw;jFAcOd_Zjc(^nXY0`4zmtV?geW
zWAv9zg&Q03q`+GixV2jHxWFu{70~rDX2^($Buq@ejK?*L``K<bAfea%7}+`j;%>?%
zn!nBumM~_FA!b7OL+?z*fc6J|?cVaQXt+xpa2bsJ^2J6Hu`chnMl3TD-gN|2ID*<6
zAnZ-AS^EX7|Klf5;2H?Go`8l37Xk;u9QYH)V1+DZ>ti@9r@otyGbZ<Z_X^5qk#uV;
z`liQ_1YR+tz0hq*wCg1Klk<<Y0#7KUE6ooTr*)8G|NN2TRcG|~Ln6v${26NHumznT
z)C>P=$?2t;U4>0CYNjKks-yeGt8*?jLm^DfXsw@+G0>X?foCbuy1?V)<eaKzsC_sC
zEZh=!of(0p$M$qh1n^M@5Fucd(Q7xx&&+57{+)!In_FZTxu6%sM<EvajE9FPbLevh
z&GGJR5}@;xIqzu$J|-Fl1_dW4LFUkVCnpY|p4WKrli1|<VE+ri^MUgMt*r-w@#p|>
z*k-~8mn0Dpxer9$n+;dD!P-vN*;4~$L|q`o4i@~_jUUiIbAjs~iGW8ub*f9o2)F7B
zBl!p@j=lRU1qiNq5`}I)6DDkg5!rdHj*epIyR=0SKAM?4-F^79YgIO%d;C9Ro|uoa
z6RzWQ6#6LN>-qo!oT*xA$!Pz(|8ic$Un)zZVLV3kS(x)?7+%;0V%gt%ez*1iYP(*R
z$Ome0fKy)%C}V(=8W1AcW<S~J$_gRCM_PQ4Ox)afN}EoJ1Ox=Y6B6bojcM9CJHv_i
z03FnJSwIH%t$`Xwcd)?<V_EPEoFvOm2fw{VpjWTVdqanUii!%FBtL*In51VoSZS$*
z*T~+!s5pNbU5~D>0J@l`Qc4CqP{e&aOLSWL$915TG|5(J3{D8x@g@V|h^4Kq9x$oO
z<-AJ*XjW9Kx<T=x{MuvBSmDP;ktT+3W@Cg4Ey`ivm_!QGH!Wj?H}Je8!{~c;*yUc|
zwC{D>XZ7V27ozH}w*{$JiEKZ6Mn?<QDF20>xP-5qtfL)B_D_%W3@GvWr>pqDXwz$}
z87>6K3v|9TcGK9+@ev02j--61?=H8Pz;*$yowRZ@O_TZ&{3xw}4->+73*wEAN!g*;
z-3%~M{JKaVT;qA=gm8Jbr;}c*kt}$3{lptE;_zsGz~*;$`oP37W-C}JD~;X^GA488
z3pH|l{D2JH;g&%WFl|unA_czsB4kv`&+(;2*1x?mWV2`%rG&YP>>U4aqY))<2~*#^
zy_I%f=)PZeTOB5XHaXw^=}V@6QNlmdT{~^Wq3NRS&xk;Q%x?DHn%S!X;BgGMI0JHP
z7ouw3z=ijE4Dl{S+D$)T2b}D`4!>dMYy~QG0sj7YuG?-W&?cVSRl1NbA*U4yQ}g)?
zfZrK_k6vj#mkO+EVqA4AGT%iY;d0eP&Ct-F0X|zd3%~p6GEj0|>5inRVE|h=jy+y;
z-V(5}zJ*2gDUpfl!{5s&!ofn;P!?w~N+CUgSIQa<JKIHfaSb7jBj8o;f+CSpW7GQh
z>6j>l{@?pNtceSRVD*HNYJ)$AkKb^M)u#G3g-~FZgvW}PV1m8n*U};g8vM^VIPih^
zD1z7J1~mehR&(ePU}0hLR{zl?-_0Ol8{!T4)C6t~VA_j-N!#G}*H`d{9tcQy2=Cs#
z19qrN+U`45*L;$X2?$W-J|b2@QIwRF2y*!!vR`z~sLnG82;5e8(r+G=qinQk2=OQH
z7gi~JuMd1OB8iEMTmHJ@yR%KJa$%L7GKl4%Ic8H;yJd;0uoYKQ>+b7|)u=(9DkvnS
z_xRmFYNq22?^?75JZ}PQpp2;$;xhL(qlvVvY8BSeA>6^I`O^$&MUc%!lUSzi;*TIW
zGp$IgAr`1Iexs4_at~-XYIbwY$LJ{rJWw$I67RU3p;)LKhDWCi243_4gZwS%<x9rW
z>jD!jK+FvJ{22{cM<^_^czAetLl{ljwwZ%&nKT;lzzGQH!1`S;Z#MtE_#s~~5}gdc
z(yt{!*nnuP(~8uKt@yNNE6KD{o%R*qw5k3VbkPd)VYCm&fA8Ax;q@EE<BK4U<!#xs
zZBGBT`xP?;9EuRou(46*LFg9&ro!MD5%4Tkf}6CpCFFYyHCS%-N|QFgB7;xdga2RG
zM`mWwGySkLR}gbqYc=ux+ejDaZdAV7x4wh5Z%>rn^9H9yogs%kfzP)FM9SD+00sHg
zGQ2LYm~>hu<TCx{Y&d-HBp*}q#l_P8qjEbtTf16&#)1E;(Tt3;p(&d?jGK#{4szW#
zs(8}-)rQHsbrfVW+4A`^c%R5DG#FiD1Hd+dYeIp|3-<uChMF%>ya6{Jm#6uE#AN3=
z&p1G$hm|M-vsq$<M~@z56U`N>Fd+bErUAetG#$y454r+}<euCB6BpNc>?!s?#<CoN
zHC@=@XH)5y1M_%b5qHG^ld5}YBR%Hbq|W|qFAPP!HaaG{C$hqrAuj!ILIyo-rz{R|
z{3aoX<tV@GO3<nr;Fv0t-Ih!6CiB}~!(E4%`E1lAaH>R%=XdA8z^6yJWeC5G1A}3#
zM!g8>u1BjyV0%<nR$gCKRQwfX1b_aFk}|8ePAW|x1&nXloop%{yE#7M1^1YN0VU}A
zzv8HAx4YTzz(F*bhN>Ar;9>=`M&488=fHBr<9O&DvO3rK)z#66f-K@7-~oe)HlHCA
z5qaG9j*gANZ}&j1A>er_*Wz{aVMU&drC8hzl&gMuawRa6FFK=07%!JQYw>j)=3f4{
z_kZkstEw<uk6XHM8)SQ3{Zl9k8<tmC(j*T$I2w?i?t<YU@dmdkvm*~G;M<1S=#K%b
z^AJQr0S_)285z-V2+Sr2g@ovX9gj)MQ`v1hU2REc&j?7%jYs{wqA+qoLe=H%#rDRV
z(`i6DJp_3htIPhpaMuhC$sByV@c-xDF8$-)N~XMZ?X1L4fZzyqhOqd+Z)MJ8`h99<
zCNXv8*o(4%QOW2iZ?mskDLP*<7o9{Rz&N}uqh}LSD@C3(3jw&SZ#lXwP<nVh3X}h~
zH=^s{R$ZfX)B>2gRc{XLLDRJ}9GEosrt&9&qY1dw?}u}wfj3cYUENTQG%*NXKiYgD
z)D57FonWs9$g^-i!sYHrFl9&#+OgmhgGg)88I0!_5C90rA2wY$I5^-Ft>C(3wOf7<
z5Ys`kDzF)S*HYcX0r_?I&+E3KA&|E{1d?9w_FzrG!501psAMkQU?Us?0@?iLrh3j1
zZAAQ--L%pErb9h+Z93EWYIhi(L9}ZE7-~~2s{i_<E6X(>e@~@`m=niz1ex((zFY~Y
zLlUCj;xRE1DdNnxIa%hD%k^E){I^bd=|K4ZaC@p|f*pDi*gWT>tHo9nqP*u&<!^Cs
z{mybuu2{{|t>uJkR0CE0)lz%0T18Ip)kvXA1T;TCe~mmPv1IXpVGShx8*?pCg8L9m
zHLHjQoKqv<@|+!RtgqhzEKQzr*#Jn1naW5U+&N!0d{pFdLCj=|rB`bQ8NPb@**ozo
zIJUZytzFf)XzCY4w*{lu411(nY?dEDot^vSdYEX{dP2lW#_F5?OOF$FQX+O}c}boE
z6iPuC{|z&s$s=;tat~b<cWp2i5a_m$u5R~7Q)<_PX=K{dlK$EY*AEq1_lQm{gm%KQ
zg!F%2k_fA+)#17f|M{~%SwRg(U+E+E;qmiV;z)sbl|)P}J{iuqed|D5<(naH8v$Q+
zrE3#a{`4OU0Ixrq2!w`%q$N;ziPsnJS@Nd`Q*cea`Gnn{)s`x9nkMk}EGYWvUH9AG
zr@TZO#bcEu|KUh^^D~7mL%Ie`JfZZnFFp#-gwex%!D0Y7B3ft0aySPI80EJ9T%CIh
zda<%n$y|>tVZ>wb6#isu9Il8cnG2`y;sV5w5Q0@CUH1Jp&q2MsrA3fESr5qZlf8lf
z%Dnp|H?MPcH)ehse_sYj-eC02e)FMoIQ%5f{?XJvUX>s$i2;Md%4qOv81#^H)?WQP
zGK;B32S|JH>E9Y_as*Hm7=aWRlU#tWB_3lRm~lQKA?XF#Gq8d-067E1W;XH}@g9mQ
z4=MO^i}==<u9!F~1b{w={W6ZfwV85)EU>15xRhVnqcmqYQssQEi*;z@gt_{q6uO_p
zsp#IvOdstS`DPPz#Bd94m(Svo5_549@ZAhPf=H(xcodMMAJ%}OP_6XuS31DLXc^RQ
z)BqR(v$<aqp+vI{2k^t6HC*cI%jY|7)4h!@)y{J<At*dMJ;4?mSM~gl3tzArgR(rX
z_W3pjfO0>ovJ)=jWLxg6>>&&jW!P4&RGbilwm{?YzdZzkAv3-9{47Y7Z=HZFE(WRn
zk+2!3a(+)gun9d}iIst8>@*{Q!~-Q<F5iLKCFYCDsgmc@s>j%BC?NHh;`WU)oWIFU
z(*po!F#pdnigy=RiZqJOpNu7ThuuR8!f6Ho5p8FMPJ$57_5NPi886DuUzn?a(r(Vl
z(Xnr?xlZs~2F=AB@cUYxy9Md!{%^WcRURs^HW_kz{5O<JbQk;3pHECfkUKKRk5*6E
zVEfG+2fA$?xbSaf{BTK~s@3tEFJDkxeQkW-aW;MTfitxTMg57&%8tTQW$l&Hr8^sT
zfmCma7^x8WhJvs9VyeLbQfWvC#^mH=&CKrJ-oQ#v9AbBGZ!gm=O!IuU1!~D2hTqu1
z?3ZoL4yMqUpL5Gac-q;vf8Y3M=^YqS#oV<MBo8N)X);ri3^iB8ttR2QUrn?=Z|L?(
z9n0Y~Ku7QVX7Ew!)_B+P|1y>5yWD`_hk}9<_2mn~8&J@PhleAu0W-yx)+8{;sWadA
z`}glTFi5;e3V+bgkJ-`C=5w8M9!5i7f5hGk<N`qP_~H^GyafPoW7qFm8rO|{lBBH9
z*<-#QW9IE(_KTinGHHF<oa?n8&*=UCUKh@3&Xq~wvYAIf02A{dL&Ne0)3oJ4Y8wG0
zF`WRlK;&{eHo0vECTtZBJn(IquXJ#zr#J`mzI+Z*dHeeMikIm{M~z}J@g6<t33(J9
zZ{S7$-wp=MR@@NTgdKn=+@3CU1Rm9Z2$S5x4`%f_tY*5wh-E(*H9NE>=Xa}wuj}Uz
z;3iug$?n<ay2W`82Erba`X1yUKli7hBcMJ($1lfRMeR}4ujS@VsC1Y5@E8>u@H^^S
zpdbn7lC-cZzxZEo%CD9})dK}ikUXjZnxp-h5d!GmrKL;?xiTx@pI-o*vb$4aA~44S
zia>^xa1c#?<<S&PRs~vDTj{R}QV(um;4c=qxE|Paoi<)hAOM(6^_8S;&EeUpSmsc=
z$ci^oeC<yn1Zc?2`l^7FK}tOej5!G08Qv4zo0MG7?x~rXd-0v|ZT(Ew`8%?oCt@sr
zIptvWerR<qB5T8Kpb6COozMH?<jH5Elr~?|K7IQ1q6TpFf<bi298H|elmjf4jpa_s
z&}9C&r8UMp&e|zqPaH6VK|=Av;E6p)5S4}(WIVa2;d=WG6%3%EzjJrrddav8A8V<!
zp76fO9cZ1z%-hp*`|i6LI$hG2_|@$E&)n_DKjS^(GO{Yxd7Efe$s=p71!RmycGokm
z;c2`BjgBqMd;V)rF9GfO=*izRA&jQW4z{Q8&<#2}5!L|z8A+yC*b5GhSb#-9*Vorl
zyVf**4|66e_Yh?J(O$ChzGJB8*CJJS_1=MURCCtB6yDF#A^MrxgpNnX7c0K`=WelH
z-x%b}_^0wSK|Ax|y(e3V_{+jClizoa_e2AL9iMmPBirHcPonfwjmT_7>k3Z;6#a?r
zOn1@|2v~JT*%zH&?k`if0W=kg1Y}IHOcvvUvi}Zn?19j{{g{l<=LRuwtx4+X>%Rp)
zVnwzl&qa`H?v9_5rJX+DX(CI-_NKlU-0T4Bz`j+%%62w%5&Yz9-0r&VyJ3NIONh#g
zL7z;u>k6PrDwh+YsOqk^w_bzYRThY$==AmX(WQuZ?iWiR6dyrzH><p$h|)#f8}rTC
z^~V`Oa7<n(b{i~wow#9~a4}5OIHQ@syj)hk0@etAc(~73K3B2QG$0lir}*UM(KmNb
zfx>(MR6Yqjj%Ibc;D8&(#Kv^a<Kz~0&X%z+<m&}Da2}Nb#x>}fH-8^)thk=G8&H}d
z=F`+XhxQW|HbW<IUOeN-J?5orU3|}#Dskg&#y8iZQ`P|qI>aXk77Hz9DW#Q);yK?H
z7E}N36*Jkkj4wYX2E+<R&;RC1!Gjg`gE1N4rWX%pLs>wfV>b(e{O`@-s;wY#sZUo;
zy6MoQlhbz}93pRDEFFu9AD!`AE74F^?OLekx)A8TU<cLh+tK*t#55{YKQc=3bsU1v
zTHgBRcb-BVfIU)Y@D$<qmScL3LZm*%Xe-`|Kb+>ART`H`BQbgUmzM(;=<iko-J_KE
zx#LF{OvHe1Wo-V{YIPS<st)YB9@@b2sI#r;w8uSr38pqxpwgyZfTNGT*famed_$l7
z9=QWT&p)X9&PbHXjWWG-aCTrc9lF7WhDLKr2vR|cZM&=~q>PXV)1cE6e<B$_zW`gV
z#UfLdfAL><#z}dj?MjCy1sj2P!}OIO4K)Zu@Y6z|P`GL0mfqaF-@(qYJw4DnlxZMD
z3(3=-R#(G^$UxJ0F~wT>VGSLk6P0%~zQ@k!)5Rd3s@5c9a0kZoI$<h#HcE!Jn9d<O
zW|o$kl52ZkF1*|Oz3)g#e72Ko>3jJS<zK@Y&6XFmoRYqN{o0}hoP+3$Uij6<?Ff2`
z&6@0(YNyj;%k@Pbjc15UNp?Y9vwhdgBK{y3a;erK4-goPv97T+*6;QL?BPgHZaPTH
z@{X2oq?xqC-6>b8sr5sbfXz$IS(6oA?I=sWg{}9eY1o`hAOQZ&IX+MSEJE4?>Ht<7
z!^X^nhcz7?9c8cdI;}zRrKw2*#x*W#PIhOflR?gr@qXw*5VRSfGvp~H**A_@VhH#f
zG|1ZWMkO-gq0sX4pMr-EgEPK-O_ec{!3KPliRr{sV{V@3@+oy1lf7#-rS*`MUWTjf
z|6O_xKq&A9%CpmSxH$lcG7RSQx^}S#6UCojY=9E5V+^Ud%W~9w-ro8LZo3Fk7>YH(
z-x_CouhO5Xl&WaZ$4|ee+&iOCp&qdOcm33A%sdJyVz-$1(sFm3)Y|$-CmgU`3>VgR
zvaJA2EBKuaTx4~R6wv4j7k7}|kU#X%_rhGPuDb0t{^el1_(py6{9ve&fq@8!q6Q3^
z+@DW_ZiNa*@#{;)GNHD~$)~`Hbf!1C4#0I(rs@r`^|rKFWT&kBa`j}4h6_sz;c0QW
zZ9h`Ji5DmOX@&bK<PX*@i%teR(BgUjWH(o*Vg27-xze|dk9C(lYIcOt2w0q~)QzAs
zBQ>DRP*zh#b@t}50by(QK2DlMCGVq9^qc9b&Yv?-&w~H=FSk3>Wp3kS=X7n2NP9XD
z5cp=kHTl2(U-XuT(ZezpKy3^J9UtN08Ey=MWA>Vpni}jD<kwi=fBv{jvkU!xNQ*FV
zx})E=?Z-Sok-{^-w1D04WL`7CROnXRKl9bd@lp1>Li|{1a4r28$?90lQ6OtwFq6wb
z3Zx6@7n^d^3`YBB*S&&2#gYxqG{FOsF~Kv&IPiQ!b-#*Ln9+0@tYl;wkA5&~U)iyE
z*kKD;f2da5{KW%UOWruCuITO>AuW9WZN_$5+|0<&3aDJ=&KUBMk0_}%yM`w5$mh)1
ztS0@HQ?3>^t$r#&Ho`NARz1A603-*45ukYqs3i((P9(}`&(ZT%*1Fb~x(?fob<!`M
zAsL$SR}E@B_C*R|>^OA+#HoW7I;L2tT=?q{2F}@9&D%iu2yUY%3ak)E+lgwAFzMc2
zz*`G!IHfRpwjl=m<lk5S$_nT`XKU)pj2?kdUiW9=E(@(1LJeN`OFoX+t~5_n=hELb
z+>+1fI~Ey;tmJV0L9^pemiQCUTm>$+DDY;D)oeZTPfws`B9+RQ@GF`+b3Cc;vIdMK
zemwT|Oki_Blc*=lN%kpad~3#XcLvn8w!Q%}Bon6!oU&4I*`3?4*sotWT5bmcYF7aP
znyxAHE~Q1tz042i*lx%iI)Mkd^TIFBI4#@%aB7QEG-dy-ynbjg^3wULMF%jOA1;ZA
zipkZ<xKCjxh_-M#vzSahB};INM@_v8YLyxA{u!&I;@K}5!MS>T&{CXI|LgztiqC$O
z6}t*JXWAvSC8a<q(x6bE1eHRNK1iw(xb0se0NfmSyKt;TMm-c>P7>IBid3kyl8vkr
zbz{$FQIE|q`@9>gap&1PpRGw78e~R8q`@K?%dzGw4#(?4Rc`m^!K*8qzx%mapYmD@
z2yRd0S8IWhww<Q*oF}!D!oeU{X>X%+BqkhgAwT*F@iP@a{kt81Xfc78%O7xF(ww-P
zY}nh+E2s+V7%+$uFMT8Yzz^pBF4!9_82uAXFqX&x%_lbzTnp8*XZU_cldS;~c#lL8
zP5&ev!+0NLeSsN4ke>+i1ImO4(WG8yjwxFftnQ^J{RXdfQ;S^3-7aE^3$}LefLrRE
z7IZl-9Ps)PgJSWhmYHkU^c3@!>yFMzmG-5qN|+lIu3ovfk+B)V$j{He7JZbhYo!7s
zXQhpGwgV`gvrO$Yn4J!L$|?q=y&`9=1{`;>IA^}tH#%&5K+;A~=}af%EO^(0?`FS=
zb7^((n5rZ!X6)xGPpPsCRaj$T+V0*Mq_jIGe9VTd`J-!CVUCYeIO4cbywvNeDb{lK
zrM{Oz$x)G&)#g~azqaSTKvj`e4{{6=0s=z6Az^@k>(29?xjOBYf*Sc6-Ing2yF*KY
z#(O=5OK)}ZEkrPiFLY~R<em|*E4N}=rDFG(0u++9V{d0r5MBb!a4PW>9>+fr5C;MC
z4=)Qx6OCJeVCPQq5|H0A`|<fvKjQ<0L&Y+w&o8mb3l1W%hIFQ@%|WWp<9eFh$Zv_x
zq#{E~%&C~NZ=(|T09U+A46AB=wg&epaUoSw&8CULSG~n0rI*Zalsei+vQ72<&#xNl
zrNE=NnvUI_geWs6GtPaX-Gn$ScJVr9Lq;2P<D>Ls&9<&<P9_^<*GI)eg3CUvX*9an
z<|j_aoOOi)9}j0B0l?k0)D;nzG`*6$e{|Ht4QuvnsOUlTraL^|m#bOrJD#xQKl$x>
zX0n{(Y$B(w2z&59E<ptG17~?!KNRGAXidiLw1oF~Y~IHW%F)r89V+8rU<8jQf!D!&
ze3L~$C2oo8LC<tIC>$}geL9Z&P+dV|{3hR$IF_s7M~{S*)Ym>Eg*R_n4bSob#vEzE
zk%R)wG=7aUjaO6yVW09zi{OxuVPJ~H2IeG{O0~5(lBA*A3*&xQQ&ZDRfI*CY*L{0&
zd0AlC1I1o~65?xOVgcxW2!o#k7+nv$mzTW`-yVo7>>lSfq9P*Q<Krq|iw(ggZT4>H
z-{ZhQd)x%*CKhPalr%OqDK)t{Vv+JFYSdWuIa;~7)q|n)JkX|91d%JYw6yfaix<a(
zZ*fd}Vi_*_b@63%+T3m1VW91j3@u%ASlYJa&FLsAegzBpQP1@0Q}qh_8tfE-;=Ece
z@YxC_Ot`BGb!3~hwPljJm@zH6ZRDo0Nyk7(Z`rUSUqUr+YP7WU`~0)qzI$L15opNX
z;aH4E-b~|?z*LnIk7@r9S}EhO?{Z7fH4_!B9OV|s_b|$oyoQ>an>V4^h8m;(<b0(<
zWgs#m<C=N~W+7<#`L*5#VAd^f#cio2MMi!*K6a2!;ZXt@W7BeVZ?Ba22Vvn)>eUwc
z@LteNwe6Ca<xH(S7%$0oJ6;Dreat7Ak~TLpdDNfCE(w0quLQ~8m$-L-vL(T+nKc|h
zM<!E<+IDJ4h1@n@NyjrUM3s{BIJ^f<HxTHf9WIYe{r&w{W=?xq2bOv6K+`_)wPA3M
zR6<6Xe%B^AJOlOutTt|~?%)$7>%(h7t)#@n&1Ex$_#q#K+~d^|a(4GlS}IEFwffnf
z)o)4aRU?vTu*{e21q80C6MiCLY1|q*44PTdknzAu{j6SDo@G-#d{a7TOmke^egg~`
z1cZ5YVwKyM;=v{`>hS<GfB8C7CN+91?j<d0S-t~5_^Jgxp<aEupDGAx1B;15aFLIK
ziA&)KL5FoYED|m`Aa<vusyacnXG`w3{(NV$;vKlVMvK%c6mw-rp9Xvb{|4m)Q34`C
zFOSH*aUdCZGNB(~3Zy$|RZ5aSC@8RlIUasc;}U`L9w;NQAS2^>*p4>rH5Y=GMYmDK
z2Izsyf%{`+Z?6>ip^^%CsDchOP!dR2uQE&IuuLpzxpj+qr68tUsB~~}pk8hewA*y1
z9Y(}10gN-6vSNy$1g<90j?TF+A<-At>g-wB*()UPL`1k6NDyA?abh+JoS?E3UXy!P
z?8{OaWJwhJW;M`BynT^{7?XwlT%v4>-FU^U%EhBLHJ(}AH16VD!fUleJ5&80>Q*s%
z&c<?`{Re)*d~4*5NZLR57CxNWy%y%HbX#LQAw3GH4<jiahFDlSZAua0%N@?+dF@RJ
zkDP%&4-V<*T3$a2=607b&95UM!OI0^E<HUxl0eN-MNdz!WE${hsm0w6e^HlCJZ5JC
z9FSqq;8be%xR7k01#4ginqr6h`-33mRBZ9`R4Z1lYiQW2-zpjY9YYu5l}kBmKfn(p
z&7deS3~X4Yy1cTI4RG=|z*pr<mT(w30YO~FqBHol^>?>dcn=>w3+a3TWWB-nm4tuQ
zI=i@29ngc!bsX1nenn3&wY$4pu1b%Nh6eu}?9i=;_#fjm&(*;j2dZ4uMgaZva5KK|
z@efhCuC|_oO|Bp?2OWoE&K^Ol({3k8`M(`Fv#vU_pE6`|Zc5-NMP4d>K~av?E=f5#
z=_-^!+`VjZZOOl~XQB~^h-toXg;p)#E2y{lB#XF5e2y(zz>MaGTDZO<!$GH(EhFIA
z(|%)==L4QK!*sp0WvYt=Zv+`bGwR=EmZP*Cii}dsf;HT&;yr}ie#NVVG&D4jLD#n8
zzbgcHfKv`wSfG3w!gd?aqJMa8uo*?Ab`t<ZHVRr>TQzIOL4Fw3So&Qof-Gfi95725
z#HU&y$frxalE(a~LBpy}R63FEm7O$!`D}evRwKE61vskHfX0)&6hh3twg#HVecW)d
zzZ5NO2qD_#w1SC!t2E|Ad~Rq2u@Q2kz*wVfb)9%kT-;0cD6>nm&Qnsi`4ro7;-7eH
z3C-I~{AU*iYJ9U<ab}VOV_(`2O{w)GM;KKTC*E&<P^RurRoIKu8Lm4z5N*kX?<<7d
zqXVojl-aq-&MrmA%S9}WOrfsqhSsUEq^m{{uj^qJFbFCF-R^v#Ln%ts0LBwty?}I&
zWmFbm_6W?GTi0*oyy$x+4mz*BM_ZtVi^>9jAw^UHj8FUzsq>8dIuB5O$_FZrdE(J8
z508$b`;6d!gUC;53WPBI$ttqoOOPu|?y5C$4Zk?R1=W*C6Oj!qfidjg8w)^|qlnn5
zwhg{Bd@i;))}zy)$j&(}x0+_fzTA%l0d5l3<I%J(8*!uNCmW4DXYYCNNQ(ozAR0Bw
zC4cn~U0*)+Ep5Ey*HNfvR1K6#8QTxKqnmJqzrJ^2hQ-D2sH|F6cTys1RtlvDYvJ9)
zJ@1-tRiI&L2MoRD3;Uz1H#il)ptw^L;<KD$6aNEFhstu9C5J|;PDFTk?6K^8V%*?h
zMrmlZ`jK{{(pOKb@&;thMjv&d#>>q@v%z#97<A`pIa{w36cofwTxVDl-&8wVG|9z3
ze{ym%3UCn`@!+5!c~F*}fCimT`Oz#0m24K!#CJ*c*Vo`q_J@7N^1hhFdYkzYq7r|d
zTEM78v%K}jyMc{ot2D0E^Q`G{l)@qZN!Xs3+y@D>vCX;ai@U|&(V~C46F+<j%wf<f
zk-kE|ee@Xq&TkIPCdxx7H_pd&NQu^tQw@K=m;8{atff^xRNbjJR52hxC!WY{pF;`#
zVh&BlYZ6tfb32%D{e|rWbc~Lgs-Ib^mIk~A{}rt!KiJ`;;2^*wM4H#7cw}Uxgsg1b
ztW}&YVca&3k+KYqOLh?%(PS|JXfzfAkss~H`{0mk!|5*l5Xd{$HbINe_A64lpGK2@
zTs1il{pe{753Y^qPRl5C5X}=nL(3s!54OS4*Q^mm&!+W(qS4xW?I9cy5$xufoZ@LW
zJS`^r@Yg~}E)lAbImAF43fh5aPM7;%L{#W6phjg?Yh7{`Rtoo6>0Sci42dX!JZFRJ
zJRig=T8-+W;WhJ-yvVE`Yo}_U*O5D*f!mkF8Ao|@eFP04HyZ()mZ&Tof%7*&o~{AJ
z$GD22#Nz>UBUZXtZ<VM3%%>TE1hNTEY|mLTz_vF*FU}3nfkFx~5vXRE%*TvKXXF^o
zV=Uxnzf`?lZDij%Y;--C)A$;d__7K+3TZY(Kz8kQlxSj&Q&T9?@ij91O>hOBagr|i
zAg&Jqy5m#iX<v1juDf<3tWpb~OhYOr{so&3!%dxXhHZ*@aza3mZ4jh$fR!TPc`~XJ
zl?9F(ZjT3OF2z!Mf`+xk`}f_BS8cwiU~55mJv07}K;Tj$3QiIMp||kOxWODX4o(E0
zH3Vo1DGJFLcp9yA;-Duw27y;D92rq01=$Dan!%OaDuKTb6%~~!Akh?nY|=IsA>D7<
zWB%iq&=b_8ah+N($y1k$zTt^B*H3&|Xyhs3;mQUU$@=<#qigbN>}`i)nS^D^Ox^n}
zV?H`OJ$aelgJSht(XkhuLb5GPI7#Ny|A)A*j*9YY*Tz78lmP+)0xBgf-JqhtP|`89
zlnmX?V*yG^OM^5+!+<o3bjQ%8bPV13ZSZ~H^?iSS=bW|9EM_gxnf>g%_jO<Uy6^qT
zwA>shZsHjfp!mm2Nw)nX;AkCP<zkbbo_+ztgj_ekol{>XtcQRjIFKUY_d2HhF;JxK
zj9L`H!oc314g`90mVHq%G1;_I!Hb~Tb!k2lKzm1TBANltk)y!$qyycf0M2|p^9eB>
zE$>V$vCT?GR8a^uwLJJPLJp_?g+lW$$e1dkwQ9uU-65Rb2B|Q&-RsXivK`@MOATW-
z3%|*@<mg!X^L>GLzt_c}J+)5M6Qel&p%dg;805@)^0alKJ8Qx#+<5zD$N8g%p5O8%
zsMlLh(=!>e*6SD0Wfnj;;V5QZF-Lm29WuZ!cGEcwM%QxN3uGHbjFv*#7k5#ivy^7S
z#(;Vy)%*0w9Apl&cVHPreZ$`RiO;bJ(BBbA(g1EWtrO`6N>Ax7yW>3!I$MR2O}|zL
z7?gIWH6W1S1<K3U$bE$EvdbL!6JF_<J>Qjp1m4s#$aK_opZY0NM1SOuUS}L%x~hl#
zvXdMZf}OJvCRm;p*4B}OqC0tg37=0=m0wWRoIwF<3_LnGL*4~3sfai21vboUJBhMg
zvadc5{SGSOfpj$34)w)BOBcI-a4gq0PdO^hMFZ1cMOS^1F*U$R41yrcc!dgZ_f&wV
zXN-b_R?^l006bdlR>G{B^<XJUz6;U;$)e=}Ay&a>ISd?v8%!c1IsmN!>0oceE%97%
zV#~tH%DUhg(D)4qHB^A|O!dPJLf|rH$V4)mgS<_nl}<=%uN@__Ny-J=HoQr&vR8l&
z7)%uPu%A!|j>(16L$E#N%JgfJwvTVdQW}(1;p)d9zWH~5QN}<yw-fgzt>z?(ow3za
z5xom7q>;``UoQsxEm5lb#5~3R#0{_0-z9Lqb78BarI#0PH&>h%;pA#A<v)?0XMqH?
z7TAi7%qZTCY&!&ciH}UfJDZ%_^zOlUV*uHL#~82}cs#zDnVE@h^W>u29o!Yx?UTlN
zp!(Q}$B@jn^U5IGRE-t83selQoqm2Hq5R_(HoLZ#-%JCEOeK%6daMdmpIH9>=>uZg
zwGw~lN1#7d0D>fz+blR$<;r6pI66PUX`}qZsU?TbRIx%HONYZ`ky}>=tT~M24n<D!
zm79Cpe~nH+@|P!~hok&QwxfI#oNpp>pg{y{fT-?5_o~{~VLu)LuA1N4>28O%_~Hz?
zu#C=k?J%GqeFaojp`E7f=-MmvA0${@K|XPG=BuRm^S#Z@UgB3GDL^iS0Jtw!!0sC$
z(dc2P$NNKo1fE@3SkP?5l8GD}MfJu_oj+y^3<Jn+oK>UtF$aDzMb^|P#JBzm-w`pc
z=sh_pGkh<Oq>mIuw*C{EZ#6%@7zC`xp*gBH$;ZC6>EKP@@FXJsF2^el8m&AH6K&@H
zPyYUWAagbkh`d(IR(=Uks0<9&3;ZlzgL*7jADH+iD7dZ;)7GCJyh%_aW=>8e<-uSN
z##f00AgNvIGw8z&bD+HvvitK9fo_A#M@L1eXQj_!<rA_Y(5hqZPU*C*f)A}yti(9-
zLewIA6n7CreBCM{q9Qv<+8#3GzB_7njz2VK$o9W#?R?#9RAbFHdhfF0(S@3<TbqOr
z#ydX%I|h(cK2Y(sG&D$%2D}71?MiUzFbOd+lnUnQaw}t!qf<<apTw1Q;2qfqX97>J
zzFyXaerJEbE{&`Sq}-s2x%rgIxJYwnui-;+Fe7#7O$HL{w<YF%V_IHbdNi(eBBERC
zGs4A?IQH?e33jUwVK_IzQjnMOYu_7F;}T9Req?Lzmlc!ADlB|rHB$T%WG_^K^5Ce8
z#RT^HEFk(mI^Cb*wV4!AUym}liCujm780}`%GXU?C`F)|7hJ(tg5_|MBbIv#-|4wc
z>yc;26pNJ0j;hy-RrrOcov(fQ*1v18Z%xsI+6L;!&HbVM)Wq<(#z#Tok5BC-nWbrV
z4+xeTC;A4r5jr)oO`_0|?T_V4yHigGj#G01P|^f|L(SAQ;{r_pVh|}6LS6cp_4j=|
z*AN=<M2FRxneYX{#Kc5<hxS+oF0Q`NC9H_o90cFBL-s4xBy}oxR@s+QCHi6ps$T`m
zMl;YvnhZNHIp(<Q52=Uy-eT+Fh^N%%>5ryOjH-%@gLr;_;c<^HWqi1v%#UV2C1nyV
zRvW#W>8{3X(S_t_A8Kze62`ymw`sbYyi}Zl5K(A5HVZIO&Q%kk*;o<w)}#Sq)JHVD
zIqGV(;z_v&`rXoXiJ^!i8MJ5G0~aPS6`@#GjLa7T32JI;iqm?mAnG1*3pML26H|G}
z94MNj0B6cW-;X~rl&=Y9n5gUpF%3(|l$iOnYj>OZREHh8tfifc9)7&Kb!Y^#T!PnI
zmL4^Vr=Pc^yW<8jPn{^l9m+D;wDW0?Y12<XN5>OW+)t~;6!f>8x#Y;asd&`cUu)=v
zIoVS2IvZO*a{uwjYP4K@$ls8LgJbs%G^Nu$>Es6=%2Sy=zJ6J}qRv0AS0V&Vu3fui
zsJ2sZQ2W1dsRXIA|Ha4HR}|;y^L|rgao#p^a2e{+D@VL3(01C=DRsUtmK&~QHiuct
zZZk|lsF(OwTk<up-@YvODW@I0{un577IsEY%K=GpHGvTZ0ZOU1MCPi2Jy;e_KO58-
zMXHAvax@uUm|Q!F2u*y*goFfyKVEQ-SZ~9+%&)A2wA_mk+n|b<g<>TeZ0XDV9Ux#V
zdhf%3P?C_$<qG}%Uo0%2S;czFd-oRq^DnSI44aPb6$9$hH{dhS<>$g5O<&qa@W+4k
zS9=$yD_FK{oCFd|c6LHkRJ2tS`CBH>?^4{Q;}2*=r~j=kF?i5hq`?B%((ClS`^V9c
zg`>#wDt);RHqOdmS>Em<8$Z%5FP1&$Z=Zd=bA<!zZtB6TvFJ9Gw1N~TZL_|or60Hv
z^<1<hxI1I1<M%x;!Sh=e?+GU_DS6jFc9FOnF1o#!?W*Si{gnLm{^vl|f@nQ-aDx<`
zxXV($KOiEMKQ*T_Hl;(dvR$yLq@EoU$ozEDEB1JlB}XH@*a8pdvVe>q3>H|51b>K#
z(GEN9TEDHsvCL&ume%|$8gyZJJ726hIK0rR|8luHQ}z2|z78Yx1unn)gamNVk8gx{
z3q11(rizB&fElg8-eDiqe+*G`PY)hQ&kdGJfY3EYQ<0$)gMa(y!L94nSkuNxm$kl*
zIhM3yK66L;)(Pc}0Jg9*ZQY#Jymald<ssBcK%T<oGYq^hKR;vI(I)g=Tm-pqZRl1B
z`%rSUUdwNkLCkV#S+LRxK`#m@*&|{}Eyp8B`X&|27O$c3Lje~9xi_=FY-|)PczRi$
zV$SW2t8a;|1YH$dv(2?Akt6qAKcl1(cj&4N-(d1Z6_B?sKNrFN@?iA|9TOXyJjl(N
z&z{;c-q)|I`~0)vV+&$rM<4;Yb5#KH9m@#!`Qe$Fv1yy$$AyL>?ocfplIr^$S=E_z
zORlOBeTOkn&GIT4-}jlKy0Q_D(Z!xbbB#(@F`f2ehy!F4Y5(?oM?)&xVUr>2B=IfX
z2sd|R*ZlW%_%g}lx&8_KOH@<_P!|1R&#nhq#)qrPAJ6^c8t|BwRssleFDfO%#iau(
z6>eV668@Hq4BTR#^Z1-jw!>SuQ1ExqxerM_gxlIGvtTx8;<_5zX??GF+w*N&U|`sG
z233CZI&RAoR}MQu={k_hDB#)v$U~&XO22UH9Z<_&uYSrTRj>U5#7p4e;@UZpzl>VK
zV3vtw2qb*N!oy5TMrN&1m6i9uE5hAlx%x^b)7(Zai%EYUDhjgltW3uft8c%^goWim
zAAMT=l17m1yu4~*vr9`$e`Ka*K>32**e)(eOKO0ou4}Zlbg04+Ie=)^{NWKKs{m`o
zl|7&eN}Dxyb9HTLNGKmlkJ-grKMxNJvlp9NSy<5cc18%wV&dl)a=e;OF8&*9`h^p-
zpBdF^VD$Y{+QY8vFy)C=PB7;#*8iE$TQaQp$JLfvym?v@LJ7<m`Nv03GnLatadUL?
zb1Br2%eh;Z!~T>>q8=Yhfv?b`Jokf5FqzjEn_yTN_3C8ai{1508-+!We=*laXhm(5
zUub5Pim4yX$^FLBbXGXp&V7FMWOs+kD_;M>WoeIj|8a$`)VE%MiQN{nVm?fqt0w%O
z+D3rJ%XY4{x2&WA4C33RSG~OcpXshhAjD&t+?`0x&M(pDfs<y3lU8DWsDC?}U0#Sa
zXoQ6poxkr(Fqmhht)FLqxvbv8Dyk}hb<CswzIj9KAKc!zMeC1EpGA}&<U%05-wHG*
zQlcOf&0r<SVf86e8MC#yH4t&<98E<vR#oPDSyux*u+S6(tu1@yJTg4<_cOW&&b<%i
zaaUfaI~7%QZSR$uD;BJ|8`e+Y;or1>mVWuA`oK%`12}w()#+yx+x1@XTBj<$wXp(~
zg|1fXdpU{>&<pt-4He+#&0)+PoU;FZ%@}-bnt5|)qbG<hcBzT9Q%Cobkj8pUKWt2R
z`57!MIe#!}e_#KNU;o<w|Ebdc|LX(&54Gd}|H*$J{IB-^H+t~D`sBaggZ|f^{P)3s
z+yB2;3Fv=h7XR1B;IjSyd(HQU{@aVVZ32xgpkis6TV^@(4Fr}UfV*<Rv>F&B1BWXO
z<kimtELJ}Ioqu&we8#}_Y7iegbnTYNmlbFQ+?Wcl1{rxc^m0Mi0N@Fgis5<eS%2{t
zU>uR);{cn26z>|*?7FI{l7!gU*cYS)I7=NZH2eT!<1)cP))H{m2Ws{wSC;?UTyW5A
zn4W=C04^3!^jH%_Sx>-C+9E>Hh{(p*p@0k+Ew?Ek)Axjw!g#<Tmx-xq<#+hMYG|vI
zjq6&F26S{`bPZ4L($K8`LHz5#>g|T)>MhF09D;!$KvxqSE?>R`q_UTOc+gk5tVhxL
z`1!SujU5*+WTdAvJ$(4EdyJr=<?uW}!l;DP_<Q9?q6X{p1&_8Fdn<1Cqdkc7&%!J`
zdfUBwdpg#6(`D)jKp!i``edul64pZ<T(~;81^8d+Hub&@RP`N0GvD}$>Df5Gf9JzW
z@8`X<0iIzF0zJagfHqfu^8}otnuDYJXi-MsXM?l=3fbKxBH>{dz$iR1Dr1L7L*rf8
zBQ{<%nm+lX^!zRn8Dwl~{8*5-0d10eU0NPy=Y;$C*@5U47@(Lnz&SFq{VAsm+ppvE
z6Vkv683cma<Ia_Ioosy%gYvP9qf{ma2FdaX=eXs;yp`p!e|Hj-%r2f@!xJ75;0T0>
zFfepOfJ@RzUb%Yh8X81odfH5tnuq7ABSl0Yk$AObwS}eyJPyP1F%vt7@J6)H*Qb)E
z8H1~%sdHoHE3+dGLP_ztJ;@KyS&Dk1a2}0ayg6On*P$bXw3)9Y>_lk(gomLP*E>f`
z&))(@vBkO^=_y@Ep@Ga;V~b@P?(S~o{&{B)zZ;t$zkpz<GCo8T&MHEaZ|6(bV7&lM
zDS<R$1R^x`9-l25A}7zyK(e`$xE}~qRMyZ95YVf2z)zr<6FK^JC{o6k<lka~Cj>z>
z;(CCvRRo;p?rKRl=>awfDlt6*ziv^LgK^l}+fxTifLpdUhpVRg0tFhjDB3F*gKMHL
zky$(u3q70sIdnY7Un6Yg`|Z!4Es=$hZbqcjgpj*K&$Z>g(pI(n9y{|T^ET=3qXSfD
zgM>PxD2+nAXeNKgS#kj>tE#)?Bu!9pWyOV};bM2sxY;TjK7el}H2igY6mYXK+d8%`
z<!$qZ4t6hX(*rx?($h2A`rB1Vzr|a??PX*XUdcinXS!aUvN8%$;0#V=ToyBCAziHZ
z`xH`E4g)Vb7yy{`%Tr(Y;__Z#cVWdBsGi+uyXg|mrkw$z<q%-CR)4*6bJ*kL<594i
zZdC|X^h@RFPi>VIl4@vzo62uHCCtLYcK4U0J4Za8t~jsv26fjYKNGbcU~Hg{&it$a
zRSGw(lCbbD*4&Et8ukc^lXn>?NMXMdsU}#jPYPosCPJ+)+MO1!C)es%t8zyd^!)lQ
z!JE|_l+R9eIo-ZK&Bk}-xaktdYmq^w=la7Ox$)Mbck1}oVpC`;N!-`0t~<XehEU-Y
zQ7qvr)xygI0~QIR=1pr}6cq;?@`sR7%FTG8^r=ZtdA%|VhsLqAiftaxKj*M8UzBBz
z5(dunOL(6-$%AqLeMs52^J7%r-KFz}DAyS+mAHpG$Am{Wnw|V_MdY&KgNS>+7w1sm
z?k_oT2H6XO9^}5`3rx$&<sMaURlX)B=FK%xIdvbxxcOF%jI)CKcY*E^NxfLmlwyMN
zTyX|jZy5z?46CD9hYMk^<d0Mn)sY#yOWFIC&K=z2oumdOOOe4=yzBWO@#M(PGQSG1
z)zTJHlk>6$Pg0eMH%$Gi+M_u%fWn){wBz1S-TG&=VyDlWyA##U-BG*NL&3$Qg&|J#
zka*{kOI+JAbhteBGJqF4^?jXRY7CQ+M>oufM@&4~sf(-3amk~=t}T;&VdCYbeM^f@
z^lX#SKRAS=_W55_yA&BqN(gT8=-HanLQa4qr+mPk&v~o?OPmW%arE>0uvA997Yn^L
zi&agkYj{(nqMzH)tMqoiqH(U$bLl>m=a7TW;QZwXx6ak^3x^!t>d?EahOE57lcl}5
z;xU+&^DYR4ybn^j+Z@RSQHf;T%lXTZq(95n`r*L|=twZj*U$4CLjtb(>`o-f(!pMp
zTW7p|wAX_H>2oRkqR?d$eyH8)hh|Y4-!4=Hus&I;(YtPeWTWL#FZ*1efT)%(5PN;`
zYLNnkQc_dr1ER9Bk|25DzzX9IQ=Ssy4G{5tshlLli{jnNozK(F1Ue4k?8>keY2!3N
zxqO2^e1=fXw+wB@-QQPP@t;HO&6;1RWI7R~)qgcHSl^u^+DNLg?i=wFU_|YnKb+Ld
zp}xD=qX}6tM@7?NguVO7(<5DvgOk3{p>h-_e-{<gRMn6*{cY_5Kk(!$e*I#6Ov|k@
z=jC;unq~Fpy-cIm_o>?nx_<qV2@9td)`#_FFTU|>{_6&%6JOzYtK%smN)P+IM1C_Q
z@@o`3Y-@MR(Up~ZW3$bBwU3ILg;P`>3Y%TPU^X-_J0RH+&?8?0$}{DEczX*)e#x38
zXnsEVp=f~~M$#T(CuvTvmyW-`ZD1MSU_{mCw)*@&^<U;ELIHRuTLVQ(66q$Uh>_lz
zn?c*nKN`Z)6oM#)QgXryU_r}U+k1Wb%?d)L{`~z<N~e32v?j&}@*!_gM)iL#pDsfT
zTIsgzrK=6=JGUBDpXF3wq(gK~Ub~QX?UQyUsKGzaFa1u5(?%)FNcz@}lp$MYI!0?=
zErf-m!K1N{4+V>!Ulyep3uuHEmGiCyRr=C+Z*f;m{F_uvefK!n_48%}2}GU8*zl?g
z4Tn(%kny>9SX#MyUjIni>~nocj7`4a1SJ`^b28x~gyQ>KTIEG);nAHL<uu;XVLQ_t
z&>MY6%ar`{sf@fM>tpsve1d^(mnpUT^ii6fh{2v~{FBL9X~w%r+i}WVT=<?=l@s}X
z^K3mS!vedO6EYW<Dm2ObT|>~w9Z>d=?vTsChFHzr@UWbf%&wj=+U9&$dEbeAji`db
z{(z{~hKZaQj=JRHcULr0M#D^g5{G1iE`__Dq*<RBMk)Ma)Y*(JfwfjzHFd|9pMOE9
z=O!0Op6xTM{`K2B0^gF=o75b)7!YEdkl(^p9JfmaD?Pew;E>(wN3PX9*{%S2dGr8P
z720x3wWh@M!%0mI?@A;Y8c>&HYHM(y{W{w1)Ie8SQ{{;}NfL^mt8=jFr`PmyR&<Hp
zrOX5!sBzj|*V{+ij7~;;oDcDmczAZko^}7S1HXQ*LgHg77FTHJ9lcqE26C*ef{tGM
z&##ImLxLW+9{2QT<h*;gwO`7_%ml-E{m1_KRB#XlqHiPRf(r{ky5w#B_yIqe)T}u{
zmmke8=<iI%2D#gGN~-N_tGK>RVhvMZpj#e3bcKbXJLM&;E40={w!VNS=xU2xR^SOX
zm|EEOQ-r!`K-eB3empv@Mxf3{%XA^_N2&JK9K_QFiFw-QPeC5B!Rl<__~Rql1Oazi
zBWzsgJLw%)dw_B2)U8n$zDd*D`VoTj4dkoB%FC?seu#QxDY$;eNB-q6O#wfVT<=*^
z%~Lv*M=`phZZeAT_Ydel{6z;CMyC2^KgE5nN6#g8kLq!Dz&iqpP0Uy6?JXW71iV0K
zd1fbi?s8&W@}P=z>iqnXw{EH*P8(=MzZOcQrp}=tonPj<Y}8Yxr+^I6cxQO;$m9IV
zX|a1A2y4hx>E$dQ!{pp1`j)|)b3tjMGxT&TitvP@w_&iOH+$W&yF+*h0xk+LrfmCi
zI7BkNlB$lfu+U7>TkcCMLbkJpb~Pmo6<7?J_hxho5s_yPA6)mjet<-1eyuFG%#YJ9
zS1n9RLbhL;CV{uWfAAk^!wBz=lZB&9ht_n;twdd=A<N+`myAxua0)c<Udp@W$>Iw9
z99FkJa2MYq4$aDOqbx}6aIMZV8(LbDk%RS)qb+z`b}7`@H5BxNU1fCR`0SKW^s;i#
z-%p+yR*D@SlBH*4E>wB}DbCSacOx)lef=)u`DSa?AA(LVdZRw=lIpnUo5Ejvu2kU7
ztrn(QF82v+B&io*g8mXr5D}qH0GKFr2;rj<$V%j>&dr35CiBZ_sOemgXp~F6PjLKa
z)i}dfOP1qqFy(#3oyt^ed0>zAfYFWknwT$9w6NQp(s1NFW=$2~uN<<VKLD*Srn*qO
zwKSru2BWTl>xw#7geb`AIb2)_Ep_zE2Xpg}OeF5k&Sfvo-n9x5%B(oPiD=5NaW`5H
zi7m49fxO*!>Cra}W|y}v$KB@z7}LhtR#+0EZ`^Hb({|{9GJohaXu0*MfRrGmRt!O^
zuF#;(+ut#9%wze3$V<Q@SmCAAO4UMA4Fi9^Fc9V3xOFbYWNw~AMU>m^;j9wpKgS`+
z_-jly^4bYHTZTReF$7=FA4zr@vg|qLhRj$P9zIJF6&-&G8j#_2j5hhTeK8^!l*Jp*
za)q1&@YsiXdO76x)5a`@dfKT*-weN}hr2Fjjfx)zo*mi7FRR~lQQEy2gty;p9RLfj
zNO5t}5zFI}*yieCu~g%NFqk>2_&E`vMK_^U9;H4GMgKfK<*{0_uQ`Qg0+AZFxB&a3
z-4|74V6rO58~$+b^91mDXuHI15v73H@mh~FL22^L?PJG^oQEJ)?ON{unQ^Vrq~dEC
zf5|cs3U0j~!!mwap7ZWXY=1$tFONmq*N@kC-R5{3=^6CW{V%(=>teYmIP<MlOM366
z=aL3wOn8?-03xt$pm;xg;8_XQp2Mu=57iy(`E|!LHOa^(YShN$fYq5paW9$N%UbD6
zrrEm!jCN+m<kF~;ZyMwIehk+cy1cxUa=TOY0uQ>Nk$D=zD3xbJ-r8)WUOhnEltrE5
z#i+1cv&0>My5!8?Isl9RHvB>Xy4l(axo5$yIc4b_BxcAzZ(nLv?~I5FP2tZx(<C5?
zl5?)orRg|=h4HNUD&TGIG4I?9;kU`1pOc7;kIhlim$P&JOg{+jf7+Ql5x+Za-j6dO
zL?STWX^PU^f$g&wYXNh+*DY8Ra@l@*nhRz$)bL(xnPUZ=x$#)kOX>wEd$7~E1b(4b
zO*b<ZKvolBhMZO+!=f^7n6xiO#iIRreB&C!w!gp9FB^}Ui+od}k*k~mwzD`MRdSp4
zM?p#er(dHOP_l=|0%Wu>i+A~~#!E`~&3@6$g#|G+oKqF7uWOmP?&hAOGAH9ND>qO%
z;gO(e*b4Ewon3_~7r@c@HCK0E^Q$8ty>dY+BnH%)j`Exod8el3Hh0~ru|R*JJFWHp
z(7=0HeEv;f&uzaoWR|$k^W9`y*w6%5J>N<Y1VG00%Y@B$rHAgM3ynh!61*SY8{hN2
znJ6+B*e6rHx{#0>qTA)Lwu3?+U!wL72<(hl@r$d6xs&N8UU*aDdsA&k#$E$r%zLAY
zeiYGPAH$fnf014zYxiE=OAuHB53t=kadWHCo?H3-yGLLoWK7U~;{tQNHUEA=@^%q}
zsVRTFXXeEt%B@23{P{p3TuqIIW`4l!+v^z~r%fV^sJ9O1ChmoRK5rQsP@f9M%&nQ$
z)=>fy0<lGd&kIesQw0dI(8S5}Hn1#b*Ln?P!O_oyNLCjZVgB;4_U*lX+|NhHauEuQ
z4B$A5d3^;5+=p$;h1+pqvAh{|`YDJfEOq!ltSu2a4Mry7&T8?6VWR}^J-wXhIF*>l
zW47$eN!{XCo7omAAh)iI9(`&K7RT3kFgJFGWG-yV;Myc@!xQ@i74dKWzT@^K7jMh8
zm*++2-~hAc$(%*P=U}BZEjnaeTr4^_#U_z5W%PHXzeqRQeq+FVLga{^O2mzM=v1r1
zA-Yy?BXc}YeZoaT!|P~(xENk=KVFSJdiT!$!~*oqq1Ke2Kz^)3Bb?8w_ejMJ+NrU7
zqeW;wDt(JDpCbhS;bEc!XLO*}s2a1jx^T#S^qNEO&D}8h_Ktyf!<sPVgtp&Z%mpuy
z_u&dsk7!YhLU8%}|DZ~e%7@w<ZeCa=nQK?|>VE~lkX%mWpXvqkjt(|=>WSw4R-AMF
z0BWD;{1y(nb7dG8BVw<5;d9ZP@04T&Q)H_<A;=Jk4h!ALiH-B+>Fiv+!dhzT$`phN
zaBaitGUY=*!$uv2lvflI2YqXf2d00YFQ&v*uJmX63Z9Ayw<qt5^|%C_b>KQv`bDV$
zJFGU@t-(Z2E_B~R526;#06T&ggfd70p6v_SNl40Sd9Y&jJ0TF?RMbDs6SI@)ZDsOT
zfc1JgKtuD~N?~a-XWq+1N?K}HPmDufHqQ{3o($t!1&JUTNm+ZFDfw2a^!T%#r3=qM
zJM{SCfGS_3`j#ZFoy&p({JUMax3H}y?N}M>mH<m|lInAKW!sG#0?-Z)A@|Z~?q_w&
zsg$Xk_njY4pE{jCg!K`2#tI>MF8c2&XBFS%;nnmW&_>6x0HeDkWSL;T?KbTWWox6E
z?MpQ66RMUR)AaOe;Q=>k)E3MV`-f<gl<E|y@OZk{jG6~7EW3=nT2?`b9%4%J7Ecai
zw}^U4;-^<VV-QHp+0G_>VZcf;JrDB^4C2RSG{=0SB$5U|4t+Y<>GEJ7*mSa-rET-K
zxd<gn_E(2k+BH^y6CxzS`&|EIpGkE1)Yr4Q4DcYY{8n~G3buLeH}x-I@M{$nbn)}A
zdg}uU3AD^BQ&SHm`YOXScmEWtU5vL=qL2i%KTv@DSn9ZI4kZ9@DZzDJHs7Pl37tx#
z&=X6XSvvoV36-asB0b|-XdO<sG_U^h*5}Uo;G3A;(ekO4{<?B=*Y=E;%Z-!?Vc!bM
zF=26y-}Eq}q!N)AJ@@vHVzDwA0G_PNx$bo*Ie_HZvF-PT+nwt!>#;rRw$%uOVxw=N
zIb3!Jg>{7<i_%Kms^(U(aK_aB5y53*s<b}fzUa6r8In@F9xTzmQo{@N5lYDE`9Ay6
zT-PMt&d!A)MW1(;Q&kpp4X&g3QEul~Msn5VE7Co>5F^!^Y@`=^?jT{c)P)f>yyxzE
z;y3ENxM8wzZ=^119LiIdpQw;3%Fd>kEY6Oa9DAj^^^XgE>!WP>#x1z*EB%_jqP(y^
zZXUOlo3Ors{YlI)72h2fetyEmQLV6@oo^;Gtu3rO_rx)x$v%F;Z4C(YN5c@|@_M4&
zXXt-+F+F;j#6-JIZB9P5lR)UPB-5Clf#{K+D2SPCyyzlGkpehZ3`X~SZ`APXnZ16^
zTWIGczc~4_?COp$O4_|xjk{(Hl?s5P<36Pgiy-fMP<`>PyIo)xBYigPjbLXLvb^ok
zl`N}?GDm%IT-c+vVL2m7J-4qLsZO?7`KRpl@y*j}E@V`J%}R;~g--3nb3XAnzuLM}
zQ?NBZY=E8&Y%ZB`czC~fllA<=RS;Hn!TY#WIkWc6JCMRsM^Sw&RAX&V2(Ur8j`xem
zt7uKQ9EF&w_oUv11@`UGPjjrQsUOM*UeO0g_W_K;PDA|A&5cAOhQ3ex0}gY2vlLD4
zSQ?lctOF>Tf^%(stBGv;ckkTi&O8D#pI-9+Pm3N*W6RBl5K^KX+ttx{xP4Om&Rq^K
z7v~79dS8?e<Yvs6btC?a>u-tL0R7yN#qmwI?-fgaV>awNQnh2MtOXjvW#Je6%&+I_
zS|Ud^dl>|<5I&1EF_gvKWjM0&TY)tu<+cq62uN8uB`?_)bs5xW0fLc1Q%J)|xSYc=
zp}X87eFgcEg`HDL|ExB-XDBWP_`CEolcB()>UY7QL8!#w=Yx;!dVs<3ssb6Ryu@W;
zOXQWcFxncijk60-rd0A8+lzl$t3r#_-SI5R-EM@L-pp8MFBYsX(F)t_MD60|)NAUU
zu5P8wQMeT{BtRVi+8_V!E!mS8m;)3^SJRhbRhvh|-u!4OXr-X)$)E_iZ|H(sHzY`p
zq<_}$3pf+aG&qgT<O`A$PbUV3vZrhh{rh9SY9<5;a{$O)W2?LJDCo}WTujN~&{%)Q
zg2!Z<{yZ#94>M5ql*Q#7Gc}cJAzu>9Te>(=v{mf(W(4Y;$vGe4KU)7bci~MdbC-<r
z-MernFIcDL4UD7c*4U$TaBdGmXW<8XarrgZ*au3SSUTM>CLpt`#rS9t;^45{9l|hR
z1j$l6@WoeRd{Zz=3p~!d?&#B13$LSnKA;F0S~r?f^z$Xu@%$@b@M&bQc>9?y9m+s#
z&a-U+pj!q<2Z$)ta6LrmiK)gZNFl&T1C16u-$5NSOb?^zNwyAM*t9Rt(Vu|F78;g_
z#fwUb*KT-*Hro%BBzZBxmi6-dpL==vi*$+Zt}GtXv)rXjN(RAFTIVFiQq9d4b9)KQ
zD>HeN$d`BB64gn5s<O~Y3qxQ-{h<N^%by=U3NewSqU7y)XOofD{dK6+dtXm}wDcv|
zmO~>*N*UA5W<S=1&CV=k58aq7$UK@{%L$VWyzrVaj5N!K92sxcmU-O`M`UUWEicgZ
z)fy3!xG?79m(cLiR7LnCg^1#vl$3lHii>5hXnei;?dMfNXGjOQD!>?ilSaC53|7c|
z=Sv740PO87o-4+=3NZV#X4{GZTfshJ3CNEYKYcq#-8a5S%m^Sl^K<Jf9bSBwnE7~k
zl;u5g8zVxQhpuv7fNy>W=~;l^U0G{SueCk0lb<Z4fjd;YsA(l;I~|MfO3%b9AdM8G
zD5%?~>4D;=UiEL~jX8=jDRDAogBY3%wyD#f#L*?X%e@S$dz2T&o`>+$=v(FFSYAKH
z;qhpcVH-Xu4}K}hEF|_U<=Hd;$21w#Y_TiB!5F}O^q$|ogs8Tn0Cl`jNEBv-<;<>M
zAXKAx|KL`qvCsJ1syo}^aUUL<?1^wfxY6sW@bTBgMD&8|MPqJ!a0?eqE6GNpf3Wki
zh?dd(8MUtL1;UEq2yu2I&lt!xCFDhFNfg}jtVjPn<@d*JPk4)LvA%b>SIzcMyH6on
zvA4M;O7ywjVP0Hg=hC(SbAz>u<y-vy{&#-j&2vo$3m&r{-HMwX4G5~BiM0bGjh>45
z&1ur7z7y=nYL&0jJJCnfe{nXo`o`79ML=I?aA@bdSqz#lgr*7+6Vt}|`7wXIj<l+A
zUv|=^+^x+|uw27FR?58{hN?RK?Aol+U>XuCI-kCMz5UUWvFEo{-`_pVldEp;Pw~pX
zIqmsJ9@*~Qx0-|$p;mKqE&|#Gqr=@C)xie-{hrkFJ=P7^XSKEdVl)La3ZAQ@1IX}|
zGQa?ZrOhnNB=EiR;i{;UI&3WjCr}{H&T7P-O2`SmL{82{2P_Yu8P&&*oOs=zLQZd{
z4@}{4dzhiXImrA+Q=QX}T-s|QqPw#TNamE3$|5)JQ%Kyyg`5i9ZESlY&TZ%N5Z8|9
z>D%irE<>bQ%A=W+${g6sGUl?`PXo%oR|!9(rF%0z;l#D&{q-YRi*cCT^X&L3_-R;G
zh)&GX>+XtUx#j+V=_ob|QH7RWO3z|zQL_Tw%&gqIkupKopFDX|-#!up^K$33Y1#py
z#b>fz`gPEcC>1jSNBW3g9jtgVGKHwZ?<(I6?UVGbA03LVn&Z8H@Xz0$pa*BXtpD74
zTA|amu)<&bW}8ojg&C`pSq_I=g*lS>PapRu+3x9XqvlOJr{DdIG-k?4dQN*|eGOE^
zw6t9enM(PJuU{uaRFgBfFf7Nuu<=ltbj{cpNJ$gkHn&6Yo>J+2R}(Gpu6B3q?%FD9
zMPzG<jLytt>M$-;&OafW9D9x=qIK@0Wb?+!E!miId&0Wvtw%E+yICbuYgTaLxTvnp
zk92yA<In2Cuk;1TeoqV!xV|`9Ie)_u<5~R$mL4xH^DJ9^Mb}TG^IX3wcvr68tIC8D
zoD}Ft#OQc0*hEQ6ekG^)q6?QaG-tx$(EUA~&T`$=qsZ!qNVRS}rAYbW0MF#fF28dk
ze#JIFZ43b!zQa~q+viU~w2%0LZTM_VT%#Y<y3d=#-FT|^?3C3>XBMCru5zmh=nm!2
z*UcQbWmrh=w$6<!z<!sA^KW3H@;W?r?m&~Vd=)=tPAGL{<z|EQy=Usv+g;8~Unf^c
zRhXE3eA#@Pl&XP@l8-ar%%wK|bO0*uD*5-2{H%K?y@B_fShlm(<j>EL3fFB?UuC9l
zN6Y32^)p8&hsHt#an^f0LBAQ|Ls5}1+D(|;@vMX1r~92l4v9@qo9E3<o1i9w<-09w
zU+`DIQ#A|>9_0lGwsk$^p~dGg%Y^f5E|UP%O~zitu;n|6JIUqqo{j(86<3p=2SPJ-
zQ)M42>R6@H)n|;Yn*tK9k<YA4;CpoXzvlg-so0YERUk?SkCCtLEfWpYCJVACbjFVe
z9iQ3nspJaV2TE*yP9uw+ehnGs4+-W4RN}tPOULG7+ix5E`F;{}Re+lM$Jtj`$s@O+
zPuU^(*-O7>NqPp`i8;(BV49IKjf=NP?Weyn#E{J{ca%_#aSk-&OL41UMId+$$+ax2
z$U9$YJo*$EklCr?FnsdEBXE^MXWfnJyJL)$*y;QCVal;unj5bbaU_qS<7o>EhXv}!
zIkG}wp<o(zpQ81j^a{E`#+}+WPbxiz+(R4=WbPS>Yf8T~q0hK+Lu4#llo)>0+uFFK
z&?l0a8+fe<_XJ$L2<li}{+h(^xrE@sNVD{L17*IIxI=Jr?TK&Lmc2+`o@<H&j^v@+
zjERxg0`<#<`I2pCR#|q)ujsTxFxc;9;n`_9`9SteLt^c}$M=2D)AI@F3jWV)Uo><&
z>iH+<|8#0!ud1|l+LNoL84uB;wsr}t!WwLnl(SrFPGp)H<}@d8W?*0)DGD$nJ6ALd
znp|iO1Q#{*Ka60BAWTf9a9uyplGitC9?vHCsOwJhCi)D$___XDX}qh`W<vl=mfa^@
zPx(DEfl3#?q#$`-?7>lRUQG}Q?<`3ebvTXDs_xC>W<!QmC2L`0%&i*Fg`6I&W=q9P
zRZ(XBuF4a3J&iUII(8{T{b+59#q$XZ>C)+0@w9JmJ6t}Y$C~eSO&k2a&>Vc#;4Irg
zby<l;JzX8zp2Y1GbvImH6q@m?<qo+<#$m)7UgY`ffkJ2e&%n2X7cJ!a{IQE&!E?ME
zhkstA=^uR#W)i%n?(;AGN=m!2L!muSnmV+lpTU?4oR?k<aC5t%WEoF#KrJhz#P|lQ
z@7`@eXPRx9mOt=sZ;L71Fxu1mu32EmK3Un=vJZ(Re=83AG1*zcxxgv5G8As0vEZTp
zVk|CpOL|N_{wrrpWUB9LghBmg;Xg;6UPwK#R52oCKW9V4($lKrT|BN6Z|)H25#ar*
zVy?iC^_aCWYgo}0zQgN5%5eibOkeqD);^i7h{zv91^wICpZ)#m`Ea>zm5}HTkpIfU
zIx~%HpgCNkpK;p6O3kb)MJzEskr$iMfm@wgq@%lPe>sq5fa%LSs%xKHS*8!KNl{kE
z4E@2^7^a~N$qXm%4$80bwW92v-@)|U^pR=Yq4Hpg(%^d&-w$JA70Gk+ftK34R|0D&
z$hrf>&jKk9a`j)eNwX$hY6+kh&s0foEeBMjq;&B~a*?k)-ZJg{YKY}o=Y&($l`u47
z+Wyl``t8(D+7a7R^_p&M9+Z{Y>i8>fH|z{a$!Y;>xxAv8oMmPy)47eNUKK(Loau0G
zGJ%AM_j1g0pHn_KHF<xZzZLA`V+8u}w-mz@+GMAMxMIQBm}2vvj}w9ZoD@v2!tw>M
z59E}{zcwZ97dyM3F4K5R;%Hl|F7fN2()Qr5r%irdmD?*D&$7@6Q_npTPIGgiv559L
z0{0o+2xgLyQme+Y&#88N{QV=6SLH1F-=~Hs{QTMSr9^4S&EpU+!Rjo?nm+T%N<abV
z^W1xaA8&zdK$^{zSFj*c-y5F1$@~E~cAurFa?Pc7OLuh*j!6DztPVWx;Oael^QcOK
zz3}`Gy{A>4mZ-bthvk{0Va*cr6X`S}+->z-hJx>VLG^S*xGa;dgc`!agD)oxx@sF#
zAFt@OB5%|ZRQv+hU>I1wY4LSiH+XHFR?^w!vt3kF_yFQ!@%v3#7OJPCT(CVtL)m@6
zR;84sDu)^J>z#0{i#=a7)BmQN{gu2YR-@7}XXqZ90kySony9OEt)%W$!OYSVew6U>
zK*i%`(eg6NJxJ+sG`GnZb+9`67cV#8x9_>tg9qZl2~YDCAB0LD9LiiTxG$zu5LwNP
ziuK8SelQEIVuKdQMEY|K+2runmR7&-9<;e`F!JdcA#?h}aOneYX>s@O{HZ$Z3)wcL
zD}hPn%G)|(L)oGw*RCy#-mK4c&bwG{c(Sr?>-%a#MhmQG?|{8~ubXgaK8!^`p@>LH
zEum$RBE&Y>9K<*=<eghh>easgQ*1xEI2f~J1zA{QWwirCV&H6mRq^sAt0rpiN~hv~
zy5IHoi}xD0V_T$F&aO{yd)E~54JHVnGc35tOi>o{-Y{Ho%c*ScvX~mGDzVs*fIK#2
zfnMfV+wN*5Re-Xr8xF6n7;i$#O8M50R)??mC^nm`Z0-#Ftm_htiMdC-by^O6jr(fL
z$@Za(<;<XU&KB&^@R=EWJU7!R!zP=_m!Uce1#io8Jr?8>i~#nqWqWuBpH@O_eX67j
zojH184ig8}jmhp7ANAbH<3ylhLxvD*N-Q4QcbQ2X_NWZE%0yq7&b_gs)nGoPUB2<G
z?)=B(S64csPOmm5X48?$QwY_ykWMdtWka)Sa5yG*V>|L6_x#EFcf;3JQRAs9Fs2>-
z@|f-}-CsXjHWY|<`*W`6^A(&H2g>-oKTD*RL#XYaRwt#Xwtai&g*9xqcYn^ss>}HT
ztDFhA-}g{aA>CV=oVLT7Tpz&(5QHyDq!ab)MII+!r2lTS0#@N0NHIxBBMHIX)*WXk
zruR5PMlZMW);YGu%%2fA_K&8dq}5}o0eW7Z!<jgI3{AOWjvo&K0~r!i{<BAIB!7}0
zyq<zfXKZRe=gb2PZs68_Z&i}Ud*NcPKNMpMNpshzc=fJaJ@v%Z+G67tJNEAnRg#-5
z!vVEB@_;I53{5_#sLIGFab|`oOS7_GP4<bST$%I=5B9FzO$<+pW5Hcf?_G%;stAqf
z7^h`)TmMCV#J^uZG)TchOUotO{yi-5*&9hxEalH1&x5F2UAo1!&CUbLEUy`uKDAht
zmtXlb*rmxZQ=x!yJ03Roo+t^?`hYX|sUk7r7l?nnJ1Q(_`+IY#+IISJosU$}iZg4K
zb#q_eWR!KMjvXg|s$9y0l;lsTxUjow^rAr~Twq^%?xPG_=cOERo@XskxUXRp&s7os
zK8(ni2sdeOrFCrLLrjH(I9Wf}y(%|Gg?ZB*o3-GoNNggyw7X;uT%0};k2-TOWjFV7
z$+-z2eQk?D$;spw?2Hd-iC>vzQW#<KnPHpuw#j-!gb#W~LcuWKmzDkl?ml<blkJ?M
zGV4xJN4gbDQ06Xc?&?xHNJyCCl+Da=FN}E4-?xh8FSt`O-*X38#fb@>-O1rwJV~ji
zMVqSyqmFBJ24m*(o%3cwHwLtXQu)dlnna+<e8K<@-{S36H$4h{4|1Y^ZBe(c|6Z4%
zbHz@<y^r5-DaSKDBwIj|g@%ug{kAgdM1OZJ#>2BMEgK|XUz@HkDiAkwX~8D7ndn+-
zdCWKUU6(dLHHo7;@P5~f#0|qNsjcRwpYbho;w78Em?m}hD49kCS3`md?wCD*!h2C_
zvg&ob8Wzu$<HGM~T)FpqWn$WG!YZwR--bq5{)JAb+m-tjM9$p;vc5N`YZ7(u)fx)(
zJ8Z3;xQNrn1ZE0fFMRXGG#x3E*VNk);V_i1qO2EO@wruWbINAo%2Stp)8Uw}24=5+
zw#KLoyK~3VaQG32g=L`5YzmRw-4H;_v{u>^Bhr?IE)Hc={=dVWRt|Sj(QBVIvQ>>|
z=L``BPwVEX3BN1z25s#A(R@x6_bOp9c3PBzLXTwVy9dwDp@5L?_=T3pCzA<csHRQA
z=2v{J0-~a^hUUgE6bv8Q`&^|JB~Co2llr;lRlCdE*B#eLCLu^f!9Y~~{sHta_v-xi
z`5Cl}_f!7?lWd7SDm}p}^(YjJZPoHqJI4chma4An9B&X?!E2L|y%SYo+Nv)!9a*z0
z1RqB*$BvBj0%XjOb1d139F3^0iRmlnKWwD86!3np&BQcIO4>W#`Z+o3EO3=5u=EPA
z$`w2r9w#sDQU0~j!(j+c8!7!bTH}pqVto3D!19{MEW+^FunITUuq}q+c!uaxH^;B>
zmfPZ#lzMnKe>KlfdOV1f-Y74jXAz7Xh}Suph@Ef{;bF*V5+E{oEi2AYwnGa*@kWPb
zpTDePHu^jKEAg~JNPv%9kYIV%@{GESOpp<7j8@i_R~E<{Xb)&G!##dyJQmmbP31zB
zdwc|(Z&RkTH=^y$YKg()$UjC#@~)<&{NZ8d3mA0P*%@vaH}6pCOX3OIqTbkl8=;@#
zewTpian`73$LO!`y<Q{gyD1i_sL^39I9IZfpJaCmJ0B^ybDH-X>?n+dXV)_?h|&cE
z7AyqN^K7KWs||I=FpDZFm9bJ|MBI^;oSIii&T+Li7(XmNhEZAa;ka0?`VwyzYpa|j
zpKsqNnsbN;j|Mzy^>+^k%ItJ@-Qc|TtPZpi1ZjQnB+Q>;E-XumL)LoWYX0iTJ%90@
zzb@@M@inc~@Kf|qUUDKO^S?Z62zb}Gl0fYt7O;D)z+m@p1w0UBi38^3zEv*prA?$2
z8Fefv%Y`t{?6TnIak_Bz8BIwT)095moQylER(kq9nRL+0+~Y%W&7<W%7HdfgV(4SV
z;w>$yPk!wtD`yYe@qH8zNdlz?fNl4FUY_~f?;im(QvGeIhI7zWIf9@s8q6*1Y{<}D
zIVR?LThB+Ehd#!Uk%nfk-}^<<pNdXwmelMYwM5EnwXu9b7r!6=F^bpY9fHR^d;jyR
zSxE(Ia(eo&6!)ePh6=Y_G=!=IXd}n9T~m*!)%M-WP4{BaeOmA)^Q>jDc$CP#w{<?2
zaWt=Yd`FI5E-Dv#bfa@SAuK^cK})b}-qbxLTwm|q^YonEkurdeIJpCI83ZzO`;@WZ
zl~w#%y1)lx{8>OmK`dc!N=o$kI=>W-(BWZkrWaO|uWwM|*oWyiRxce9xQ$yftbsd8
zM!VdZm=V%{<L=vMi2zL;XoGoqu{vH2<}E;y%JzA7*Myv>Q=3iuP)^7?TW6>3?9toI
zBHSxoQ-HYG+L>(lHBUD1=~idUYa2Nc4jQ2@ERI|1K@|Qq*|-2lA+<)N>A|(N_s@$D
zhx~L>rVrO4W6bYVRpv<E;G|FA+c-S-{uo#_mUnj=x}VqXE)V?}_5azY{FA$60|lY!
z+Fmha6iz$-V-~JrVKvm%`)=RE$uO{vgJ>ENHNWc}0Ovz&J69Z2qi#H4D0u6WDKYHK
z9V-(lV<8ME2tC5JmiS-{C+C-k8q5`;;Q>V{5A&U!dH;Bn-!hm8HF{ypcSf?mkt4q%
zbc7!N-i_|Mnf>DV^JjMsNl1d=yKv}qoUdRmX4}s0p$J3NMhB~}tH-ja!V9#39(xx&
z#G~r_?K_lGFC<BcwU3rW6^g3E7OSrUoGX?;+FYsc{On;;!BoqZ*Rh3etED-jZS_ur
ztHtQu)jZbS#HkO{nC0h=2#Px~T06Ej>MkGPs^SjlB(nc&i+Kgnx%&9A({dzTummwJ
z7FgE>M!A0u#gRJmX_8Zh13Ko7Btuo|i~t2iS8tDPBBf~9%kHj0i-tlRLxbk00K0k^
z{Rw4>6e(H%hcbWS1)=6Wa<1{_0P3|)+oC5@`Kn=xiY>nnUMMi8+Y7MY>862<4nO^H
zg=>eY!6gB-gEV87>XeJKb*r5Mp4n8(!YUXxNfuXIY-u=7RxJy8xE__HMX;#tKQ2lB
z5|acqx@ea>Y>gExk540BL>ok{4mWm<aRzSe&I7khi_GcT+j;e<z^(0+gmBTljEuXg
zs<xuS(KVc#I-PB%%VjkLd)8gcy$!=|Kxa~BVSzD$NPTa(+%OozXx85AiA#@eaN9p1
z2jG#fsfl2J_Zz^uI;<T`Dvv*LJf2n5d9)Cw;<f@n3}IDU#%8KCJM%Ybb~8Xc<sk)=
z?umR_s6BmG7|2S5hqpQ4KyR$)h7B;)7p!$LE?tYj=o0RtrKQz?U^DlzqsZv*Nfe#D
zS0GpW#oKwD5+8(1FdbUX<Nj8v?_JE$0uAp!lu|?PzZ9M0MVq_EidYx>b=L{;Z#B_B
z&|u!Qb<EX}Uld*7xNxbM9y2v+G8la<AoGf=nlUxM@<u(3=9S#Mv?mcjVKe)uI9kVD
zl&HjJpvceS_G9-eo^iHJC8`#-wh)to8gj%q{zEeUoSK@V6=pawYfLwLVz8!7XJSG;
z{ia*vntQ*#g;!shr*IYr0Qd{1;bw>CL)6v`j3q0n*ie8*1}2Y{e7u4fhg~;gVhcrg
z_wMTlkzC~#W_p-Dm_@<>e-*YJH)DuPaFQul!mn1vY%^Eu2fV$`pk<XTbmBHyOyM*w
zz)T6c@dt>f>u_S}030Sjn)%;SNSNu%8*}Y;jUQuy8swy`{tPj`nr=?08gV@38!C4A
z+|tq1_%puyXwO(Cj<ki2x8E~qXQdU8SCb;n{t|YdcQk9>t>$QKEhr>rIo)@`!EwJ)
z`de<6x5;w+NAE|7F3^N)5`zwket#8JJ9wtBZ}6zZ-gRY|Dfky-uNg*CB##GoY;-MD
zjv+PXgD3z_o_Fj&r%T`rxUW^RR@%nMMB-68IOJL|1w656c}kkA*Q~L?fp*VZwtv8v
z2<Ac{0pZskq~G9I&WWn6w!S1SO1A_T_v<rykq4HacK$#_>Mvdx8XMGqr=CDH5n#q~
zRcLAR7iNpMoe7Liy(mS3W%A|&9HOzls)wk})~${-<BPMt%v0Ys_AB%3DM1bruIt@g
zlSbHa;=f8TL!&JMW6HlI>3P4)^u1-Zai8WRZjJla-WJITDRt|3eu5oOe{Gm<^vCzl
z&jKqy#}siwsg+J%SS5G=W1^`)`dYbg{3D;0l^bZcEl$V868D+yR6x*3hOJeX@LF~R
zqj2fmcy4y7UE$1|Hv;F!XHo-*U44O-QLlqxf~VnhMVy?bQv-Iaixct-44P~hs}U|c
z9{giRDZ&>Eb2FAk4dD%?1!*V9iz&9$0KHPoQSJEWO=qtZ<B|^58@ljK%gnv_7(o(o
zZAz}Kt);`Z6`|%0TgNng0}nti-iaA;bhdkBJgAycZlz`MQ~4@brhhtevkA60AF<cH
zedI>(-Kj%<Tl1K~fku=M|1EGF1Y%?C4?=OP$7nUpI{UDsJe=jJ+kuw=SQ>kXP1tX~
zOF?c$!(+>K?1E~EyfY+_sOZreI;x}4JzEuE2-uhTTDZk{Re$DdLqr>DRi_rOVqn_N
zzP;TIvjM}>yDywEh|X8w<E!SeW+joK@1_y(FMgLK2#A#czhD)3_H;!rt(`;R5CvnT
zjGIv#9o9UxU^+_AUbnUix+eqhl_>**=gu$Igd)F!k6V9k$#O=>0G8vitl!O^*zvr<
zj-T)KeQy~zwmfX1BRbnZ4Y`kxZ*wuY^9sQo)SQ8@-RL=}L?k8ZNJyAT&V_9w0H2`2
z7<@h!nfxC|{x*V<sQRZwiS4e?w?D*-3rk&Xy%L<<Q0~-w`&Sk$pi2TuD#(#HsZCv4
zd~ihikFr&nL$SEBBZ6im`^Fmb^e}Yw+7s`;j-j{_$%UnNC1O^eDWE1LYs8a)FMOv~
z??lM+LDO!0bX@{r6+_66!CzmIcAK76CUIS^E*>~f9c)|_s8hLHo%w;+cy^yZJ~>AS
zS>^=gva%IKT#J3NGx$06ALl&%1f}d3fP$X*+FwM4cu<#h?;6$6H=inDX}YWdL<VNK
zJ#weRJ?NS(En{@6whQp<pFMh*<pW{1!(`0YF+tZgZJ9svh7PTWDtN!dRw>8SsL;B5
zzQQ08$QoWNF1GOrf!lHQEPlbMS?Q*mrx6f>*XF)xWByo?Yn8zEWNN=-Q{DL@&o~v8
zG&Ae#KZ6sqQ5FS$$WGsta@X}=T6FbcWQh$=rQ9w^<-`NT9e_X5C62k%odVoLxNEFr
zYias!0rreo-%D1@r+U(gChd@$CcPq@cFekQqpPStYdPkVMwV)|>xO{TpYS&m(D;W6
zmC}OuWdB)!ml(<QF%fr(W~Bo^pvQ9W3tn$<A84(Pi%Z9r_1(dlt|37#C$zYJNjy2y
z9;j#e@ovXluWoj20+0;~+BhC<v-V<|sIn!-;iotp`oB$;xzw8^S!p=$nP1aC-!CgN
z?S0zGJcNxLO39&z9%}07btN?iUn{gcc(7_QLYKz}SoT~B-n(F}%a(tuh>TaZSdF)V
zu%;8Jf9xu(O=6H&YBtM?G73v0Z1}A5c&+#N0|2deuhn=qTB8S=F!jc*fv-qfCu_|e
zNevLX7mU0>*x}w#W&!|POW#x1B5vcWW|v--Z+g)kW$5$A5szEjqrbyvHB=WxLL8XP
zceF}A1LB1TDU;ESocTYry>(R8%@;mOn1rY_2&gnjNH+ok9=baOq)WQ_hzLl_p`^Q8
zx=~8HyIZ=u?i}B5-MfB&+_mnyE?qCLo|w<fp1q&_>^<|ulyYzygN-?^p;r9r6}%+Q
z>1nvC9qP~8R(u>5y)H)2x@RHf)r3>{o}Usjb|O<0av}mH8=AYp5vZK?BaQknMOgWq
z;wRzi_{z|P)e`Od0DN{6^1}fKv0+jA0cvL)v^^vk`xQ-yT;R}VcKG(P0^kQKZr-?z
zT1U>RYwvjg9V_l07PGM%?zhTOSx<@R>xK1lRN5U9mYd$Ef0<fb`QwARvfF=ART4<t
zef$<dEiQ~6r+a<8VK$F_m$P%u50L0ulJAxWdo-o_nteMM=vEyeTQo4;-P^fyZIXay
zcd2J{U{WRJ8jctm>Z0BpMcq@76W@}OKH4xai^D4`t3;6A{P>J!!RLBYGB^?-JK#T(
z2tqxZ)L<n~#sxDRI(nVMPW6I3JU#!Fg;bDBiCwbJ5?GtMF>-=WNshZY>{fj)J=LV{
zA6|m&HV2ZUb2;y~L4h>8+KqedIJhHQe!YaF1#%_D_C~*mexS2YExW`0HcTr77$1I8
zzeT<y6<zM=@>!?1AF^|iIDUmJQE++56Um}4Ll@D?b&h8g_d(Ly+eKYp^#z2#C_hma
z+a<?~qN(nJX+GZzXWM;$aT6=Uj*{Y!Pgxv_AU3P@o!*IE{DIEq1W+NfKeJ>A-R5b^
zB(C%ag|M{r2=j?*i)z9iY_~^)Rue>zTBzoCM}t`B7*fHB0}d0rvXa%+PQ$A4$)|C;
zP4j}4EC3dGV%AFlM#{965`-K(8pg0{ahbu5XysQ81OabfF4T)>mMr_M*`E}t<U#5s
zx6hFd-JJ5q;>&WZYB!qn>Mu>qNZdWR4re&es%na<c|txJnmgO}y{MQCT6L9SB&5uE
zYq(ataKT`-Gc7xBZO)zeO$yNe@s~H6S2IU*$<qIEy3SlJROY`#<P(tMhI3EG`Zix8
zQ&k+S*}75zI=eqZxr+*l)o#A7GQc-`|Fx33f`<bquZx+CsT%{c<98Q%W7QpE{_r{J
zg@bJqBr48DQW)h^s`8_?@KXFU0Kde{C#p=FA4+46(-s_)i4Qg0r6(`6S(f-dx)fMx
zb^kqRI_izBbs*f{4NN(rc}N)Z8y77IW9OQv-kmpgFaWJCzu*tFu_4tm#H!|e!Kdq^
z^%5jFp?zBOORSS!0U)U|x_@tPX5JGCgxp0i=&8O$1L2}X_Qliqu_M-;NI|P8JT1=y
zc>z9>Uz5LyU)4YD9i*S!Y_3t)*b^!$1Ytc`>ZP6*x^6=wPFvzP^egLbnuN7ouzugj
z@JgLG9UEq$vKY)U>ym|<6Ed}KX(?y(MJ}B<He`C%ZQ|CEwvpbDGeMR8L3qYh!|h}0
zr{UaTYHs`n;C56`=up|U+)~qcu2NCKUq+g|X&`a*#1S)gY-W<1rytOV=KlUInHfuI
z?YVpRQ-Y1{7;nx<<}4;V7Yn8kR50{*rk}{%5Hf0;rp0z~xdPvQbooO3ze`sIQdMV~
zLTU<qitvGfi4ys1dndb*qv-F$I75R|8Je+oR5>f#D4nG+KG@sZEZQrez2LNBewA&f
zC{^@Ri4J$gmKzX!WyO`2O1d+-JM#7txuZ)R>Jz6eAdh2}zMOu!U6k$zwYQ+$-gf`#
znjHUH5pwdy@PiFIV`A@(=jpSeS^K=&IQkJ$hSk%J&dd6mgQW~+&qu+}%d-ip6-w#G
z<mg%_rh@?Ku&_<4T~+(Tnv+wNG0WI?!%Qjoj&-$`jD-ddsY+Cp@(8J8W!e1Lg7^2i
zJYtZqaYsf71DD)t%iRmD8l6o{GJ}k7mb8btCeQaH_z=W$1*czFJ}}lZ`*j&Avy`!K
z!1ZyR1>_F!Z(%S;y%#4cQEX23A3m<hU|?=cXOq!lq$@-<9HHn6vc@GjZtGp|V^bI#
zD{85Pq<CSNv)a{Enzqtow+&?|Uz`6?fppnUVo-bRoLH%$P+}MiiUaJLgqi!<nG{%j
zuWNiJn?DjX{xRyyc(@Di4Sk{mSw(o9^Z^qi5C5;5C6fVewowaE5-zQ+&T8r$RfU_^
zTHWk<C_{$_vHE7Nl6^h*`)j9LX0-Bp*Bxy;MI`sy3_l>v>s!ez?L4khGwIDrhSJ||
z?*GVjE$Lmw)Ld{8WEDZp4+^4*FrV9_%S%o+69{e;XIx!<gU5*-Zbsp}#L7}0LXN6~
zhPIrMd=c%}AA|~1hV9iFF0ajD+i`MAqJu5N%*zaXZk0+8JAb8K5;;99E7j?G&$2*0
z@o-^EH}J_G2X}Om_)=gOc{fl)zvDKM#GLu5YDo#RvBqtK)aKl&o~-D94e<?vGwC8E
zB6uDyL90q*77TFf)5ru74V#Ca6&%38THkevS*%D<B2kHnx%P<WD1@nhlLg!LQ3cTZ
z<{EZhX%LXg)j2PcqdAUkpULg(5bjfUXf<3(Us+8B{@<p&w0UwG1BQyd&Y|pIv{B6_
z<}$+mI)A06Gk3?_peTW;$K9M))j*^<_k8$fdPKEaVaP8dD!r__#3npgQJy^{$EIg(
zs-^!=;`g@f#5@?egz6RUuo*kIk0^q`HPQza?E;-7kMLY$p|NuMh%OjCKy3pQn+0Zs
z(AsTLg6Evje^SuT{c10N(qF9rW`WVMBHd_B?p0BpA{KjAemwpo(iJsfmh%<d)3UH+
zbz05>X-R>tArkN0^!LI-FOYhO<ir&-R7$sx)JMs9fBx)JrJ-eNw3DKW9HdWXz;-=d
z1_jaAtcAM#+MM^@kQVb+hX|mNwc*iHPhtgRP(V58Dy-i^<JL{n^6^VUC9ic{SYHr%
zcyw5x3Z1B^yrY2UDxh?_>8t_arNffavSHWhU&9f=k%D}GH{W+yT+HU7ZN`Dj%TrH4
z0exw0;TYf_jPQD}caUq-J*QnNuca7-t;-$~D%qoYOJbS-`di04=oO7{h6<+{yZiAu
zTWRH3ZRq`VkXGekk?8IE1hPwfyi;lM^ZU-scUCj!g1+(V+&)%;+gUezkW8;Egl&a{
zkTW*UCmQ@M`y3iMzURH#w%EDSPTGC`pN%49BEcU07Es&C;&B|=UgQt`k?L&!-06va
zWjxgb5gc^U7qh>9F+LLGr9*1}(YEFH8!_M<0vWY_uGi|Ssg~>Y|5TH{B(<CxHlGgU
zv>sSFkB+(6H@4ul!^=p8KavT-I68PfK0-)OE+X?*Yh{HD${fKio%ne5q5h(vFw~WX
zF~v)DQ3nx?KNyjROFT4NP&IZ3+t!*%R|wjqPF&~;j;l`m{??9O(zS*&&DzdMV|)T0
z5!nMu?@y1ey1GG)-WIKVEW19*yBrrJSXuD0vEG6rZ{@Lcr%xxg$X=cTNY4T!Eta`r
zPs&GxNJa`I!-KFBqGh!_kpV9Wko_2!K!dRpMU#I^1aDsSaf(9TE}!hs{b6^gUWYEM
zzrcb1fS#@@A}ZySYwz|nTYPFXcq1!EG%NG<^7bjg!HHWwziq@X_|r+Aan`0G@4_Cf
zk1U)^pdF-Jvzq9(Jy9PNSF=YpcrG!`)<eCd+QPn2fMiwFmK%|cg)|l|`AO=vtl-ZB
zI^7zJ)77nN0sLg)i984CqLJ@amV<nN0C1*s%5))A6o3wRvxXNuMe*M<?BgR!l|3}9
z*6*a+6ZQu<jlQA>6BB_TS|Bqw_)Nl?$NqgW0Ta_*r*uhH{0~kA?_L!sePm|z>gzMr
zPnxy+%G!~$RfQY)_NCs3v3Z_$EvIFIK6mJnA2%`*5x@#|gfLDHTDtdc1@B3e$9{`o
zYiJ~B3z>S&cQFyI)!-1<XDpaw=d=7phoAXdM^$`cN92(dPc_%NgSKbrSCr&(W4m+>
z3596}R>35Hbisw`@=ZWW?}4@tP$>>hD(K<2SqD{-I-58q5yt(slBF+NjlCgI=;~0W
zi(SGs*7~%+gLp^@-xeKeJ>L<y^5F*y09^OxNBoz{+xR$LDL_-09w0dC1{fY-d;1pQ
zsi1cKdWrmfurhS?JGC$7LMx@94IQlyF?V&@pk3>V^+K%Ma7oliKiIllQxovSxb_Qk
zPM125Yfty8IZy>=k&q~1gILod{Z2LoGsQV&#KlTLhQZi#G&N6OUiBTYhTn+@Ul$u(
zZ=wpCwEy}Qb1-1MjElcyV9-;`|Cw-Q*SkkMZ)4<*vlw3E7ac<|w(e^~?zN)1<<sgS
zGX`_{9?(oBu#Jk~PK#1h)k*PSN{){}!J}@poHew})ZcHU9?J8=Mh{dYcJXe-FGM6b
z-y-Zut>>aXqg9V*OZ_(pWVfK{ETjM?iCU+(_nE&W#Vc}jVr&CPS;DTQC9(dQp5w%2
zMiY*HB9O_HmS!HnbbHJu=9<%f)6yFN(z}Y6>(4XmvFby<n~|5Gdk0j|1BY0%m`9<x
zQ}D)5QZG}o+*mn+$beeAK`{r1(Lj_XXa!I)z6JS|4Sy}_-e}1X&1Ol!o+qtn>nk4U
z?k5BdOsJ{y!nnK6C%)D(L55N0Eg!!()}?(>*z{fL>wdtfRiAXf61%*79G&>HxN?q~
zomzE`Kv(dP4bTdfmHJ%a`EwB6ES79imYNQ1bCvY0y4oz5aHfDhgkzKfANTM&L$1CJ
z@O*gqLxZnhvn&^whL^Vy5E3wF+lecgvF|<;6lval3|iCnmR3GeG$AU_=F@fJS*Uvh
zF^4!xCUZy9(V%DUR5t$Z{Q2uNvtTzswpuW}Qp0!f?$0HI5t+?G{&o@cI0*AvrJ`~@
z9vBu*Gaz8KegPmw9Rbd}am7#;^5kRw4(<L3tB3ZE%UOD;T&(7FJ5hKRJT3uQ4d)fT
zk#%;|(s$;246?p<_ND;um1@+tSYRc@4mziL3SPNXiZjdib*EPP5v!K|aJf9AMVkH2
zS7Iuog;-tapS=I*If9(gvoLw*m#ccn5FEsH@`vJ{%esy{TI|@M?(U@jocJSTwugxi
z>)kdi!oS6Ibc1Uc3RFt<hac-a+)*oX%@C>M=RWxT_Em7u*-;<n7B?hhe04CU`*Xbi
z8_1@+d2K3$9=5<cB@h`b%#hNnK>*{yRx-qTbF3R>vp#BB!)CXzcjT{ut#De(!lK_?
z^+d{g3Jyw)nH~q#y%gWmg4shKZDBQ$^PYrhDNTQ3{VbSjwI}&U8AR&5x!62b$MwKI
za%8+jw1s!>@ZRxDQP7e!eUpY%$LkT>D}4}Ef5Gz9|BJ`#eRszbeeX8f>~E~5n~>GA
z1Z7VGIDx0@_Ar<`O>BPky)5WOV0GFNPf-|gdj?1}yUS2O_u!k`*$KE?h!vC|Dv5FX
z59rEcX{S!UsVht-+g-39(QT$$&A&Hn)f~!X%ii`@&{FCBa+ge=Dw?&?iL-&IzgCAm
znU{yx)^!zUnt@^5c@YP+u4F*0o*&mjxF*yTbCh0pw+GHBwk^I_Ii=GA?TjZ<gkA4<
zSC#0LtHvh=8$<f?{$23={G9I{Kq-7U87sJ3{6(<hYXMv40^+lptD_1A7tm-p+{D|Z
z`X?x%(Ut)Kt-7bvVL1hwY$ox5eDoxF{ooTPbdBlr<0Lj0!lR<HNW3KlGb4&$xVQMz
z;I~7IswCi^=E*ccXN7-846S(Mjh3;yk~>P*$xb`q)c_+mtmcy+sPu9nD+Seb?N8H0
zwz&nrhYDiBb=r#5S%GU?FEt{M9%9{mCV~ml-p(O6EY!W9-Qs2p#&VYKMWo$CEDtg?
zdd?t*+OxFWHe2mS?bIxLp$rj9DucD!nzJ4sa(mCDsbxX0Ll}(0-I?bLXJ{f|(b+*m
z>CGmG4L>!$|65@dA!I_AbDpVqts`8qd-t9NNxnbIOjwSJ+zh%;1MiAy@F3ePn)&sH
zZsm+qcUV1lsnsSeW;hEU<|BHfQ|k{ZcaTF!nnK?B(UzzVkNq>$m>_Agx<C(pPl_{o
z_gV*HnEFJ}$?3xOjzhFgbzGnicN|raFkRT*Laf6w?$}LQi~rZW53Bg|pc!)GwaDxI
zC7!vScT@(TgRV+b;~Q}KprnNvoAuvyW1|e4$0ZGL$x~6xk?Zdpf)eC!1YD2zJ&Ov*
zzZ2N;Qbz8Ls};QNn3%}tb;aOS#GYLr5G}H4E8okA_9NyDR~&a?d~3Y5H~&1UZts~w
z&X*OvFu+-mn3ESN(iEM?dMfBWL6`OEU#Khx(OD04_*U=Y>lqql+69SfRv<b@(lj44
z;;mu|kYS?|CY?Y-`K~wa*J24Jz|o=$GPVs(#0)P!y~v;V4a^?Ed_@BvTOWlE5U&7{
z_Is*bCJlUsRrkBuB05#M=s6w82NV`S?a>7g6k-<Rz5XrBL=V-#uE{9FAe|`*l_z$R
zlam)Rh6EpZoX7OuD4VhCT2FtzbEankSp)?`=Qj!j37)s!APbj&Rg^zK1xR$K(4>^1
z-OZ|~1W8C1D{-Sg3V__Go#_HQK3QYxN1o>o0jdH`F{#@>MnDZjbdzps`iHG&!&9NA
zAMBu!**it5EfZ@tEum6XI~vm;w7A;o8oKw9g0j${$H+Y$Zuie|+%D`p@-f2DPk=D&
zb<4f5Wc}V7S#)2V&SpCCZWGU6k1V6z4Xm_i2F3&W&a&MSJfukyNMb-E-WB|5)b@<7
z`E)cfCu(uBDrbI>S<#p#F|mQc>gmqCCeW_E{#x;W@Gp^mIQv4%IOxwo>bp}hMDLFp
zd#7u7l;Mo>2%?*_n&|E1)X95N%DapL2()!e4gCUxAxdqECRQ)xco_-3wTVNiKN+<Y
zE%EZ?CTmMQRp(?+qjj6WmNKZirP)q+_OYBfOE`z3r1&d%ET>8Plhd`|V@Z1Q;+RT+
z3*Py5NboL$Z=j(-XAPD4cP+Qhs0!$1`xX%y>}LRNvj-J)cURTh5-JLw$8Q_=mb`o!
zfcr!qB(#g>=E0F+Phv6*wR*ZU(-p!g*x6YY+oKbWfpQgKI!5pA9=2Xczfvm=0acD?
z((itawtN|V<j55p3s;rz?TP1O$%UtbSUW-PR|*KPjjhYY;JEa)m4TspLfHyyNir->
zC~*pk0P%sZViT#2g%I&D9RKdQ`je|%1OCq132PNJJbvw<fA6A3!Q1OoEGK?`=aY`*
z)YJM4&U@V+aZ|K8b#7M}yCi}ln&NK2yN103B*ho7y|Z%xH^<BKaqm#pWMM~Tchd;0
zG3deEJI-Wqzi4;$gc_dy>{ETY_oL!hdm9nx?HiPo@bB)-EK@)R82ILu+cD+jrSq6u
z{Qd0@Ez0181mWtEPf<bn>@^=K5abo=7_3b^o{eY8Y1;=SeMFzDxqf&0=nY`w%N@{l
zc-`UTDS8nR(YKan1nB=sGH)6FdhZV9A}9>&ps;qQkP(E>bIKz+V{;eLnv0S37qf}M
z#O>lkH})Tt>KEP0Or=*d1<#F!RHAk!Mf8n~f)DQbg6XQOKU<wJ{xehP*)bUh>sZFY
z8)&X41iX&}XSUCs9`0)**GN{-Znrt7YFwv8XE;>n<PVcH=v^e^g08868-yru@n(jy
z!h<mtFHTlBjg1ZZKdH{{o{W}t@CjP6F<li$fHLnlRi*BUOAeP(Huel_KgX+sPDw8F
z2{oBGMdj1N3!RIJ$&Vi|?mTLxEWX%q02C?cs9Z|q0Q@fD${J|z;GH)zO*xONX8>@-
zA9Up>r{6KRo5{=8-rj<$ZZ3<qz=3ZU`mv9!v@yT)pMm=4P75IaKx@%{?@<n>Ou*bs
zdz>QG{@FL7S2NXd+r}cGwV%2nc~2$?R1gi^?(F|*x(V8yK!x#}$=^oqxa)Wjo!0jc
zOgsTChP`uD{#q&7^3mrYTT-Hnz}P#sk~8(aB<%QbbZT!5t!=9@&eTZHooo}aTmN^@
z_9HrqAZ{NCABBuxKf9HRRMU1>Ez{IjqE%DtM~x(8z@c{Kw6>Dh9Bm*X4#z(mHgl-Q
zHYk!!Y}yQ`mSc>@enQ-L-0~K1iVyZq(tYAI6ONU4qu~WGG*?Z|9rv9ba(2$jSxd-W
zTbui;H4BO2!-uC&@WoYh1a5YA?DvifOY7cxsDJ-Htsa)3sU)Q)g;8R<4otcKXFIWK
zt?!7%y?0#$@mZlgw?!C{$*>!&{gZTsjF*7*K7v1$!o;vCXKtP?Tg}vb%GRf?ulFQJ
znK)&neVlZLa4L<LE;1nd%iEVr*m4kEWZCgSdO~RB?8E&_?uqf@_oEni-LXPCHm^;&
zt8pTu7A=e=S>4<c;`;e5Xt_uFDsS?fgu=H@DoS2b7X$^>T@ld_@4AOr5OMa-c%-i`
z4?m&(S5F(FwW)3HRQKXnql#&@Q7LUH)jKw2YvxQjlezBvvbt2d-ePFLAclwlt_qrd
za$1VL>^NVuq)O)D*_)0OOV948YRG#CNZV|)jip}@u}5T;Vkh>>1Sg7XZ@pq5cs%Zf
z>_tK}(wAVa+EvdQA+cZlqa}?rE?8Oj_D6uPE^m&K1g~|G>&TepD@}BCfV)Z0xTG^^
zldOkwFSRt}?d)ndY#^z)MLQ+_ki;bN|1Q4CD-fh_Y*mbm-cdXSSVrD^mfhKlw_J^(
z+pMtc&c~9{yUz=%?l?(aV`~Ex*gdH5EI*$pe%exkQF-OTV7bNch#csNiWSH!9B529
z1mhe0(~z6Qo8=ni_jsedwc_G%5~t5;o!`R@A_j$u{XXGi<Nd}1?%ZIogv*ndC)#zt
z3zn5w4i8J1*;{*lCJJn$5X@3O;O(k`_@4ZyR7ZqhApL$#w|oH&XTMl~Jg|2hu1x}&
zGZJ4ySrLUL=<@<6d$it3Q;(0R@7ffm8>h<Um@F(5I+02L?CU{9M#yK_9;7mZV#9Kn
zFxdTMLsX8#3u^O~1>~mlVf)05$H`8DfJO*Y=T?h}F_k7oyD(QjKCk%}+grwQQRLW1
zOHa<jL}si>Two}21H}gswG(Z)=dSW*`&>3(TRE$N&n6B{`f!t5-jbeS8hma!ha7}~
z0obHFyHZJ?!-JtSy|@3)GE->Tr3b`?Op-Sg?rK_#ZO-@+jVp|Jt|lXE7>L<F^sQYy
zQs(aq)AlAFku{jmR38Mi30NKO=v4j~1v2#v!*~gapCIK0=fOA^Fm4-s?>GokMgE}X
zNo4U6-t|0r+YZXLgm}}0s4@8ujcG7;pWbPLK=@b6RdPKoy;i{hYZ<0pYf=zB)7ea7
zjc*J(v_K^ZIDe4SYjyQzfX3t-tDd8swWM6ood81@)N5PQeG(oe<^(wFXZ>+$p;8pK
zCsM+sacwmv%>XJ8p+3AS_Mj$7bh+s@a{Z6{7an^25M<NNyJ4mf(v>rJeEuLy{RJ06
zToYIgFg?!t<De&pf3l)&Y$UR@BL}qE4VuRX!4&}e-qio5t?b6*cq5ILpX0-0By?H7
z<k8a5zFr-${TBzku5Vdk61<X<VotjfGU|ka+W!4{Sm0DhmN63&v-w_AW_%(N$i|tA
z4W5HcIv;uvI+z!VmI3eU(?&6C6bE^Lx^faYK9MwL{iDfG+0yvtev%L98@y<|!Q<`o
znFrm!vpGwM^bnBH>3(jkR~DAQ?mmFt&!}L6mSba51F<*epgWl0Xq5L^mHoM1(tn=&
zSEDcRs#v1H*#f7Sn*IYm0l8tNDRI71j!{?WJiH%*vUfE;Kro_!KlX-dDUI3)8Dwze
zA2^v~^5t@zU3XuA{(L@OvkwIs&u=6`F-gBRYO>0J?ztmdXYVPp5ZUFm{s)LEj`79t
zedgxIA}q2be^78s8(MqV*KPLk(uKOuhMqI@4G_A%=irINiX#BI9K1%F6?BOL$O+R*
z@botf6d%D>`>KO%X&&rm_=^iPOoF2?cyU(KR(o8KPHo3%+yE%|9Sokm#<aSX%>;0G
zF;5h*t%EZ?k*t8zXc`hRCI-#+i0R-&y_Cz5W4R7zs)~uf!tYW0x2-SCNh9{I5!C(*
zi<yP_J!HWL&I_an=j5x_mMuPQvzmq3%D*NJ331;)!sc0`K3#8ZndOJQdZoX$ns7X;
zmLr_=;iRD%u#V}Ge!S>L{2E1$Sg|Qy$cO_d<fy2#T&@9#HiVgz5Bg}KJL93CjmHl^
zv9@%791<0>c?f_eU?%qCASdjw80{iQv-z%XlJ0H`Xit;NQiXR%hCW*fF<iOWbyA?w
zF9%K3g@Y|{Y?h^vrw;lX4E4ge)YMCsTW{BcDnVU(!kmdb?47gGuAZZK`-oczuT}fr
zRs%V3o>7fAEP&WJwz{+~zTV=lNZ=04D#wkZ4r=Kw^IT&>PK0)+y~b+nQ$aC58oFBf
zDWmkC84?sm%n(RO{0q?6UW7{^I+uhXiv?jb*MfbpRgcH420TVwg;{?Tv$3uW4(@)F
za1d+*20XXcATg%GP#vELI~^e&cfQ7>Bgyma%gSVZKVyXGI|9c<rtmyJl~-5KLTi)5
zHRV>VKgM+Ky+4_4?w)Y^!bal13F`GABVg8Fazvw}1NIbg`@!D}OQqm$a$FpJPHXFt
zW~Kiwqpw0?_I10@_d1kJtEuMu)O~vT&zt9Zqcl5cEqH7;iegWoUUAuOeNePS<Q~S`
zR3J(5(rU+g!)a?vSm}bF1SQpe^~re{x7MgyG;#u(p%%I(WaTsEm|ZrlX8BwLrhgo0
zEZKE-I$!^Wf$xw6saUG9F)KJhUSxL#mSe=d*cKBh@cwb@>~Psn57705X-#bYNcZ8&
zhqoUTW##`no8qY|oO?58w_N2K(jPxETQ{6jDP~GRF5^Q3aR>8>e*8YXjMG)E$((a3
zWvM8i0L4tOuom1{fC<ihe29v85*5}__(x6u01asmobdex-z_f5h6c;?g?mfquA6Ge
z+mZH4%k|0{u}I>d*|2a@Hr9RCtajN7_HPU~d1SWG7j?e7wEBLo#dwAd4A1a_S{IyZ
z41)Z<kVWd^0I7|NMK6hAsGtxi!4(tntKmBr=r(3%`?lYVe!B$9AY>=Qx3)Rl)ojoQ
zJW%$X{(eZHo7C9tm&N7~l+fKnu)QYY#EBo@(*v3%-_g)4zt=)(e+majsjAd9iorQ8
zfa_z54e`MY>s|iBB0Vv^Hm6ml6M)pJeVI5cP?|=5#lOB0us2=(4BC(=BO?J@4m#+(
zu>Ctg-;cYN_BhY22TEN%5!+~Hc$V|VW^06s(tEuRh(^&1mm1La=%s-P9O^S!DI;Ae
zKJWqHU{Ud6uxd<yp7Bu9*_@V|J%&(Uqf~M!`5I}QSP#S2#@gu~4Rqr)Pg5$6v6L09
zYcj~q!JeT26CJrt!oa{}W?^&v&H9v-%16P!E)_+QRguCKLiqxD|2VK6Rwn)&8xzC7
zJn;Za(^Q?%*VmU|RDnb0%;_5@Va}nmsVOaxQbSVH(9pNAxKfqM(`h(cL}%2hY67?B
zlAb8t5po7)y-KgQ(C5>iG|^oyJGGQoE?aldNa&AWNlC{D9zM}Y;L_PG3Hjo?e-kjG
zRp*j7N+3&34N5;mrT4I|rmTgd6xCXyi8=O%3i+Mj%ctS648*CmyQN0+OB)CHeS5ol
zD?gk0u6y)X!5IrD?@Q#zF_FzzifiM<B&y~ecR{wSvoi>V;q%_t*Zc4e9H}YYjZT5T
z`wcq3M=$z>hpKFOV^38JSHVH-^0skJbrz=s3LOGoB1MC1xHX~?vTjFKg@i9iE2W>v
z0359YM@{x`g4U}hC)U|}1!s&Qs}HlCBi*|*!A>>zH)mv3ObJ2FslDf?*O>!-oP{UW
zX9UD|!hWce5(|C;Kg{0V1HccQe37~$dv~VE{F+6-mAI&6>sVC{I^u#z1%=EUAGI19
zf>Q;HOB=sR)5HfWJ!-{5*+DQx<-PYbJgUtrFxp>QmYRY0J}*6eEH5{p@9*<moWBga
zV!c@~vNr6%FHQRJGoSDoI@a1^ux2{6x|%9SXKG|E{wW$?#G8lHr2Mc^UdoIPDM9Pz
zcH|lmb`??tqG0To220g3;o_j`3VO1t)V>c56%Zm$nK-KwJ~oI6;&|k!CDrwuf%DY5
zWPi3%of;B6@jQ58T3x+W`_X@jmNtE5Sx?XGb6!1($K>7qdg)EE+s8|*(EC1FrpBh|
zKEBok`jjuTAuqTkOP?h9g=3RkyMyz{<ao#VoR`+h*W4R_w#KtPpY=%DRQ{l#<&6vI
zlnBDajMM)wtht3DDyId2Z0IIn`$Xt{k8iI8q&Metx8+WA3-Ze6WbJ}Za48npgY?G_
zMYW#etbewLQ1b%<*oePANZmWkDy;ccXjovbT_T9K@e6D8bJY_=<$&bpG=mSHr<Ofu
zc#(2f7!Hf+o}MH7uC9;v&5<?cbkpU4FS>dy@qC&$TyHgwbq#~~rDCRSLiLaP_s1><
zYF2FSOtDwBXAn)NEv9X3{79pgmE9dEj*#cT6d@s+N`Yp$Pe+)C7q-Jutb4u&1`Lv@
zV&$G(=B$1EQ`6vKuRSL)D<4;GdgL<_Vw4D1y4W2!55RQzGoSN{q^DbH==*nx)4xhN
zHvuS2z3Nt;UKQI@St9MMN^*Y`w3H$y+01i&k2f_*FuC%+&Y>w}9iY?AjOC!HmIxq2
z@<;J~P%eBfZOCId^X@CejTcS$^7F0g*Tvh$^(X7lk@O(d3Leg_oHc>3YlUT+zjx7S
zK{VFU`jg`9saQ5Xe&LT!ZT=F+1`-kiHVW`zt{O+ihWkp<nV7PIYfgso{hwB)>8E9N
zQZ(<&8q&XHQ#yVYk{j@Myq+D7Cscc?gpj&=&BFQ`+dLE2W*57;At>;Tq*mW#N#4KC
z(KObc%Zy}vfIc$0&C9kw!a00g=vGeY5;{`irEpy$X$%`wZQRLI64V5@Zn-*!xRzNi
zogS9_y2aSiVs8}hIwe!m(M!4KxUO{fudqi`<Ll4(84>~!nJ+^5bCyL+&4}b7S=p2A
zoUm;TdWzz+2il>gX3Cu}_{(hzRKN|HT~z$p03p6>zC8Lt!!1|y2%lZGtBT7~LUJfK
zLS$pA%*N-!#cs(6Hu$xnK97ZT9vg^dW#7}&(>U1l+~)HXwoE0bAMT4Iu6zd<JK9!^
z+(+E~(!>OYK^gzMceI%L1}X&OsHqqsD_alIWGghNecGxG`$6c-Xg>P3%c@#PmikB~
zG4D-f9kb}B(^lDYZaEI1(f)@ajRJ<{S*zgUvvas(;L!Qv`f@EHAS64h))%92<LjFO
z{AW1njUjdNq(j5HQ65Z{rio5uw-e=mCj~sh4Bzt<A_Z`5AB6M?N@)swy5q5Fxh(-3
zVP{|(q#o1Gdr*6w-%bE0-5JqVyr+}t9xchqfB4z<;2>A<)kjjR`o++iGHG*53BlCb
z=vTZ&i^E0Ktn3?yMIR(wcTT3%-eY=h-fa5PHoy!RW}M!H7N}($((YRcxGUY3{vn19
z(%<sR3wgPRJ@|Y76I~z*QvhtTT+^SvVeGD6imUZ+Wi;lXL!O(49c?zY1|Nk`vN6r1
zzM{;CH?4QZEiR5a+Hvr4GeV)+oK1m*B7@%PQqoHMIppKZJ!}#O9`^m%JT~yh>^XxR
zug|^v=CgB2a^Qli+3{laN+z|`?yJwXp|2@>W=b?6)$VhtiMx#a0;@)t+cTv`C<w`>
zHzeX-_^mj&hPR`a3L{tom%nG`SSk3U`^A0JO?A7Yvru&B?&d0ne>Y)pJ0~+6D$8k@
zt<|;6YAniX&`m2455Xn&|6`A9CUl^Kd3z*@dA^KpxV*Y9+wA+Nz->3mCJ{PA+K6?;
z=NN(pqjltlz7t$uMr)gKUVGG68F@Pz#?y@MC<%E9OvU7Pt+{8G=rXE=bY*0<1rI11
zw5vt@y;XRFw1N0xk#MDE;Hb4+a!=;K_LNFh630N;#sr0ox#CbnnRz9n<xmZpB1+^~
zjVX44=b<wH{;XeY!ydh~dab<TsFjk59Fz?9m!g$kO2{f#y&@M|Xz&O+cAGiQp%02q
zbW7#nWKPkh_3bHBk4}Xaq+1PIm#S5cw;Wr2xg(V42M3;IonldMGg@Ru@5U$RF>PC!
zS|Ln4+ni?am-*ZRdFSs=h0wNOIyTo+sc25?%<>#?ub~%yxzcqpbNzW%?Ge;;OwP78
zg{x8QnBYh%H82|<!-64#xjClV_-LCwUsiS|;$zHb@6Y)m7{fflFUgOK?o)664CCBb
zTA?rdICGCOi*zs%I|l3+qV~7+DF))0yF-%X7eIU9O-C!w-%ETVMZEBJOxr^x;fz7q
zs_PB0VqYZTgV3pic5;O&^knQ^XlS)M;?Lbel!Q*dD+d05HxqP*@zcoZrKF%G6Womt
z4~)~zHf|3cDqU*tV3S^1Ug>&bq53H}CK>U>$u;uxKG9Q4Vd;lDY6Wb_zT1@oGm$&!
z*|3xsKUKt$+++0!(44-gk^EHzySjH*$GKHh90rFJe*B=HQ8hbR7a$T9Rjy&ub?jUp
zzGB}jd%x{^T?aKjO5%Fq_Ws*qYT7WUOJQyfNut);M+J-4BEA(|B!!IQym0I%7L2p=
zKwjbE3AeTn8fNa<(9$Jp4m`b=;YLn*-(4bQsvt+VdrVM3W_VvweRj&*f&<r;PonF0
zKa@Gs5qq+N;$Sr)%yivHQd4J+kB<G1Y#AaL_(bzF%c(l-X7;HR4&v9InAl%(;BHV;
z(reOVWzyy9@E<o+Y2#+)U$gx9!FF*@t5M++hymj)4<3`=*2uFS;u7V4X6wc$(a||N
zSbv|*e6w$|Y)w`B`1g|Q!KmiZ8B~x)*z?aN>Ywb|HmGS<3txCs!us+xvfQ0oSNl9x
z{mSW%oY@v)g*+?uT^+3w3!VSSGfdTq;<vvwojDu0_&7d7gf}NV@W>AX=EV`mm`j%8
zugzmJQYC*e_Du*49c_@cbnQ>dI%|QzKX+6kojpdYqcM6P-COem?Eh{4m>jq0SJ5ts
z?YCePUd(502Vh{0iKJQOMg^5mw2aO6kLsT0S`W2p*lbE~;1DZXxrq$^37j%%E#RbM
zp-{8#zSJs>_RwV2`)h7U^M^Ehsb6m6u09coifuGoPC49;b*&sY_pVoQ3-0zO=rTlt
zE34-h8tGX0>Y&*0QN}I~>w&Df+vLnA$44Cw-ne>|NrxpQ#BG1l7RKJeNw3zV3|CfF
zRHgn|^5GXa&ho;*_jv4ghX?q7?GIv@ir8DA)@ZnWACu++f#;c8Wf~wsZ<J4IPs=%-
zNi%n~FB<4B|2ag~8%DuD*BF{g5^^YT6WSAJ3Mti&^2EmEIIS|}Z~P;*g_P~B_5m|v
zW0Xh7JgcElzOOzzUAxvndU4H6M^IQ*%RYq>+~DyO8CjyMcL2n}$?uM8@Zn-lkNVqw
zUAF8?O1!Khmz?$x^W(n8Z1uPNGs7GWzpk$`j5#T-<}zd(Z-uae>q^kD|6bG&Atf`W
zHU+Y1+yA_)aky%J@uBt9^?1l-GQLin6`EE_)dtQ{n!r7hw950#hVuJmGi`8n5g8@S
zTT_Q3PEV;SDWv4J8J@gsoFnz-S<Wj_@!`DOum4O`wl~e5%D1WQsIHYCJYqgER;S#R
z^|UETRuX@*BGTFc0l=668GoReTT#%mFPqdJmrA(#7#1`a5W92ur?P@3*7etsv(muP
z<&bQseghX(je<G`Y&$j0eorat5m1`Er<W4Bfwk6tbmjE&X{f1$z?r+)$m^b=zWbod
zUu0ze`>lsIGe1?|eL>@!3P!J~ZTZ&Q_`sSbzNK=nvAgf!rmoJ6(6CKnsr!xK;eL_j
zF-mB=;+6yYn;8+W%FniTc9DScbO=I2@9h)ZH6Rij9x9J^q_=Ss5)x8=`$Nm{22~)|
z1;DQ76jg)Vsn4=5-}U{Li=zsrpCO}bQx;Dg%dWdZtJ6ufvaOwQclM*m`80Mn-s_h7
zIn1X}7&o_ngCJWBI$zsApIf(AF@PVc%E>+N<7P~J)Wj)2Uy$72FUM}u-?Tx1yZZ29
z36Jo$UTkMyA9&F>3uI*6jh7rO3h;=qIMsEo_V}#JuYvyZ%%O_GboSMCt#Qw5wK<44
z<t|E7UqV1e+vcfS$I`xY5;`(6G8zWj53VQq%jcGFRX%=%ik19nW9Bb`a&yL<So6Lo
z=K1+K8U}{k8zi!#O_p|bf!no}6<Vg<Bjai?pb7H<w|79<@)Mfp=-qkZGqUW@794RX
zs2-;W)wEhJF^$rL@5e3BvgaMnMbgCJ>6YT5sw{y=hb2oOs`M~@E%->P6f|X&ZIbvX
zPVeLOwHCBs-AbHz-*YFZh?R&#2meZYU`*2znS)SjaDEJ<c~5}MS^E<Op;)V+WE1C|
z_?5%k7;h026cjV%`cGc;Z)V!9x#jLM?oNvO>X{5A*4wJ;TcxF!H6yl0hNT8Q*x~;&
z^uKnFuH=ROk~ka63PHR;e9BP?s!EC;qk_T3P8#!bQ~z-w$&b>S(3Qfn+Gprsu8Wr=
z;I-VlOZXJKBlxwa**5kOJ9VIQBIF;1%uhUDQJCaJFC44XDluz>{p{jvY-T?nYJ9Z3
zPO%TxEq)6>YU97n>z|+B2>3^?=@ZWv@D88M8K+R!a>zU{{fK<yP$+dYJu`Bup?_&W
zkn=X3UNA9h3ne;{fK_AcbW>Br^k1%R)2#1D$QzQ!`jwk`5MSOtQ++OzO3_GDsK7|F
zAluk)$Ni0=vXaGL%@Mnfn{SMw8i4A}?(I<^fHdEFn2$PdWrmZAh^>xk|Jk8UevsV!
z-CQ+(KM}hmcPTRWQYl%8SyOj+rl{z2hpF`1kc8Kyvt?=|%`%Qi>R-dJ`2>cKY2ll)
z?_Jp6gZi}0eh9Bew^pWcOE9^aT84&;=3Q_qkc;{!TFxwLb}(b)k^d{C-IoX7C;xS(
z@I}3=$-5#_NJadgM|(r^)@w}mE=JAh#)ER-_9b~wG2gA@4cg~i;~QMX-*|a7&s|*L
z>EAAnaUbJ+uz9JLvo}l&Q!sh^r4{y`GPEn>s#^*MfATc-)85BG?4J@o=jWY$oeFA7
z5f&VybSK7z*wo98wUn@9ef?K5sCl^01Phz`<?cdU&k!Rxqn(ZU>kq-DxuL1xp7wu!
z34pBbUq7DFpw<2B$N4mxzc+4!|NZ~`Vg&yF|LLRuUwtz4|NkTYKgWSm@Mu2I^K9Y1
z;fD`u5DBVs6F_sTt8`pkYQR0Iw0rrWp`j_MsnNiIvS?UO+g#I?x~FFY1uYlX_##Cy
z9N31AP9g;@GqW6!U|xJ%44>Lo0WWw45J+<|&!bln(b3eQ?c^{tOw4jF%>1ahIM7M|
z{O-Mj{mG<iEk^Jgmg|tjP-!`3<N)148VS1fqI=H5qA*oyJ$AA^RS@!-PCu54Z1_3&
zkRdk89F_VP-s}O1=ye=toI`B&qR*O{&ziBE`;cID#&2xuj6W$|Is7y#9jRFpDK+ZJ
z4?#*MA~YJHe~ffiot5ONak$nOP0<s}E}6jN06Z^Jo+Z#EbDDL@q5xWH&=c3I{dNYK
zJINdb0`64`ii)_8A14DnIdzizZy1>ER;B6>dwJpM#)&Rs-uJO8!26+J%s6095x@52
zasI7?+$0Wm)<1uXL?-I>Z)r?-oVz;z`rH58iwQGLV^#b>@;UjOl2R0iC;{CciwT94
zloUK_l=tQF@z1U=HjCit+1WJgrh_euZ;{Nr6L_6+lat^0>Sc@gJyK~$mP3EWZTr>_
zi!6$Q-EvYXEG!JXcV6~?Hiv$|z5mmf|IbLF{~4-M@(40g6CJ&3;jnp8y^2*F(8rSp
zla-SKzCDgyJj7%mRTLq)7xIid)fbaE*Ltp@2C&4q#Kh(fd`Rp~E4HS~zr)ssPr%T}
zfcASH_mfYPzu2TL$BTvL+z*2tw<j~eleM<B<!V-15%D<}FG7op`5_OQWPux;uap{!
znA#RFCq}wyXOs3gnJ~VX_r?mCNrb~wTUy>yQc;=B*14czEEnL+WA&bRro02<YkAxc
z!FR0Kt}pnJso2`umP{!sC`5pFFg2VBJIy#m-@_uckQYX}`65e+bTvEUG*@6=SlzZk
zfMga1be<ryldTC5Z-;>zVX`Dsy%JFDqkHvgI7P&-?EO!aJaUxm7XWWu9CR^tO+(AS
zpGe1Z$%7RzA1h@4j`1l{;O5MLlA4;HjZG<3;39jn+&rtP=?$<n33R+y;Kt3=S&vg~
zZQBk?Ae9z2VuL(v)BO{@X?Mb+4hQ9|$e<kxF$Dtv3Pf|Bnhzg7wB>pR0yX$T@TAo@
zM_GceUGYCrn6?Js!qfAL8Z^L$n`A~Z>rNF7q=1D}%c^9_B;erS0GUFiMo*7mLRPWu
zsmf>yFyTKtzaw4UQ3h(QpIEyScvT12tJw%`P?PyGO%byQZ>jV`64Z}{GUH{kC~^$D
zqB1JhCrYybv&dW9W{bS1``5BP4Pp^GZ&WbEIy1<iCRa>V+hu{bq^6|-B|p2oC8X>Z
zU^;+gJUqSL?KEnVik21zM@%14*WD&)e?}O|)4Bsrd9jgKgjkh(m_*qgmnSug0rcQS
z_X#0iJ%s7du(B!wSx0!fSTK<(5Ri=Jbt?3{+86in@yW4U>m}l`H;GAeKst*?tSa1o
zDVXyh5qXaln@UO!2-WF8*AD#qE!Y8}p{j4*d;-I3xY(e-K3fVk8_A7y%Dsn_E#uqq
zbRX$z_;d~^uIadBbYGM3Iu0w!Lj0}nuAj&zDy@@t8gD&YTU!3C##xUJ=O{{AI$a-+
znO~jlvlBmhQ?O{y<8au+DG3+2_(MLDr`m_ul9T?})NJOXuRKm?oLa50(9wq&DyLI_
zdf%si{=6yG9_i{Kxij)v0RV0@ZoBmxv-Pz^Y$nf}nV0Q(%JrJ>Nhk0?%7WKN;W{42
z%GcZ0P$>`HCn9)K5)-WR_wRl%2zLFJ2Q&6V(r8#%s=&oIT<-8n$8ks>uJ!;0--MBJ
zbJaj>)Z$KHc9byfTE~nk^s}6Z2poj(ptS1i=#aOy{qdu;R2ditIvjbA6D;akL8?n>
zFAg4FCNk>1YzRmd)~qGEl^OLAWn^WQ_CKAc4iKAg<|94cnox9_w()b)efREbVa*2J
z%JT9bAh0=<uU>itq#xCBaX+;_#m1&gARWMJx_zbo<LywUG(ree$90_&;mo|eyli<A
zG)nY+ZqEB4GvI<_1w3lN47_*HLUMQ(n2-F17K-Jt++aZ39WQBKWJ5BOc>DH?liPT)
zUPc8W*T-ZE)7zVC+iER#IQpAycx-GgkieeUcEpoIxsn4hC{sF)3Kq*^7%*93i3l--
zzA44U%-U|-<u%tQQ<2fpSwNCeJs$6qU^0LdhBO*)Y01f(7b%d;I%9Oxf&v2Q${}er
zweL=~CPExYH-^AY;pxD1SsJP4{Z4l0m|$ju>HfgfB`t@FbOm8qW7qH3$&p5a0V;{m
z+nue4x2WbK2|j)HET2j`_8nX%kx$)>IBQ5fpC9RV(XqB1co5iNvXD1GJWXh#bMvn;
z(z-cBFZoDigFm3m&jC*jy`dc(8Y=g^210XF6_$uCMuppxrLNR;ULQ%XDgoChH_K>x
z6UB;j)e|asll|+LFo*|lZES2n0C?|}dd#h$pwM~cy_c(;M*-`J<7BFQjDe)=1xDBi
zwDEGl1ene>c!0#^bT$!82)p^{pFZfOlwut&$tb_n1_c)cJZwU_zC2ZRcCJz>(8!o5
zHKqq=00-9xcz}@LSP&sAMb0>i1=3Z!agqlWjELJdy%mO}2E4HtgGSjm3jV9T=E+jy
z&}pm6^r0l`hVxb8o_KD>r$x9(X1;nXgcwMI_6N%_pbT3$&0#rJF;dzm(4|_8ENK6y
z|AjP^$X7jXk6~N`)I`sl?wzH83(JE^cvavL-@nuNHS5{DL{i~)IY{5B-4y`(C{Yuc
z(Ctkup9>4vSW3JA5?-kQnbOgG^|a&7F>|maon2juK>QHlTri(;$$=0Cu3JVDM1T+<
zbo`Cl-G(8sKXC~ObF5}+OwXXb1+&ron2~)Xwf8qy=drI1d@qhSv;0V1e{XOhy;`-c
zF0);cV9;&Q>gebI)*KKVtlS<#;&42ong>vViuj#*>i-S6kN%I*{(l1Q|HNAaH2BN^
z{(k`ye?K}q8TI}nAbi?oRcQ`lqDuR|TO6nLi2!L%P7bgWfC0?zAaICWAqWqf6UIr(
z!or_ZQ&X2eLR>ib_*p==asB3UR~5u7gxL?V6@+LkfoR;d0`FSM{Ya*+u0ET(oDK`m
zWlq>dM1a%Dt?BB+{Ovw>zN`Ingnr*Gj)?>Kj~uU|p&<eV08L*Te~)z_?<Ffc28<Ko
zGGOr4?wGWJH1RR}Zl09vH(QRi5QB_DEe@_lU<G275pNN|DMC(8E*a<sn}CV_S9%iR
zjYtG&7C`tu#&h!i{P_q3WOE>OYc^d~0#s!UV-a!e5t%@Kfo7%I&(C)ObS>dx2Oo)a
zMutcnlD~}Uxb@T~L`0;AknrM7fr`$r=4M<{h_n#Iw~d<3!o2l=Xp@rP;D-9M1hCPz
z=Nfh4!yu|XZJp`-$dhpKCoR;z8IzNSkx}N6p9sR1`A?_QFo`)L+_qMG;_A_@GBPtI
z;Er44DIMX|o2h|}3c$O#9&coD&bcdcb8{Q_z`>pr4a74&Snk*u$_mwo^!1lNat+rw
zm`i_W76O8@d(|$wc^4O%nVB0U?WBreHCMpDSKAr$-(DT^xEw5Vue_F41TXB*mQTaM
z#!ds~Y7QKxgp|}#FQ@y6vT6q~0-(z}y9B1=72pa0q;|75pKGY+)E($E2<3Mi#faob
zB+tne26rIDw6+;t?#@Bp-rjlnb9d44Ou)N=skd6qft7aLna%-VSsXx4Wndo99*O%e
z^=W<VTG}^h3nHum;5e{cz0xWHXg6n3Cprt~sTMJ+Jm*ss67m9`UT&jdrq;>)bY~`(
z-8{<mWGknr@kZmxlP9>`D@4kXV`C7TG=o-^oQN+5?WhrW*vT4)JRRqSyOz^c^4pW;
z^sKTV>{(SU+yk#F9^VF7e(dV}5E1wabm}!2qb0#_01{d!aSdq~_nnvbodaeZD@51n
z+z{G7pP1g?`R4&kb0nh~;prgStv7k0M()@Fmw`YPkpJS+Zcmb+ZnUnai@)&fZas(`
zlmWHIJPsQXHICaL;*_DSF#NEZ;lU(eP96<iF4)+-1bZc#CMdCE-QKL}6EbDM4^gS1
z;k|`sNgSuXGM7UmOcEX?7Z;Zmqnzx{L`3F2PmgdCpPnYVtP<`z=BX4M{U#4J0r3(t
zW_TV&#m=q_pv*T4z&E|wZM<!GM8ZQ6-9;JNo;E$LMM+6%I^X04PY1r+WM{g1<>Y)Z
zkcsa3^Ea%<y~DuJR`#7El9Gymhjq=0_*kul2!35#27mffE`X~~18EB#C+8TiFdIF+
zB$(nM!)R@)b^8~#>w_6<kjE^022bJcQmA0z*RKzNS9dKc>RK5Dn*xW7j0whLyZp5Q
ze4aw6E{Nk%uhd_zAHDQ27LWmG?P5JEX}Ho_mqEKm$$T`Q*5>LizSi?0w&TM-K{xBm
z^R*<ySfi38;DSpg|Ef&w5mJHL5-q9oJTf@6pHx-FWd{NE>Qz57PZMD9iRDTAOYL#+
z?%%)P9k%#m8&=rW&a8s)DFri%<FJ(2l9iC4i!uO@ka40YC->fQW4L2#2Q25B=U`*k
zP_2`-BHRUFw&+PC5b|a(&-dHO=<NjoL_RpNne#YJINcwy@c^<CF^P#fW`ZDbI;nta
z0r-)yRJ5PztYx&SoRWI?Zm)uVn+T*fGqup=$+lslrbGNcOGxk&@DdWo09(-mkn4R9
zlg`miLnBt;Y#tR6vVh>PJaDV<@NmiPW{{A0V+nktsJ}YsGVFDuPe>$M?4VAHuy>su
zKfsfyMJ$olRe=fAViE?nL`}D6r$On2Js2en4y4evQUr1}9kk~V85q9SN1>cceFK9L
z@PshL%1CShXty8INy6`12JBSHD3OrWC?Y212i;%At4~ZR2nMCozJoH_1YnxZ&d$LA
zT>8H*6L<<Vbo9dt^=jLU#ib=lh~aZ)Muy9pTm9uuEiN7&m0$D1^}|Px05w?*+%x$7
z^y$;Oc2Zu){BEtKwqRcKotI!O=t^gR8$Sk7+3XBN(1BDRko|K8&RC5VRL;_^AGsF*
z`&gOH!p+^hxM|3~bpzn^c=mRegvb7OEkAIOwwL|@tecoXCe@a7xh!j7k071F-{gUr
z{xr$N$ESrrBmn#!k2T&XXPVM+bE_k!Y1U<p1UC5Xg^2ypA9+!5+&z@H)ENZnW!viB
zbAc}UY~VY5R!GfDCx&yC=|N&<-j~RqQQ-n|$$D~)J|7U!W~>C)PS{gw3a#J1eOnE+
z`+oWI4nA$=<nLjO7XqTDj+Nj)8MX!h&DGm<ss9?>Q!&I2J~Sfm1@m@%GG)C9+8)B~
zYvfMAy5g9tYPYgBGt-T%UZ2hh7A;y?8m=5O1e(UOJL;+&A^5Ce4n#)vCL@SUaJKdu
z=%M3G;S&=>n=?0H7C*$Ua^JSJw@14q-GB6qhBLf<X<%mt&}9H((UcUKSj~L<_85R~
zHy~S|^y9}5K&63QUq=X{j+4w-6Wxe|2R%6zRa>n`MLP^$VL7G3z>SzvNC**7l_d1^
zQk<sN?eV6@%1jk^Pk=(Rmf`97fH)ZxSqp&;HM^Z!trGZ**HH?DdU!g(fSVOlR*E3c
zz$GPRjO$UcST=T$B9V&(>!~6J?4W+iZnc}XayyaVty6nzTT0Xp6R}BXLJ)B}J7O&f
znysAyu~`37SXh|NcKID_5J0ay0BKb^JbJ&s7U6Z=qH)mGgY;6UWa9r^pYGyM@z^C!
zrn;y}+U?B?)q{jFQc9!#@Am;@TI00K2)o#8#ySR?+7XkzOZ$Jou?@P)B3AtOyx`+U
zd=Pr18=$g<i1lDp8<DCy13-m3GAb&wu8wbaIu}?GaDc=d7X83}*bLe*c5TSTpdgxo
zF_8B!&&a%c*AEa*ncbSq%G#QnwNvd32)F4R7URjZikb0>dgUMyy#P67J^7m%t4hy{
zb?W1t8BG`r22TfLs%sI}1xVn?YR+-WQUmNRP`{{ObG$mUL7-uP75IxpyE+F8wRu4T
zGqHVQi_ejly-@=-MDG7WXL7Er1+j+PZq=C#P-kgix0v=P3u)BYn^s@H6&H7_%#w;p
z4GId{c!&WW^NjhQ#{`gf^XGjsCE#_4IIUiRt((x$1p546z7*znJ?j7EgI?c0ZPQLt
zbG8^Ld4$N<5hgSXHSV_&n@v>|fTe+eJR(N398ju%NlreJCXU5_@SB{+{*PSsIY?>J
zKYvD<vF&02>F4RmK#n3cA}qsk07C{{Z|YKpluvEXhNZQKlBt3r2#bi+ui*hi0+Qm~
z;$pdQD(T$cEA*M*N5<&Q_<&aH`s1id0{4ofm%t6vZwAm8_BTOZsFX0^D~MQ)f=+GF
z9z2i(-lfc9f`yWTBHOqx5l#_Tusk&g(mIE&ad~)|^S-{QxOf*c#xA?_R!MvAF&~%@
zB;7ZE&3uwPlrMFCC@Z@97Fh4lV1^{X`RbIXoTh^m_28KecZyVt0Gk9K205(4Y2(SH
zIW+rA&lJo<nPC^+^?oRnZ$Emdtb`#j)fba{aiJ2WN1#z=BJaQcx_7w2!;L|wZk(5e
z@8!$XVMpVT$4m<Ln80(4+(EEx?rI474C&2wctSz}SZ~v2UrdevYn#)0LHD8q+mNq-
zy?NlpyjeFlEd>_YOMrcl>=P3dv+6G_I2ieR53rV)I&a=O;4sX~`~UxK4_hC%7N`d_
zBnwPaA?<v!SN7M}0~;U?vAaq%x9bC!z&yAA9|<fYuYY-Y`75xdeSKo0@~cZryWRTb
z)>^;c6a4knRqf9e?S~fvyM)=mMgyql4QiqRo3%hGZ)LZh3oV?&B7*gT3vH{v&H27{
zvsY-`ec-jhLW_=ei)Y{5lv?%s?RH=<-bhqLWCgI#1&Sb$;SCK8@0%(<e*C!f+vUya
z{%>z?b`M_ecXg7g_ZHwFkk(8qP%*GVn`d*z4Hk(xzVZwVjB7oCmmB~?Mt580WwrN8
zdF=dhA;6}Ou3yWkTlE{0+kvAO>Yh;>5*pX-{T9_N4C)B%Yu5zb6BY0nSXr6Y{i%4r
zzxwN|)L03i*nxI%m&ajh^sbVX|A0~U0@!q$c=_d&%vnah!8dQ-1PwS?mAzR3?4Ef}
zYU7t*_u@qcaFhgOYNsV=xHjM^a7g4UurdGYqI<o(?)F&Vm=SO&A<P-HN~&R3yC7(y
zVS@{BBx76N-BrMB5Fra1+5HgN0~+CVm<lXHW?Wkzzy7@a{}@mg8Q8!seRV|>G^`oF
zzi!vhXR}RHP6)8Fvq#H{g7js!WzJf5RT0PpgXjHVzk%2X-0N|74I!#gMvbYj7&Z}w
cJbL~0zqnU(p8X!p3m_+Wy85}Sb4q9e03zTh+5i9m

diff --git a/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-48-1.png b/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-48-1.png
index 0663016850ec1273873378bd600f0a3ad1169142..6f29b40158c760098ae4ee4ffe657563465459c0 100644
GIT binary patch
delta 188
zcmV;t07L)Iz6Z{}2Y|Ezu2!edSOEcl<?ZeQ;2z@-)^&FYm4;n*<>utd<m)SMqtV^v
zz>UlWW?2d>k|bYWUm1BmRBQS9`H~#Z1yo;FtF`F4;O*lrlaI;G6~Ub>R;$eU`SK6&
zmq7-1mjgG-&C8SBW|JfZ1O&Lb8n|0dZmxWMeB|TJ+v_T=nY#-Bfb(MoF8~0y)K>vB
q0=HmV0cZvU9yQ4Kx6EJxZhjPPT9VL@2%jJT0000<MNUMnLSTZm*jb$b

delta 193
zcmV;y06zcDz6Z{}2Y|Ezu2z4_+ua4gJ;oob>+TXN4ZG~h&B>L?*H_*~qr1z28<`8t
zvJ_Y(Nxr_mGV*+=*7Ec7B{`l8sJ^ULYteJT+s9ibACsFaf;(BPR+;nj<saZLgADF2
z2X2&`mnXZ;CP@ki2yk;XaJQV?T>1F;$j6(v*Hv0GcNYKv=f?_O006f`SpgjZm#$d>
vvX`z|0kXHQSpgac1VNy1bGOW30d9U24?dF6QTfAE00000NkvXXu0mjfm{3~V

diff --git a/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-50-1.png b/dev/articles/internals_pipeop_torch_files/figure-html/unnamed-chunk-50-1.png
index 0ca545e8b1b5928bc56165e1e47ece21023f43e5..aa4af5c283685dce5d0a0f590641a17fd45930e2 100644
GIT binary patch
literal 124094
zcmdSBWmHt{8$LQ1s3`ENgra~-gMf60BHb-5(%qecf|4?nbmss=cZ#$F!Z0A+-QCQ2
z#`yi$`FPG+=fhzw7yEVXckd_e=f1D|x~}bK1vznC91<J|1cECm@mdK2x$X*qT-Cpc
z4gN;w3Q`yR!hR<${u=xSKPMt}KH%3aTL}#Z2!u2c^9L(@VaEspc><Aq{Zhp>aeczc
zGujE=v2*k}$%$)h`2nfa%>gPUsjqL>tg0r7Xn$=K`aQvZ_x`35eclI0m#~1lV-JdX
z`3}zfk3T=L<a=ad;vqmDel`?X*O=1KH&Stb%VDGT4pVFo)$b8NAil#a?!14mA*rUc
zu2_Hmqgh&U`^uk7-`Mo&>wj-y7Hzrx_fj{=kNEFpJ;8P6zn89bIG9;tezuhe>i=9p
zrnCN^Jh*#(1v1&bU~kqPMVB1*{@uKNjbYW85o??w`)-`L+oCP^ydy7_q;Z)Bwtq|O
z=V?jjIvs5Pm(-D6NX6-wCdm|(U8d}%LQxKKIm&o}{!2GIt*d$#>?NU|o}T=P*vZCa
zBd*Nv%lhs@r4}y#BOu5OBjt1F#rD1T&Nj)-$nT0>jkft{d0vjepV1RYWJyJ{>wmi@
z-vLp0#qC%1DcN&Z@?c|Z7|Eh=Cs8g3{O{SRsZUN$Pu+R(_-Pp!q}5=mP=b5+zTM?n
z82@~qhB9fCSiyV5va;;V4-SX>`T6lD-b^;E(2S0b&f4DIR&p^UVuJtp@uN(9Ra8vO
zog0^*k%B_x^73-`=;+9iz*Rv`F3jG+fhupXp~@3|aa3WyAcYx?AFXR}rzK}>{tZil
zB;-m|+q(!joLLBI-)mf!bm1Var>BQAX}|en4lDfjLPEbO<_5}SnkSOoeSNXu&g(Wj
zCM9oriR$^8e2?yZkxSa#*b2X$W#%(Ype#Jn!Jzzi|NdWZ<^N|f^8ddF|BmYJ1^ccA
zdpBO}pq9843L)il(zC0*Z-r>5E>HGMd}ya!5x%$e^?bk2UFr3D{?-R#Tcfy5r0DEi
zIe<h8kY;pxqEC^?yHbRDl9fob$jPF|Y041h!if8#?|FE(u$k?mT?^IWjl@xMdw$X@
zWF|xCrifWPEFwv~66PqUA=ntP>sUd-b0eeZWPxYQ%<_1Ftw<I~)DxV^XS;iQ%C@$}
zIR*qt@U5&WIvQ=JSj&?W*0Zlh?D71hH^>aDMt&bWK|w**2Ry;DZpKwfr=1<^b8|D4
z+mx|Ly0I0nmVYNt#5qAnkSO>Lc54=K>JLo3$&n|S%s4sK#HOS$M|M$0cGXQuzp>)9
z`qdzQ1Br~bK&V@~el@42q!ncShU}NR-a|Es)+!uQ4#$O_UtAb*#wORcb##b}O^!i~
zG*{fd@34Xl(5QR<u6JgESuiM4nrUR0Dx;s0?MS+2(UBKKwOB^~nqFqB0A8IA)K!~y
z>C(F(xZlvwhjwW_2R&uU*#)M(`@{G};>Df?8Sxi5lKhJMN@KD48D7H~@|vt(<0~2m
zhrToSxmA>-N32}!(Wu3|R+!K-yUy%IQ|k=3)^<!~W3wUaP79?%``>)2&}+D`xDp{B
zxHdN8BLS5$)>F+zYP9sL8dYASU2@G2+tf}+@hO$Niw6lNzL3-ty7(~4p1U+N5ApT+
zn`cuW^L$NC+hoQYI@PILAitLpAaDaF*28Cbh|BLH5-V-yp=OH8Y#A-nudU@jN*DB4
z*Pj@vC^PZ93Ss&bS^WFVD#*=i?m12Lh*WqFTeqRiVQT@_DjSFRNqU=7(3aHPZdugA
z&J@ai`X-MK_QcUlM|`7a+X6lw<XHW0){eT*?8%`CtK3qMselbGe15olJBa>gt)>+^
z1GYuN7~7MjHR-7?9!}0YlL_(l``_3lBxO-1%0AKIHeW`Cw##G|zv<j8&<xg6EkiW$
zMkEzhd7tNfP`U!a3I0<*eC?<A%$j;OZF8+-X$b#QOOs$DIU;*j)~1)?olvB0;mtyG
z4MK4JZ-rTCj-ryLl0<1uD9%5`O52gjbqTdEP`fBGmSv&33aR%0`%u=+*K%Q%TSBSR
z%jP&?`Lja>I2C-7P&La5-N*e9i1okWhDakizffKmqpT5F`w_y%h6PFfTjGQtF4=-A
zgPqZJp&y{E>W5sg_aTBbTm4%eyPdmtp_z#WXA{ZM32xhoaS%w~|DJmL_8pbBuco9z
z<*Gb6w@1^ZCV17MgQ@u*OS{)01o(ezWH+~eODY*-yfoCA=_0MuVb>wd3|(us%aVXP
z32JX6#D`2{|GV|Q2fRTHRJk>8AL$dGIMMi$&8n)X6pvR`&ZaqEe0^teZ$fDA{qOEm
zRr<9dybU9S)bKeocISd4f%(PzP_}aYc8}e%H|(y{g;e+NK@?2?&E`J2pleFLYHq66
znm4Lmv8ppUMLE^!Y)B53(s~9`mihNNeAqh~$6GXD{Qt`R!fRKBFWL3HZzNgzM=Q>0
zWFNYJJS<A4jO}r2*(hT>G@Ic%GC#~|Fe)=PTPD5Ak!ldwlSl%wWigh&@u#xH$`^ir
z&3G{{GI6?Ccj|!9L_BnExbkW)T49Uix~O`@$i39@xj;usqmr=9=TL2?kJqrx=fOyP
z5{8n)@FB-H{!KXUYw-(&=NiXFtz*sP$%+Z%WVyx|vA5Txv)#-L$*&s!<eIdUXMX+y
zUkD(!bpK{80v1$2lU+HS2;oS{rX0*=)!(x3xbUVh=r+E3VdiU&JVhNb{}!Yo7KHfa
zALJv<Lwuhody-Mhl@}Y5dh;)?0lDQ(nVJDQF-Jf=jJ7Wx{_B&ijz&hn=9XLGt1DOd
zN%0}JcmMa+6X!ZIQqEUXz3~(<g!AMu4bz)k;-*Wn3h2Fq12*|wGekU@#4SiQ!T)AQ
zl~9%2<G@Tu!z>RaAkx#ibx@>e>!=#Xh}F%LoS#7_?P$AaX*E(OuP$&Jh3#kw@9*PZ
z`M>9s6Sqsf+69CQNa2L4CsG_6=;T@RiiYXRq4hh3XB)Csi3bsl2^WEt>k~PmXb(NH
zj2+e-Y{>Z|vWVZ9kd}sXeKJF9%3U46Dvq&n*k&K&`LMk@)blFEU^%+3b}>(57dPZ*
zkkuH+bfF8g!8=HEmckzpsUYX*V^{nbFEF3IL9q}!EUMp<?q_Ya=EmU8Oc!gs=RUIj
z`l$h8^qw3JgqHT-tlzVgEiMad)}D*Z4z7kVQ~IYVtQlBn7Y0n|MNv-dU5ZoPy$eye
z_HS9QZk_L!QRe9%5zo&(QptSiRL4&zEnIy_#*?oTh(~O~4-vlmFG;uaB*|G#`3SQ<
zP{|qz6-}O-BQ6LEb(ew)#&1j8wSh$P{$C=oHd)o@NtFB8M%Ba2-Cn3s!Z9iVRm;gF
z5NC(@Cj8s~InJ#K21%To^0ppo!f&oi4)p(+v+lbBsqO<{_V>^96ps!QDT1HL|78#k
zZgj~PV(~~yz2Lq)4ZMUix5p<?%CfHc!{-p+vwzPtIwX}v4zn4Xk3x+!s~lq2@d7iI
zv&=KhRZ$tD4iiqY#Hyb{Xn)p&vn~xS9dNEoV&uCR>~Ch7(f|GG@V`?~dd4iI2M;;X
zO_7Oz2HP0yuWWo7fpbefs%Jr|x8y>z#Hvosa;&JTBW&2f%iV0dS<8q;?|j3{^YsT$
zFVCpxMsXTE8)XSg#4{8AV!Le0JmtA|(PS(Lt>C|<zmoUx;VR#}6XRH^DmA$fYVGuE
zzMfOUxBIJ-CUQN~_H+hl&5bwgVp^Avi>lhghW&6Kad`i{9dxVi_-=))xe9VtR;%2q
zCAPM7YD~2IYf_2&4bI-EmbX7e8axQhz&UfZi{SXZiu~(_)<s2b-d|#@M)H^G5ranu
zR_3E<osTp_Xfz1MJJ_O>97<)6gcMSSD``uW?lX=TOHSx{YVfG7i8ghdp=;!!FB29$
z3nI<xf}$K2z7x+o&cod789!dT_UBE1Lu6Y2$Y=jE=jLR_@%-WH;W|I45jys=3MxiA
zXH(aM@+PHVE}6?S@&S5|XU1I|DE^x4$CdmSEbHzg1sWf_tBp2qW<5XA70XbZ5<*+1
z4PAjq;r*>)WaX|%<In^?JuH2~!OP1X){|pG*Fuo$!tiCId~8G7f5c~~j=5+%o;|c_
zD*D+hY_zTgo0>AqSvNRizL3>sDnwscF<xH2lpF{=nu1A{uw3s2$zS8|Zn*x`GU$+o
zVv6WgYUeBX(d{tBR$c$TxC0x%^YilKIedc0)9`%(hJ&8eT1~zUj)f$C_sL4R938Oe
zgXLUW=4~IAC?)0h8+7pxHZ_(*AimI6(V^eVzx}0PoFAcCBPDqGG~XL#C7P^UhSJew
zxqOI0OKvA8ptxLmjeWniadP>n!snEyP<A9wH4@tlHQy4H@0jJ60qeB9rO>-0A~jYa
z12T^!6yBDu7%r5q#aqd2F9s3rO8wJpHj7|(*t1S2n#2*>f4-Hqp7r(lWsv;l?+Vpz
z#W$F*n-pwKe(%53=vJYjbybV~?<N!#R%ooO4c(&hiJra@CVp-Su5z;seDZpEA~wP+
zkSH;J-m92-Oh3IbfTWUq%7U6mCiHY~ZwcAI0$&--j2H?(m2u8d-Wx6)UKH?f;^z}b
z-KWo0o#2)FM1M|40fEF4|Jy`pbOM7U>)G`hfu0WLH&91bnyQ}`2(b)acJo}kP?FlD
zs}Nha|Aj3j^>Z0$l*@VXSed<iV*K4lWQ&-ye)PZDnp0Ct47e>dG6t3F^iWEI*H#PM
zhD;0mO$OUv+@l%jSm>q1T^N;}+b5s(@z@y#vK|(yqK;irG7_}yA0WO1VSi3M1fS)>
zo!SZm56OY}Jdt326GYTgXeJivSkK!R1k#)kUzdNc^-o@%5`X&s{pjJa`e@T-)hqZ^
zopGsBjAzcLFaHTl9@XBH0Fj0GH;B|v({FJ-=CQgK2I~84hSCc6aOw}Ivq$2E%zY_W
z%EZZ)B!q<}+pa;P{`LN@zYrT^D_L4LO5U5d5Ux#;LvkLn&K8GF);0hX97L(J?`2pR
z{=o;RDkoUa*9Q;<&94Hazrzh80UzfSCVWP5uv~mslf&!_XWz-|=%QtCCh<V7jha_-
zZuiczJ`wRylT*-Di+^rh79g}!y}3Nlh1|h{kiGg_v+3;gj;zhciW{-RFV*f7Js*sz
zp(r=n{Iple1{SJB=Mip=S8Bfcrz2q{h_KneN>8W6z@j*(>#hs3+o{!xeWBz?-eAa`
zhlUECh95}O;n|yr1bXV{KQQnh3-UKEvJ^nt+}s>Ofwl^g_4x1K(mI5hm{_Q<J|*qq
zFM8r5MD<!c@10}$9c=%c3@qW2zqk=%<cF{x<^omkjK{}(CLq+lR(r{_K}jaWy~;eT
z<4hYK#L`bDbqmt=7dt~-8=5#R2L;H_<@xX)M^T<!7iY-djo!uo9L>0uDff6JP2g;3
zbxtqabiRD*z!BX@!oN)v7b1pFa|3epw>CrIhp7EH_w70TQohGQ^44%M0X;R#P?*t+
z_}&qwvtQm;Ck9_R5hp3SX#E0Aov>pTfP*REQ-Npfun_*q3!2Kr%qSnh)7NsbFV%4$
z6j>z4XH;ui^)a|m(Y@eTRNwbfZqU=s_*tzPU-B-?+GnX@Pi_GMnf&lKVqLSYnyru=
zA9~mW^2K56NXq)iSH(9t_v7~=m7^shdhXhMYG%#!aB8{QJ7!p^hW{J<{;G=*Ed!Uj
z6R%)Cy{rxGpypJngp;dn`W@^H%!tE4tE(XOf3joqnqeY;UO$^j&n$Vmn7|EUGqKl7
zBC0k{fng>_7W*M1ZHfOunEoO3X{T&vJ-WKqw4hLUtc`+U#L}RpR~4n%DwL{?5EsJ4
z@E5C0JArIPFDaEf)>g>VuNsd@3DR6P(ZRumSlh4uCg$d~$1;7O*$`E;w+L%7ReI~1
zUci~aT2>ZJK?2cM``=yjV^&bgjFH*H0yZ{q!&n()+Mdnh4@;N2K3eaRlKosa{bY@E
zS=hbSes`CElqg&C^R3!S4P_nVzSQ-TGdEu9dQ2Qr($Vatf{eq8-)+cF&&xizq4Aqi
z_|5DEO~iPe9ej(;;4s=bAubw?bC&=^?*FRcKyde7I<a-*OZ>ytr@?VmxwO@_(eg?0
zGuxHd^MB>KL@Yf9eT^GUc)s<56zjti5fjK{pou&t5QIxk%_gr*988LL_f*k8DeyDb
z^{TqJ>Y^+m;}qw0HQHmgc)*zvxk)%z;emQBAh3J;qSQ8`9pC*Wi@rWU^vwMYZfpXA
zo|(>t58w=^^@{V#9P65#YS6(l(X!c)8<1GsKPM{nS4c9C)=@n=!th0|VqZ)C)J7Al
z)|K87YVQ0Xzl{7`DdZz3OEfP6z3U7hfS6UXr&{pkQ6r`Wh$5TJl+rHX<+RS^feo9E
z`7uY0U$(sJ>+F$2wrh}g8nUle_9jU^p@Ky5UmpIBZ-wVE)|2TsE|2o=h(+=Waw~><
z;Pb%8FLJiOW3OJ0>q!+%$$}F&h=4aS=JYtl-D@3dup8chW)cbs!3@rZ3C#<UX2a)-
zS1}!%iw;th<NkfAXU`x}+<yYdcPx=9mh5z&ZDie<qi*M_s->sS`{8)6tEf$(ISk-;
zP=ohZ*X&8XpzDS*m=FD&#;WQxMhyK3uch<q>%LN0wzOn=nTy3+Y!?7!pk68T*nAG!
zas~UpncF&{)b#Vvx`d)(7Ipr_%nU#UJGa$>wFDz58_p3&_I3o!ILldBTx*9+Nl}q*
z3Qdhp6+<!)pJ`2Y-L_e`H$$IJacmqcj#hXk-pk0~FzbB#5gS`~@+U+gg!{&C?zh=W
z2$8@nY5sGA=iwbuz8Km@XK_(=v+Sh07_);g0?<h{a8sm*u#((4o$cySVW3m=7x#go
zly&<99&XX4>8gvWTGKauc{8#nxB5MT6ci(_B<@lX=wPk4K&aOaPg{V4z7S+G^iRH+
zCIb9J9D{n7X(esiB!@Ku#AQ|3re^j-1~tX)^k$G)lN-`fPAjdm(xxs`Ki?)YKkB|U
z+F<|9ZHzNtSGI4;(f?F?LJ1JaS3x>){R{fW?Vr(z$7zh<gu<xEo8DF#16UbLxxwRE
z&7Gc>m`6kow@LPrN)b3LUM$kdA-$pGM7j6X@BFIDQ`T2jB!kKezuw4qT=RYRmqM_W
zKK-n`#CL!j;^+}$qYy<&xAFZ}l}Xw;&h`PS0uc{LXNWeDwjnMi#-?Mor=LhFdH1qZ
znP`-;QA*xz*(AMn2O{kDw;tI_zj>SLcX8pwY8CtVq{bT2&0j4EjUu?OgJCpi_x^1)
zsQs(IG9v&QCH=+TQ>X!&_-INe>j5waEQrG5zo`zv4^Ys|1LRbMcF{8gH?&kDu!>pX
zl}c3nP`*au4tQeOzbCE@FPc-S%M3)2-mdWK=lZC-)!B!Fl9NSH;*!~x$%suBU4bb4
z$8YdEt&AIU#_e~B9QByKFx{y_E3U=}-L4-p>l^liZ>gLntD5WbX;itasQ5{O4nphi
zE85b%G6LL}8tWE{mn;@<=#-oFcG&Z75bH!nS8A~duBYqtbmYA<%=9KZ_xnMOG?Kal
zu+#s^iLUj1P5^oCA8$u!p3WJJ?k?F<Re1=1K7gwN!2U4Qcw|>@b-BoBwBcQSCG)wa
zT%ha$Ol2rD&7Y9`?#xO8&_UGG{@_jm*qMhmjKN%Gm@UO84^v$SO2RExesxd*g@Ig%
zXV;<6`qA_njK;cNvTyR2{CjV2G8tR|ss7e*g1YPq^VPni$$?qf1P-_UdkC#641nnF
zMkrg#_?c5u=6dX9*A}5vgKpj4bJT`F>WTiQwAT;rU{@4qi8HPZ?Kl!ZAl=WIN-@e2
zWFZk@t=@PsODwWS%@iv=;^qdbWOCT7(_^_D`@)mtFtC!>!hVn6*M8B}xzhXGAvv2#
zH$Qr;iecmUmZZ{a+hP(`cs>~-VJ!x4gMg4^b|m#<v281LXiAfCnOf1%W^15P#mM~H
zTn2y`tDd6w4=lVYG>72~{w>ohxmi+;6GfOq)ICdeYSAi2+RaU!)06eZ7lM+KQaM^g
zI7$aP=aY~xt>gk4^KZN^d_XP?bTr8ftqFx7OCo8gB*ii`-rtpk-s$E~&KFXHqsU>y
zs|zesD{+rbSK}_d77UmgY)H8_Mqo68XWAg+@tjRP$C|;p6|SPIR$EEVox5=Z&-^&b
z%zU``y`bl`APTNDV&EnnsYk>#P?pg+P&6n4rSxy11!?3rGvY1LQ)P|ZEgah)JrY_?
zw$?dVO{${fr|oRb!-Aky+*7iCGaGG-fGf22I1`Z7DwP3@Kw%pMRedVgHe||}(IpO@
zcN}$MH!5*{eyKKwK9$0xEzIDnW0TblC5i)WT28z*swJwL_TFP^@nap;gIfNqOtA)n
z2g$C^35h@aN9&A(ID0+y9EOXOd&E-)hWq%_yse53-`e-tb0@i92Ez|pKIl<O60SNb
z+1K4rO|v#RIZkG>1#|&lu`XY1IC7CE#q@jm6uZX(i7W~0n=Z-E#Bq3#>hZt1H_DE)
zPRC2vO%@sRT-nx<<AAJQ%4+jOZKsR0H>PeNFT>5iqFfC&COCW`!bQuUmW;_>7J84h
zS9^O-$4VD!XHZ<xk+C^q4Z3P6>Rb#p7ye68bx?x;)LwRNYz`m5t(E510_eG>@)LA@
zq$7zLcNFaVJj}n-dcBw-*0}O*a3z9n^>qHZs=56II<^jK_=1+d<b6@~rPF~2o2Vpi
ztwqOse{=M{!>9{8!eIsrvh?S$`Lb@FFpFDkjymDU7r>L58I@lvWk?s@3YAtnS|X;)
z9#z$>v8p4xbUZea(0oCcmsh(G>H^fj!NpA$^YJyphlrTqB9?g1!Q74~6%AWABnO(S
zrjV5vH<-l7l@<X1zKrZrULYpOBV&H9rqr4ot35~jxG)S~UM*D3W!cm?`xMVj);VyF
zH4~7U*Z!QsD1EJ?tznwT>`s6W<?!VrY~H?pKYV(89oFi9ACU1d8{^&~z;Q{xFl%Yp
zM?ElFSj@lV=*t~dy?VQ%(!4AIm3nr&yQ>FzTQ8I;&NoZNSvUFetW6!7>3oFF{8>GJ
z1<vt6)||&SGabf`%-`T+cVAqG2Hk?tQvQJgw5^k5e7bl!$_I4<Ie@4K*_NCHvo)Yp
zA6|sFJu13cxjqKp*(2^0qEriYZ`96wh+Ef`>syQyUv66h7z^bex1{0BJ}G$=9jvM+
z-I%6XVfjF;OcY^rnL@b|O5P4+C_<OlfDiivqlF(Dm5l)uc<-UCpp-TGEkL>Tr%W2>
zBJ36m=3s(^2H(CV)SK?)$yjYhk8l-Bu>FXEpv+yGC|6=X1_v{0nH}|f6NP5hCZ*8s
zhAihZ66jQX#PDuF=m=sUfi&a(!L|xwg=00+puPqb_G=QMe?IGDiD%4L$SDITH^y>O
zJLJ~wtv-%5N=Zpv_i7HBJntjq-cW5)*CQYk0T8F?{ZcWgXmq?#LlReh;>bZQHv9|^
zBE|Q2-7*eyvZSG}=FDpa7$_?jdDsLNgf-(Sk8G4XR}M@Tqn$t&1`QQXSRRvqO9TBA
z_DatQ7l^{>pMog-5On)m(D}Lc{yLLHo-5pJ3_Tkfo7^4==dZ@&=L~K|0mQzz;$}3_
zm8dOgzb3KC*1Im>(bMI11Co_Ag2ZGJ1!qHJBa6juRm~=gSUjW0k>Mi+)t_uDB;?e_
z0uC0|=H^yD3D?}a@0|ashtF=l9zL??5LN3IHsA~gl!!{#0@TMe@ficB+Q(WQz+pPJ
zQ+NFO=LP6Jtf7%de8R<NIGHBC3lP8m%EN9xH<imYJJDnVMDLfL6$P2ECN_Nw_Xyqd
z|Lfy(yKGc=V`3Z?`+wG`+R9}0X#?vb-vztP`*pmI>s7Z7kA|-b)UCxmIs<ue*>4O&
ztE)OUM<u!9p`LVM12DcRqI?)obHw)ZOS-m6{g#RcFl*%uU1kQ*)j^|5Tn_T$>_nMW
zgGDNiT`^?OegiD(4SMuhpQMn3$E<*d2U9Gd=0Q)wz~)r8e!>GldJzBGx|jOX#_Cxt
z$hd;>?MVvpu;r>LZ{+SmkJEhTT~Tpym?}(db-bqR-Me=f?i`Z(<H5<=mb>S%1OUh&
z<ERws%eSE{GklA$@sM&F<!tr8mo0O&ox9VphwMAGKb5&)Uv{goQkX*azh{Muy}GXY
zIUHH!W#P2DEN4VEp6jM}=L_S}2N@zCRr3LHLUR|Dw}zsU*b9%x;Ddz@6+Rp6R$AL1
zm14Z%FdOtMkeG+kto8^W-Ya0RsPPfhbJ0fb<H(*{q#YKVwiWuh^yPnsW?CcSiPX!L
z5QgmeJ%qCCJMfRZ#rn1M94pG1JWM$v9_sQ?$5j(7$aX!e`gW~0lkC?QFAds5c$S*B
zrkeET+C!1mc601@v#;Y2Mw{7yr_y$2_T<sTu-YMFWh4g=GTn7jV#3iUOKk-cJ9mvt
z)=}M`ndyzXTvkuVrB4rQa!^gO7WkRi-S|LO(i*Kog2Iae>d$ItZyU5H3!~h%Hw1#L
zaahk$F?uPEA^tB8oB;8kwisHgTEd~#v*5TKP-$hz>VBxi)KCOChrn=^?pYS|UvpHT
zZKdB20Tg6~*UT^gTZ45*!S^IIcqhiw3aKwE6fANYvO$;3Fwiyx)U1~@;hTf1PUf(C
zH{`K|^Ar}DN;$pHT|*zTi2>DCUg)B{J@ipdS0s(aw-2{Y=Og6@D=gII#jgR_dA~84
zMH(~f=UDoyr-POpjn>Q8dnVYfP|)Wi1g7#An(vnKCDrat;8)m>9Rk9LHYIu4qrva|
z!imx-0xZ?&#8HMO-ebEK`=dKk%{MvfX1tu-4TC_=BAu)hCMpY0ZaQi2`Fj2>{guf6
zo0LiWUwakX2_B%%->#Tka5M#~RkwZ*l4AHkMcdW&78RqHdsV$uC`?`XbP_fCV(#J<
zyic%Fuct$ZLehPscPxmH43he(Tcsr-CMJ8U@k|cN?s;rmVTKd~4i44_cXoGn6oG|g
z|ECDP-98h*nh8p5toy%;@3KCSUEOZ(3xc64n=LFVF2ak}&nkC!_nbQF$h&78W8Nx3
z8U+PQ6TR^Vfl#Yz=B)gFn3`ibZOA3%bWNgcBvV*u>R#8IHvos~RWPUxw#b%MEGL}J
zJO08c?_DW2>vzFlZVp(<kNcXtWL=x~Au@(Z%>N|?na3OOSWT9hCMk_w&}J?^mPf6J
zd(a<!$LPpxIU`uUN|`c*^>ZEJv6KfZNO3zd!5tyc+-n@39vMU%ROHKbh!g778Keyl
zZ=IIE%dZ+(Z16akSQnCd^3<12Oy*4XA&oqJp1wSQQAMxda8FS^RU5Yi0FU-Bk3Um=
zDc%Nc42RkB6VZ<W=4apF9CH_Ppqbf8w_sJ~vS<hJ-|CW9^uq_!fiEv{aPQR1e3lN!
zG`+F%0!;pyCwg}Vq2%Pz9DH+A#MRoPRo3y@jYBFA&xV#)-JK3rDDZVF>mk0<3~Yg=
z;Fuxe*u<fOH4Y~DIUq;V*6nUMUu<`1KSU65EMLCr4|s6Le8A&VcGHxkfia?gLx-2p
zMgn5u#b!}Snb+k6*wYL}xO%xk|AWvt4sj7{@+bZ+3fenC1)84qF`m2ci+oO03#T~s
zB6Y1(GXn%vOYCeu(hNr0yv-~ac?c$@=rJraG>M8_u;)>&dnYPO!YhHAofGi7&%c;`
zXhlf!AQ#B!;t5Nh=}~P{`2aseVWdxS)rhV?(x>Cny$#LF$N^{OSjvfkv?MXX%Uqz!
z@Oyv+u|&&ueQRf~BN{j=B!hy3k(E|s?BJ9Y>eQ*dfB)Y6S8<yA;d-7qD&LKp4#b^{
zR<(YH-26Oki8t5Dbx}e$IqgDcAnOn4J>sNiW6;unxv(X{o0_>HV6XDXc|%RY5?7Rx
z^Z9P#bS`~R(SXdE3nLO(5>Tn;(eGq59HR{kDwci7O@Bq@xm&CsD~oQ%cM*?sQoc|?
zH<YB1)fB~E@s#e!qvp@kDr2bd9<;^1IJ^-G4!71~*iXQ~GF4mIw^24YcUY}Xl#1rj
z9syuT&vxC$*f^V#j-Fo1Vev<QPKvktp@o~9+v!}m(0&KGPp*85VC`HO9}}+yGF$d^
zCXk$|R#sOxSuUA>#o1b(C5fr^DJZ0rBUYR|kMZyW-_!7*o$(|R5Y&8UA~nv7iK%L2
z%sm{#Q)I-C@p1~74sS?~7JGzK($(gBs3&thkOc@s`)ueUpV8we-|k@G;^leV#nY*H
zW<}d5iy|bPJZk{WU!p(EB$?x*H!*IL0n^FK$hkcU2)MSp@HY=|2PhgmFuewc%LXq(
zw?fP#{IP0#S0oNTjuLPLI9ZNMY&_!N)><c8>q9I+TdjqbDl)4+;oa(|!JmM?teVVR
z=}|YIX-7yQIu(jg`Ji~BWMqC;Ac_VFIV?I&U(ime6+4i|iiSzoqpQZy=yyI~bxiPu
z8E`E*6Vx|TAUH)o2<kBow5W)PG>~akhIL$$PLH<Wz@w&7sD?=3l5Y*bMSlN$A62X`
z!eKprtY<>XWB(eNBj2=qR5rF@(v-usTS2@jRkrXRn95uSTFut+xm#Q2sN@`YlC<xv
zm$$Cu^r9a8w|bevbrzUhYyHaHy_Q6!X#TLi=RN+19ha;Nw^@Z-=Z$1i^32gyWAE7W
zrRT&}2dev<td*5&8ZIgf3>CguTTQ7|RW@y3XVbw8>^`H=)1O3-Nom~OwB_zSd5a@F
zqHN4mYRno(ndJV?YO=m&vv!T?`i&bv!|3emlh>(p7VwzgRM+!bXvqZFYZB2n{7kZR
zX{lPE$6A1+b?RG+bgU@omd$A7OfNPCH0vB<mDs|%B)b%zI0uTHAqOge1&4EEBF8{O
zLo|NQgsyQ)y(4Z)UCst%x;!Q0uxZ>|w|LIB8r%<Rr#Uyqs)m>QGb&8_$VMzVPtJQj
z&m^#$rL!2cYj0bCw*dH3cjd|z0wSVCZ(Dr3(-Sw|S1U92UFH?tz_ZkjBqvDqJUuhx
z^?XkBFw|&~?FGU6jr|nVNusww$Zpq)WAX&C40YrWZ51)73Jc&n9RX0OGD(u*J6M{F
zo}HW5>efr)CdI?N`}EX4C_ws`_q4BI!f67g&2JD7OWdo-{i)z&GpXX@vX;NtRly(i
zQ$P)Nux{2E<j~|)qaEL1UeOrGBaT!r80<=PkRF-Oq&%q4uOVAU=>UP{HhY=Ud0vKf
z@ZM3yJ4?$vv{yr7fg0?GaimN(>)s&=@wCt9BQ%J#?w+1V%Invzbt(nvifBHMDZ9~z
zMVrw5gH6HIba})l{L`oEY0mfW(LOU_=jZ1>jIJWDUa92BCEfPk_>sc#AXF?4G3jx{
zLtE5qzcSc2_?Er<`-M^wPY8C>$rxKkOY@~d?}&UX*ZPfx@u@I&_6Kg-+;bqUQ%Rn2
z$*4PUExsumh_1AVd-v@dhC^&u(v45{A~j|0P8~=>DPx>KE`#c)=fhKd$pe(n8!G{5
z=r(0#IQxZ_Mg9hk>D{>4Ox9}E)XjcT;L~xA*Wo3;cyPz$C~($Y`<LAdRrJZW0fw=G
z8nSB<0Gq@~V0({#52`DpuG~m32sB<f^3<~Z$cwXOplQXN+{==UTtyKAK1f7s34n&k
zt{l_0f}WM1z-wd3N`fl2wj#qvx4fNhc&%Yl`~3{vME>|4Yn+s-Uh{hf;SqIrGQ08l
zwKdp>57$V!ZH43F;-n<F(V1u-pR;S@u|f=hQKN<C@7U$5XO(iUk7f@t`p^p2h(s(}
z>b#I*id7bXo9nLj=xC3&?B6*c<9`h_X-waWt&@Ujap_<1?<8)3I)veEF~VCU5fkWL
z`Pkh$pKuhYp+vd_K)+>yLWA}{E;{SWuXyu8hhlopO-AymREJ7FJJ-6(-bZ-|AJ6>Z
zMsgT1p_q?L_tEX!+$cSdY=VpG6j`QA-X^x_m}SXFEjB(g8ZSE~25NGCCGIN2Gwx~$
zHgJ@>XqCz5_5q53vRjRo-}d%u@w*PqTt&fR^W~A?;MOC`xb`C{8OhNhiHG0x==gzT
zF@!#h2>}%DvWJ&lN2^gxbV3dpb3*}WCkd1rqEG89RYTqH@WRKVDFMF2FnwNAzl!J)
z)&rLr;;4<b77WjDwM)Rps$1@WgkL32LnD9e<d%+*qkn!y^+7%G3A}n|k-LJswNm?Z
zjS$f5W$J}P2%CDwGb*O@BN=ofG?PO)pff&6aXEz3B1TQl^CfhDJ@f*nfeZqPDqvOJ
z29;Lr<;$0v)wXGkXPbNri;GgcPS<;c(C5k-VxdsV38!J5<d9o{);+K`pHh>xKw-E~
z&Hah1ONqU+N}v&n-&=9&?h`0F^Ke$TA2^|UQr~0HDsDQ_s{`<$+AzO>IB4x+wGAq-
z<xVsR<r=TmInBSxw)WgW*t=j=K8C4!1=JU5!&HG1p$PmAU-}(a9J~U*4=G@%6LuyI
zHuZvmM?xWW%ZCUHnzeRkD-5si6IPw3K;B)Qh{t%m>zu@0uY=h`CUthYZA}Q!tE2M#
zhToMeHylkW3k&;tpRUkMxXy)QHtY0o1BNejt^j3!+YCX$)caZie+#Kk=zsfD-d|V2
z@y*vt^g;CHW48>;p#tfm9;vB}smWSRv$|NA+lFC-12cKOM_mOS-vQ#s2!I?)XJ;Kf
zhD_h{e%`^>aaywlu8o0GMlBBSNk8aqO&l*)v!KJrf(74Chw~M~Uw+iIxOG!GN1VFH
zuyT1Kg?jnzX<PV@i}jK~)f^?!EbHWql`R*Q%^i{0bFvej4)(P<ve7d0bo#b-^B-mM
zJ<TgB;2b#tgw7yINU!p9SkI?$%(_lCnqC#!Z>nNwm|5$%lCIwxuoCdddfO4$^L<&L
z((y{MO}WcrtZjfbrsmhv!PrK|?+5mc9e$tL(BzqMyI{?&95^ac>#22OH|u<I5}ghU
z$uONnyQKpgQh4k<@nfCeyo>Bv6X+CN&Of|EsKt+v;&~nKF|$4hF?VnEP`Vq01m_O0
zE??~T3QsLpKguh+IFI>4E>2Gsz!_BytU_nqVjda<LVk2~<ypOzLm27L)0M5|r6t(f
zUvR%-)0m(8^UEs}&^}SYWs*M)O-<=toE|7Qc_)&CuMtuy0aL0WEp(y~j`t`nWeJ^?
zB_x=gWk9a}M!WR9mA-SKn2y0<Yvf+nJI=M0OD~T7u=l05ve)1yWn%|`9|R69l?+r8
zFngpq(_9x<N5zxUje5-I(zg>d(jIb3bJ{45e!h?KW)vTfiY7;nZ+c^zN{nEuzE!=E
zK1MAnZgzQNYq^HGacHUOvIrHaH?-Qt%)nh6e}w|jg%6^<)_vt3Pf|;QcB!^On#Ri(
z-9X&%<q`Ob;JL0UfV$0$Kp-SJn|06}@u_P}6jTOOC#q!-(JeFK>6|c17zqSIhl!~&
zj~d#o?qJ5VLk56n?J7pQI;KK1_;9hQ1~>8<9fW|<D50lKu+$WpHx>TDU1uWx{yi<_
zmx|F9W_{MAeZ7r_cONY;<;atH3y14YX;I`{Waml8&@a3Vk|UGlsguq6`ryE{U*^bZ
zE4wzDp(Ywo6H)xX`6q&4Oosv>nkgtO^mB$Wc;tN4WnG!aoj0$aoq6z*Ni_p&p`?<n
z3SpJW^`vCGJcq6r!W+n-t;XA7`RN7a6TDxO6{jhzEGSiedZ25>w=R~*z=+h$+gW&Y
z<l4kv(>{Ok#bM(LWLosMXrfR6M6PEdet?x&U!kSdcRS&fgm!iGZD<(x<l$q1XtM6x
z#RirfXwXt4+sSbXTUqg|t5<ohgVodx`7NUz6PtC&FIO4$94y6GU^Wz(6c{-9wE+V(
zSU3{B`ZG68U5B=J#olaXkuko~FphI`Btzt3$GpXk;>S-C6}gcG$&B$Zb5HH}OKkyx
zl2H4woZ6&~A0gG-wpAu>{?(t9tID`H>&DrRbY8b?@d3>a;c;ZvJ)4E$bwLw64<#Tb
z%ML*0=c@-Vn!hQe7vBZ*yZ(EAzCIFs-mh%Vxz#ALFS9-CF8szkIS<rD13#{`X;}|?
zP1HrN2DjG$9z~~Ah1r`!Ea7OgVVZMpiTS58h1kle8sl%apNDP->R>@l4UqG)Y4^Jt
zZMlbc+JTX+)F3TWdC7ta9~vn)(M!Qcl&6+(7Q{~@bW|O-p$FeiebV;fBcUk*P`wA^
z#CjN#n)v*5UR0WWe~f&v-nn$L_6)D0NDtn5f+=zg4e(bO3K0KIty}Qv9k+`?o)3&z
z`6H*!RG?x1Qmt8f97w-4ub8!^ktYE2?r(HLV$>=UE|*52Y0ZvMXiDFQA+iGn$+Cer
zK^mG)&!vho=Az4PN8#KIjYUSlF%bgPT_qx)9E{#!(-UJce?Qz*xC4(kiS+K{0@&)%
z&^O!!N7pe=;P+O~4B?5p;pBRz7ld<5v20#bG=9damb2j%b%x^EkZh!Ykhrn5aKz%e
zj({X^eoe&d&_(XdUa#v}7^s`0S#E<jMgVsp&=OQF>|nDNE-DJIMZ4ZA0J|W@%FGw*
z3C~L4f?Xi{R!$OZetsFd(cRe(zV?Zd`pxt<o9zu-DO3?-cbS*vX$E6q9C{~#$sXN)
zfTa$SFwZ4XpUT=7fR)?+Xwa-Ti`Eo8j3T<&x=>Qx9bI<@Kt=@=<QPDP0vabT2-iaj
zO}h%LV{|}M=2$%l;{4$%Z;^ecpLEMJHRQgA`~Ime@dcTS^G5o|N1SG;%$K!?BPUA;
zrcan64k)e4UQ%War5M7QmoG8GTYT}`G)}~23W@faXPOE3#)&4IQ!eF77*i5+%>~er
z1_SR?cg@e`WOLQ$Xj*MX_Rl-?a=wg+#$XQ;M-MqZA|Dt?p!r-r=aM&&Z2N7I;db3X
zsFO<D-B|@CA5#5cA`l-o1+(ciFI(yMwt*og40;({&~(6$h;a`Bu{su0-7evcPsf@P
z2U_kYlO9AH!!-+wahwMCvKGJ%!=ks}lu-$Xy5p%Wb$Ca2t;2G}!v+eXVa%rj7=y2-
z<Y(Uc3uddYC*Emnn_7ITTErnrw!*%b9!Oju86X121r@+vTV&&@j&3e|)NQgHkK}c8
z%)z)G@IoKyI*x%jTWh@)IsfRtA`5h#aK)IAqi)Ta3wDMihvq~YMlumLtp_cG7P)Mq
zHgEw#Qs@6r<<R7da|WYC?zH(se2n%%bS$9)hHABSrWir^$fCbp_SC$+;GUf33x+DI
z-%<+Ki&sfxarw>L(kaSEAs^g_a~G^qdRFXLTI2&R*4#c(by>2mB$%W?;o;#i=<|Zg
zAEb)YTV*~MTk%IHjDWlBOL2W2@wDZ-t`-p`4b|#Fopo$$OwI`T(R16bHG7_l)!5kb
zlRo{Ro45Kub&%IA9|8huwd-yMC@rYV_d7xPo<{H40#tKhc=+;Au$C_~I!t{uHQx_k
zf%x|P`DS2Y5>&PL<Kb7=KeWTl&@Q(dCV^A2XG}~A8sL5AU%;tlC;q%Vw!;&c9PB4x
zIAKMa%Lgc}$?OlcI^_%pU&Qp%%Nt9LaYMZiCF4TH7%w-m5EtjhMp+w|!yAPfE1HK$
z<@gIbd|(eHaqkd5utC=TeAwfcb9&I<Fsz#afW^U>R$;~A@UT&pF?#K=;}AYv5MFDx
z0f#gEoYK3G-IPjiR|4Xk{G4BrrnF2o<6t5PK2h@XVFrv{IY};@4!C@6&R-P$T*Cqh
zY4$PZ<2%};c8n_{A9?cmncg~mzP&QlV4c_vPMP8QiZ#?iH+9>?5`_Wz#K9kg3=2RT
zi!a=w+fBrt&>I8|42Sb3|HN>^m}u3;RCdMew^pZC3mdcH(vs1#S(CQm%~TvGUWOOX
zz$X$KH6YiRj`43yz0h&Cs61zEBsuv`QJW=F_snG98JO%_$-N|irE&Qqko!=~%w}^T
zN727h>i~7A0)h;n;clt1!iz`QYujN>l(EvF03Gx&Fy{e(5pmL%ea-${O3vX{z;L6&
zeBuGzP3@71ck0d_(ehm2X6%R|{qA$1z107yiklaWo>`Q6N|9?5@ydB!2N(|CxZur7
zzn{xtJ__hRQszHJ#LBWd9p1sJpnA6DMApADBWr4K4v(6a4-40Zkuh+wRsR&9{$e_1
zr?+!5q`$cL>Nn1x6HW9N!mx3xMQ<rai1w<h0kK{H1rK*GH_eW1I*|jdm0CA+E2?z9
zZm%*F7ays7akko$Jh5O>v)|{uc6oKw;kX#1?I@@$fJMO#Qn4(ol$zidXQ;GN4<79j
zO$Ogs2K{lDb~otc1p)ym^UK#&Ys6;>b4tlkw`B#p@ARcngDR3rhHdTE2$l!3IP#XM
zg9wN6txiQ$p2`l?0@Qbo`9m!Cm7#?F;eb%hN~Z*j3^g&Mp#8d<L?tghOqPVMd3d<c
z(nt|}c!BZN7y-p<C|?!hD#J5*47xh_ECq$5(*&=1ytZ`vm1J^F-U=IOr4I>bew3or
z97?N;JisaT3E&4`r9bAq1<fK3J>W^ed}ed_96kL&av%a@ezzAt-MJ5?Y??L&L`IFQ
zcsTJU=gC)u`BZM&D3|%NbHFb3bK2rxy^@}bOKOkL{Pa5GF8GkZp$17k{;f-<YVVaD
zcfR9f5E{Nv;uJbk*&!q&02X<d57^jJ15K0J8GNzJz<tUaSE*%8F7)xcn)(8Rjf(J_
zbGNbN8d?1ulNd3z1X<7%o1**KlRCQ;!3SoTZ(IOZ$4W43I$TQoe0i`B0bcGRxALnA
z3({QkCy0t3-od646=$I0z}Vb6xAyPCMD6&Uf+^|xTbH6`&xH;*`HEIgJ*JP>bf-&?
z#9rI_1yxsNSG$B+zKJ($eQq@}$X=lzhl_|Yi!)o$t+5Y&h9mc1;(EhT?a|)Z)dMXd
z(U~iL&dKP7C9mEy&&{QQIeXWsqp^N0--7Sk31~I#ud>ZSlHU_20xC1NW4_KcH3fbC
zc(u`gGJV09tNT(>(Rxx@;6`9Pb3L?ICwP92!)IIlr=<OD9P66RmDwMWUhaFV%Mn}&
zlW2|u(jK)<W9ko6M)MOBu0j*8KDVBJ_%Gf13pV_}S2v%2IPP1Us3latckdHVT@v*K
z^=cB$;<_{Be5|~vXkg#oK<mR-wy|Yy=eu`rG||{1*d0gjL7tO7ni<dJGaf23t!SIi
z7K7UAGk>hfOMJuU@4zQyv^^%SP@XlPshgoH!u+_P(0niEGOM`yt9p!gpFE#S1}zS@
z81<k9&wmU{)DPm`D1A+mF`t)wW+2Cgu9`YA?s0B(I}vRcR1p{>8<+QN;4djFr#u<B
z8??vP|8-<JT3k!>wdXRK3_Hn)k2;t^u0duC!w1Y{F89Qow%SWc>uyf$QH9pY31=y(
z^NQ=bvq#XNzpG^E3!hbTC>XW5)o>%T?I`4Q8h<3i1OM}|rb$(Rw?Rh35(7p254X{k
zOh3(}Kk?VZe332dl&tB_KGK?AMc(j@n5Y~2rZcXq>5SKEjCe}Rf5qUo$9ceyOpQJh
z7WPJ`kJfVCgmN}X9Sa5wZnGKJKZtj1Un<KFzCu-W{A~8UtHW3JZkX)bA!GKzPYw5l
z8rNlUsUZ7@;}vX+)YK^LEX)-0(|1#}KmOhs*XDhT;Q%~DJoK(d%=U#eLyE8>_vIPP
zR600L(EN4MFTA&l)g4%$1gfq^YsxOs5Fydgq};B~)mb&SQ#oIx>*vrsU|G6@I~w*X
zhmmUYoas42#?9<|v4_s3RdiI$?(%u~Eu8S?ei#LQACGjfH+;OZ=C%<-v<Lqlo0s0C
zU&LoB2lc|mzzx1lJ5<FMLY%w>%wmsEh=T9LK5K4|ZI-*iiD|p(C>k@|RXnq891vNr
zKB4)erP_#<lUu>vsZ)AXf|l}a{)8XDAhzxkmcBog3#;Gh{2I?tes7fB+I7#jq)6T?
zFdQviWwn0kBi2r|pIjPq>TM*eBXZTB*CBU^#;0JRIj-zi?HTKIXv5c$YE{}IZ&=#C
zm)mXgil@lNld$ctzetM8t<9ABLe<5EWF+<RJc#jgUeIKHD`~r~v3_Ap`H_cChK${p
zGxQPaZ5Af#e+II<{#j)R@0sh~!65O^6{(TEg}2`hy+hNHlPuZQPl8(+sO^o%JY_;d
zN7zk>IEgBn+=<_w`}HXjd<-BZ&oK+{iVjlE8{}tqBg|G~GMt|;UET8HP%&zwGBGZ-
z-M}8i7CV;L2#>H^i0@@Wxc9psTS`@E8W+CN5K$e{*-IvqZ*sQkl&NkWIw6&)A#xo4
z8uH7*@~rrNUIfkZ{2U^dS<nZco)C=pwUJ~^FDMXy{#^h`e@xhem0nAJ#QMo5r|E-c
zmaQkkkrS~qH=9P@)fa-EdLsIiVv|3o3Slt87h3JPPprMZ{_AW-{FuZ1jYk_=MBk1@
zT%=VY57(4$RiwfGL7e6cmnAi1&s$45X<@}^A-Q3mkl<(eQrR0Gu6_pb;7ho^`NEHO
z-;D2B*{Tj7of6XCW35g4gey+eMI;clRpRrNpzBIN#tN&fo?hg(GP`S+R>!lCbmjD#
z7FT<B?P9FTXzZNlt0{Yh%T>${(8fM4rT%?tzYFp6xjym6ZE)SFBfnMupHy@A666It
zr?`Q!h`t<7?ZCTKON&67v*W$~%pAM2Dp*@S+UlftVwvFjura>=%5;`XvRG2`^l|)(
zwzNcMbvZ$({>H^=F~>GH1SEl<e1U20clr*)ZS|+_Rg&Y+);#skht;3?)qK9y)aRK%
zUb!{xucv+=d;}ChGkoRkyBP>QA;@N_w2)tK-fVB)4`=n7zj<e{lv4teY+(PA&381i
zzwTjR({ac}PG2d@H0_yu%2Q5`=<yKl^}^>n_GZngcpGk+>iF><u7l1sRlht|??8$K
zFIBrUSjyVcUh*jwUB63k_vH~0V}3@xlrq7`FTt@atR>%D%DkfOYfkoOzBN*P(9S>C
zH^9GAA=P$U<i?&Mxz7+8?~lsnxodj$GVjE&gVGiHCh?jAa37A=P#BP}YM4|$zKaQ%
zlm89VDp!U6eC&8;svX}~sW9gJ-19jjqk`R0Yqz=bEBBGZtBaKfu*9ApL!@9lWTaXa
zxRdMWo;X3lCC+A_q*N=SGmR@ZA(WXWD&l5GjJK8VQ{ydne8;+g{M6B_fE6{r^u%mu
zuGYU7zP|pD<iVixEkgSS^}u|RcqcA<T3(%KhZUIE=8~8amV^N&daG0ZMsa#fV!i8U
z+`y~9JWlVhbkP|uF3!HS`WOqleRZ_TKqh>X-3hKfW~BB;Mhd6QN26_X`oDqFQo4R~
z#XOY<4@nfJ`XeGHYIiRq<=+~=)6R0k2?myRbKFN^*ZxaNd}U`w7nJ_i@U3rBG}Xmu
zh0&q%+_t@4Q8r&4x_UHv<3?iaU9AOeE_>JESv7*8MPm@3^G5B_P9CtIA@z1bMNX!D
zhQE(m{+FcP-w9*0q<Gmhpf^!uDOPU*DKvc7bLSr22v2GWrJ#Vn@1A}=dJkTnczz{P
z5DopXzdy=!61ICg*wyt=n?&&74#Uwp6^nrXy+R@iWro23YJ%7(dn&>AA1#IrgiaMF
zPwi?Z9U7|6iPy!#j9SPoY6yts6(w((ix`B+6w>yaOLlh}qNdJFPxlQ2ek96T>+5q&
zu3TiqrZ7Wp&78raazNG^?0l<2Pe>Zikry_wpRSa*Z87)->SsBSE|MR$)^0IStJu0q
zD+{*Zzx1Il9Npiu^Du3VfKYqZe)3Cy_I&-(G;+ykwTdJtTG>Fw{3@vUtnd2ndYRuR
zK4hGUo$x8!d;Q|k(vQpxm$FeNG!eiK%jW|y!}&$<HnTyu$GCg~?;Vh24<2gyrYn3+
zkn;RR#zc1c`NOuCXQTDiI)9Oy$Gdwt@I}VwOxa5?wwpDc*CDaS`2FP_hN~6J;h#TJ
z<f=pwEkrn;+*A*T&vcjQjW|8~iY+hmWE%F-br^oCTYeO|6mVzuU@JD_ft_#t{(~PI
z`*mmwnoHsS=-GwYkk4Yu2vxP-;@%`~@2819;_C0FQj>If%BeoSPS(EN{6&rwR>uE(
z=9&}?e+t271g`j%2Yg!wI6*<C0_$F4UJqXOT*3`D2JS@HTH2^ryuW*hJ36p9QxkGB
zPY?3S6FLfiLAJ?-!GlF(od{^0=-QIrGYT%^M^!{2UmXIHBYoN?Tog1Mh=sGheIJMj
zNJ@S?{m5g5%Zsml(K)e>NOFxhI~~^CSj3wY=FBWdF~3X5o;|ehX2#|Ze8m~P^6%cQ
zIPR^T2DC2@TCn(%^_~A{b?7r#n+z(b=O1n2Zs>*k6<Jf0;F2bAh~_ZP3@#oWbSr8`
zb*wNtV7ac9l%O<{b*cBa{PV;_zPD_3N;YJxhZQ#jgYsY^krU?1&7<b<5p#6^s}Lcn
z|9x>J{6>ZSzM#&vW|6+aY{V!07cc7fR#ijea0RG~QV+IA2ndLGw%6S<jh4oP?HMxn
z%7<xb>A&UKaRoFw5RADqZjsDw@_p53lYeWz(T)1>y1f@YU&JyebdmV!>l>NpNe?*L
zZiWxFZq?p^g~ISfCT}%oX)uEmyK=q@BW#4<XoGM40HDo|ZrY^iT2a1sak{PR33A?~
zUEj*OcmCy7^UU^c+lsO|R6j+33{e?+Oq`an7p~&D8*Pv{%6|W9MLa%D7(9*s&q)>d
zi%7<=vUwlpSuSjDkFc*FyS80br60YM5UL(n0*cvg*Ot$qc0}|xVzKZw6_viRa=4%}
z*IPq#)tNWa9RXlJ?$cL1O$aBW;HS<WF9E@RJlFhGg`UGJ+q${C#_Zz~3u|SwPdD$6
z4gOEkvIaAw2ZUi>|Gj?Qkxq*P5m)1H&7vS2JA0$0-JF)ve*a-To<f>ZUT1d?tVN1p
zv%xOBKdT1cOtEGC#Qr}3QwW2C@y`;Pnn?UA%fEi?lh4kIH3=P}!xTz8=6@+O;QB}L
z;x)q;M0m;NB<FGq_j6SoJ@;?l;mGzZb9+Rmp@Gsfq(W$%Jc4Y9dA+HWeMQCN?u0?9
z0J(If_^57HFA;f#rqi;o2;R+UpP&~AFZn32(GgX6@6y(3!Zd7(0RTVj=DJ}50@03$
z6q*v76<XH0R%<?8MaS1ElRskIHmdJc<FJjn4<0>|GPY)tuCi>3&5G^$)%nS3t0_<~
zBY;wu|NA13_v^=}SUP4qRvrHrQ(qZYRoAt<5fu>uC8b3~8brDk1f;u5x;r+Fh;&JJ
zcXxL;NOyO4*O~D6e%JY>mjZjOxyBshuDMv(Z8bL<MYC~X`ICI}YR*=Rp2q*JO<)ej
z<M2?h=!|{<k>3!NT$mw0*PPPlt(82E0N5KW9A>7*S8Z485c`zmGQUMxBuU-g9uySg
zn-xaq&&rhrjSq*_eeJxU^ovG?8Y;Zteo$NltFyroDCfz~Qng2AF{H_Cc}nh#h_e;9
z?ffP#w%OauSDw{&P<w{Sww$0Pi6G$vxfYQGAx@F?d0Uk^ld$p3-!L=)Y6O%w@LyU6
zyB728;M<!Z^0JV%#!hy75W&<Q+wShSAS@Uewj}W;d|u5V&Ygq)Orc0vZ%r^=7^@|a
z74@h6T$cQf&AYS{9P&nd%G=aD5?Y|@STE~h@`!m1eE#82gNeCeT}b9WQ>mcc)-~et
z(xUbbhjq0qjttjb87k?t*m?V+*;S#{L34ZFv#7(Qr>ZY9N&?Rt@)JM@0iywY6qwWL
z=j`NR(N}pMU?JF?ZwxVoRHKuj`PXgr+xslghNMKv#jV|ejX)Yp^?eO${7Y{DG__je
zujTv%)7q;9kR9Rr6P<I4LFg3|w12a$BlMwS%fgD{T>t4RcYu>9=I!0GT0D;q_g#uf
zF3`YYnP9CO=yojE5oolhXquIl=er9&ZaRbb|J|6pu*GKsSfX5992v4e{~R%e`p}Ur
zQX$ib<t3Oxb3@ftWyX9?t#(g~weC?R?Ai=!wXAZx+=2!hIaHMF*lp+6amwW=<L##|
z1#tks&3`prWZR%BMj`6-Lr!@|0aKPltHWr*Y;WB<(R?@WElr-Q;?ybP`<|Mite@W2
z^Z`GoBhQgkB>qV}c(F7z)Y(7KG1>0XZjxVWF!K6T1!)EKq4X-}|F@~0llgG-iBe-D
zz-4Nr2xyuhe6bf5wGK_M@S9PF(CaYD^jj~-FAWcwR)w40&Sz`lYnBw-eYUX30oVG+
z?iLFx;&@7-hCb&Ydn2Ym#bnlLmr_{dz;qdamxm_HJ@WEjI~sWxCg|=~#y~c?Ny=&*
zR!&daD*%&uS`f{qoY3&8VvEAtbVdr6q(dp+j+sq`z;c`2Cz*(f{C8-ME9itN6jbda
z?dVDs7?t$f?yf;y#_Q&#<=IyLVA)vA)+h(Nx>9x57uJ4=@yN+tl^grVY0oMS+Pl@s
zOghQ-&~;Ne?BeE<pHFR~xuY(=iLJEt{#|MXXbIp4dZT|s@aw4t-*y+gf1C%SUpZwS
z)B^62Phyz5!E|(8!j(Ub>%#5VySzA*!uRc$OUf^7M|7niTCK5pitl-=?8NYDO;tpu
zlJun(uAC@dW=iV`y{sy{$Ki(3yyv33E7wQ#@6(^rW_NDUQP<gh`Q_@=F1or#u8`gx
zC`7b?|Fc-GZW(vH%^ZET^EbW<HNLe#VwSbCucz(V#k|wDkc1VeTb?{giF-zoi|H;3
zO0C26Jj%J!s8qk+>3|}7hSez7Tf6JSqOv<UgvSYg9rwSAocL*D3>fdbtC4ha2HO--
zWAdKyBo*th^axB3q%|O-ymoMgn+k4LlI@X|=9{szkj&Re9$eU7wd*-;KUJn>q9vpu
z3NPJw4M5SE6P-`&h#}ZmBbf*$R9Bkn^N=!EkDFfGU>X8KY9r)Dxcb$jLA4FmJBQk-
zK;ko?siX(IwHdFL1ZA_d{A`C!#zs6U%u)*M`nIVmGHfZsmzI*>eUS*)dKm#t(YQW*
zr;5W79QcvLKgsouqBKWpzB~_rUeJVmMCE)G^X5D5KuMr!kqqW(Tjuum?y=r|Yvpb4
zz4fXb8Hw>MuVUEut%`#)faN)&Tl!vW>-8JK;nHAP$9bwM2Axh-TdN}9ew5^~9sR@Z
z63h6Tqi%K$d4xv$<MjJ%ctj!HE_qyRo9_WL($Lpw|9Mt6y^y#5BQPfp4}Np=H^D^!
z#;xig#XwX8=aWQF(@PQ4`!Q!DME3r}c{M^%vqMXcTV=T4J^48)G88!7nrEc@fz@5u
z(6IKmIHMBd1P%}_W<{mt)g{M{=;RQogM>rIAa8+*%B#QXJKQr$NY%S{T~kBv-y=B>
zWc8IOlsXeF(iY_=+~V_D|Ja5h!NIvvx-L2&(Wi#H5gx`5IqkNzahRmTz7`*-37ECt
zVGeY<fafJ9wk^f~dQFR9QZrDnpIAZKDY$YnRs}9UoPW)^wRxT7xI5GHG+W|fqb@aw
zs!Tt7+_LN&fV&;o?yI7$rJ5}bFp=XG42QoiJhDTJ3&+yrm(wHG7P&UqI)=o~efHd0
z=?M!61R|Q@*M=0?oJ+fQ!_@$#vn#7tBL>0L&%_mD#bVwN{B^kk@o{)#jm0k={0)iw
zrN%%3L$S2Mfh2Ibtbf#1RVO4`?<M+73m?dcOlPI@davHChXZ$@@oyxUinKd*?{Z^f
zj^zHrt14VMJ=?pRyZF%@0(oj`8&P`_DcHduuPoF%xjO>2&>CvIgW@`zZ85}OTuAu{
z3T}L|Aq1GEcA3y5eT>@5&VuZY#-vePvm0lgMzg1=QE;5}|E~peF-Q7MeW2cXGoQff
zLLDW`GmM^XloI2rT(Wz=fz#Vu?K%k}GAQl$!5*!{*^Q=uP)r&c@cC5vjNCP)@}FBA
zcy>S{Ssh9H%NbV#%1TOWomHBk3$GNB9U5jlw(r;X*t}E+?vcpR4a%Z9ji+bKZ)Viw
zEWz;!<><52^JAAET1FB{cHnkDEsUx}$}=C_p5g7VVksw8MvqN)#OT5yr+y-FZvaxy
zJGQLmB5+a6wT@57!+eNoM4G;(kU?{0Gc<295~ReXzr}d24;9~UqdW5fIV`n<t^sn;
z+f38d*_UVL|Jdr1{r+`{J=d!Mq+Ex46+eH!7pnJ2p3c}}>Y_(@n#Cd1Mc2MC$}i<z
zLGuRur+MaE!>x&7a}#d@V642d8q7-TWk5VVV`I#>o<gDCQR!7fz)w#76k$RYecTqh
zzQ^1VuG!%&Ce$%D8)1{enQ!F}p057h{bxV{kmc@G_bMkwq`^rTAJ-swj)YspGWAQu
zzUd;fgVY|W`6mZ^M{45a6O9G>Y9daqpiv71vh)Gz$P4-UI$*ghjW>4s603%DpI_e`
z=9nbl0pvsd6iV(P>jk_^6NRXGp1rwN{_(p6Ndgt#fMypT44j4qU_RXbMM21aa&Rlh
zP@Oiy8O_a8T_q&|!8K;5{h7<Rcr^xg-;tbRw_QeQk*K`?Y>Nlyg4@xq&?oR>UJPO2
zIqtRnpvrk*-8c|ssqbq2KAz5yc8V_T7kC!mhU_U%RQ&dvC_!6G4^9aC(mlEyJI5I4
zUlNU4se{dGOiNU4qki8kw#V8b#%2TaSLXsQbL0x|#~PE>$LAjHH`Tf{ZE0y0cqR7p
zFrbVN_86X7DnaYk4vpBbYp(LZydWoxAa5|{75V!XyZE*R*|t<p>}W`HqAZhujdPp#
z!gasXi-TXnL~(M_N-n^{pa+++_hfCQnn@xDtv>D!#gyZ!Q)OMg-jWGoOb{=(7*=zy
z7W2kZJZt2p)&aSibqM!yjVjef?%*GV+1K}kcy%%6TQzOH_q$N%iF&sK*m)G^h3V;l
zvQG@F{`~lWpdgQDfghx$t;4rbyl)G=q;ae>r|Xgn2zFf6t6Dauq?A8tF7ewRU(c^X
z@bQ_1UcCa^XJrL;F4X^HctGXV1*(GyoaaDXhFHv4h2Nwz`wK`xK;R|^5Cm26+Jo^e
zqm5m^mKsidyH_6V?45{|F&+3Jzb@K)wJMj-i~E6g=||NvT&ZaX|A25~fkaEqT7PFB
z@q(lLq~XqP0n~rZ2A`Q>PjD!WDAup?JOBRC`Tm2@`1|$!N5p@cKyO#OM?pq?BU@mg
z_Vpp^4kC*P&|spfs+i>;8g_3Z(k|9S65iO_TA9k8O4g_WXB2x7%W}tI%Nbtol><Rg
zSJw(UMzU;H(I<u^cU{<c9Cj~Sj_M>?xg8;`aTHWCO6vB}8aD8ZWwHmW%%Kl8MoE5E
zjyPA?sW;+*jmthQXB_`=w&#eKUye-7S}IGLee9Z$?|h>@$#Q)X<Lx)VR>2f{i-WbO
zV=AUgVCn(g{06EKz2AZMBSxMme;uF7sj9k@o8|J}2NhOsJ?YYao1FaXKgL_uI@%x5
z<@WrO-QS;6)M<)9{}IMv`K3Q<^;%iO{y9X4F<+)%+hF;uAogBbT9TX3gc__|rnJm-
zQ}50dJA6e{2KA@U^r;UI$=G6>h?WR7GhSEA##8shV>9h0`L~+)FL)&Y1SMgiL!h9e
z!!^#8*sjvus!OCLzs}62qRO|RR?&Y0Y3JjsP^P2*^d(O+X;{_?-&o-CnBB`S=MYQZ
zl%2@mxgbtd;(J%`dp0ftT_qe!5fO>e#q)|s=42i8ZwJE)OQ4-SP0%>4huvYf3m(bc
zi5aK&@HjDA?djZ8!J4l?PH&G^%sr9KKeK@V^G23s+#i?8;x5Eh(Z3{*3LZ-!6k}a8
zlXlo8zO>kG#)NLbt+Q_qKXXl2OIK+z@+SGc^IVxHKx^hJ${P(xgO;V`I^rE?e%r%1
zVD=B~Jd{x@tkAy&M$QCsS-NvwXv=y(Y@#I251^-@3XuRUuM0RU^Mn}}Y%6F4d3SDN
z2~HmT4yU=R+4mg)?19@0PWeUFE7I25{B>X1T)hTnV`<rGdOOakcJi%RD8MpV`%J?3
zcHGeU-gC8uT?qoPuEEp<NvWimM4AY<C}b~z0u9+1n#tfMY}lP0nC8lRM@WSS0m9pk
za^0uTMXB%DqDDd~a%V~+!W_<qdJ52mRGr`KU+rD}hl(UT|51Y_m{O`wUgimoe5zkz
z#;E%aaE|a#Z#|M|OR8A-GN=eQhu$>b-T7<`$6%cwW&k-eNbDn*+t+8v+&KaUVLnN5
zY=^@ED$1<Ixbv>2TVW(r<dLPBud$igA{>5B9OTD6+>e(V*T(Du6<PPH;$Liths|E9
zT)?$|``<rb@urVETzQmPs!{gmXux+LE)f=o&<qdtpo;0PvjtF%8rUsgYg10vqW^qu
z%s*H=Kpz`z#glh8U%<p`w+ZNa^JD|(5^zwgqRy}l=bS=Z@v?3H#6K2|iqv(&%p-66
zR>o?3hO4>e{2kn*Gmq}k!OSuJ*3n2qbPWWsIORHr;q^3;6BlhbY&x8J`3CPO3g@h}
zZc_%JjgOF{uist${<V`T*g@eFkjOO;Vz0DBJ-5+!L>x6z15K`xCDY$u%xB2uX)}C0
z)+>k~fsVM89p0`19S4TF&KSS#GK}k;2oSMar?4QkoH`>a9SxVuVh=Z^Y037ywjAyr
zb9{MdC&`5Y3gg$et2eZ(&)1LG2K}qFva(KR(eD`G;UQDa_Xzl_h5uK3e^z>h@uSXZ
z(-f&x+MRf&J=bT4E~?|~so*}Z>_X_%WgXAf?HQlp*1(O<sT@DYXQwOV;=RP$QDVh5
z7En)Ha$~9$(rASN1clD-pwXfuN)Mdu>a~R+-5oGj=}Uw~ldSd1_TQYo0V*7Uux++v
z(Rt08NX&u!z9)c~TKCYv<l%5>c6o^xSGO{fL}Gud&57a<_N_Y>UGUj^$8jot80qx<
zfd(v%)$Xfgh~vB+bUq<@P;vwXg|Ilzi2w;FS50|*Qh5JsBygSlIrr~4*kXWJD|@wl
z#AwpDNxLZrnY%H+Y(UD*7V%AI=ihGk|6Y&V*zfKHuA|n_PgIF}!IZ|_q~i(TqzYWu
z9P}Pg6RW`E{XQV7fk~}%O)~r4pIss8iW*O$zdkMHtHj0Tshr-<nqpEVT-T7k)ZtmC
z7|?NTTxkC!&y?1*0~3<@==cNm&i!*X3W~9Q`||)LH#7{)<FVpZ2>Z#5L9R-)Om;hl
zQ^s)SF|*`nI)#JlgfO4S8I#%bk8d3Q{65gNc)iO8+@6ftjlk3H#X);!lfG*U20~Zx
zMjmNP_bXaDMxCCS$x`E?$Q26y$xl6=C2S@Q+C2K^oa4Wob(gQHOc{|X`V!m%`EP6b
zQzWx4uF}XrGm5$KV=i|Zi2||=2?gy^c#)4Nc$QynoaZ8lxWu^A^b~G}LRqc9&(?t%
z-F}A5l?X-@O7nS<x(;3c{#|VI%!pEZINGYk<%S?*|H-0c<I7lM<4H~zBvooCIKpra
z>**yR+B6IObgCHMfL)_Zj=sxJ^MI$Bl*)tR64WknALps{pMO5r;m--X_mROvlF?T7
zC{j6WOg($88RybVEq6Kqeb+c^->Cv~Yq@b9!S-Q4UmKD%ZO;?hw$w~iU8@6j2}`7T
zM7J#9&u(Xl;~Af)cr?6SFjr=NA%@Mv;~am84IsGsS+@*O++4kQxPEL-yfBPN*Aq4-
z?G*S(wAS0VrVRvvdY4sV`--!iAcqDudNYfq&IklM?alLt9*F8v-_%uRs<A%!+_L5>
zbJRn7fOP8vaNi(TWri(*1tJ}@i+c_jrksl9#$Ld$wIA2hQ9=$JL$+klLv;8hMqAC#
zRA>30T!22b@c5){zu<C0@qdC12Q(n1W9Q6~=S;_==sNA~lyQt<5d`@C!6Hf%*tW43
zk;ADSvGpwNKqtG4tESZ#ce$!CRN6~2fC5t?lF@=$?BaYWrcY_|wOQ!jPKuRO<tmo;
zFL`T8vi0nH^S<!6S3j&Gb_wn%z)-vdrE9A_(gI`w_uUS+?>;X-QjuS*y#)i}xz%o~
zCC_u&Y%=GI89M+nKx323;)BkARYp{7Nh(?YuQI*hemSE2yMxxW@LdOk)saj=yI1~w
zv}HDgWJCeMLqn|NwI_#}15Cr|#Vf0r-U0%^*ojLHMp53~Jv6hZ-BztZU;Sa+wsW`~
z$>EF(L|LFNheQW?d*;xc;^6_MA}`zy|69Fk{LR4@y_F9)z*g}3o1q3}l~dD;iyQKr
z=h9y#Xrrp^<XIV88#_}4%M?`aFTKQs3MBFW>vPE_PKuAv^tF8t1?6;f<KKze?kYb&
zDo7^{d2@9DgRQA}mNXFhmljv#mn~Tol{5|2znV+DN!Llw%)IGdxPM<J$O}tCg3Zh#
zLnbCtTVaWSkR2=Z00Wb$Gh-6Q{u>QxEYnjE@jqc+Wxmib3v?}Fpo;v^){(_alc_m~
ze2bAc-C0zKkh;Dio;|7#KB465H1e|}DeQDb&MpcsLvcxQ>zBOrcN9h4RYwKuW<_Or
z7S#XU!I%GaYA>C=JxHoDLNo9K>W;OD`NWm~xEDlZs5(rHr8XigvfpmLr|#&Ox1ge8
zi|IK$G~#l80v;!ZaC~P8Pj)#$Q6>(ULLysLZ}3rtM7J0y@oG={!z^~0AtE!|GIDm0
zTd_>=f5hHvi?&h<4)R_s{~7Nfo0!%C>%+*Er?QrbsEEi*?8u?3#3cx9?+s7qNJu*m
zEqfJ%)#^2wvB`8xqljpSNs-a6wZ2Zbx^22~*NuY~5k6+KnT-8v?;`NLq8`n?Gh}>z
zK5lF5TIo?mQ2W2l0O0yl)4;1#zW5JoGivs$q|}}7t<K*AyDsqhJbUaaTcbH6p~P(0
z9*s9i`+nSRB_ziR=N>kGw|!w@5~k+Lx$+t*dA2kBxhQz4=zm!NWGt5ZC$k?ZTRK~V
z!8+NBIfhHd<(Hd&l^zUnx`knGFeD^)XR%-PVl4{-9PGm0ZfSc}!MN<defsgAB+32`
ztln_~K5}D`npz^D(-!b(?x1s~kkZ6uTwM*%w4^-G&aG{WDL<H=yUiT&03nS4Bu&g@
z+7dEot)lUFAicc3nlta+-9iNldm~h5uEmE}1PF@00Mk9tbj=*q+!OLsppQLqae)7c
zoDe0xUQX~}Rsa4Q@eh^bzXMcoZBUh?tH(=wK+dgeKrmB0R#Z-KAzELV?U3X&4);>)
zs1;;AG#)I~<BnB`6;{R50h<&GT*Kupp81FhjZE6r#?6u)b}VHnZ661%#p4k3vWjKF
z1@d2G^{&aovZQ+nT*zXlr@G>vXAH_&53{wkj_4mqhw~ClTk053AAYoC2#w;#<AJAG
zyKK+_^zt3mkpzV6@E^wN{pl<vX2oPK?mKlSOG~DtWK?aV(+N5p2{~4^Do$KC9-(=2
zaYw7cThwARn)x<1M$vQ#LUeRM0@XN=fbH@_rh?WcZ&YHJdgL9m{lT~N09^jk=1Qp8
z{xR2qjMg$m{pGQAq-p3hHEmpu2x%P4BInQc)*H_QEewv0uuJj0o!o396<H*d|A^QO
z5o-(>liW3{ilfzXY?YYzQ$~+i=Ikm}>Tee<LrU>EaZZf78TQ%YO4u-7hXiv=@_p<t
zw@$I6IvLK_M-6bQ^9r(DGbh_IW$ZI?S30f-W!65FMT;#(@v=As-~s{ZPCg)iJUIZb
z5Okz7CI>X7%n|N!PmTT`(0?bTiO4lz>;Dkc*gHyKv1W-aadC2-E(#bMt390{s{Lg3
z=_7U@M}DazEg6~C>Z-cN=5Nbo6Z>^qdd7}f=<9LOn<nrg*hnsOM+R9k12?D>lslYS
zXLHP$@<J02fO&6(S*Piu-5AEj6y5mt2LE}Ee94DpTAKS8W^08NOp_ceP{9Eci|3mW
zCEG7ifFyzd3Iy_~8PqH>LXD!j`~bw563ct)+71!kKqeeYzoz9YIstFje7BY0Hy@u2
zzkV{(Cx)?ut4Il92U}*CBz0VN%Ye#K`r_kb^YWskpcouC*75==ilP4;k6-`F_YO9?
zs%E%%wywM<GnC>`8?H4<o<)Ww0vJ^gfha7Qe&Q_*A|4+5%d+GMwVWwx-iK>Xmi_)O
z%`+kpFeVELMO%|*krPm7goNwpgayc;x%V8xgUJyT9_3zBf{Li30{qV$3zafrI;KD&
zbMp`Bx{0)(TLD1>#N{nkcfo&;<s9eo>3`Dh)`xjMRR~aRvMus~L<ylc%f2vJhzynH
zRI7;>f$8uP5vg0X&+$-YpR2*7K_URL7=bXuTz$p3fcbL*Es(RpWg+m9&A1-7Hng;S
zZf!JRn8H6)F0iba{kZtt(&7K^|6ANhzPYB~D@0J?#7tB#5w4$K5lf<J44A<EaJ=V)
z;^j-Bt=*S<JZmUO%S?R=fg3mj;UZp50}+t&0S;(voxOjFT*2Y0hlnGa6M0LG)mTBj
z!@^A2_jVfj&MLzo2oM1m#xfB=?r@{DZ6%ctXep!(`&^y-*T_e|U1;i~7}15Kh86Sv
z>$@}m6aK%&rOM_U8g8_8`9{2$zGlGdaL$*DpEzdg6s`BmD)QK)Qk*e20>d`Y<ph5E
ziQA`7pPQh64Icz>I>T`nWNEa|Oa12d+wAY%#%T@u9oNKS8>a&~lI8pv^p$?R7IVBI
zyL5L4R#a+V%dy$<re#l)HE6oZoaMoEO6x^Hw-QyW>Mp+|SO9ZaDC|F5!bg1E2Ig@K
zafhx?Z_>C5wz<~{0dp(Q%28hepJc2bve2f^v(GFmf%cK+Dj29e@+=HG7uSTP1h-vz
zeJYT!mrCcd4{5nMsWJ6TTFo^dHTiN)xirIE0h0IB5AnV(8p>4Q;=jEzd3UOK(g1RS
zKeRNw#rCuOVlh2(#0Z*`ld%{LkyxHRD-*AJXobHsN&lDPW6cp+=DVmcaTitUo%i*W
z`V%L;C?8k;yQPQZ4;F!kzdH+onEh42vdfJji3=|C@2WgDw|Q1SVDz~IX^ttnez-hY
z66fG3*c9G3Am7ZNfvlkL53MCva%Sppq(Dc$ZS{ROLF;j5!0)aaiecsU=fK`O+#1^7
z|62CnvixsuzP6rHBj{^s>kp3&%`u9bcOB_Oh5D0qifWu=!g{|_4Fpj%PYfpyMU^}L
z`I%qbYd2RokUG^1R4Ky<!O@MaU9ssmmSkj9+G9RpVLo|e;5F~W)C}k#hoC=se=69P
zjp(;=+MMt<X8N!vz7AZ)sCd}p6(nc><0*t&^gLzJbn`AePW=22(&Cjm2@;h<@b80L
zporh&WRhyN#*t_(3y;+eM}n9OP^5~sLoVq}7tA|Y?wHN~toRADPy07zNnqA(84d2s
z^cu}fIMV;gf%mpw&7@}KRX<Gc%ak^l3RDnJ0;*vG2>a$eZ*WT{1rRpvc{bYP<&1ts
zQh4+%Ut7o5g`vyI-nCW+a%tZ#rYOVT_*1wxsQ=LNiSoyJ&eslnLaEpUI}UDup-86J
zT)uF?UDvG5W832XOy~HxsgmN_9tr{}3pe@hMhLGoraBkx*R!*MANWCAYiTAGZj5@)
zeqDdY`U%NcT~~SIJ1d`Gih+DZ|CNYB@vr~#hclQX1`LLC4l#x~vIpM>_9m1Rk}0S@
zkgUFpI#fqn*VQmE>gCCEoV=dokG(Pxa0gNX2@y3&KbY%}<p?rVGwdUT;+e?C{K_A2
zNx^rDl4sS}AM|MTd!->MnVmxa01;nE^l3f&FQ{Bu01hYX_Boi&-|VOQ30oi_99Y=Y
zFLTzcl8u63RB!7@_cK_c5b{wS6S;EqZ;pzJVMo$3+={FR4ffg1GyQHrb3qv%(qCAS
zUVV(Ly-v3OC-893j00|G*3P$Y8<(6&`jXr&+l5{h=PitQ_OL9AoQ$3|D`aHUPD~NR
z*dBqC*VaEeE-lcx+EaTs`&m@A_M5u7xk8F?S~CE<@opS6NI2cwfqvOX>_EQW;lyeN
zslC`Ldw*cJ0!Ik21ikrE_?BubFbpVOoya#`iuoj)v9tf8;)jKQfk;Hn;GZ0<0WM2_
zaq=f&9P8&wO*l&7obsXLuhR0$z5S?7MhNWvRh^{4Lv)|$KQ)fd1&Ge|y&X$K<#1FQ
zt8!9i@bCZ+BseI{D|r~~#Em!#E+QldQ;m-?qJ&iJt`SV<fM0<{KF;}yhAGIvF)ZGc
zeP_hyXa^+7TS$bMjfQ|k2?x8TiZ)Zz0SK#pbz$1Z5syI>iT?$z>sm9?j{h2(T2Rr#
zi{Dv`NT2~DIXO7k&T!H~=k|yqfrHIEi^xOP9-mi2zU2O0iSu1pW@@TZK0h%Dsx4tG
zI8i>-Tml2$OW&CouzOma0T68M<+243b7n252szw_<?@dJ-~<lg@mytxt@_4~SBNL3
z&pAo0e41%$?sfXR=RxKM%r7lTl#XWV128&D^=G`Eg(NgJ_^VM#E7pg{$K$d)1OTHH
z$ldAC_*P#y7`?^G#A?A|AqpBX2nrF>65D}D-Ek<|YB07n3Lq8ixOh1kl?EvIs6A|4
zc|QPzssN^1=fcjYp%F7mBSkvb!PQ1k?Oi8nugD5A-NW9rJ@_}p!^#;xc?}Zu@@EVu
zN#gv?JPxrf$MJX5g%hU4BctjkVY~cXw0`$~X#sg9ji1hG^Zdap?GId+v5o(gtPy{I
zWGK7O1_{M`I|{zFlHTbB9S##3FTW}u<)ONZ<>kh}Z&i-|LIq1(r(ROd=o2e<xnpvl
z<_H|d<FP=lBq_Ao-QCEEY?j+v>;-*e_x3L$7S5dfnyB)a&fX%J1qa<j`#?8ujXs}i
z9&6?flsjztqa@zhI)G6G@~e;sGiqWJ9@uY<BU9+SN`fZ#^oQyWl~!6T8Po~M_tX2d
zhBM2(_FLYpCDBm5b<u~FrRb2`o?Vc!Qp}ILfc>c+$0n<57dF;XK^&f*{y}*{=&KD%
zx@^Dws`tyepZTScFe+^7TJlh5Yk7x_PXmr;#LwDzBd4o}Pm<O%_t~BTX7v_3Mhb)m
z!HIz6U*nd<X6PV`q%%CIX>{`=e20h<GzBlX06__~)xDNhM4#lpNjz|Kj>jVH0eDK&
zs~aLU4i3f<;<}`yXe*eBw2a|0^7rO##BHJBp-+Ao8Tq-^65I{@`1t%RKL2%50jt|T
zw4Zk_j`l2rPi6D4cK`qqKxoQS{~Xs~pcG#^iKA6Bo-h8zOqjKf2jN-ulC?sRf6*yF
z-C)rt&Y!zmg(H0#M7RxqVfPtC&4oASBeV-d_+?z5F99!;4mxV!hxkf<B7<b$qO5TN
z$vh)iGP2T`X;FU@5HX7jU-$biDx3#z_WsP_X)JO}b5n@t$AW52OKAx!Wsdr2&4&6f
z0uuOAp2({?Su`^(K0OhkK>{K0A*7U}v<A-1;Q@#gamz=8VYsTCt^sp>FD@+9-RB$_
zHb+feejKJ#*`7NGQ7_kLS=E-8dI85RB*Y?&1A#LYx?NqgWj{LS^uVhjz>!Umb(?Gd
zqN%ud0cI?&0&P&Mucuj2$@0Y#P3F3I85RA@mk1L1`<6yJI<jfXOLq}!;y;6eB;u}!
zyfU4i-+Im*+B-XQUS3{8<P{aivIhPjU{JmzCG|Sp9v`<M<MzUqEECy!=;wzC!Ajzd
zKPz-WIu7vKPiF1)Y>ZHtSAN0=IIJNcc?Y(Q8U4<aed_zwd`s{;x0k@nXl@_%1fjy2
zDsZqta&gk+Rg2DZkRIuNXaQYodC)jrsYcqhd`17Npf@!H1Z^dGN^?b32!GorgLuR^
zxS60#;FOHk(7M`z+D*(Qg-`>@g5z78Qzg2cy2P4%N08yC-@5t0GsoglRE0NhG}-Q>
z;;<)C23-RQFDa+%#NqHl5N_A=MIc8{2zx8KuoVHh_;C|bHaGLkcKxtOJ~A4m-i2&1
zgHeWBUV`~-8x15!h6+?-Schj2>utT)8{|RGD*Zgvx)kfM2`{Dw0-hVz!4y8Wjbm)<
zi4~vArv#8ta*%!;u;2Rrd9mb13v4BSx2ym~Nm@@0d;3<_(b|W9d(q0V)M#|JnQL4k
zcQl-hk&-;RfYFE#!03dCiXS?=ii;1P`4X2k5zy^*^5q32`E;r@fyekD^g$>IrUFGx
zsL>PnQx*q;+_FryyRKSbcHlW>7AN4E06@MtEQpJgXB|3UrHYi7b##$cf~whzDR>H|
z>9iwr1x3ZkxrNEjv{e?jv<@Zznk=z>3tR?tA0gmH=kwl@k@dnXjD%4z$ZBdcPfir>
zlx;$AZ_IH)VvHOuV>k>2SiTWYWEZr6qBw3j!pJKTS$^UGc7Pt24ov(mpB%HSac0tW
z>ncFgD=Te6KnB%ClqG+t9COou(;8h;T)FpwyMYYDrixtOYyeOrA9!2_Df{VR#%cg~
zPYEI`Z_?<W^Fj*)<xM+W9Gm1xo`DcK`ObC2o_p4%Bb6F&5fSb0nlya$`qG<rJ3uq6
zl$*>RourVHVZC!^)Lk`&lw#3h0wFXk0D8J7vYfx%)!VnaEfSMcQ!9wTOwfKetXo)U
z`V+)h;W5vOwM~Fqe&><AVS*zi8QQa%XE}Li&IF*MnJL55Fu}fGTw}p5PB9TvjOn^i
z4EF0+X>FIlW{Cc&_AvxR(DyD<g*-2}AFcAIN<u=6E(e@@NJ!^}OGmp%NQ)QlW=7j(
zsk{wx6nLgTw74TkBA$G5Io%KKB!Bd3+23h}ynBF4(w(tt&b(in()7i^O^;E4r`8T#
zh{3&BG`ER+RlgevhD`Z0{Eoj9xCHlE_T$?iQb)a`)C@|1()KKB+F9Fqx2XlvyD=iV
zZ^y@A=ojJ70JsE7T<kATsW!Rdxq{CYmXoNszd&A8#6ft7EfAz>s?ZMTwvq4_Z%Gm|
z>z|w!?j5{=C?dsvUG5h%w{82D%*|+~HyILpacn)~1gE(=gyx@x=+4rW%Ip4`pxhHA
zghmW_z}c(7`330Y`R-fr^@U6;^Z*%Uk2+K@p3<i1t!#Ct@~T-J9*8Nwyue!jIBRY&
zg9s+06*2EhDlZB_bxdw+G1n3}+tN9uus`&-Y$<2TaLPYnJRDXhQF6ONtR9FZl??m*
zz`=OJ-LkPdp3N%s2&yyxDb?THb2+5PAjf(f$smV$iNu}8_@N^4pnM#bn}>{)I{oW3
z;ah_fuG-F-F}&;N#@;?J7d4b|`(|eEtvNu%FeOQyIWlUf2t|qR#sye@guqKdarEG5
zA{54^|FW-q?(3%8pESC8aDwMI)%bm%9k_^v7AAu7BfLwJ!`Y4pGBP{xNHC2prp$cl
zvKI}N@qk6d&anoj`g&HP6)uIp<?+V?X;*>m#S+te`g1*w*vvnOJ4i@bbcm`yJ!`C!
zbaTRinZ@cncgh5UFamzewmp>?@rVyNq5X8z-?iGrY0^}zlMDoxPnBjb6H{sq!4J78
z<N(O@{eSWo`|E$=_sg!%u4j$usy*bZgOQ4w>ib_syJbM~5!e&g<ENRaS`eezY)0gV
zB;dnCj0?38a$};VLXS)SJJW<@jgH)EuqZzk=_e|hB^iIs3$U5^*_CV<_;Mgad|+j%
z8pE#jNq!v_>eoJUr8X^6X@9*8Sem*CIk@z$0k1S>E>u(sU2w{9X(+XWN-1o{R>Y(>
za2th9NQg6@KIjsa#O(djdv%Gy52+cNMq0mP>NtQcyglMG>q1UO1;`yDNQn4RWuI|Y
zUHS!Y&(WbgfVBUe`H~=H5{+ZZ(y)BiI586s=3Q*H1s2nFz<qVNb6t?~=*ND4JIGLR
ziD{Khin_6>hUBW<j<85~-1=9m<+7T?E7>;x_1N$@sa54Ga!Tr?!pgTo{O8lEOhsmy
z2d=72^Oh`SP%fxuqU<n8DfM&Cm8l52Yp;djwaV4qt7X;<JRBo>o&5Q~#b$3TK00D{
zqz=to(E?%8qKe|y;mIf<2o`<HQ`VHy)cX!XzV21}GWj{A1(co%5#v<jJCs!2?b6Ha
zXR@*t$|ZH2Vv38q6+T}5{fqL-PvEQU(89x)G)w{l-SBo*;%F~iD4d52OqtAUfZf2P
zT7y%t&qd}F0N4;vH6(zlFKFxqSYrSp+w56mEzZE?Tz9dKZ#^ukKd?1LPV3hx4fmb*
z_%cWzf+Bj-^xm6TdVa4!Eu@qLBCAsOm#NGdPe#VTj^5t8S6}-*v}!`~|4tON!jfJr
z5p!NxGdOUI&Yq>(Y7!AWhwe1S9xgv%aBdc3DymL0(sy)w>_DFwuPF!TR4j&*Va%Yl
zr(cSn_ltsDX~MWe0M6$nZkSA$7`K6~Nw!4Utja&p0iqJ(Ino<w?p!p=8WQ54bVH;}
zV}|EOLJ#XXLHpWU>z*wfvHE%63S)Q>sEc;qperr;r|)t$!Se8L;EFEs3}Zh46(GEr
zW`6FSIjT*-L$}-$ALJP#g(#=O8TM$4yw}hKN_5-I*%^f$#ZtS&Ub~@m1UTmM5S+@L
zD2rL?&7BU=2SK=PO}@k5E(NS*a0Nl?3aG4v9P!DuR=}s;!<P>5i`lQbZy(5cSQmW9
z|GWVFcOa8kTYTaGvf3n&cAav1BevHu_Eeu3OVU2k_ye`MHEzBOoVGJ!wfB?b^IupN
zK&7RxJkB~vaj;tVJ;m|nEOE!URU58p2QrqpRa#_3$*)R7-?c~s`d0TnJ|$(d9{ddS
zu!uK8iqfC(MFZ$93!pvL;P6&;xn6(JanjQ^k|iDersWvu>sayfhTJqwA`rH0ZEiw^
zB|<$aK>_aZ7my8de1q03rptY4lE_6A!=N+0Y6v+o9_K4i+-Iu0QB+TqMRnMt0q28&
zZ)&bL3FN)hpMvgEDf!GyyRKCt9N>Sz<McwF=Fw?>_C%JGe`8*4@qWo30%F{Hdy5h|
zaJeuKXI#A|<TXT1n`+K|?bp3oN>+91tF?NHzI>#*2?pV7U`@<!kEa-N$SG6yl8$Y8
zb$9Cwl<D=W`D!ic`d-*C>)gejn#3s%PB*A1InUiP(dw2DR&LwY9k<5LS<odsEPSim
ze^yf~?E8$wpM?8;YkGQbLCy*QG8mB>F>d3`D(zeV`@u0=7LZ3!%bouQN*zaMp!J*O
z_5{uv=PjCA{`@y!AWP;y2VpwPG(%}LymcE&$d4b_HWhf9I!3*J`tcN?8!s@=TzGn5
zOr`Dde_FC}J^KNUuw{0$d@~jv*aJ#x_qAu>oA78+1810r<3#Gc8KC>M7%zc{E_2x_
zEH(aB5DDY{!r)q8Wq1^?r;doexNna5KCu!40tj=r<cBojDsTqnd^HqD!vp86zJg4{
zq07!JRDN`npz-h=Z-*na?%BDP*V3HBfUuC%v}`#&e}8dqsW1h9Rj~!KvYtsR3@P)Q
zAYz{IoxZvTE^(LWClLswTXr0Q18#Gc{pZJcnESVm1Lu~nwwSq!)Fi(_)+-zZheTW)
zG0W3U>dL=t+S!hISnG8R)SQ`3?!(PYi8(G^r@zPY9`{C?b4`lJS5Yd@bR``v(V7<J
zRH?%LX)c2Exh9UO{Jh)fX1r^HIi28`<Ur@?+`dcaocr;CnWa<oVz|D%r%moDTkooy
zulXUq+c(8Kkt`Acf28+&J_Oc&{`%vo=7`_OUi^CZ;@82+X#9Hq)32|eeie`5&eED;
znm51dmg&V+r2S#@36Zr<MJHG=JxOF4Uuu)<n7v&V@etwFT3#AgiZpr8j7LEFM^(fb
z$cu|(6~ULs<aq0+L_GrjRX?TQtGApCbQ>`l_Ue5=fhVj-=oAddJh2HT8hB%~BWG1s
zc4=!n>8n{cm_s6)o*wJvO+w6sT!9p_!NdW{@lZ*KoO$+_p{PlBMzFfR0u_;p<E{$(
zd1I`(Cislwox3}}*jrR@VbN>&?G5J5fzKf|5NnmGzw0H-^t-2I?six+^;Qrb6_q&x
z_62=+m2?{s?BO(R%i>_77qjQqY1b5jJ})LNkOb3y4e%rJ6kucQ=O9gsT^?_Ck8X6X
z8evbwgw@rAMa?|ZGu3Swq;0Xyk)e~MD|t)rfL|dMubjW4zrDFAy6u5ji35$I@QjeH
zb`da#fpMuvXzZm;`TNJWiuz%7-_^eT_Dc?|?Q2y?&Kcehu||bgTa|y0iMYZHy|Bmf
z_y@8)|1uP7EI05^SS(3b?IKev{uCxBW?+p(ZD{aFt`uK=tI6ZUay`KR^LL;p;@6Vc
z;Me|mco#nozPvpC%v@^(vBUBvHzhe1W#)Ir|DBxbby*PD*4$dzTh6K&JuyNH8(L`_
z*rdB-N<J`CYD)Vj!1N9g1^#oCU%F@jdB??jvy!r}+jD;VHU$L|mqB*k(jFa!ruq}B
z9A5-fG*jRZ^f(NK=A!)q!Z9#y1_+mW`NFQr(21WKIePxy=GR~OF>vh`@6?R_j*KIY
zsh&dQcmD3Y8{sK|+RB~)eX3-*)pqB^M?`^IhlnmA=j^UeG9QD!UmV6U4Tko>&m3V5
zHnB<X<&cssjS!%IueZj4GRb{B-IQ?)dil6|>?-e@C?3jmpepvpam{$*>6ebKn7=~>
z%2ZM+s5nz1@(7K&WjOJbq|;7UtgJiN2KXr6zcU-Da4+g9q6C<F@ETGpfBvaMxP(rA
z3H1SI<Z}ncQ9I<`jwgFBZ;oXXZyw4`mNbU67HY0S#mT{z>yLL=UD5gQsay#&UHXq6
zo6{<}tNP+b)Wnd@O!sHq-SD=21ukssIBo#}c!c|#8cC-F=)CEq=vQ?@YUB2$rCWov
zYX^VDUq#kLCha5(4r|JY-{kGU6O%umlm;7tL^hDvY?|H&!J>ONBm|=e{8}CJ<rM>A
zM8ruJC}Xvvq8QoI5!ICgor0a6u*qboYd&?@76pv+-Xs=35zoyNHx>*OMY^xD!JaV`
z4NmmW(j=RL)I2QA&y@vzj(T=O!ce-J2<*+acRqsX#^l-REBdw_m50s!TJ_!k){<qO
zZoSg-ocfFkPNJhrD38_%+q$FJng9LLGK+oD%{S5UaH<JTT@1;rP`1U43C<G2jJvmD
zMwv913Ie2IFMVDX3-}@uTFs9eB}gPJ;L%Bthglhq{t6LlkRtn7CEOmsQu21T3Y7(F
zTH<1^dlP=LvbK4;lG>I?N|p>BNAAjt=74*T{c&TKVhvLFs!d@$*2qx;7kQoC+3P(T
zzVX_7%9ZHqY`oliy*bm579uKR1XmHYPeyg=xdsH$Udrxsd3}4IHri-s;7=IVt<Rvx
z386UOE$BbU*80vyQ0~Hl%7KSRy?gIrtXBE3GLp$p$hv!;rUl>Zpe^NJyeT|1E_YfK
zoLPmfoSd>S=+cPIqf9uZhKS2s%r${GH~yCeQ0^swLrNkJp-suNZ@?_o`mA#mdZtoU
zaTc3WG+SM`={qnu_IIUukHsivkq`9V)OO`WQZ9rv$*<0`C>1ojEiSX)$_FyVG$v@H
zSy?O$HZ_d;`oPCLcHMW<6z1EC5053@@?4jz+Ey;<dG5DgLff<-Jn(b}qDoE<NL{Qi
z@8(oz6dOE56NHJ#X}!(v%YI)M)G>vkrAZ!scMz$HHT@`(d6KdXc6#ppji$X_HJ#z^
za}98d(6;@ry*C*;W<+of7$Uf01T+ILJE$c{C!l-ml3BPUfp5Zmw$-Pu*5sM?zV^Du
z;UdCI=uZ#SG#27~NXyXN9D{MP<S*+eG9JT3G=>dOenCjW)i5@a2i{%8;M(J&)xN(y
z;oO~oP_u5PevZ3vxUHll6%xq*N^@%C&D*VPXh`D6*jlS7aJJT-g-s(V!&%vFX6vxO
zDFv6A;G2ZGd0v(hTe7A@G&9RVTXfk)JPZFpK-AOyNO*SVfXCS|@`)3RwscHc^RW-S
z7|pVA<VJm*QpQO^K@j<B>&zkM@Kd%YW@MiQ)6UARx6CYOv7fXwUuvxy{Az9iW*v8y
z^t+(%e+1GdBBN-lOiein7?gJ_!5-ACw$T9t68F`s=JQ?_ZWG?%$Z!u~^0uK7G-M{Z
zD~F17p`NwQ)NGDKw1j9r104(<q&>Z@aFq->^l%eA@2bKOp&$tw-d{{7a|{|1ZHA&B
z$CLAga2`tQO~MyC-3$Y-eEZ5N`4YP^tNixX!eOqTDVRJ}ak0YoHb@xhADNXDhkAzN
zT#ctgmDVp6;HzY=9FpSVti##|$43&wv{m|BiC>4r`uT>*SpSxz!k(NivD`D_Hsvio
zwt{r<P`^otemEv-?eCwZYb;_)gZ#YouzjZUK*95|SN;;$1T#kjyk0mpBZF)E(@1)d
znsZ)pgoL^}VOv;k9cisR8gk!LN=kBWN$HbUDR1J-uE%-qEDx?f6EYxX3BwxSvdUod
zcU0290r#(%b!)Qpqr=tRvd36TrK5<y2jfI<GP!r$7igHLaX2CJ)n+mA-XdYkLv_#{
zjga63U&Hil@F(OiLw>y`X4oIE80%dD{JC}I3Ti1hWUZ|Q*R%%AP{9w9WXLWWnf~kS
z)GSyK9H(3IK<zxW>KCfQ&3Hz_!rAPq)|MsZOC{<&7e*6*K2J-Zvq*BDL?toCyDXAP
zk%M=dC#S|@pEVyBWd7Q}f7y1QfZN&EM@&Ogkj^8i@bzmG(INb=gQYj{stf{qd6BoZ
z!I{i8n`vhIu+`G<-*l`JlSq)_JF17gn<JsA3r0y?D{#AMet-1x<)^XnoI+J2g_X}(
zg1lXIQ6*dV^Qn|%&Y$Lb%x-)3ISleiv#u~L(vq!S`IF<;Brl;9A6d{24NKW!9F<ty
z)81ZRtDv6;Hh6f{j;TQ<-JiHidP~30I{D=&Hhnvg)z+Qul0;J<f%b<BZap(G(+f;C
zztdBp0V$EHbC;)H0|-yd_RoJ+VyGISJKeinv<FI)^D6r0drqPGzI(yI^TBpBH80QK
z)wQVviN}~YI(mThw<3>saP18H3O06U!g6YzxV1a4J92%ifu0`wIFtGBYzk6R$d{aG
z7qvA^jHB}+)E!p5pZBXdsoL5wX$%IlJ4Tz^t6h;Mar#=O1xJyx;Fa-d86h$Ff$>>!
zQ$?$(ZBKp}>&Wh2tKdv3jXZltLY3aFIm=~JZoQr9xPvB0T(8?-e7m9_pj`W%drBo#
zkoa>J8k}$S$%&b+Rt8e3;ZSAwp$yYBYg$QL&g#T06$jV`7-e_W6x3u%g^3UmDv=K(
zdru2y>nzbiqGVQ!Omwwj)ApmLYyy0Q1Sjm5(AdP_%-)}%2y9|~+4LvJq@b{OusaZp
z5G}RHlgpd#>ltwQ)0<`7@y+(i;`r!K*J@chUzs?a7P{m804ND2CQQ;kOg(*l4*x>w
z<fJ6R;M!jYzfkXC;X!k0mO2_A)D%{#3$F}CgQR{&Jn6#=GTaJ|iEmu(B&Fu(JKSB%
z^V=PB2fa2iAV|V}*?tsC^kE6(a2#oQi%xa-VyOWCNn=TU&Xgv*7^R4CVUDJ#(TuhQ
z9m-z@$p-c>Gq;-SYw42LQV|6QO+NE4yeprDQol^oH1bampnWz;HY50bC4$4@QhPwP
zy3C+!2s)SQ8kaW(7OdbZ?HbGF)GVx>tb5=+*?R~P2glZ=be?SPLFV=P#&>B-x+(iC
z4pBi9e9Zk1^9~Ktd9Q@6r$LKPxT%iDF*k>Frb!whp3wds`9Q`LSU6n)U*i;$HVyqe
zkwL{(%YjwHx;DUU@Q4n6Z;_zXKJ;4g7O>_Q7vnON^ScN?RZ5)cH*Sz*5m||bFYbNz
zqM#<Pu`1mu;2LyqlU!Kv?W0sIbq8C!ie?V4EA;!ao#0+@ox<fCtI2H-nuAezu~`Or
zrMIlIc{f1h{f+ycaV~daw@2i@jDDL=CC|J#*wmb4hdKTE^HSrP!sasyo9m-N8t^J|
z(ofKnMDxI`3C7LaPHxjGEcjOCiMCbPc?;%?CxGj;CJb|>nHrv=T90N0xp4q=`mIg*
z9o5<UbQ|q)Uh087OP3^4>eLJ|L;AMf3WL?jL{5y4?N3{Y&4_lXP13hU`)6%eSzo3F
z*UzakC><rDfTBHSx@)WTkAXZOeh2)%&8;7S_37<yHiKAEJGF0`f{!sL7G1FKDdZ+t
z9iAKGFxkr{Av{i}kr%Z5Y7evA;E!aMR`-#bLj5$Ru@jxW=xNGyD8~fj?(KIA%*MaX
z{%cOUpNs~naA!SHoYs-zxZ?Wrez#gw47Svvo0ON&z;g)*cbb$f)!Nx}Kuy(}OX_jv
z5OFdY+K)HlU~c~ieo&seM+6`&jNnVw2Y@2cd^F~8CK{S0xgL&WgswIuhj!|*oylDd
zFngAaZ}11#K&ts<={HU=crcVR6cRSI)>mn@`kw@-zRGg+WBBqu=%8uA6%sptL?+0v
zRg1BzZD9`GTyi|`K*<i2e!p1f3M)<;*30aSt=*=A9}*rhP@(%I==+CSu0ofxj?8>J
z97bBx<&P)y^cR*tOmFy*)f01ih`LlDa6~+u2S#0C0-8~OYiYr@J&ESrZtAuBuhcNz
zzaKo{du~L<ZD?ee!+DCaxTiB+zs78)gS!*Gwg2YHm=3m;y>E?z!mFs7pUoDsYvDaw
zl+hQgkPK%6F`c14!L~>YoaQL5`DY6;-l!w0JG*uxjs4-?PWBR!5lunpE#hdHvf1}#
zDLvjVXSh{RH92uvGiS=z^F0rnj?{t#S5Gd?xK7#X6&8&Qvy@TK3DpMIM%73qNZ%Iy
zcw<hM73NcvF1#~~8YL}wGurZdCrT}=UXT~QL#~e7#GZYP&!g1Y+k~85MCv(mV|%Y3
z$703B=8e1HDU~Ftt@gR|(9sdrc%7lU7z#U;!$7>|U4lzeK~~eEoX6)<=en0UgxmUh
zg|)hvI#;$qz`$EzIR+xJ9r@uu)`R#DsWW__pb6YG9DDNjayReoh`B^n0RX<I8wrJG
z#vPzLjqPyBn_R;f*G7G|g*>Z*PTq^a0fL=X8V9%;%Yh1}4}#^T#mrk)dc$N3KN_F5
zkom+aUpi>Bn1z{u*infT`Ch6t_<f{Iv6qG~u`O7Xbevt2@=aL!!W?g=4L~0_z%F|V
z$T$VMrkkiK$M=hQ7s^|O>$L=`P*Sw-A0(-F;(7j-Z<3-Dqn}Z9_pHpXx_Kkc`hI&)
zOibiUH=fVs?xn_QKQ(0(b^A;a8{5(O1WgStR!;w;siHF{Y{kx+{E0C(8*AZ79Frlt
ztPC1D=5d#44vhR{4GtBRbOAgijLt==&d(e=T7fRg$?_DBWVA;JXnXaKR2y_|Il&2Z
zbv8~K(gSElKC4DPoOqg<25~$c=OPVV&CscjJ9)lb4JBqGa&=O>Hd`Aw?-cO0)0pl}
zjfqvfeh$~L&-YandmnLU04ReptnHYt@#Ty)Ykle7Aw=Hl9TT)UZ^rLb_#~XiG>yXq
zpTUm@7N5tQJvE8(DEIj2B6^RO@NsqggdN}B`)AJhr*acvuuwVZx;ae<gCn#{fw0bH
z8W{USyKFFegDx1@tRBDj)LH?>oT)EGc3(QgT5of&nqjW;z$?K)#VLkssXjF`<boGJ
z+ABsEW4Vg(lykefVOX}bDl3h2IC_Q0WU~0(ijr9D!YNoqzO@@6La%B~EZ|$mQ0H90
z>~TZu7yfmuZ+De>Bw}yyF~#^Dl%9_l$Zg)CEKF1H4*n5TC^jZxxx5@DNPUndUNqCa
zlP!7ZTfec3k5BX=*w!)0XKw-6m9H!(A-wspUf>yaO<JS;QM0duV5qsxU16KiY<I&V
zA5SG9OU2SJ4?8JMSJ2t!C3KlO2;T)ALgd7OA!hFeUuU0*x9r<EHWMo)RKI+iAXm6}
z@TUKU!QE1cGbvw4=~XXYkiK3>O#BzgZ;ft>b%?vW3V<FfR>{WdO_^t(DCz<I;lt3;
z_DMLe&;VLQ`Eupi{%6O5_~pL(x>4*5R^>{puQsZj_UUJ(8~4Nk@kA^|_LNzJUvaOB
zXcX<!NAlL)@{1UJg5O&VX;h**`vfNs{f_3zA-t$92Y+{L7QJa-%_K#8`3#(fa1ciY
z_<3z)X#eeb-+8RBPx1g48}0aL)ASSyg%ih1K|!zn&iC)AU7Lf46ed#Gpc<E<(C)aX
zyVBhgWXOWb`5Jn-HmPq)5aF##+$6qDhzpIfqdrF1W2r8-ef&38VZob_UQ}8j8ar9`
z#?@xihZ~o0f`;EU_(uqRj6Cf72VpGspAvCwg<8`{<+04X(nc@zivp?pS9J_~yaIMF
zessPQJ~_=noA>2b<B#FujOz|<g1uDVMmmdQd*GNTEqC*L<UxP)1vJ~wclvW?tKUh{
z653Iopg?T;u4KvhYSNt^Rxv>3l~X$WTp1j4So?4;^7jvKl^9a*do8~`Ga=k*CExpv
zN5ZHzK;%SUdQxObAP8i?tnHz*riPfwiVqBHPdaE+(k+j5#H_hgKla7Oy|`K>xdW>D
zJo3e&%O!a|6A^*FW@b*$J25CHKI2m;)3`30{$zmr?_|QTq0zy7!Q~SR8!d+k5%5nw
zPYWK)kmhO70{UYW{;@p=$Oy%)n)>?(nV2v=`3m|$l!WXn%}c;_F0cF#9T^s_O^83D
z#Doef%3`c}@J%R<Gz~T<3)0qgDvnvIx0H&Jl2fk<O4HLq_x`BK|1PV-^2hyuJbiUk
zRNePCih@ch-AH$L3rKf&cXu<O2nbTr4I<Lr-O}CN-QD#be7?W;57%<Z$jsb(&)J{a
z``|IKGUS}c+f!0@uWJA)M}?A2D0tM8un2idP&!ab>eZu<wV?Te8Tn8IhkMu`2KsLy
ze;H1hkbm~F3^SIsP1k*3s+}}6#L}pV#CH6eFF!)ms@ZO&Kw0c^(3Y=VZ^zD^bUbHI
z@@GL<QYfr40SjUOU^seeuo%|;wf`RzlNCLqz|;Vdn9e*?gI6sF>!!N9pPyIVt8}jM
zL0P+J2yvX;*W2^@lh9mAuc7&FbXbJDZGZ1iF_IcJ^%VnS<qZxUHn2TQr{LO`)A71E
zAaEvPC&u85<|;)dL<MIxaf*#YO73-}u8Okz=X-VGCMcQNk)RM#$aR+zxkSe0?Kf{9
zAQFS!IjPI9V&b>=)+T;E0`g!ybCWrhoNltQ^zF&`mok@#%3?W$;29MAepJHvzndaA
z^%wI?LqE4NJ~e%L#b$m-BcrZKZh|^s@?zVCVylW{fMTm9C9EE9N+5-+OjKbi=?D;t
z3JD%Jd&SK4^q``m>zkVhqL7o{y15%zo8+CUAqTyJLqS14%K9ybN*t5-bh79DpbNJK
zJ~_Khm;JKW`wu54B$=jqIoZLXwOA=h>>1J-0o}%VQ!rI5tS|DmZkw8A2Up4d!jW^+
zrwO&Tz6$V10i4)A=q0rxN+K8c$lMsTBk_domjrCgk1=93iU{oNkw*)^a`HRx#QwYV
zY{+kQp3z9oi%k&`%-!9{2oP5EwPL9lFLq4A#(VRn{Gu-jsvaw?C2o(cWaC;KRbm5W
z&a8K%VX>UMTn8^SCU^Jk^!H5DtrE90mFFJb<okQ3xeGvfR<7+Ua<+bt*3eC=ko?Jv
zKzID$D2h~R04)~u&Wj<H3P8>8-{_ef4?KKr{P;@;8W56l46XXNh`*n>JT;l+9$$Mh
zlap7aatK>E{oHNk=lrB_7`mb@05M;=vEQ=cf@6OeNPK{$q@eOJz{&mG_Vbdj-_x4L
zriu+Mj512XzN&29#OtNK`ph(D&J#B)Mf!P`Z{88{f7okQhQ#gl^RovdMvCm=QB%J0
zSKQ|NVMmYSAB{Co$N|iL{fS=oY8U>bw$$TE0Gz+^TDXHW-okQhX(9pfl5Ki>Pw}^s
z12=Urv3%)%XrR%?;)4~JL-a+OkHStz)N?kw<VH8bi5)LN3*`<Sy3~9W1=^9^2zwUD
zsy|T=WzYTbfbz1cr}w_RtEml$<q;}!9_ew}uwZ5pA(xqE406Lr>)ea&1N9S35Cm^{
zSR1n9@OlMB+;_`Sfg6&L!95G(3<wbk^9f98x~KRX?q2fdLX!b+ewY3f;l_8!y0#Zb
zef!QhLG`Pm^4{p!voXbz@IP_g!{+PlxA)BQqt{U^nju?*XS-9uAci6=EA;IJg4WKe
z+4q$67Y;$u*ZWp7ot@>}_vo6k$p+azQ2Y(7LDhFx!V0Z-In{CGeJ+P4TcK|joo$O$
zyd?Bg-CjXK8h9qKh5mc8s(|9~@5v&3)|vW(qWrGuT1Ft4$ho`WElaM<deG{W5r7k8
zey?-3{Mb8deL0Kp3|9bG190%<tEnV>s@%7;%B)bafj-0ZCwG-s-x&aCkCk<gW52sp
zG3WMtIml$tItAd;+CD3GxPI_3bEQu~O8Jx-kEplo^vTkj#mCoYY*wo(UPL}#xAL_;
zxgYT6m`xu<K;@QMOC`uCv@N<7s38X~Lws;OBRkV?%Y5{WssU<>KU?Ps$xRMu+Qifi
zBDAd|hSzIwkF4K1d9xw`-kxdZU)GqmZ&-Q?Bo1!QN+?%MSWCj*UNcgU>;B8Fb#nI=
z9lELD{1HlPo*Cl{K0f`{KrqGt0W1=udygzdj09P=RWD82<oF1r!K#-!6z5ob>hd1d
zEu#2-h|&0__%8`*x!LNqLzhpcBzf$?{J@pzw`cyqWkK_1%auB@#gDsYW_q}{QYZ87
zf6PlnS#whJ`MUctU05H8|Id&43DH%q^Kr3Y`Krket+#SLmB-znVGBH#>)m1>hS5=j
zr9grHbMuG4>xU2%F0T9GLg*xzOA$pf!``80yLCY^|G8RRtd4l$U`i^wtcz`e9bhq$
zler#6^;DXTJHLZu!~QZg=<D<eKfYK>>RSjXk0G@^-xUB)TlQ3QkV0+_L(z9MjLp?%
z8zLy?t`I?0$^8B*iOfiGN$>UZnEZIoir-HKpOSe*`BLX0fdET+czeYMXfG0jPUNq;
zgZXwN<uQu<P#>M1h!iruQCIRqvC9^3U1mOX8#Ada;Yhl9r@U$GpW}ZFks{L9eoDO`
zZyZC3s=wMXH)CiFC5IJ7==`i1d{kw%b_MOdVbczXB=oEsN?OW5MMgFio}q>g9|0~G
zOw0Yr8db78{Xa?ixZ(Z(J-EO_(dBnT!lMOorsRQ8lzgSbB;Zx7q}@iO#q_T?HJ+G{
z=D-jaLL!-&+<a!}!B)bCxGfCpIYF>QpwNup=i&7GbNX`VQ5>kjY9@1|M%%LMn_FWB
zEpaTnT(6a+ZsVufG66O2W_3;|w;@(FH0ZVFzp5<dP*Z!4*N{@~;}fVL|9jL0SPh()
zdtwTPbjxFZK15N+$24#+)A$hZhNTA4E@bZuPfcrN9*AW+G>AU=WpvpA*)gH&Hr!#(
z)n+l|@UW1AvUlsa?8TohU3qhEs8T}DV_o*QUuxHMI^<MUH<|<w6hZj=FF~w=mO|U8
z4Ib9au=GZgwRPc*r5=G1@>_zUYY;H6aF}~Kz9i{~N4P#{BW*(D7>c5H;HRUdPfMSJ
z1c4c%KV@+ODlGFILjG0GH<c;>M{g~5*SiCT_yx#fR-3(<9{0$=<x_Hj&5X?XRk*Ld
z+_O;A=VUXIoO#r6Grn3#+w@=x_iX!EjrbkP7!PI9_E}2fjLdaJOs}JQ&($~gyVOGu
zI`HkokW~!;I^mL8ZD=v&i~eAmAxCxNaM0;6mxzN^W_d^xa(-S5px%A#B^fbe!91V%
zhwPcVN~F^CCJs&YQsYSsv)D7WI$-v866`^lF<)Xhh0st}3n9PdTT}U+Y1tMa^rRAB
zu~@H%I2aqdp4C-qs&CmQQ}7u`D&-J)9WsHrEXV7uuM`sRutpcZ4~ZfaYlR3Fp549R
zw;?3~FfS~KHc_e0ha6UO(PkoK-uGqGWB&cjaTq$fRRSxtwKT$JDo=a7_Fd3GKUH}n
zG6yQwx1oRmpw0}pbtF`#d28u6>i$=*wt{m0GXPnwz*R={;!XPihyHAs5e+?re!G*i
zN;(&84r2x$5x{HWb!eZImQ@ZH5nt<h4ISR!KXPx7o$GuiefJs%i9nbv;#e5x0z)8i
zV=VFEgFGlIen>D#UIs}Q-w~mQ9a`cxVi_o8+d(dOr=@zI(=qmZ9~TI-y;gI7dm_H1
zHJJ{_oz>c3yyrGy?*t~`e&SWvbRHOIKn4OZQ7+tm+sk;syn+5DR8D4<dh;5H1RI^m
zPQ=N3jHU719(q$1#Ts`Rxt3A@Qomi~3~mm?HGJG#>$Reek*K}xlBk1lr2vi4&f_5o
z$UL+%sl+8tdtd&eF=dK|fz7+3>o>Hh>wkzM#N2F$p<)|&)eABGXN`dn_ac6w$N>(A
z&q@PF*1O&=2>kA~@6}|4aWUHE4dC<OK0Rs%^sBdAmLy#-h8X=!f@@61vtwlJ{G=H5
zCt1rEIcnOS*!S#h5EQ)C-ay?DpMb8Y<^jhuPynh{M%*4w#xp47Ax_a`**eYBy}m8>
zxwby?wn^3L8W`J4ovs41g08e%DE?ugPOrL&0DGJGVE3=a6ol9f5=rXO(K0$jkpQg*
zECDSEDz;6N@CF^Q@X~+5feoP4`uD7Xl`PTJct@4GF40RiI|`2Y6};KFxy6CWfEG0V
zGOI0{rW;dgMhw>vj`mY1bF=YD%%=00oi2L{Hf?)N5Yr*(<!wHx4Fwe0)osR*<&^Si
z&r7e`w>nU@O1-5=%_~<Aus2Q<M@A%~@9T~%1+O3ve=lWX(eEF7lxEqj3xKtB5Rn|w
zdQ*a<wzs!ttlfjPh6EU3I{5sezYX8kc!XX`_ocnv2MD*j7tO2L8He9ysbkZUecJ1w
z6J@S$W}4yXq7>fx=uF{J=9$X8TIu+ooO&Fs7LQpD1deAN1~D0FeLRyI^m#-i>3+$~
zeG4$#x#fW{u#Dok-hOc5Mueib45kVtexaeMiJTJAGjM&~h1>x2sC9Hyxv5M9rhzhy
zZwkWNrnb-5X-#JzZDe9}^(>-2fywM6fZW&Y*F>c~l=B0FSz!@`GO@Aer%u@bJ5!Om
z9MykNt$W}aES#7PzYC$+&B!(U?52eaT0ArN>;;cfrM4HMpuY;-m(9iPl~#)n4T_M(
z=o|bBSB0==A>s%sl6O&ckd4!^udM>R*G*khZemHJu#lD$8LuyN&K`}-?FCpX@lUrL
zd^nuRLISP0NVT7}EU3VMBQ?w<xuMw~i#qsj<E|)hdH)E;u^`HS)yAzC&{`~D7XxPP
zF_}It;d;3Ex<yd-VW$2{eLIG3F(%&Ss&p#y&^mk6*&=NjIMK5*wGS06J8K)_1oRh@
z@;v*Gm-!8{$ZxqsVvR@wK&U!co^(){9?8|iy++mguFY|6aKUN$>8LW#?by3fv-T?D
zYZ#)?><z=86xkTVf2d8s4L|6AbhPqEs*5?KKl%4>s_+<XHLlJm3!B9y8G8Ylm}<r7
zH0@0nr4JE5Tl|mTAdimPRg|=3G*N>DOALOt##Qb|oS@IXMX@Rycjy=>QOW%&74@KG
zgMQf(R(8)^)2t~r|BjBoBr&PP(g>WuX91I@9&sO@9yBgjJ@R_#>Yls{XG3liV-u44
z1^}hZ&NlpA9G>5ZM|OKV#6|OOIqDevPaPC2G_Y3DkXV^q2%!o`ONapH03emuUD+mx
ze^6eOQRyulWt1jZl2QMCM8fF4BMV?%XoIhBmk#^rSVmknceB?qen4H`Kv;ur01-3g
zOJ>*$(-Cxx^`CJOVtB;FEL5f4OA&B{1+%VSzI_z{6VCWf6Dt!a9~waS(62|9vT9Xs
zHUS6#HPBn%t{=V-cnh?G=`sALb=r82x78<`4+7A3m8B}?QkLbTI3KD<^H&}cQ*XNC
zE5*I_2`v$?OK+~<zAVxbN>}1L^`M^e0L!O+?%^))-uvGQ63C(Z|KD@8%3cWgw%!pV
z6wwlumZm1?=&J-c;Ss6$(ztl|o(uiMUqmi1wL*hFo(&?v)6S2FaUPC(LviKQ*iG*R
z`{<1|ixQ9o?IR%M*qZp3#-dZY-)Kq55R#IA_X$hUn-ljjX}*1@+rU)TW<SSj-Va=i
zz<RI5>@=VM%K|`=4}GvWlD)5N2Na2l-VZ%0x__ON-yWC|P#R2FU>%e)4rhy{e<ULg
z9{yvba=0)Agvt2feAuOC)}7XU+4N0e5)y`!I|5Gk9VRF)2Tp6Nf|AjxjFLZo+aWY`
zq}e3~)%49bhieCq5Oo%Np4r62jj*|vMMOs-p}|g4JN9NMl!M-(YZ$qgu(rZ4GN<gV
z%dYl07d5MV0XTr#-QFGk6UMp8H9J2*zwBXw-MYg57O}YpNwfOw;19c7p=h-!ZsmV!
z3(9}Q{#Ck4uAz~i?3ze&otMA%N;T&G=I==kOQB+gKXSueedYb?Ina=XT5Oc|$_#2V
zUW;6fIdDprtiw9z{@!GIZLnjlHC%A;iJR#=)eOU*P{nNK-QQDb&9%DrBVOl8L`#5=
zT|)8keRH>&N%34uZ3%<RV2bwPbz>vX@Xx*=`1z2o5@9o(ab?T2B0P1(mM?6JXjQzP
z2VCE_`pGGIp~(%2O>W#6TP9vZq(en#Nw+pj4@Z@oxN-jOKEQJN5GQM&nHXBy6sd&x
z?uLEWF_@nB2isofz_?!mpU!#yR!C82#!;}}KJMe2#@<NB*82uaq^&la62n0n8mW*z
zTuri=(V@BEO7HLLC=Lm-8u{#wUP}XFz<OL$Uy?37C1Kr2@~L7t3%>xw<+g_7G@%++
zXEHW?-c8|UT%>}U17Cb|vENDOmE`k<j&xl_4X*!hXHybObB6j43O-%jbgw}5b9vti
z;4SplL`C)e{&FTpf(-V^f_|4H#QpH1Sakg^Z$jP`jm+u{IaJ5Pv|K1JHU~HV3&}a3
zYf5lw8Ky>7j;V#z7qW4rAEV}9{}d;wsH`-(I(|`bFoi;!01whUV8+d9)G)Boqz=b|
zSAB7Pt9;~l9$o|@%n0322yv*ZGc_3AIm`ugb&T0=;({JW%%uWGj3vx=K*D%p=lw_3
z!_z8UwnqP!UQcVFAlnB&GH?A%iF@*k@ZWVHrAqu-zU2T^uB-VZ6rL3{5f&#;8qocm
zS=@zT&HFZi0|tp2ZcrP^@HqKq9RoeKJ~aG0k}y@o@xsi<)$(#`JVP6KM&__I7`<7F
z3)589YF)?`<Jy(8S)Y*=?r`vd(A}D~;f30$%1f4f%RvH&r=8xoo|JP1u+ie^r>x!B
zzW!r-$D{va*6%G+bW#tUTy1*>>USDSuikgO^Y4W?2LHe+HKaFpzmeLW_y*E###NOp
zj4NmI?Ua;s>r2Ma-9s|+ykeP=j!#k>GS`&AcsLd*#@}E|yXo=(i4n%p(besE{p;yG
zl{F_NVs;nAy>P&;8kh84QV=PsA3_*27+6`>J9}8Q_%DNd>!E+9+5u|R_44@O&s(!t
zMW5=LKYlA`O{+F8Y)76IdcwtYW<5q|HW^_>h<BohzX6ztBg(ke>nM)-Aup#bApb#V
zi|!*3u;^z!PPL~HyED;z?yD!KwuZL}-4-QJy_<Hg9Ol?7s@Dd*A!0-Hde@hYF&ZCB
zKj#{qKDKMgN@p0BXTh6|X>2sHQUX63kQD@gcMeyeF(+G;GrjQ+tZ%Mj(lN~}PA_zm
zW_;*M0*L?HS^vlL-dhj`eQpE%q@!olS9|3H{Cr&!3z~^Xr-{kPL9ummsO;Z`ftXC&
zW_A5E^9+&U>Pid{sXNi0POPt{SDpr^BYa1(?6f~vGREB7D~AmBJ75s^(bKGdFPwX!
zHdf7Dlx56+`bd(3w|`yM$S$#bKoA!<S5;KXsA(cCN*4uUSa^V5h6>7St+0XYOzDv&
z0f;nS8(qu6+*Qz4KX5g-YlZ!2o9p~h%zeYd23O0I$b{vCv>TcbhPa3HAh})GEpd*n
zbdi6=2cmYv9j##rd@+*AyCA-aKaU|qEA4%;AnK;&ReOLT>-}vK0q1>m{O2wHYUG#*
z^%b_mO$-L3Is7Ueq^_bcPM?A6hBozM*}e)^A2~n<+rq5PZA}$(n-B)1ZYo_>q4;zk
zxJhGb1W5mJ<zI;Ze-mY5XIgLh{h9^br==V<>_Uwpu*Abj!=hU*##u~oud;V`Ltl}T
zjbzCSUG_gA;}xD_1p1C*{|4Ss2ypItB$_&k+-W%*AEg1)Us)tFyV@`&aR3+9>aHy&
zkQt|xjk(=+<fT?YLGME6JVFu$IOgG7BLZZPQ~6d6O6tTcsPHSF_|6XQ8+fW}>n;=N
z8Kylr5TUBw=1$YrWg`ZOO>kB5&9miDqcB~Ped3ThZN+8ZIUJn=>z5|^Si}DWSZmN<
zgj2kf647pTfgU1dx%O2d*NY)m4g$1F$*M(e(UiGdWw=QR-EbscP5Iw)C01L94s#Zk
z7JzM90SGN<rtoe+)z)ZsL{UpxT<9jPiZMZ*h7aX)M$8_E8#Ok&CJrK<hM4wR@4|Hs
zWgKw^2ddKxkY@z+P^yqIV7yq=(o(WWq|2_v_bUHinG#f<m?p9$B_+k>c7fN`)%E@M
zO<YXuQ||pySACcN_wQ3mV*j>S9#e2-j-O6E`7~<-fD6I6wT`u!%=4`={lz5AZ8Syn
z{9rE;AvY`(Edl}`9%D4;nvSsK+6h;Q@yOJO$zB}uwf>$9y3OtP5)|sU+@;9Y-?+Au
zI6saJld<!DEm#kPg#=Wbt#r9ADVYo89Aw|Wh#xKx&orsylhUL!lv;!FawPkSR)0C%
zmvP@wa9eFI-4ecZ2Hd;nq6#(#qL^OB98Dy=_Bzx}<H#6?;NK0Sezm|OSdzjX+8#`y
z3L)LUv@Yse%!;U`Tx8oIukK^6G8Y1&ZqudOGjN{CN7eMd-+m<qf2giWiS*5mAsyJw
zYyP?5XA&Ckx(g4Kz~eP;+gR@Q$@%V|--6K@eAdWft>#)s&oI<H9$&`ZdQ4+&6Gusi
zcMlTYE=pqqxp<Ga?qne;3Z$GAxjT2OsF&g8JRlZ%?$Q!p*b~`Tth|W8Zyf~y0Gs%K
zZmUu#iJR?wjQ}`!UQA3ZCurp#FLVn6via%W_oK6w;nso~A1F|V-h>B}lQHiJB$3I4
z!5`ZU{Ap?|&Ck$bk_5$|=Qus&Xy%*p9b(B=zd9%v#QfAZ0yvtEj<@cfCmutGZZ)|o
zQ79kTa-BjBWn3&eTy$N@J)1aDy8V-?l`NK~N4{n((YDc3#7m2Yl|E^GG!L5U5o<vF
z5}@@QSjAi)3z+hBQZ?^n25zuEl#^qHs;%fD<K$@D7&s}l=2hF;s=1kN<r?w*(oTM6
zFpSDYV)^(~hPt++V3Y#5xEd>Wl8SVufRelA>*}f;%nhZ2FBDWGyxN$=Am_1`l~#>m
zd%h!1e3z#!#JbX7IeH~}MWwT}kRoxn#E<d(G|DX)7#QTJ7UM&G{rZ)~GHWr49K_RR
zQ5vwYuqrGjek3P9`2;$;x^^ur0Bx;scQN@tNEQU+Qbm)>6VVYx9;qrPCa8hUlz=LU
z5+yM(Ss0j*{$fl7!E`ofWgP<*Imzj}ZyQm;UZSCeUe08^a7V|sxtU&8hGORQeyBus
zT1;HN0|7|d$O87`Q3EpiAXq?3$%`k%tG$bfMW~q3An0X*4TUJOix>W#6BFzG`?yaA
z8i``ZDj9y`O20e?Bf!c@7pEl2%FOkG?&zEk0V~FMutpWkmn9>l4PA{yE|8KQDco3k
zo40pcH9jRIL`BL-TgcUh40!Js%c-JL=RFlfJg&bReQ`Z~eeKiJm>3us8kH9Q;o(S&
zi;L59b5n_RhpghcN(xFr9n$_7s+Ht!arC;6A8Ysa_d}YS`D0^aCmON-M|MEAl|ysl
zehUtv79NM08EPGS7h<JAflBeB(1L>h7sNJX75M_)IuaO}dwa?^_SbEmx+HmqX%$A9
zU98P=UYxlyudfCp>PTVmvmmxFT3@m@f3oP^;21B|+~7!!ZL{EVLH;C3hcZ<K0x7KB
zy}bs113%z$*e#cWp}HvIQc{1HmeP#-VqcV)kCBOl<3CiiqxzX<ft&hte>U_jt=eXR
zDzg6HUeqxDM}d+lH@X5G_#`2LFaCpUW4(->C^BQDItYxJC2odd*2YT?f!|r#|0HV)
z3(xwb%SO1eDD4EnJIHP#Kq_Zoeev-moxEz?*#|z~({PgQB-i{|h%&?w3KEwgBj&~d
zwiyU3?`-S{$Rb@?a1sEy5SNGG@7J@h^Ab`dq`WA=g5TebAFCC9URAkDM>l+MQJz~A
z^}Ac!-l2PUhE3ZvQc^YOD3S&;h-9<7Lu`XoG$i#UfgY2)Q=ibmE&Nqd@}8cazDT$E
zu|O!-85b8<N#jVu?L^tr_o_1#2g%!Jw!+*Q>3^E>JmcgPIV*D%v;AAAifU3)_Hh5~
zq0-w`;~%I`386{vvJKXz?QXCG5)-|~>(B_?Dn6?U4erck?&0w4&HM~Kj8Ui|2(SBr
z`VO##--K~128P;r>B>od6)Yg^|Mv*?p5kVRvku=gX)QhOdTp5yXazT3<8~nRL|`pF
zVS&j{Q~c{E0mz-6Y|<L7^XPaGLrr_#lN>$K7#bX<_Iq5Q0kBVvTV481$o#416<?l{
z5>kh%mXw_VGRUP&PZcWqN2C&j;_ic3y+q5wE9}JFRHdilFEwCs3w@teHB|#zj{E6e
zaWkLTA&66!xSbflAYl7dR<b(O;$|v{+t|>#9?lc8u%Ln>RZ*FCFkmO(Hk-%WYH4ZN
zH;}~s;%89I&;N9^?a|*@nli}-XO-Y2=ymv|((u3gxF9>_)~HSl2*yZD|0`l^S_Q|m
zx8f2eve&c#nwTE;Fk98*55^KcOgr%6eqq$+<{x`7AIl*Gf<SU$C@~{9tR$->A8ybO
z(#Bt3Yzj3D<T`HuPV9W%HWulB`VrkP1z*X6HCFQiqLpmN%GxO>85KRxLpLvOauxc<
zWj;`Ec9$PCo}hw*lY>HQ6ME-@z<UE2@Ar9LkSc<q;43w>R*Jd4JglQO9uwTVw~~#0
zO@U&M4Gt*)!<Rlp^UpOUizkU7^x@Sjo|$Z0nA>wu1%T&YYU2KP{gnxcdq)Sn&4F@Z
zWszmMcVVMi>0I2vlr2NLyJ2P5j05d`tK0L+x%IwY`3}nu`}#F6TY?->#INGw@TMj{
zIyyRde4cCb?-Xx_v^G>Ha#~9C+wt#{V`7wLB>!y>esS$_Yn>qFt>(EyrOj?l&dB|J
zBy`u&`V?ftx20?`M3X~8873w>C&uIT5wz5&n#nPi++u)@sU;(mIWXJI3E{>EIr<)m
zC%CQ=eMSwF88{v9Uc<?zDO$q&EA-4OnnGbNm3uk|s6Lyv!-li)VEQT+U{6!z&7F}p
zcsqwi;UM~vH9K3AkjPM$j^l&4FTH%NdR$gTvMs{FN8oj@_qFK9U$X4wvsNWRBAb!P
z$wWAm&Jz1ipk3anePH~RqO~<{DE!XOeoOcQw>{+4vSIcQA;`=*Jvec72@3Y%>5585
zjc9X4Xe#xuG7DJX1snMLzJAQya`h^nbqur6$VMLkI@B@<Q-yWSt5}AVjMrP$IpBfr
zCm1mGbKnC=l{N7E2o4Sk3xjECZ_kjB{h?c9_d;4)nuV1$C^#5;`SxN@DuoM8R8;i*
z{QUUxax**4+oPl^n##e^@#buZAB)S0LPbSodwYBMmjLvl`xf3{8h`OA2gsOP&6N3%
zWQeY9ZwJQ4qEAjvo*ynGrTIMG2=pTb2L-v8$fl=n_^{<zBnt}*?@yP0dwwz4=fR6b
zL_{jAW+Ho{$jeQKQYG-%(><PdnH1MW_v5qHZ?O5EZvWloEg5KTZVmQ=aKZFnKryMs
zn?3ey#vh*;qpeQM&t>!!g*UK@3iyM83dN9XOJ6DRLa7=M7Cn$d*$s|6X~a_Bd4#(k
zLQanV+~+|De>B|$LsZZZ==Vy5R_(LB_}eAdgBJS6Iz0^m$b!gb7A`igU;4$KJvXU;
zx+8|1nwFML+I^c~K!1WPNH>eHI;^#H3yiZ<H#Zg579r-}I<4)=@S&n(FdIBheO{<l
zv`JGh4inK-{FXMTyKLrURcoEL9&Gps7+ZU`II)MPhf>|snFN3ui+hGI;%kl=dTQNc
zcGJM!wQPc=l&89J`pY1N0B=Dg1#HbYU!~ipikR5gTj!A%0F8r-h>C`mE=QqOX@Q1-
z!)DYIN%~n-G%6{nZ>HSzX89XZNmbQmZwxhB*v{UbWD1wF+r@<HOcR7VC@5%sYYPP(
zy_3HA-Zw4|0~!YYBYpAxlcdYhVrXqGC$UJ#@=@~>85dV#Tbm#<GBPNFP~ac<s;a8O
zN#zI@vlBy@2?<|#dU^)HW3?^$JV_fFk@)-j8=06moU96~6l!#Qc>%o*K8JsoR1Vfe
z>T+dyx#8Ae3Y1hTPr~~8`tG9NKfu%G*flwiOZvPG@SR<L`fb3EkQ<HGSU0>l2m=MT
zHd{IzQyW3p3_0hMSgz!Sox%8!slZSAEGIN^LGU4X)|1{`Xt|#!GUUtoZmd~%I5`0i
zYOOyUP(I8ijVjHuMdB%>;UIAN2C}7bbKW;Lx^(j)Er3XZp1$;419j%N#}TU3ixv|h
z6$La|g^wBYmsOjaz%Qzgricldt<~KiDF*J?V{SQcr$3-TAS$vC@5O03MP>W82l22b
z)A33Z)X$g7f@=z_Vr~}(==G*ubm<nwJHh?s!{__+k7&D<if1ZlRG7W^0w4_|BjW_q
z;SV(Gcpi6$gkn<@lhHC`No3?+9oyxrC7*1cr^oe8J{_Saeb3ANbbz8L9_tU=c>Ch$
zO_y7~qJ?D?<#pyNQSD3?;(+_eikv3LJ1%ofMnV#O)794(><@<#nVK5b*vQNObPs};
z+QfW(P4nIxjrPxG1ZXl+1s3bH%P0Xe9v-^0O}2x>6(6|d8@$}aQDWCNeQZpY+i;(Z
z5srpa{ZVv?vvUH|=Pns8j)NRbxpicv4Z+tl!RHUH%7RI1x=!v4LzHD)EqlHA#uW53
zTEU+u%%w|b-dXk;)9lG>;JB}(`9$feUH^5def`?&3{-)Z)6^7YnL|D+{jMsoYK`}T
z6#W*dd@j$Qxk$%$gb^z0u<aS$b1N$?e{=t3T%;mqi7`9ag@t4_lLpGO)-?LiTRgFW
z25iaq_&$v+w8o7b>&W@3gzcQF!`O>c^(D8>=V6xze%zq2%=AV)O%ur1tcKY*o)51C
z+;;kt*pFI=E_Npj0d#o<3#&>=SZ6juWW<DJIaxr+%ZukPC?Fv40$_He-cMk4F?!!$
zlV64h2d{;)FG96UPscm&O@1ICc+zvR)6u~PuXQ@qx}Ev>@#7a-%_<%?8L$kF8m|_=
zH6!8k#Q*&H{`B-zjsl$ShY^*S*c(ACipgrBlzCOowJIegwZ1!%55;0V8{N|KsjRFF
zybh@|@W&5>_5cJH@Nq^N-%SYpR-%g8bs8TZFR*)8!!-`SmHX_uczB%7n5HoTd}@jp
zHY;&|$+!330)bTG)ZuJ+d{mGQ9mXpc4INEEOX*gv5MXFEic?ZmyCeJSJid2iJWCP)
zl4`;I%Z}~Fh{RXxqnxqX7`SlTHks|^y6_mf66p0^*gd8)hU(+Zg02GRySr`sMXkCJ
z(4SJV;ZVmhJqng5H3aelq=bx296_P~GIv7p0yGg=rSL*ONB!__q0<Vr+L7j5j;-S^
z!J16J<oH&7pzre;wFz!>c<-$-f(vZ2es}U~cqiwTv{uOY;vlGK{ap5mI~k%3PZ#!R
zRZjrrJ%^5N{DrqIx_36sWHmJ{3(hYnXebG}h*6rmF$X*11_$j+%M9k?n9Ww#b6i5`
zeZfhai%&Ng-u%n3UoxBUA==4i8eQ-Jv;djMj+ZsGk)0Sg^{;*b4!6g5he^zPflh#$
z%3tg5Zf0ioixByio<%*pX`?AMDe3OxikY6CTt1(W_t^k@GPi4(^;{M2p4|B5h$um+
z)6Qtu<3exmkHdxfu|`kVI;S1qO6M7N(r@3s*}J;>9A6C&50g)x(^65jwYRScNYBN`
zWA5+o<C`;k-@8o~X@i>h7I22P>FMjBIVB2IP%d%!`T2LJiif~>CjxiKdUu4?VgnY1
zw$x!pI3E@BBRQzsog*V3zWn_8vkSRk;=>lemG;iG5B=yLK71`Ncg!?EO%1c4ANjYp
z0+)98lim)IDzsMq<=2|0#z$D_2~C##tdQOLh?2^DH`7o2Mh={JM8D&``_rZ%NKE!i
z`kN_S?K^WP<E)IE`RqDdy@^W}$_^NLN4N-Xc45A0*SD=_S5<2&l?U(*Q$A1b6Gi?f
zy!DO`;mXe$`OMF~hsT;y+#WxsE^&=Uo7!oP`MYYbKNkS$?6FYXuR!Z@l-%`21q+MX
z%?PRtox0Q(`amtv!V`b?dAVJ46d&@n+<A+t<D&{HiICyj6kXHBw}T$i&kmq*Iz5qA
zz0B{|ua6$*+b7-4<Rl~|-VYvNJXvsPC;}dr)AHs(5?UBLSg}kPR%64%dIO29NRNkG
zTUXtCpjH;>HsgQk2*!v@VDTSCMgcej<|7Twegt;i>({T5BI{{2Dwf-k_zM^gdvl67
z4#0hq^5$}tY3y+JPt3lg74h+T+70<H*tO!1784&RMU);<YaY{St}=!DrTgzG#*_y-
zX!^O{S*X5t8GQwuh*bwL)x+G}tb1K8F>$o+D^4oH?o^^jJ(40-%xBWY)=!EkqQvb~
zp%WE@502;ST+R_!=pyw?GTKF1WI!<fto$24weaR?y8fsntBC!}+S2<sp5MMezn5=C
zpa`6+yGnJooH(?|pPa1Hu*55zm_*bxO>qX>&r4JcqK`jlW<1Gn9=;xz$9xb(9Y1r&
zPs>`LcU(&HL}~LJat}xA<G=YydL*+N-OTSw;Qt}zXJAnnqDw5gTYcTK%3Q5NYQ~RE
z?V%V$xUCZs#-KItskdQ)AK56e91V$+^HegwX`<}stEoE01K3{7>gwtmhRZ3ap@H4h
z)TCAK66&}$XutY5sLgM^)^RIGyPoYkBHnLM%{nf@4%kz!JHN01M{sv@gBHvA<^GR@
zK+3G^5+p91?{-_)*%A&PqhBtrZ^M76qoYIr`kYM9$jE4AC|fGw_ld_<;txet)i|(3
zn7yv;0uXRIS382S7!6*5wa_&)gC!{`8QMvkBYN=4w%lY_mi~Xzl_D^2RwETlhNq}q
zSNmr_5qz0mH<I?d6x{AHax&i)h|9vyWG^Bqb8zVC9;9=1z)0m4r4;5PxH{ibf$`Xw
z4S!X_G)zv!8p>1KAZ46&?riLIWhLS&dje@Hu<9Rhs~7TQDle1cduX3-Y5C=w0YnhD
za!f3&kt%D=*(&QkQ*aK8LM{mhM?9%y>~R-?57cnEsjPvaAtO7zhrNcU=N<A@dS<47
zc{y`yYpb81-#s|UW-7dQV1QQfW;EFpOi6M-eZsi}REX*-ol3gnErZoGLn4cL+i&4u
zwQ3tYK0ZGGA$1K6CgWaIGO744p`p?@&=})HKVH4{l{4=_d*^io`vLpG4>Y&+hlZ3C
z8%USeIS)^a<Zd7P#fjxAIb6wOeqCAN5tnNFq#9?T66;ns-m#AM(9Xrlo05NZh=N5R
zE8@ZiG8;A1Q!SMu2%(FOLp^<KvyFFt9i2aaCI)1dGO}7*ALl4)j^oh_o1LSurSS;M
zxxGr@<m8xEN^jIe;i|6(-_aw68hGOg5g#bAuqe{|;R<-E8L6A`OKKzJ6RMRdVy?gQ
zd?aD|5&wro3LYWH#@IhwQX8I1KBc8BP*SFK>k&e<hSD9=SIWws-IM`6sC>{n9-b-%
ztBr>jnKr3O7d|=|ks$wQ7XPOJH!t>q<xx^Wg{|}UcD0cPNcIgQ06SD#PHjGeEg(kq
zC$hrTY7I^nY9_FlE0oUVjM~bl;_YbQ4;2Yy?u^v`s^x&IsjDk98NgO4(&}033a|CN
zqWg|SU;<zS_|L)W<tC*r2O2;SDja!ump9_(>w39A<Bu^iU+btl%?94F3BRkeRPPME
zDvev#OJdh?2xn(+EWZ{*7oNa39}-IB@m>d`XJ?0Ip)|?4OXiq@Y_oRDxgxW|b~GMK
z2xg>8euTAxDoK(3UcJJZJKJWSq&SdXJB&>kbE1|w?fgiyX#L{h_)XJNg@0olU6jJ#
zckLj_d~UlCNgsgJnl6D>d39DNPl4*y><%aCdPqn~dr-TDVY}oXR>gLJ&2F%8E(@#$
z<=iD~<$@C=_fKXDOfpn9iqW1YK{pTT`Fv<RY@7+gnpLGY>sY^0P|`g4eEt@4>%!wG
z?Ch?h3+F18U5%o}?oW4gaJ$N*mVPRh4BZXcoAkf(xcDTia@*zxLWJ0pDz88TB?&Jy
zj?l%D=M%rEaGYL=kH_I*2YGb9k*J#m_D*f>csaEu)4vd>j;Ud1i2h`OI#e>hPt)`(
zTQxOIR?A76DguDe<mBGB0FO`EeY5^7f!Wk-EGtx_l#hFsm4QKSjd_C=ZOaD$Zd^P(
zdnczqo>zx7+I7)s-nT*KV_A$Q{g?pyDXv;lH(+AR9(zAMT=};*DN9f0m4||jj385?
zXZ3*lX7dQaXls*bB+_7O6#r~DbeOq}h@3d+PY&@gsnh>u0qQ?5aCdAfX3sd>!V_E_
zwcno@c5G)(4IjeA#aEoiwUIQ}Tcah;%;<q>Q{aHBbn%{Kx9j)}K5yKMlMLD5&4ODe
ziIwe|UNS<~7<+y?{ilt4|F``YPCIgQL5U`0+m?IpQ&I;fy3K5Of>!1(lCB{`-YQZL
z@j8n-#<SsC>C5@7duk03atHsHpgS!)@p2rT2q7gK=hshT)79yCX#Xl75l{c26vgZ2
zQqBYs=d^Qbq2Z1q#3gOu<H`=JX?iiB`JTAa(!&$6$SJXdnSm6~|F3^4e79YM$+e#%
z3kgfjVhL{{ahc%s3v_F`*K|eDEF=CK#DSwB?Mgls0>0ky4WH)`>aWan1`u77sMK(F
zqY_hu<P4C>3HVHgI{Nl&-4Ym6xXbpxT@6ps@=|JiAX;5@_k1K8Txx<|j}zM7HVbbI
zV&qbwlsDhef)6KIw&xgo(&#AD_EdYiVbNgC(FLDiKJh(B;kN#i>tv&9gxDhnpe7Sn
z|GrvV^s5vNcl^i^{pg%Bu_)VGIV{j&`H$x;z5<Gbe>n3G%yl`vwa(BA0NtjyY!@5U
zucrqNmYT=%Rf~15-Oe@#EXZZ@mGf1uueZ{CHV>+oCbz|Oq&w8HX=!P*WKw!7%*UXn
z8$H<-v!vV!?~s3xORxZWpRyb6UFSoMUnsc@;lyO7el-@$J`;j=#ui-FmPxk#(YozC
z4@Vj)x69ho$7t>H<~F%vuM2qpUc-^e1FupGTB!Ejsg1o@opBn2al<0P)~YXeUXFD9
zoM`XaVua4NRO&wDrjN*OvpJ%-M#yu-Co<GKJHphCi_{02KN127xzR4c=WsFmDfvk4
zdXOUvaXWj|o2lFp^ZI}%?E2b(Ch8<Nj{C@yrL!jRLr9SN%)QqE7r$=(AiIz4vdXs5
z<j^%d-U0hrerviV-}#*hox20);HF8x`I__+SLILi6Vf{!ua}O4+9p#E@0-*bW0TU;
zt**T^f*z2}79u<5Tu+mV`J?8si4@nel$12Mp5_rgXBA4j7DtM@fjY}~=Sa$bH2V#V
z{)yBS;=8g5(Eh;wsqA$99<oQbq+QcCo6%K!ar^;x3lf|eWS?N@6=B2oCbKJU>3z)4
zFZr3T9p|ckdD3<f*-Fj-2wh7mkK@2R{TVjatk*Ffjx}>qh<tanR7dKA&+(Uj%7!6N
zzmaGNVlEgJIfI570wyJ;@cfWDcCnsBg(QR=T=~W?)N!(oyQMgyy^Fn>?+&k54ra3b
zCtDZiyPs{M@m(HMc=LWiqgG*X-W?~hT-w~&VAOB(YkNAHuRUu{of;c6JYH_KTBwTw
zR>m{x(5SZ2-c)mTcE(~cdp`%}M|E|52YfpU2v~8oX{%;bG62XcC@7eLpOwDdHjUaD
zF;;_88OEJ}>u&t?I>XiZv$V8;n;`@|3OKUKZNm**ye2xocD}q8CUbq^<P>T)Z4J}d
zeA!-L0g=(UGMx{Xd{k~1b91)la!;gJDeU+pnlu2ZFV7O)A_DV{xHL52i;vnWOWU_s
z*IPg>xb=RZHBV5d=iFV*1Omg>9w*p+M;c>^r#cZ0)f(%b-^C&!sjSKrgYEqAN2*fq
zd9zCCMSrcr6)=htD$X&$Q>4(<mK>ZBa9rzJ4&LM^zptCi-exiu;~P&D!>Tmip^9de
zm_0RJWkPxR^Rx#%@CZTgayc6J%fXC0)3LQ~8JM?jHZ7>nG!k<u&^~fH2h8lTMb{w7
z#??95><+erBy-KG<|bxL%X!pRcYohn<05Y<8*OcWdt^%ejr$Zn&G|V8Lk&H@tA(nO
z1I8y=&0dn~04*G5_@TV2>qR7kZr8IOh{-D{Vb93;&qu|Z;Mr-JyU?y-@u1*Mu|((I
zZAh;~<Ayz7jgvX}lxx(snwea;qPxnsXfYa(ezEuAeOgPgY*e;!I>=0)|5bNDdk5B$
zs#Or0BQY`G{#=q(ZSs&zvQWyBmSnLGIR|5qODIk*3zzz{jG|+#qoZ?wy%zq=3UhKO
z0Dph?bO~_%pR6p0gIQLO%l#gp@w}_vo2#}BdSJ7ej|cEHDm4{xeX5|*c@G`<U)Q&r
zKDv9u>ClLXh(p%(W+T7bx~yj_^n0TzU%|jcK1lKU)N08nC?EqB$8e?1pGK$Qy4HJh
zC{6eJdck#R1MF2~A3y$Lkj+w}5*HT-uYe07;BR73Q8p(UjE+W~t+tgdHyN02_Tjg;
zxA!mhadBZ|w_W^W*cIk~fWvO1m4*3wAm{j%?J9HSvc*)<z4Ov5f#C*#^wmL8VS*^u
zdAq!ngq={)VXt!eJy)TKsm@ihHvd`$pX=0@7)__5Bwme-?gooC$Zr5rH&!e|8&kzR
zHeu}flv+_FAF7#-bmO(Li-DWF`@A0jj&Twzgiv+4_ciF<#itiAb>k*NM@3>^-Ui1Z
zf56Z+xI-Y9DcxB&&%?&CcQWI*(2UQmNE1;m;9#mCK)>|>=auT}zYSs+&)(7fnG`1~
zyOmgQw16grX3G(7w%nU{m;x1FZ4P(UPAp;d)Xu#&64w*&`S@<`@O3V38gjhr3KjUP
z{J>~rLVG~;^dkTIi?m|3&?2SYRDnkGExE?Om#?it;W;^K=n)ENIS+iI{57lcao2*`
z?0e|_x;hyDRw7b+Itj{ZR(=>pS->tUaO7Tp_qx4VSILNX_a%Thq)|%(S%m`_+4Ob{
z>*>8GlwawGwAS^yg_8HJ5|~QX<*&xR!?yuAcKOPm;ch>ZiPZh-mH+X@Oz`&Zc8S}l
zA4VAk80!HV?#MuCG6pCGl%9Y*<^12k0|6R2Xq5vc3Ft=0=jZD!U!ncyLU7o_JT7+8
z=C-0K6}x+S^skRs0OKHx5F4JH%#zO#xxT)Jx~VjoDo|HkzXVTNfp&e>@lza~Hn0HT
zJUl$aNFv)F`v(S2z;6WFJ~<Q9X|1=#MBYWQ-)@$4(zAcflu&SdEhQ~&K9Pr(uNwDY
ziwl^?GSHKYbsB-bOy~7*aei(#U&E+cr1b&_WHsY|>y8bgsatVyai!$shMOOG>m_vc
z^p+imFz(^t;n(kP&Ym>1kDA*TUkkWBP)1Ae`Ea2vd9D-0PmfcMU3zp*R;1Kmec{|=
zk4XQDHthf7xmyXavV_$g5sd7uqv*KLqLRVVkUIU})P^0c?XQGUA3D~J-@WndPiz-u
z92+x+9MN@z&C<#sZ2_whP@R3NO`B;YtNQ(Ut)?4KUMNdXr_QPn=RnKEgt7)pSSV26
z&}#+dvkr8PlVp;KXvkuz0$>UbXKMSM17YhY^|jV+K}{3s*Hto@dp^S$oQCox!V~Fi
z!9U+9uB^`9!38l25pSy=PIWl0XPR{@UqO<*U#k%vm`1MFn#28rS%0gZj7i4y9n%iY
zEX_~Wst*2=k;j?eVXVv-&f(TLpn(nbaeMY_Sr&5_Ti($7DK7Bya-+K<<T;P=+yMdt
zpD$XuM$3Z_G)=3bHk3!-0l*VE;O&ph_kPY>c@9ox()~0uG3qYjyvg)O*SXZ6|C&K!
z5_5k?ElOv)VoVTl$Q6jJ#&}_fJ>@`UkMj(W+)yBoH(U1?Q@W&82~5WCB;)94FWn6d
z4ZR?cT=|S^;YQJL{DG6TE=G2CRc@6pxr*6kpjKkA<u^9iSJ^Bif=Xy8*L#&jrJQH?
zaC@OJ_m*Cy7P$eCq70c7uBY2<w@t~C;NZUJ`|fCOpALl6{kiJml83vaX5Caf@CjaP
zzpd#CU4Bnv!IExL9x0lMmMuYRyT5RYobt0JH8_q1H~qK_1fR=K{!&UgJ10yw<Z3|F
zaJJ9vnc2b8qejA?wDU8rUR@?cl^|8*StuQ!Qq2#?T9jst?NM{xf0dh%AY8?1w?U=P
zSn5=?^9kq9Lb!@jx_DNGrM!IOFoR)mu_051p!oL@+XHk<$fmS1zV?HzZsoZkS&Rek
z`U`V)(l}VG=`1u&lY!5=WNY9Cn^ibK^XvAd*%Z5RP0-4wxA2q2)@~Iq-)=1x)T4bV
ztmW)NhlG0`L}F~b3;MB|6HdS?6_T(%DOb3bYpvfoEpj2&*Kg2%qG)lEF5Y}JKoH8D
zpNRp(wL1}*I@bBz!g=}tE7|y!vCdI+QYz_3l~Qk<Dc!Kq1dR9D1vv|4?J@2K73#G}
z7uYr=Ywj=DpZB!(*&mAhbiY!+HA*nDI^^{P5o{_>jQS3^zO;YXvo%Sp@4p}~UFQvz
zPggqqNSR>F$8OeOLA;dQ=vO559U)ED@LKa=+=KSnA6QY2ijD0Aw^qH%3akA)QrEx$
z0y8snMMcH%*jR@5!!0}xTL@5IqT=Jb!9tFREnEEhnP_D#oPQG-hYfd!_0uymgfS8)
z$H&V+;vgX+63kS92IAaDU?qKnMPjlTr$|moiIB*|#Kc@)Uj7u0&#P8p_ThQMYxRBI
zamnnoEw{S1rrzkm+UR-J+1vXrR|%XhgNDcLVsx_F;TI5KV3xD!JGv1BYAuubC<%D+
z4FK-jQ^f?Eo10*=&&vAx-{X~bU_BwNtgMXZWzND&37k=NNKgUgq(G|{+0^tx6&d~2
z%7hXF-u#Rw33AZwRdRBpD6$KL{f<ssU+vBX`-hBM0zg0!<Z)vQ9azlT+Ld=$I>mI3
zL%RA)nbb61rpYkGH1)oFo^FkG7FYV|*`liCuvH6itMSDI51N%U-uxG7x+H?i$jb+}
z&b8=h15JIZW=rcA#VIk_a_@0J1jT-C8}=|x;ytF}2H7`<oYoasq#NriJ`;F6C4;*c
zg_^4ty5G!w3ReTe>6v9xOgnSWz}5k_JYq)Xe7oMJ&qVr4*@#~3E-cElv=X72%Nm&L
z3Yvxf^Y5O4;18VY8P^*oMHr=dmipv|XG%a6C9$Kc`fwBk#&o)=7WdwCeb8#e2z2G0
zLY@oXzeY#E!;7ynV9mC<KHCW@_wrb^ir5m%%k1D@sAn=gf0GxVtAd^RX^g9~_I7T_
ztI`AywB8N+N=~WS?O-FfXz7MW?<cnz(SH(kW%rKfNDW}c>#HF6x6#C4VPmguZ1@A8
zk-+QttH_<r=%2UEPmkQ-PyH;U+lZ;oc~1lg;k{_VrKOAjofrdin96o-dHL6=2FpgM
zqO5E<p-=#WQJ@EOxgSQp*4ssZUU*yySuMtM>`;rFn|(@MrGcsg0|Nt2NieX|s#P%u
zH+bU&4iTnhG80mJTk6=Oi}xVX{mr3^L@5NPl@O|;n%)~bn#v+E!NAs_{~_=VQ&~A9
zg=kp~j{*~u+j}%=T1Q2TXr|02mIK_p@&lds?7jr33CoQ-$FsR=mX&`VT9VHAPLbTd
z2<H9?3ay+aE`Y0W%ozl?k<;^zjhTY86;5ubH5K}7(~kC-uYOQ>J$}g<P+u`A$@am~
zm#{cEb!=yPdPyqdg+;gSv~C2s8m`qGsmj(PQ<Iw_b{#r!=_4b5jEsiMJd{I>Hj=8Q
z2b;LI2t@68IsS~lqW^@lrGcw`j28!eSh{yFEHi&h&chZKeeRT}p{_7C8Eca&{k(Hk
z?(dI>^?~UdXo~zMw?74}_&rnG`S|=qgi(@`LuyMICkhI;w<rC)X&X-*l%L4L{xc4a
zZ>dy$t8;uO6mlyVTIX`)Gfti8-?ZU;q$2c57Zy4@B_%e3F)`c2Mn+5f6;$H)?)}aA
zyfQNzm8<SVNa!0$aZ)P6bgiJc)8X?IsL!OR5Qu08`LDuE|4M<I$mrL1K{8@AnHzhA
z%kr4NW*pD|gcNH7Ez(S-8gVI{K=BasljBs15>e}9_s`)!>+c%n6^;{hw4t|h>aeG(
zKNGdqr&{U@a>&w=lXK%hpdlp2MKD@rvlDqtj?;QT<|r8P{x4?ZH<i{Y$-?7tE;gq9
z&i5F5Y&xw=!uXMYB}TcVDyB?ouLq;h&cl<lh{0b$y=bre*p0<b9sQh;(&R<Ss<+PO
zgV~=rRnb;oi#!=iYH<OAL@tcoNK|~qo4e7!dkV*V3RljQJgE+%r%yw%KlnuR6ewbh
zb<SlM9-_B!|C^og22FPDtxo+rY)aBMZTUTuZOH%?i_5x&q?T6BKCBy&MZ`sY)pou>
zk4L6TO%e8nj0BEXI9w?YiwdOWiS=~r-OZfs*L(Tz2@rlJCnI_Rgl7BV11Wl2dYCG6
z|I#0dH@cfA(;D&MZc0gcz*LFs(l#9Ldt1@=RwEzh|2`aZ__%`2BU-hgC0yIEpPE`9
zIL?BtrkS@8u9Zu|#fzsX)Xo&T7%C~Fzn4(H-|P^hiG+_F74280^aIo~og(L*jRWE4
zSlVGldx3giOZkkgoJgG|*Y>t)ms{NLMK{MEsS-h05cGTH>RS(J+Jx>*<-NAXW}|Q=
zx)e7+)|5%xXo<<6cfSq<Vy+Pv*R*>4*!)DyB5lL=SQ}1)OQu@hf0l}X+}G<(vNwYM
zAj87e<yDO0d_)q9cxZ8jf@Sye*Ergye{cTse2O>f;C8@J9F&5xm;qC`qJn*{JRW3W
zG)fH2?^?RtH(c!W36vD4ys!0d!iQS1w+Fzz^lH?&Nv$$gojBmcrWijntbm!3p`|tt
z4+txMnSTs)!?+r<+uL5)InsG3YlP1Z&cB0n=a$$7#25-xD~97kFFyv*c+A$sqA1YP
z+I0N=5El`o%qgLTnw(J;ieo{=Q&}_RV)Aejm5d0N9US#-_CYHskce7V=Il1F&@VYE
zuIst=W^ED=%>Q_K!^~0uf8L`=;f9LRo)M2(mCE@Yf@ZlHhShOgJsXo);8a%GF+L2H
zmP7t$b6u~%z@UaA8-F8>_{v~W6pfUKE;S*4vgBqwp-p}akp5|GtjOe;frS!=!vO!2
zy;Mkrg<M)1S#!qZQC<H&?xyWx+h|HX@H1c@eZm|z-U();y$24!K(@@9k!c^0E3t-e
zZxWaOMHb%(ZV*2Ys8PyE)k*DNYKU2u+{;;Vr$P>j)p&TNdtG0`zcMjio6F5Lt8fYX
zJ@RLazkh0~#)DM@3e=~eO`XgtY4E&`w-L;M=`YB7ftGjyQ?BWu5Z3dwFrB{cfwlcL
zJGS|zu1ik=FjC4vMyI)Tb|)>CE3w+tkv{ftejs4(>iV}Zb7dGgA9IutA=1diNP0%S
zX6c=}=Q#(C4Qx)zvhqtD+a>5<ajrYX3wxh5C4<4DcQ~^De>{B!P}SY@H;N#Nh=PEU
zBHi63AfTjlcXxNGh|-93D+ow8NK1EvbVzr1zKhTA|IX+*4sf~me9zglyPsIj_$}l;
z*+Fu%L(H(amUgjFFea1S8WwJpjgo?QCV4u~P=-9Y9Iw>fKS=LZI5Cjp?M8QCp><|#
zCMk#Ev*gw5JK-EAP8wvXko^5{ZC11U;n-lalf0cB=jjfUx%Kqoj?d>Ht&$iP4(q}V
zALssIDz9Bu0!zzEcJq#?LE|wm2EW4KwQNUTXw7K^_@xBIQgv$#)!j=mjow0L+?jB_
z0afI*laKju2Qxx2LFr@ihLp0suJ$4F=e_KFfo1EN@52<yyREH6jF9<79%DYnk-6rY
z`Rgu#606Yhs{GRf+gu-GMRkXgJ@0LDLu327zxqm3zq;YGKHHn};9nUGB%+XdSrsD^
zeiQosD$Z?f<%MA=e0clJl((M8Zs9c$5jskOA$SfAMF8s2U!W^oU0uz{!ZPS!0uNm%
ziQaMdr##-wdAW;?D@kd!SX4zQh2XM_hrPgXP7a*bbEH>P)ep##k%QKT!W2J;6zqIO
zN1Z6AwkPU0UA>sIM<lSM!gP0_DYTT<BJ`}iin1@anl~`&XyKo-DR!-u^m`w|u9=9x
z3PSiNGS1V}_I`Xg41?S875HaY>JH*DExd(d98sIoodf1x)q_9Ys|9}Z^xnNe1thn|
z2uLnd!l$Qd;{KZGH~N1L<vLwqvgR71-1l-qMxMtb5N?bMDX|*;)-_R=WfY2A-{*>|
z;<AdHid(6d-lC^x-=j^gocB$2iu-A%tG<xD+4y+hRPnRB`C9Yn0^aE^QJzLkH79re
zOjQQ@zrRJo<3w3#X=X&8XynnJ_&L->Prv(2OEc4Q_rjSIGd8H!(X78*noe5azVSjW
zNe;h9L`jz<dryMa-0Nxdvje@()nX55H~46->!=6`{o>S4uJIGfT&S6@(Gu5GS3E&N
z8gy-mHdtC<tkmW6TIs_LzYcmE^`(w;5*~DrxkfT%=qGz!A}tJB_rJ+~<zS)<emKZa
z!O46d8eci<!$@NC;mhN(t82{P@>p(KZYTA^Et>)@`R+$yNs-Ok^86TVZvWkY)J(w?
zF69<uLRMB*73a@5ZJyU%>;?W@r9sK(0>#gVUHui-c?o(;4<d=X=v`S^{Col|TycZm
zXoxyJE03wA+Ht7!^Ti$*@3#+L*llv>nk%Gc*h@v0cSm(Tt6Y9$%2nYpTFxQuR7O^o
zd^JMdGgT>-O09%XI4+a+xRJ0-@fW8VN+SFALdmt1QJz|7XgK+kiCXDHGs%@dz5?j@
zdM8Q3`1k|mHYEwvPBZ9#Ikdfl4h~QDHorHQg`1kQIB&cbr9TNPJP)-qJ$RIb`z*6M
zdA5mkTEGtD-RXR*LfoCvPR0T$8ne72V%D>JsmBDm%kQ*aciLRwZBOb?<sy`jVz755
z9Ni{SoI0KtwQp*Tw(ikvsfr`$&d$8jqz2jN_bPN^W$IWaheuIq5!kT>E>zBh{3*Gf
zj!q9(GUfJ46NMe=g9(MI&?ZLRmhiVq6HhJr-sw%akJzUHf^&mz+`Ai6SSVA@teNVq
zS=Su$TL;XYPk(r8N0z&XVaS11{y7h`z1>MxB#w5;4y~iCm?Z!UE|u~l10Mu=j`z2E
z>jnza_5BxdfzwB_bp6uC;K`I3*PhJG)=Cy5Zh3Y%$+8$)`c%ti4wv17Q0~&$$zu1h
zhjKSvNGPIBhvHE+KSa@~->}Hy*e%cUlRFKVCnqOu?CpiO1-xImySViIc!TMXbegJj
zezeKS$!W7a%_jv+S|PH*+Bf=;FiBb|QGdp9etuMM7+EIaYL_f5P(nh&^g(xXf~iuz
zouvC?M}Ax+S++p@2aKnM$BgQ?x3Wt#g_N*8%eUw4UOj<TiRG|z@Za3#^=PtoKi1F7
z=nT2y>r{ByAfz2ifas=|mEFkzV-2=#UHmt3z=WD4BQC)mU+|jX{I#3fc@b;;_8n%o
zna2o!q|Zx+kmf&?nRmwY5V(F#5^R;?x#uVI)>NRjpUL6M80U@Zm?S}8XkK?*RBgZc
z(n-o`TS+O`-t1#xXChKI3bC|JrSloxf$tY*Gxn`PDN1A&R8!y5NdhVAghMo>lQ+*f
zv2oJ}oE~SII$Rp#9M55{Ec|WBKs=8gy&6LEc1oyviH51qe!Qxb9#%RW96ZYOjRG#(
zuBn39m8K?)35M$hVSLN1f^{zx{UahWlJbKwJ#n4bUNn9#FJywu2)be;Y@Yj9hUx{b
zG?0Hh__>}VXLqbRWzNN%uPI@-w-;0CA>uDDb*x$2@R&$Dp}YH}N*O2eYZX#Lm*m(V
zhPvsKoMG?q7^N%Ut$z>r(MKHN`Fqpy8PY(fxR;ZY^X}r~_EeQ~dq+n@Z!bE)49v{T
zC=VX|(5!Moh>4B0T^%4RDlP`ng|x06CkIFL(dI;W>Ku_8;Lw#zL4@tU@)Fx6LzZS6
z*fma<F`2_t11;ji!<vFUXNr#1(Z8l5u8!wq#_*rv-qa5_%T}h*v*Ib{;@(TVg%zID
z71gO~zyFXho5OakUw@5jwAlWL_JMe;(e>pL9kj`eTsTrc;RYsi#l`K!M&E_J>@>bS
zV)V~jnb0Zw+Lkm?#}MC7e+%6(y=N*V$~3eL&flHf`LxO3Hi{_Lu1I-udf9$psp_`+
z7{Ze^QH>0wx`d=;gLYHcwS3;OA9HeEOtcyqFz+8AO^^RdY#U;o-s(#_#cYZ4+Q}!r
z8&AEF%yAc<gQp<6NtqD~?5nTYZ=Z}!L!ah~=V*62EPR3LNzm40*fZg~H48r_j=Vq{
z@RZqaVZ-9gVy1JUk-{Y+LC#>}U{b@Y)+y1*oP8dLK-gXxhuxDYb*M7ip2Mo=+BzQF
zHxMIAWS;tuq=WfuHzjNd=MDP8+|o=c_vN1S%1A7A#I5-XCl;RDKJ7Gf_*AzbpF03e
zr71=4ce=<Tr?ZR_i*HwR!Eq5^;yU2Kq-NxHqZ6n_Q&p(XB8IjMm9j2hZ1N=3I9821
z*>fx;ry^~t^C@2mVreN_J{$WT`QyOBYob^?`1SwloM^$6Asb&)KnnoR+}+)+pO}aT
z0uC|V68as$%Yk~s>3PN)jLX(AGJ<1iX$f3z)6oKApgAIJOjd+LU&oYLvXm-PJdVx0
z<zRL28Lx|t$mdk{(|$xO0-?0;uihAmk~3>+7W%zLg1!s1;)%5Lu6VqCt^GezOHZo|
znS4wNV;g(sPbz}6L%Q?ez6jyC991&fT5m!>Nm!KbxVx>Loqd}-H?^d)a`|X4{oh4S
z=*wY7Uo?a_ks_?As}AI3p_P65^kaDmBQBzAwZLY;Zm;t;n$<+yVzP&y|9e6IOQXdF
z9TpSLfNI_bfrM{Ld@}Eb67aL=)ctFd=w;P8h-toKGrY8O<Gy*-5PKX1#;>b=)8CQV
zq@)=8`iZ&u5@{ay`;u5&c6Y0uuC^Ky%4WZ%j;->-L@l!*<B$m{i^8VMR&E~dxQ)uE
zImcZ5ivX{8(T@^Zi*C#WH#r3aiMm~mHy=n+OjsOonckd^5M#g5`ado}f`496IlkGf
zj@gGK1#ty8?&udODB^A2TIyNNDb6kXvUU7e&z$_(6FA!`>Mq~Nxp5<Gl_}a3X9{5<
zddk_^p@_eplT_gxv*Q9N!GEf1Z?FneLgVDY9*$Bry8|ivCuCk1=bUjvz6LqD%wi8Z
zsO*20FXhYS*Hiev2~s;Dp~#y-_^lxmee{s6fP+a?nVO85E)n%_81HLw<lpT*QI0J9
zsTJ9QYP*-pNg#P}Vn0JZdp#uH-Oz%W@wCQ-338=9WsYGS@$-Bv4y*mY{qof_Rbcfm
za9XiU#P9lH|L5)yGsof)PwG$*_ErnoY^@tfusPS?^K%BX*7M5tt`4Eo17sgTsygS7
zBif=yKpy8N`?H9n<vGwp#_%I9EKE>T6d9qVrDY{ccH~+_^_LCtP1~<EgaC^m;5iC|
zUyC>5=4c@z5fKr>6$q*@FiF+a)M&M9@j*3aHC^2f0BfV)W0Jo9{=YZe9pTRxtAFA-
zU~JoseTWu51dPv~?kqno&9N=VTFO|K{qI-2mTYVGE@c5*Gq|ih_$^R+6lq!`Oa%Ac
zTwGyfB3Y-?wjuB%CjI?pr8O?<#(@iVuy>@_@<ouE!p39kF62`p<!T{jKw1nkWF8;-
z*|Pgt7ZSMWN^f~#;j*r{cHq3+uyR^2tqK@c>g_i*%oJ{%&LJ&w&Ljy7Od$)~>Hb;%
z4468o*JB-ds5o4h*gdfY-qBwRm^k!x)|DiQnEilkWjNit-c}Ju5T?#v;IM3keN`L`
z{4_|4T=Np4O-oNmm~&-u-$?Wm)}zcdQ3sbmLaij)=`x>gir!fq0v~OMuDbE3tLyzE
z*<{J+1d@;sQy;o7Ezv<zF_nPdsaQNV-4uiTCu%YzU`|%@K?plLSy?#~$NbJk1^(gT
ze8y%Kr~N)Q528=9`3Tgy^HokpLdSV&(;lv4($hP40gWFv55h_NTueeln8rAEIq8*m
z?Jcla;W%$*_&zoB-W4(W4LK{z4{7{=_f|mKu5~pu>S#SS=H$+ZCVG0VwV&z%6(i^=
z(`^J9ncjxvWC3q29}uOy9UF^_-ydA?zu;UYu;N2m{2Tss?rS&`2Ybq=QZ}oe(k7F2
zj*nrrge;%YjFt-O?})s@c=A}#LFmONH#ddSlHaM)o6Hn~M6E5i-?lbXPwoyEGA^Gq
zGB)VX|5hxuR2#}~-Qv;?Pm1)_e#;oPd(XhYSYO{tG;wX1&1o<tJs0hU3#l#O5sF8Q
z?t+geAtX*6MNZB+uJB!!E`!62skG<>5<33XndqM%@02BQOT2aAd^R>VR8pD-q&ASF
zX!jFG`NNRZnP7tMNv2d30Hb_<n&^G~%E-j{4N@n_kR%8S&$UDK=MjGD>nd%Xtf|D<
z^ROrRjh@J1R)6tkDUgsyuVg)Z&c#Pi_f01S1EamYv$3ZEU)kLqXqM4Dc{n2S$%2Z(
zL5xHF#Ju!*ZOe_Qy|SlAB(CGF0Hs4`@qb3hcfMl`xO_N<D1N=I;rke~H)O}tNnN<{
zI@Hf~Z^2(w4-;D7ur2%GDZ}|ybf6R>dt0TZXr`R4w&nG8U@s#p-b`map+d`xXpNTN
z;Ae&yc<FTz2%DCq$;JK2l!6|x_DQW4hZ8n7>~fqj8Jw2EKmDTV@<-h1k-o29e1{p`
zhxi+vQ_6dT$6>Krr!!(ki!Ma5xVQtgZ|3b<cbLhrYMmDAqPh+o{9j9v$k2OlwHVvE
z)f;P`9?S2jD!*O(Ur*?#25Zvnu&dFX*W%*$)6>(*%4EEk0#l@!QpG@w`TEvP0?d+=
ze5Do8*+%QE%_Os)qsT;aO(`nq(hqCTIXB<x)AsY?arGx;h2o+V&v{Xpbj~1;w)8<Z
zp%B-F2<t$$IcH%z=$cHK@|VXkVf5|WegQJimFZhS?1~%`;$UO`Aq*F_uI2$!vS8t}
zSOqp_xhkT?L4E)g0}_VyP7m;#4bA3?>Fdzf<!VR%Ga^L6y2;ndr`-a9adfM|{n!ea
zftq^fm*>}YFSgaKYZOm7_hT7S^YW;B-2a)~%5jd3-JT`FRNtz-(BI2=IZ{IBsaEk^
z;Ysqjphu|_0mPQKljD`8*@}is1;o`;-)qDJm9VQrty-RF624UIFg~++`<)l@rJ4~o
zkZ!txIpq&Mu%cj~^9oOtFRq6*+WwSF(_k_iFV5GTw~P3S5+w6nj6TDQDOo_eOh#Zm
zDPkMHJyP7NxrKOmAVrjn-jf($Xz-#X9|SVv-a@<Yy=hQleFjKauFk8!bZu*<t~Z8B
z|NiwgXdP%gE&y)<>cT|X@IN+i174Ob9Yt15tR2;hfYBhy9JGVsH419FMDwtIIIQ0U
zrNn|;C{g0RP(-tMHdA|zmEs_%lL)&FW%+-$#Df0Vo9=dEPGWx_V)WfmKLg;tsMKCR
zpQt@c8S*_IXE*yOsUDyDJrpP-!rI-wT~;kkdz4&A9Ls8>G12@OS8B7K$UU)QuRM6=
z{69K99A5qZ3^Qm{0K;ry$4BhK^MLa@fxD4mTgJ$voZhVY_yur)>xpg@>0T%#cs~XF
zbbALyaI4(eY4%P-wydCEga6<jnez9SV!lnDvh-rFpbvN>cVzsd3|?RVE5-_8zgo`J
z@<A@X2y3#iv@`%BInfaOcDT0+%6pTqj!$;=*)7MLAkhMD2yj0D*rL6-2CNI4e`_Q^
zIlb|Qt~~t~dX32CmW@!-?=~6|Vb=_+<$7Uj9tqxiU^O91t3<BA&E@KPL);Sc@L(Dl
zG?X^CZeh>UHu$<7|EHVlnDYaqnzn~p&$|T@8gB5w47XCfa}KZ>iY~|D>Cv1;uZx#F
zbWo=0LBh$u#rQ)rUT%W0a;@>V^>SEZyaYg`9y4WRrGY97o0`X>q=v@DCjkrsn!EY~
z1BSdFdmk(GMKiHd(;eY}MEc*m`$4z@3r|egJe&DAEH1o}IT;V(f!SbI8@p0Exx=2#
zGJsg-7u$^ex!=`x>3=EMgfzWFy=hY|ekJ02+fN)3TVfv?@KDf2IMbP??hYhCuz;5|
zs#U=6zfo0WXVU<7=dd=ks@*;XMBQTR8U7NeXB)e_gLBo(i$G=$m2cK6Bc#J0IYxuN
z`!MzmFTzLE@DHNjKf-AsQMH|C7d>->yLJcZ6(!0d%i^&B*WA1W0jfQH5N-qXUC2Ua
zEi_YcGiC^z3eFt;#b1A$;jl-=L8R&7NSzjyc%ggMY<{POB8sbpPmp+{lITOCFW2_r
zma5)FRcOQ0qs!YxppUH8lmLE<tbtwmyVyK>`-zg--?OzbtE<f3`$~TZ`LvdvYTxnH
ze8%z_UGQGD^9iY{>mIXZ{vCb+$M?U2B*8gSe(u2su2eAu#`8%-LUJ6<o)l6jVeDvG
zgW}5AgEj{;yC(h@Wf(rUU{)glhXbF_Ednqzz&d3us>r3{Y*zY+dlI;OuDsJd99IW?
z1$~g2O!~fgll=olzE^M1(w=x-JO#EB@+SB1%~6-g^21Pn!?rCSsEh#FtX9#H`somI
zs<qsU_!%hg#@Dm_oSDB|Vw1`uBByxzE7B2xCElgB@9e+1J~qMP+4fS--fpGNUON`U
zZ<$bx(dwEBA<A=y5V^9lxeVCcL6^4ofQ#nQpYlweZ%Wp&%Y*0hn#I;28lc^uJH05F
z`2`oBYX4~Gs~XI)BFGcryJ0U-Q@BR{ZUA_3(HqZPULITWsj}XAx}({?V6qmefnmRQ
zEB}kZ0fmrV6_cmo=*cf7Ui>XbpAkzpdiL<<R)D?L?5-Ot5B3L;UqVGhOh)DlOkmri
z4b`3`K8cqv5!&0^HR`+)^E9iXOoxqbeyH!`vHP)sP&s|m592<kM)p6`)q&M9NIwzw
zVA9y>O!{TZW|N`89)JjvYyF#KqO>S!Pz-fxL<&flnYNaV#y^cVb%zhNMwrM53<TE|
z6kx_dV~oX(w5BE=^J|P`priWev!{}~+b`OnjP_^_wNiHqPm1q^Drse6psLApgWLP@
z%()<IVWE+dL_RYC`b+oExE_5a_kcq0QEqD=TkieX!O>xKx0^N#{5x3~W2EDqZD`6C
zn)^})2A<VyqZIdQipr!n9_82l_;i1f;0B9*XeIRvFZ$i_j5_Bde6=~Qj<#<7>E`T%
zvD#>$xDmlH8+-Xfahrc3AM)_e4nSWo);^GSaC9szDr%^&zjgQSUFbaY(dqsWH7wU_
zzOPnhp}H|=n}6Hz^;?n8Z!;Gbmw*L}o!UHOW9N(SD|G8!*)KvZcR0vfT1l_J*j#)P
zIQJ1Ad;Ij3oD(~=Qz+jrY~Vu;QE+v{$Is~&a9F+_mJPFJ4^V1}I7BDEc_2b3VPWs0
zKT=JO?uVQyl9)nKJi5@SQzJQ`S9|iUWoDYG$lTm8GuD=!79Uk#f1I7V)+sv*=E^A1
zy?!mVIGKpxbO-r{861&-Uy|(giQK@@mWDlK_{!M@<~}%0p^HjL1ZwlUL}wz3Ya8pH
z1q=g2a!)<F{HM98*3$f~or2s3YcGxpOzf73TbF%G`@GaiTIf*IyjCC-(_qy{^aMq9
zV}*M8uIrPQ4!rW+P=6BT@84!iM1G{(!o5c}a^bSQ^X5WdFVoDl{`=`n|KzYp+U3Fs
zfwj~t4PKlrq$0Liela567jzsrnXhhR_;t25UBAJnxv<{d`Wudq@=!!fJw7totKpR`
zN%m_XGw&Rve+vnbMiL`YD4DG01%nB>-KpKubj56G|K8=LrLXbvs0+(4OKAGLKBMB$
zKYIMQ{tK}%HFpHFE7m@CnkL1EofL&68T^~A?$QpO(ON>T1lSuV9>ciwQ)oIp6`fsF
z>7Jd)hboL-w9|(g+GYtKzGYo1=iO%fo#s@5eT}yw7mpO3S9rarSWC!M-BiXsK2Cc+
ztLf&#HXk_bZ9gwa$&u8AJy9EPEDsOz6OaHTf=Nht;=@O*by8?j<)zM<37fA~SJr_E
z==ksx@4`m+1l$kuRSZ&Duu8GX*o7X~g0(XT=;t2@p`H>9vostZ_ZX;SfSM*KGV@LB
zQEGN-V@D>@;6xQy@t=$C(t=ivKsPyy^IE!^SRSJ%c}Hv=|M3n$FsA4C&&YA7#eD1=
zp|Q7GQ}^BPfTl*S*Vn&lSsYo+H8b>gw={NRJ5JG+5@Rw`{yB8J0RTVE;XIAYq4KgO
z74mtlQT+(<H1QG75hJP9$sxq#-{&`gEWo$`V!sko$Kydo<<(s4>zoLz9L~<h7lZM(
za5b0YC?TqXO4waFyQBWg=Q|(yZpzd)nOd*OeDrOH?T+^KEBs2g;Jf|(U<jaNWb)dP
zhKBnW$9MJObVKP4eu)=heHkM5ZtTYk;=lQhv}-#O#dHAzPRUO7ftv@O&J8k4SGA`p
z@E>~D&`I{+c!`%@d#R9q0n_XDa~>YN*+_aS(rVWfOpIFvY|NC(0vNwr`rbhkeW0j_
z@v2wvS9AZ@*E<fGff*80cAIrC9a4-!mdE&$csGlSnEHm(jO}IMO?Z4i4>L3EPx0b|
zTIyKnnB>1;QO(OhTTc_+$pzh|^W*K--ri8)!bXtMYRNyRTp7+Td~x%AYi(UU#MH#^
zyEZJcV&Zj8_2lwQ51ZTDI+mnINeEk;=5%1L>0S2&1LolEWkb-nl4cm+@W2%$P<bR!
zBQd|^SAE!zi=<$1)>d_OeR+j47?6???#(-f%awG&?_ijsmGwW`f}1Pxy=izOgL#@f
zDdMWG|ERjtth-d17p}9<-RZSjj<4#h&D3IJ`nrp$z-IJ^=Dg_QxC~5?Afi7{^ilTo
zlYzdDJ;v|xuAo}Fe<f+YWYMwHim(zsZ`W0)vg;Gf{ru%MrF>C_?Ip>Lel!Z6iTeQs
zbBd)U2(!9&_oXayhPNmH{dI6dRw5^FqqYfqT4fxPQBBCVnVOYmSaE)1jst%_cD$OO
z84V8)2dO9pA79OmF(DzLi;Ih&X&x+D1D?cpg(8Di`X$SY&k;cn5P+JRX=?TXFtmQt
zXOn?4IsiE52zY&k)b!%2|J;m>#u%&>8@$WHXnKhXO=!bqM+WeLw}eqt1lc|*IhB9i
zLwuvn{`oFEo;WJRGt&bXHYxy?sOqkpTf#H$8zT;vF{Z{9KDP7yx--OjyO9KCvdkyh
zd2?w>i!X-5>A_iLzUZ7<_)RIj)+ok>jVlU5fuXHshf)2#Cjs}9TeeNiVP_(Zr0V;1
zw>6EwybK!`6DGRVmr|=jD*NwB7IdjG;jKTPu&%=*cTP-5_+|6=c3GMA)@$6G6=`eh
zF*;~y-l(*B40q^B3tG&=pD&*jb6m=v3+pMdLqOQMY|h3%u9DtI7toy;(H<z16^=Yg
zeM`A=)~uXRy@mb~#4P-Wb{UGnW-BJ9SY@42om_`2OadB5i$X4>gI8+JY46-E64~U#
zKj%J%9OmHiO#Q=)7lnNz^DGSLpN6vo6T|-2jr)?56>lja)SXPuMr*oPxKO;)w6AUG
zX-N6-BLC?WUp&;#1tg4q+v8=>ZmWr3r?H+y{uW&xP2Nbu0#s-a=Uh0>KFYS9))n1P
zH#@XvV~&ewvK=ZaZUoI7n%IA{mY<_wNn{_88y16~P(v$l#BthhX0!ecPJc`NtaydO
zN%NPVPv*U7QS`-f`c2b=<OTY76-o5aq_eYcK$*6;mJ~9>tug9eTJm`tdBnCevCZW=
zMdWE|33UFBt9nn9?YHyVFn&ujGa^qq1fPiioj{f2OR=ox9($aaUkf;B1f@m*N~f$Z
z7o!CD{ZOY9+mgm_R**SfhDl*;AlFTIw705Z`C#N!b~1^cvYa#a!`M*q<pNelK<qO(
zIHXiG@vb;|KZjdm3v7T6AUD+!@O^XKjI*KW{FgV;4`|%g&Pgb!xvk<|h$RWt2gxtr
zI<eS6AZ7U0^Pl@@q_U71wK+In?@XSu#$Mb~{2cVHC+1$sxdXCxzGkTQTj)STrk=ig
zXZ!N}7?p^_;^(;EztI&gnsj4@fr-lu&}?lu4f8QoXwDk4gefcnjuwqMQAyC<8T3BV
z0793<<WleD-7)u7X7FNa6h3F_OLNLR4?{@B`=f7stp#TFfAIM`^7e5Q`ATraGZkQ#
z0y=L<u>Lm*%d+Eb|JeI!NdP+9TBvxdDdh)vw;}Rh9EL_evRvGQo*wueEh}q#m))6U
z@bod}6_@_IS@ZeUiPhv!aoMF5tHU+PwXwvA4SeKn?w6ugNgWf^?_+@*4!Hr4h#{9>
zZAPNL36DZ)Ut<#+%kkL%vf9yZUmURT^lHyS#DNBw>m#i0(+TGjuYGJ@fs>DUEOm0O
zZ+)LDh=OuW2?*Fbg$mTe4#zh)z)lCW-*kch0<+;PV5)o9p>UnW>P_>K*8`)mXv#g%
zGUF{}Oraebf`FrmkMeL)chaiDVA5D_jqL4hy7OGgiCW6Lc^dP8f~~2=?JJjuyzXRN
zlLOU7RP9tD)B#h_Cq;Sa`aVpFe>gX1l+hj+@;iWLi@KJP#j(n&aGyy{1W)*Pwi)P!
zT2)^vo#PN@XJOZ{JOsf9+|fcA-8_D;rDz=Eb9lMJ(Gzb#Gw7q-q_QQ=&hr1V5)xSa
z)b1fu4`|D)j&TK9%y#76szJHdxow(UBhwJ!y!(tN;pX7=4=tFwmQdyJIHjZrd*^}_
zA1PK+R6$_~kD#&h8*CZi$D<Nuq^Fxzy8nA+!wbiqM<6#1GF8w{3#+;&yM<k5?F;&(
z-Lk}mB`iL>-B8cFz|H+p8%d^*H=<1QCp3<&9$KcOKNiH90-Qm%h0tX=cpEXXh~b%z
z;kRBKy0}|Ajb#M5!JCSGjh*4c$x?#q)Q02j_9d!y!9ZVWcdgvfx<zIKNvHqkkU^_?
zr<B0~&@RzM)pBIYMKbq=w~z8I(^P9{?gWQ?@9#kC;`8=xrRwz_k?iTw1?(C(fBg4M
z`Ier(Rm;`3(Y(g+*5ciNBZC_xjzN&kF3&&jW`i<4ys<+h@BwRzXCOC<L<#u}Iu%b>
zOa_0Cf9<X2{l%qwXQtFrFiE557g6-i@_S({Pn7)N$w>`>fM!mfY9R+R6dg^wZx`y+
z-$nohcf&V9AJ&?i;#|;$4AAXo@7`FgMbl>K>q<k_b|YU5bVdx&u-V*BZ3P3wAA+%>
zmHZOQ)P@eJZZ{u5;E3p(b~d(ItA-ZSn>QMjc<&O!h(V2$HP!nQ`S9do2<f5bAu$C5
zdTd!s-I0PB2~Vv^qM0Hnj+FEGJYLrkoI~mK)7tXQ|4<R>*Do$sO;lKfg=+~|zB;=f
z{#0n6x~<sN#aX!GUaN7*QMWzOTOFtz>%2?tNY`EXAMydcCJ4C;)K3s@x;lF@;KSlL
zW?k?vY1vt85AS>iIyn&K_Jd~S&(YckD=HQjh#$$Wu6;iPur8f9mgi@p<Y(%Wl+cFo
zIEhiP1m{Uy%!hu~Z%tL*is7_<QCL_QH|=|qbtt&Sj&Xq+sR77}pUi@$J(X}M#1Mc8
zA-}Ny{_kT1AfCUiz2uz&!HHnS8C8Cu^X<1VBs*{Qoe~#DuqZQunqt1VxF~YDFp(Mk
zZ)LyeY}8usi-xmafz&Ev7Q6SIk7YN|+<vc${r#Ac{#R=v*19!_)MSc!dVQ)`j@%g)
z5DVI*i-*oiSM3y+=W_<e5;`6)T7?=%iSPrg9Uh1xBSXWGkHRaa_r0tAKOY5XFFjum
zNt_*+LWmnxDXuhLmJ6@yUDFtRpGGk8l_Gk--?~$lQjyAzEMj4yJsc{rjg2PBEF39l
zYeII%%f}bXVND2vBk=Fr2Rk&3M~_~D?TIYZx>q-InLmyQDhR=5N+*xt&|ESx`#NLn
z;SttrytwfE7QSR~XDJZ{i#!6PpB=+8t|I(=Js@>m-0IvlC4$cU;?}w%+3<iKkgzur
z&(U^y${!sZQ`28n$Eed>J@;D1&bz9NVEBYqKiT#(MOZO5JdsX7I_~9c9M&tVv3D~>
zoxiFr?xAz!k-1HfUnO0chNVvZ!$@5`K>!ajkUPJ76UV>jw`}C@O5uVu;}evXW1fE(
zqv#}LWW>nBb@lXww3yPy5JHFqLL|xcFHUwv_`T1MEa8msB_t&Djf`YKN2M#4<@3jn
zb9Lxdm6b2T+h0pdFZHL2d4P8rCl?oScsp#7I{HEldgIP0I)}}1EL2p~2WV)fLq8u6
zX2?*0XsgXjBHJ6lVgfKcAkom!fY)tpZ}$gWD;Xdy>@lrVfuk)R>a1+@*&XZqH9i4;
zAN;V`R0@ti6_OtP%sg8fCdHKsid%^`sw`}-7S(6%Z+fM_O?QNRThIS~0OkFFcPJkm
zVq(uP1^r)fQ$}6pG^@=|+8=QpGp}{QoN?JFW!5v&*H@Q?<>jp{EuW@q+>4>v_JGUT
zCQ?aNFKWS!&jQOUKP5ZcymlOWb#+xThVgsTCD!SuS180yn9u1*5E1t8Q=)!R_hDzF
zSgL#W1W7*_Ik?he1_0Y2{l4hUZds9bJ{fzGRJ+!Wnk!$;s*GU&u*X)z1h&~*Dr54Q
zkEkO9%*x6Xic51hGEs;>K1%tTh*HsXI66_1ubMg(|8QCE)^@3Bds{b;s`tWEA0DTl
zqi?Y58yY^`LPWkX{s5iIDk$vVZ-O9BMo9@9dSKnh)hU#cvCO_8R9STAr=;`)w~*Sa
zGjk6QkDa}}N``dl3=$fez(^WpX0QqIXgzS_bE`?B*Q}hKpAX+21=0kk^A<i>gOJnH
zKLkexk5<%dZ*#_{F2_^Oss%cNwaakncHH<afM%h;zn_!)YCnmMg(V!;&wi!vyN4&V
zP8yT=+~K#)K?67@K7NKnOSc2J_}9APGinc{{6|74#J~G{9==<Nh_Gu7c=HD7TV;*r
z`_|wRT;>BejDdX$Ci!9q3skiEDbLT(ifnnNl=6&MS~ex8y+779G`tyE^xhBS3CS$9
z&0Nahc)G(y)CE1Xd0&Rw@9PWbE7$!GTD*=0l(qz3dr)y`c2_yIV>xf)Z<5}k{0P5#
zmvq8@@5#%GBO8Pj#R+(1<P6h|2S*duhy^x(8*Ycr)VN1OLt4ksP$XL^caK25K<5@P
z3K|<5eZcnsq#>W`b)GXZg@J>`m)O`hAS8pGPwxP;gIZZY*^!cx!ua&TnBib`%~+K)
z0~k_3LGuW7U`9sd$Ug564jjO{Hx_KKGCMrLoFxzm6<f!`q6>7AuK~}|VT9LgPm_p_
zVYkFx^QO~!Cg8<;;~D}Ix0BP;u*gVX>5O?edA}+u9zT8B7RXSmGRLLxs;ePS+3@EM
zR`qqy5-MyY8Irr^G+8R@m>ABVc~+l|#>Pc(MQIxx(McVGfq)Awjd{x4{Jme#q_(;9
zei*H)*z7&nzmWMF6Q-(@>`QUoQr{JXKJK~;$$N|m>xqac;x(yCp#Iw@e+ao7GU_ht
zBZaK0)XLn^Th!>R54JlG!Wbe<(hT{T0~-F13qTUBhHx?yMBmSQKNZ1?i)n4dfnJ#w
z-WogslR&=;q?hy54;KUUO)WkKmSRP3lYWVa`V<+fDk7QxirZo3U1lZ~IHeel6&b3i
zs4Vc*dY(HWfYV;0$GyG<;EYSfef04`1oB4A6pJyKOSS|(^U=y*&MlwT8_rTZ&b{g$
z24@kVggI#!9X0u5B7kr-Ml=|Ak<S{UTdHVq42v<ejKGL2Y;`aL@zJA4?=musr<q7d
zNWhV*m1D*uJYPmq@&VxO_R|_>)B6=|gM;B<QUzMX7I@iT9;eKp=Av^V#Clv@aI>+Y
zs@t}H1fS0@1Ef7EGUg0kGtTcUzw1_ym&ooW-v9D_q<C{r5#g+)_S%X1T@qc~?s+Yu
z1<JkVsh&^Y@tkd#=m~?xn}45e9ptZzrcytIhxRBsO8Tt@zPfOJ!FmVVw17>7fDe6<
zm5n$1$B(xe3xuommLnsgTC!()bSG@R5yrWAn0^UrRwv+R8d*u(DUmC0kru%Z!gm+x
z>?<h|@phd-9w3Yv`a-|_r!tNhk9vrh=;PRkpkl=_vqogmW}8bZ47#=9=2h+Tz<W0j
z%a20BL++?U!JTE$uC<7R#h2%i^YNCj)Q(xH8u*{KR=e)s7y}4@deH9d0S6zU`e@wz
zG)cf^=Q-?lbpDg)+IGiU@a4ZZk7oT(t}nTZaaWA2P7$+ncnL(A0T9*6C6CkqMeFX%
zI<Lz;GUvdr>LX5IJ9~;zSh2CW84DN5jOQ@{kMouRxapvi8R?E>Gr#snhW9sj%eA|T
z0zvy6`HO0cnxsH7myCnm{ob;@6gOfTa{)J3QV6k;%M1fv&A9X}&4W{=*csa1{=<mp
zdv@WXB{^Eb+-2+~c6J6sEQA*Ld-%6f#Rk$v9i>~F29`zy#;|ceiRK!up$gx}r%<+2
z<DMe-{#`bIHB?xtJF@d0-@}v=vpTN7XY8Ic4<0W-lQuq;v_LgO)z_YnBoFqJlKioJ
zckq;E#~^HXm>MhpVB@`@>`4d5H^~vS*?oPl-vV`a|2q*2MZKy|9wjSXq50VBbE-sp
z56%p9O!3lj#Ky0KrITwK;#Ex?tF!MW$YB}T+39yEINpG`RAXfnw1zZ0r(h4MZ(z^_
zZH2`0TVo(j>ntf8%ak8WW4KDJrWfBzH}^a{pem8-O&Z%il}VyKD;5=Lq1|v(M1He#
zAKn7`Yvo?>5GlAdg0aoKsk18a6lo@8Ih>^R)`{6cwr`oevt5dQc*_cwZ`HHdAj|29
z+{UB~FJd-#d$wFPUp|t%ECZ>xY`LuU?v_}qFW{+)$UfzBCp6Cc2#e#dpI`fuV?<3P
zWZp!#XP}lZb|RHUm79;G?kwsH9!J0i%c1Z0QI(kDi|yx5S`|tL6CY^#?2~GgNq%mw
zw0WN&&^vnnw@NOy-D9tgJV4ttM`~z<k>rBIBhmS}(v3qI4z!J(T?2R%tCw3X^3{P+
z4Fdnktk3z`D>36}GM1+gT=C*B&ncYu^HbPwvCtbZz}4oS2NxMlesu#}R!`Y2o&vrG
zv4#;=%445a+QNeNWi8rJ047fKB~Pwgn)r81I}VpB&+{YUK-JO1tOw|1_3<DSgvGvx
zib?`DTDlLY;KV3!i^zg%<s}j=_aTS&*`03@1=f9=oX9y~IoFd{e6QK!&!5p@q<6Jj
zrLi-`sA`VwSAiV;CMpBv6p3brBR>yq9O;TKjb`R8_I8_xTjCCV&ULNT_S9!v$`2N5
z{?Pr3kL%z;^`<V$^(n;)B(Dv=U0t@xaHJVXxyzaD!@y)Zqg#TD)|oEZ(A4&`$<z1d
zwr!ekuM)9GDAT6bo;K|rMk#uLTwSB4&{QHWl_tmJxTb*cCtn-gz&o!^vtPZhL%Aw}
z=kms!Ern9#>f`6nI9yJ;pjcEb)I%~gHBI36j01<o0btJ?^(4FjMxA)Vw!KYTURXl>
z3Z4JtD|0Kkvgh~i-8-EV2#N_6cSUx2cTY=1R;A91Pb!YJesc1L%mm(7(Dmyb*#ar7
zG3dM{<Jr*?rH_H83Tw#gpzd+W>3xXnxM9J>F|d~wD&WImK|;DE{oNB%1TW504$Ui-
z*2AhdB_Muz7aEmMSt7Hr$qCeXpNg<3k(#Ph`b9i6lG3>5-lLb<O8b{B-IG)X#+o9s
zCkrDpw_;%79~!FkGp5R`Pg+=xI<K*c^&~MAZSLeH#_q0l!tUC;sLb{7Xo~(A6eKj!
z`r`-lRlZNq_!W+u2-x3Gr|+{!iSvp5w=GXhhYN1eg9?L}`Th(ZN9U8qZ<Jk!y|o11
zCd(mMTs3kx##{*CLU4N)><M0pi1b6uyRDp1W-*3wVvJdKw;l8dM%{5tR+F64$^6|P
zkWl;94?Rv7qsi80Qbe5QJRpN6BP<5%p5Fq8?x=RNzi%PLaae1aGSLhozS&97$QVqM
zzyTL9(YJ5khFQwX%b!9%bMu$WeyPhRCIt3?EY%gS3(7kN?(+Ga&PEtTWHb_K&Z=GZ
zwbq=S<+{nV?4y?M_Ats7r}z8vWBecWGIdOb*N?xl6vs^GVV!H^3lXT}h1}JiRg;u_
zKX1XiB%ChYx?+qgKQ)l@o4wh#y!fxL=;0f23H^*~|JxH6o@7C&k4H?GR3eS1%%e^5
z^@BfUS{BW`ThdBS@i-VjiIVX|C&`T(^P<SCJ5d(3ybd(<zFWw%fy+a$y_hk${Iu4T
z0?qp@So?}UG}xDGqV^=Bzs5b%Gz4}tg#F!ArOcCs$Y=x?<L$rIJb%UY+gI%q0%}EA
zj=3M0XkWB6q6dVBV}c(gDKBqAU*B7=%R^;^bjesb9d~PMYXqenm3qj43IDL(S^*PL
z1jyp3WzV&0ie^%+$4mSHI*Rxy1G?OI8~cG&vn;ZDdN1~ud+!!;SZC+NM6SLkbrB1C
zS`X#U`wi`bzm1<|aqO~9wvdoG9aj4OTkqYzXHR_j-e)lOo2+QIMgl@*b0)yBE3$kv
z8W(5t{DOP%A))?BdLnPR73%d=MQCzw0tGpj@3f=FE)m-Y8J<$eKKC_hxDht^Wo>M)
zCztR3ac0<Yd9!!~63T+#^Hp{&&)&`??o9?Az<2$8r+Eu0bl)1T80np&WS3H9^nn96
zrMudCliGKm{_?^TX17{Stob~O;f6G1M*h$(DX=qo&w?i3uIug1f(r45XswPVhDhPs
z(;u;?mTwblD(?k{IBQk3L?DXwo3YZmoaf5TU@7<8b+X<Reh>g|Uz3VmcmMel1@=?7
zXIEF-$4kxMh>LgdBqb&Fz}I8A?8(6hFD<E4S=+N2Ja{W?=S8Qdjv+7b^Fsm8LiL5o
z`*G_aTz&{gG<UQ>7cnz4^QDvL?9Pr&?a5rA2N^z8n=Wt;ZkbF~vVkdbvkv1o%KOB`
z#NcEUm6Jp3cDyA8x_b%M(7TU|nN4Z9XTf}CKS~um$YUW;e^<)xH(^EE%5&Zpft>Q~
z_ZiWf4;vaj`tjA*fdQ9vH7r^?v)^%erm^E?WZ`NY%Ps~_TESb4n#*^HmLLS%h_#@6
zd5k+xp_s83Cp%VO5uw3K6pK?yMM0s~$#9rA<2q8PrNVCC2>z!yJQ)=CM@gg3Gd?IM
z2V=O+F;^+rP@bI0dlGNs9&htlLH4WW)>Nfz+oJNuY%fnYR_jl(?Wx86?{6aXN0(o$
zW^CpHNz&*3+l6ebL0aBc3(;PDfs-Db!uiHpzp3~I);`~+^>V(w$&1TZ{RZcS^YF@7
zG6@8i#YkmFnv_@L5J+X<(gM!2A$VZ;mQ7SSTW2nVadCqfk*C7SE+~wnsFRQXekQ^I
zShQE*x~uQIFsem(JZDd-!>XK+u<-8jaqHmVe(g~(9!F2Ms)ii;3nRnKojcn_hMjwa
zz409ReUViav~+a)gvkQh4s`rCN743iy!@BCtkX;Dh7vQjL^G~sOADn0v!$_nGcgG<
z*G@FVkL7cbv+cI#)Yyy@Mn5*)4s9i$?jFhG&5Dh;g?zKn-ya;HmdCw|**T0MVn(WA
zO8vJdE3?MN?G~!I9whA|QPBGiTtr7!&&nnX9<C2`w1;Ia8T~L=MUi;DkY8YavFH3~
zWSd-9&!B1Gc=~VimL|>llXn{)WAGXGF@k_`NmEoKo~8$^km}tQJX&x=0|5ahK7J5b
zJ)_~WeiI8NeDI8!G+)cwlol*D;Y=6nHX_61hF7m%-2t<27gyKawd{OmFg^;B*j0U%
zt5gh&Htn%5g-x%~2DPt$KtKaHFG6&gc9};5<F)-mCN{R_k&$SKLN`YFP=G^NzuCD=
z`Ynv`I;^--nLX=*O*yn!NweK_FCUD>ZfW>#(q;@_vG@*be51HImnb9p`eUo!AHJ+}
zyv<Y?%3qb7VbA@kSn2>rhOWr`?{s$??QsHzJv(-BDdn%)JP4MLsi+KA4r3zb(b}dG
zY;)-DOV6q;lR9l<7xj+?|1h9$rLdu4!;;X~PoSkUQ5(Kxc}x3<d`SZ*UZ`hP_DfQ1
z`;2mfpPxT5A4?#vB7uqA1ezECd<3oEC^b+*XWcAl@3m7Y6c(mC9j^W2QS|=dw1o9<
z2Wt%36fE~mTF2Jeq85+j&G&wLboJmqRf*@!pWKtBLs{0m&9fbH)!R@7&^RVTMau*Z
zztw2t;xAZPOWNjv+<34)l0Q^xvfcd{`v-tRr`Q45V!28zCVkIg%7B8dJ*7m~+`mHi
z&NjFg?h(ev#6)CSi$3PjzbSgr1L}2V=e50!+j}%8--y@N)!!~CFPjVY@;u!bjt;2d
zx#=?y;FiYGIAiqF+>f&(U8A3A*Q!|8D4mVW>ru>lWJ?p(>5O5b(fqw@{^L@R{Aj><
zoMN&5=N5{-?ow=oOjT~|Wd6>qZ-EJ+!BzR8iJA92r>0^C%E(}3!mZbXtG~h`f=Vm6
zltd<tlvFPmGz_-Rq%N>rzD67zoQITyE*0)Drb=gqpb|rG8Z57MZ2Lb}I)hBwDch43
z+rO&9F-a;N|4tDj7FI2XP~3!N6j@$*ef{*{r!A1Z7ER40CtgvY1XsW5O5{<Ajh$c6
zky$p5V=VCGNy?m88XLq)(SSM>%3%oy2i7l|_*~AuR=Px=SXi!A#}?MkOg!aZz=>^l
zKG1m7>&@%)ql6V<LsDv0Ase|zY3yEU+Yf^fjZo4nD~y?oicjGEQ+S!(YoaZwNw1$9
zbrj6U>9)m7VI?~FXDqYJe$sIf7h#GwnQF&{qt}>|(Br+Pa$Y1&eLRPB@%g%pVd#AO
zMk(GXmQ!$HcYG1t?ABZSFQGU?*mAn5xPHV{aJ8$dYE*>#Z3oBaJg96-t|tyM{qs>3
zT>mLAkXN}0YSabP@V=9!|JYf&AX0HAbx&1c$&J#oq8}<NY*EN2<K{w|X`__>s4@4d
zjch4*rp<a=CLP9mrs2X?^{CjBF}kXVl1zM9DlQU_J8Zh56PlZOuUvkR(|w(NIVI)9
zY}gFjYOE!Jqm{9vYT1X=@@B>TDB^O{kQk0vdOG<km5&rYy6C`?P5_t=l*Ky+(Hb^(
z)MtpBH8fs4aHI4O{_RQRslGZ}rwsUzmPQ744(`68H(_zN#G#eGg+b+`pe*dC_q~vJ
z!9{W$`K0gJgN<st;Adi?m>xVydEqn58O{{UmCpS<%@5c5l2Vs@il6UD#$r}er08)H
zJ`?GtRy0wq)2g3$u;?A}&eMfNM{>tZDV-V3nT18wi~VV&c=P_|#FNqVKx26eI@%K&
z+Jw-D=gcG5r1u#MYI%s9Or*lI6i<%17n_14y-v@0i!X1Vt{!mpG&Si(u++VW*|Kg)
z=;9~2>Ic_5^7OAXT44ZMD0hj`4>94c;JrFUIfS}4N`9WD*6%xb4)M(YF0~Z(XpG=+
zh!ug_y?X1m0$n9))7I<Fck+#p{+K<Kcx@1+{hS6kmUZ^a_E&3cMxmaX9qj+2ItuX~
z)fM5O94_c<rso0H#F@B)`3fEGykO;hKM`t<8#mHK>CamEV)-M=CLXYwq4_@E$=woL
zwPSP7E<}UBd25#6UT=oymz1x3w{ApqXU0g1@#&N4lPzz2)ce$SK_llb^CrffpT2bB
z->;Kh<1$)1gzn*;ch~N=oNGcD?%nH^`xZ0|FSf{;TCWW<q0F6&5wJBi4chpFbSzYx
zz{bR^*MTa`M0wx-LPjrTxL~egI)`vHIvt7B`Id!bkZZmp>TR*Q*MV!pJ9W>x3)jPj
zZ0+=3jr17CnTtq>OU$2!3d{>NIlBFE`a9&UHRsq%eoW?kOW`GtF@+NvQPaj&V&Cdv
zP#j-qqj4`5Pe=-7HP`jsGQ#83kX*Fde6;2Lp3VD`TeojI4ZFpBqq_|^2Q{Y7mZf;a
zl5%_Xre=qjo5Tn=2LagcLF`13OyCS_YT8Krg-YEI=2V@~nn5TiE>4A;4U0P({=s<i
zqzN!V8rD46=B@EPU`R!g3{Ab3(j5b@Uiznz-}3SbMH_E#TS0~Y#3!Zix>1Jd=>y-g
zaz?hr38w2?v(33xzn*w|Ck{6}ed6oyj{ew&CC+HeF+h`!*SyP{k%fJ!Q~%1M)*-EF
z9^t}?J_hN%aRI0h_GR;*j5+CiF8irSt?)Y;@oY~h^^6R+ajfHOo#=yxRB<PQ=|5;|
z(b-;>B5~C&RDvT7t8;SUQo_lNJtZf6cNBiZzM|dP3Xsk>Zi;4kBMHpz-tF^;JFQoU
zoT_RC*UiMPFu7w8Vl<D|Q@>25PUr_WUN<yj9Af^f7le2TZsMY@Zap;j-f!%oK<jR?
ze8vkJ;WYWd`S}c(<dSX5Uv;6Gk_NpEO5jCyB?}}&Utl0xxt}NiifCTVbXqNOaX0R6
zf^6n&_g1+>#as674(X=D)T9|{`}Mt@ffm<xp<C5_vs`n71FILU+)hhF6?EbnYPgN;
zo3kUu^$Mkhy7NJyfl{B`epT*tM!cc?9jNXE8<4c}#n;|1H9{^evI5hir9`FT%~{vy
z@5VzyM2=!lm$@GVG%T>wt<GyzjoMd4M;0=f-#u)6oof`IEWpGSU#XZryMy9&Y0lu3
zPM(9iW8sf+dF(b-e2;{84KqHbZPF1B)iN56Dn*@9wEx*<#Wbv0gjoXia5htwJH-<b
z-EoGx6iV^TvztrwW>;yzu0ndssPj^5)fZ?P(ANIxac>%Op&c*>U$}H#9Y}L=cUNA)
z2iPGrEbKAB4v}h(@RNI|j=lT{wJQU)zrQ|cW|GT3O_AV1@ESW*F5FzM<@$8|>A?$P
zKA(ELvVQALBPcnEx%QK<@vvq!HZ1vh*2GfDXhv;MsU6KZZFPDiHL?`x`m4282K=#z
zB<o*iFQzuumT1%lK&ALN{A*mkrhy!DpiJUoZ1T0v(k`Y7Seac7qOWca7qy*~V{R9%
z{T94AG~;C!ct3_qix&@#q+Y*~F+SCH?|hQ_f^M=1%_oI1Eg0va5mK}`Wsfm{Vv5<S
zC6+v@+?ww#*-uaG=8r7TDo4`w>6V#;S<arJuZ>OE9Zb6ReA1+ofAHU8HH9y~pTgsE
z+E6(=SZ%z%oVnfw)2An+MqrlNmBbg%4c(QMQJ59?0_G#g@Sl?AYuqgIEt@Mf&URDN
zp@NbhVwTa&t-DV4X>q<><#KWr4kTaC*idE0O^3{kf;|(<2+?AX_U(?qx%A3>L^fMN
zfsLbavdu3h;)z_bgX?DN(RyQas&?~qN0!3ATcNu$uZG#~R;UE06ThFWH^IBQ-_t`+
zcvvv!<ObigwVf>UR;w`{2Q#tSWhz>1Xj0sUAcweCrLV82k>xtIK1?30;;}|Mad8pL
zuBQ?^9}%5GX>pAH5~+INhi>fruW^;Z=H}Q-NbfM8NJ7(a%e;8=mjKMWJ*3>{R4a`O
zBSbsBh#^wHyx4WfA3c`&@BI0-4kFTdDQ7Fuf!VOL*XiPonC%T9rfxNa;J*vtA0y4Y
zcx=j#jEsG-bR4L5wXd%3D66Cd%dBt;)af7O$5F2nrNvtKh`tc<6#dmmnw-h3|CTZ9
zs61z?+jnixQ&q3?((0WTvnB5mB6J($kgV=2$(cn4Skbeo+W*DC3q55t9Q~qq&O4Jw
z2mN8ac_!^LqT^+iw~xxVNOj)I6|%CV8E%drm`(*ep5m4-7ezyGFg!V~^jEh$@;b-*
z+?%8iKXR1F-F-N!GPnIhU(etb-!iU?d#R2|Mo7uNiNc)iZTe3LeN&|;1~#&B%K4ul
zJ)~gp{hB$5Rp~nEgH7Q5xoyIh75b%S*h`B6!&Ae)u#A{CyBGxVthL)NvuSUiS0!1w
z9{kh{><vAA{M0T}G9OKrA6fW_ZDjSp?(?Yp)-*HKp0n)BBX==ZI^HpdW{<0pDSMpI
zye~X%)PY+0(R<QDynhKhw#Y$ZIFzRu4<$K{%Va~ti~NtmMn*;zEo#*+OaNt8A5Xg*
zhS<WG8ve_J!50&-?%@#;3QJ7X@$K^o30^N9RJp%A$#1i@R(-Z>P+>Puj!>E-QO;cU
zoGq7bU(YCNZf3ZCbDZ&)Cf;)%rY~uOx<~ok2WkGwF-K^4eYLuXI;;UHU*f{vSNOlF
zZBj68sc=1Ke|u$~@xpzubidH{)l97vmif9}Gfnn76BG2oTld<6AG`7#)tYBf-j}uA
zRl&oHHeY0UDV4opYt_=M+-CkNLQ08iCO?28@Ivzmf?MLlOOmJ|Oe~tzC&|}?0|SJ!
zlHv?BG}&#z?|mPI)`{tzuK`{A{+o1-YL_s?yAK&*-h_pP1+|XRO@w2uOu%x9yRYJz
zA)C41OpBT;>Z6oOb*vGPDwkP6X*~AAv*)f)K(!OH@M11`=IDNV$oCPPi@)vPC{0)w
zXU?8G<DY&Z{}(JjwjI7Eb)gXg^i|vQ5GyY)Z*jVy8$=nl^DS5|lCn>{E_TI&f`h@~
zer7Wv@{Of|!kD~^0VU;1k%I{<?vUZA#T#C(12<2tfmf-+O5>LEi{B~lU*QOkt}XW#
z?$=KS1(I;l5dc)PHvF@`I9H7svi+=4g`R+RF+l<f3Ir=svS=P91L+`cwgg7%zjfIG
zDYKYQx|He=JbXS>Jj0PO6Ga~0w8+#hCik@S5_3{o)K*mMH{;|)7#qoR<eH;=X<)|;
zMbX;BBNOV?nQ16^IPwE?-i<VL!zk^9Yz3;F(zjF7Av-bZfD|%vP@hlaw<XvnY-~zg
z#9?B!LQ7_3y&u<9HZr>{&}Gta)!~bMRWUpxRaMoW!)C1Le^J$CU0GqP);a%;^Y?Ej
zIdtnYE8SC7_M@1@44>76$xSfC2a>+*Nw$6sP0iC@?<@J)LJG3_>}ZY@*FB7CwM}7F
z5eh=h`{G(3l(YMJ<*Ylz=N8i5rqf^gk{I_cyHCGL{W}o+ntsa9K<xLu^vgY(?T3VI
z=Z&AE+(nk>e`-(s@hEnq6U!G%c^NCudf#TjjTGT>qKvBl?!|mPPE6$j&IP?jm%B95
zQ*+jjlYXQNq6SAT`gIo_3%Y>SIxwcUELNq}y<p2;fB3t9J?w$i+wv}3z?1BG*wAE-
zxnr)fS(_UbIt#%(K9>usi`~`L(dhZ@#xOrT1lIgzty%N>VrhZ~eJk4o&_S&KigHgd
za$JA?|JeJ>sH(RwY#83aQI3Qh1q>88DxK2Oih$DHDM)vBC<-bK(%lWxy+M>Ni4D@-
z-O|t8dj9u#$9TU!AD$04<D4M_*n6+FerwG&=XG6kPWhah%locCz(eb=uxXEumCYO@
zTCdw8kxZwh??Mkh9y>GQ$ZEdj;^5$5{%U4!9w0`EY(3^GP^K@Rd+*%D&vJ>CAmBlg
z+!#?=eJu`XQHYZNYJ?sVD#lW}04a?#OjO5R6FmDMU6hqr&!`{Eu>dg$I(mAUHe)GD
z^+t{b_{W7KyW8kksN(pzbfq-dra-x%B<oVP++y$_h&`Kb-sHAniyh*~jiT+@xyM!Q
z$#u<+kT7<MLI5|724~7@wCPmB^J_AVvtsd5LwD@-#^R^EKn}PG-tFUIT8frl=+Wf8
zW!hb_^7=Bt?2Vt~DU6SO%GzC=dSEs{qT1F1T~8*;u?!8){yjC4Jv6lFEobcIncf}9
zM(vx8#i9#Zoxg{Y=T_>OPl}=cW33}8u7X01tAK)T+=Yck=|L5_*Hs+`ySFMBLz-2e
z35%p+^!}hnOze~AbZ$68e4L3@9MB&yg81lSyjJ7&rX5hw>8eni!ph?kiw38()@gR%
zOIt5hg>7|g>{WhWGA4NVD{{5B{9T=v&Z1S8RXbx!U#YB|Z7Kq54!9Co4FpXX+dH9i
zpEEG?fztccW`po`+*CC`jmaBkS;$K2%SD|{PXrRC>Acp&m02Af(MwlO!vt>xjRj|T
zxY;fqM42qiw|xweuunJ>uyG;}l!68mPqsB{Ilrf;mn_eZur_~?mY(qoH+tM}dIwdj
za;5jne#@c9EpMx!q<=F4PBGj!?mexMZZ(_a$JMgoFSC1bKfgeg(EVd+rDc>!JpS$H
z+=PVwS#G52T;{TwcmBJFse9qbB?x*608hraZ3Oj_5oms$pE}R!{Xa5qpL-=4?9OJu
zF)+V($Oq(R3S%ZR5}qfo&Yi|Dz6T!y{PmZYbkf{9kiXP-H-!Vr)vv!8*NxQC%{lej
zaKlWx$HfAnQ{VR;QExn^|7ih4Dkkn)zYX@%dIRJP|D&rr5Uzci7|eq`>GtA!ijEdn
zQ)=qn9h4-V_=1I~W08!G%Qc8<1N2)#qM((FjB({>ww#bMjSN~EE6=D%o5PAVXfHPz
z41vdRBO<=(UhQEN&rAd5K8|1_kAFrCEjjW@>JP`E#YIVS?)z7qUDhmbva*VE^hEG}
zl>BM;kGItY+iG$C4W>U9-1%{0^j}tTwr1TgyYD{u9xiP4__5859(diDTZQ&^1xmBX
z77~1klf(3CeI-D3=7;%D;kqH`U(BRpNMgN4yuA`S*LBFNmij$W=uo_u+2c$uNAXO+
zLOhhimd$}@QmyKaH>HPbKj0?c?buPKzY4spNVtF6ZFf!O^SYOP_aILPGbBs&o}vED
zo3%=tCs|bHHaFQ=?^c*#Be^Uw|FM0v8!D#G9Q%YU%2Y|si{&#5?@?$W{m5;ydqPZX
zrekD?NU4~2X$cmg8`OC2iIdul)kI!?7iN{#Y}K#hY7_kHrII+}s=^b$*7{)2ZlF?y
z&;LyQ9~r-_hhOPIFp`S@!sGAj!N~=sglJ#R!3&_4<N6xR?FDkL)ZQGGi?TbD1-JvE
z*I!&g(_)@xoOV1^6+aHC)2#Z8x~af>_Y=>Uor@A>t<v{<*!s(3%M&_S|44IW^ySP~
zS>;wIxyO?A!U!RtQOv5-p3;##uy*W<GDV0w4^j*8CjbkzH;H^SE-uFSC0JWdmDK&?
zP{T#L&a0Wh(#i}ICF~xZ`j%T$a|CuL?{f{!I)~7G+9wkQdvub&ed1WZUaM@}FUYsU
zRJ8T8Q$oWrG;@H(ZQaY?gYl_j_f*8dR2JYue~C8#I2lHr*`M5xd7q!$5Es*ayY*~k
zx0MHt1LWIL8(uisbN49zfhl$<T=h8B2tm%#ee$Rk;aue4&$_iR3F-iz!TdM9fpPJF
zR|}7c;?rBc4SvHcS}&BQ&f(&*>3;p4PS%7~o(W)cD#OsMS1pz1pP2YAz)my+It!J|
z(^0!)r{OLs63`6R>d7KjBZPLHe63{VW_?}7bVd&!U8}hzO;|Z<5TvBt>cjjGv6mkc
z6UTA6msL&GAoRL_gyxk669gpu)H#e7=3jEy*@dx|0u27T7iDdkm5K4Aq41vI(ONSE
z3oa&|Oq}-*Udj3$MmP!_na(QTYq&DSvb$!rgO6i4ePRK_>+kHoqH@nnl7l2wL4K$H
zhKQ~rJC3fEL3?lb)WWf&g#=#^0Ai;`9^JhDp0iO}UOPRL&(T{zyIjk7TGy{2hxR{A
z4f|VBc*<}Z$~YP*)Uzr7ZyD@9=y(#)Fo6Z}ZRF*}DRIdwHRZcp9JD!*{0lj&L)bjN
z5L-)rO8v4mXd=l@0)PpvO8$kz#CEk!czu?%pu;t7@7r$nX<oIQXHJtx=7O8O@2i}d
zzE3pl#21G~Vr-0-*zeXtZ>d}NRI-Ncmb=vQr2<wIJF7MA)^&yh*;tC#`5iK$zQCz;
zLVFxRBXE`j!xMpOo^Q_X-fs&1@G?kPcyTRFE@v4K2a}Du-Hs2=wbi=m%B4s1Ieu=K
z&DMppjWg9#CGD+I+-saSty$jb^j@7BF7)-`Ppfldarvoh&cGnf4{|gX!Ft{|#U4ou
z{t;d!)Z0XGHm8PrI=TX8NXPd2Bp&6=Db`CaQT%6gb^$Bp^}0C2PK*cm+@_&6Cudwr
zD&vh0>7wXD?cFpvVmCzL6kQ$P*YKXv@%VCbDQw-|c7=7EzvtEM53lMum_F$%k}|8|
z+G9}aisrrcdexr2KpsC$j}u8R{oF5pN^ZTtwwbnQJ9>SGKljNz{}QM##>o<5Q1ygx
z@oOT(pGu0;DMOs_r7T%XihBJr;-5-Z&dkViVkj}6ZunZ)Qg2tbH;o1qG+#-1A6MyP
z!|lgSX8-1cMrB)D*Y})HSiM!%Gb`FzLaQg0`8vVfNghJ-(ERGzSIyx1?u7K$xuc`g
z$`hfEWPy)|I=YqryMj{Nvp~AzjMq$UVAb<l1LI@y_`JsWMD0S*0mZiE2<BEKV_RDr
zqA@*S_?NW&O+-Z|@is4O=Q$~51h2Bo_uDqoBoUuDIdTUwMcb8QN8@5{Muxm}m+yZR
zF<l=OUeAv;TG{Q!XdGB}=N>UKM_WxTCZ=yJZ{2|Je<rWemeN^YKWP%B4fuKt+?t3F
zGBV6g`C}#{EV(~_e&!vT>q(}GuOsSM=~Sr#lpa<fRw=JPRa;AaFi#`*B<pZ!2=P5j
zsd0-Y%hw_zK01PO%R%MM(!8!x_5YslKf$%|F%}ee|9$sQL{mo>9<PK%A>}wFEgK>I
zTuBifAGsUQ&ZbKArmn#W^JVsnR^wf>7Q7#jc3F*_(uaouZb(5wZ2n-8Yo^=ErR84-
ze@06{(GK8sjM)tmbwCw{U<yKY+)sO0rPQmP;3ja$SwG&Peaijdk<UG`cF>eIccss&
zeE$zA{63KGE`k9#mLg7s`1?u~T#$eqY6jA%<XfbuU^pMr){C4iSMRak=l8u!e=)5r
zgt-x)fTsQhot@^ew=5;ChCfyGNDH^N?(G!O8}BC-|GElCaHzqAg9MmNQE}y4eoa=N
zT|9BVY}np;nlaU~wSmI)wumZ<r0?0JpIOGc>bms(KQ!)dIY)UPmA@a5DKOiwbj3b@
z5j}UiE%{_KgN{68&H6jOw6KUmYYtEM8p;Dv>RwhL;2bSix|A}oc6U6PNypwYxOU>)
zDa#$_T?_U0&J7(-of@IwqpZ?dA6mS%@tn3_-)ih%`*|Ts&oFli=fu@27Dio;>AqIZ
znaJ02;+yhBJrHU(K_v@-WMkHCjtXwrexzZbeCrQ-Kzp$;?u=IJi^$0R2UaKYl)U{t
z)ZTKl=X*7q@j)LKS`~~WCIo5O!o^OnV{Na{Ge^)qJA1MRCUrmFh2|>O<r*qN3G0L?
zdc0eycQR5C`*5K)QR~MVen2cou($sMksy{-v)d$t_AN^-Qp)l4o~W&c39(+?)i-Gf
zcY##{#fUTB{%n-{ZifV-uJq_Jt(sa!Q1QRpcuV?o#xQUXjmq3*+CNV=R0jIZp+!&K
zTvO;iPwdcT1DJ}(&fB_x&h?`57HWB=KD+?M!ux+_Su#L2F|$z#E~WEjqZ|~Qim0*z
zw@Rh*Tx5I0xrSr!wQwa9fpunZVn9BFb_};4Lqg&RKMzZV%fQV81q`clXJ4xt$A2Hu
z`$rk|P$^{5m(?q$&G=oL`;m<T&+3E6u(-DLBUqVG4M+5`Y~U<#DYS^YdE^m-))z=i
z#aW~E1r=Kl3O?*zR!T(`ZhQH;KL(fa>zI|2Ym@pP*S*@Fht&TTV-tkYpmg3IkP)=m
zo_+?x*=v9Qtviq@B^k^95rDPrehFr=R29A<I{vSD11r1d@M@ZB&OnmgIbK3|n%}rK
zrX|QP^$0Ct;g*k=t~zQ}`vS|BCm8Q5s<m=S%BSdGxkYg{#OOL~vrup=F=JD(Sjo<H
zuuw!*kzH_WT<i6iWj9~U>Ea#M1J|qhV;$xV;cx^+4!KDQC87j}&_oqkArrWI9go8C
zP}!*bb7vmVcY|<Hv`CQaraLdPO#oO!!}|yN^NR@pnkP|7HRgHN`?2f)h7WhjpyBj4
zs8j~(VM0PeEQ@vBT;8+wYxnA=gBKr<Vfk{u3A_PJZ|Vx0)Ce@uGmtq?S%r!j0Pl<p
zzl4d`Uo8RxTkU9kHit%>vIq1HWP)%l=~?PEt=S>xB$5!3=cHZBMtsgg6A_?wRK0VT
zFg&d3WA+_5HBG+$DKt+{TgdWHe_dLCsmNVda9;KtIlhfZaxhiNtf|V4N1hgGEiGwz
zv1}e+h*WsFHQE8p%CdP%{_EEVW{KW-{Xw=sfG$&jwlEY`*OQmu@eeT6e`!2vy1v=X
zuXaAep5NWw`M481D>o4T1v0MtM{h&RWs^u<WRc!WpBg-K*(hb{AVoGVYHqghFYkmM
z{j~I<Sy*bg!ecI$nc<@;ee2fi5gs^y+=y!tAt6tpm_fDk&ci`YKiW;%sfS*V=ouB+
zFl5Y}tEtXxd_FtK;oW)c4b803tzkQ2&Yow&&RCfLK9|=|SO;&@vGtxQ!FW|Ujv(s1
zWoXXRec@iC3KHW=MR{e(>(RCuG=LvaE;HqY^-8YquNT5U8ove^1}Syk&&mjY<X~>`
z<ep>12aVz?MS7kgnytjLe3hsbKngYq-?)2r1p{zL{Iepq0G}Mm9>#TY@GP(Ivr7;`
z7~D(PhF;&^c556v9`Fol;TcRc{}37@0LG8`a*qz*bjbZfDbep)$`JZ|ZEOpDR2p8(
zXg)n~qO06>YQlyeU~|O@_p=o<o9VtWv`3YJ^Fh{={yrNKanX*1X4CSq&;EWu_Y`#3
z=J7i>zTUWDesHkVR&56W5;Zpm{tGu=0Fa(FR1`Q@MBG+3G>qV%5GS-}yzhOT!|XGC
z5FNFss?fu)Gc(Op&KR*ZD8thzwkJ*pBxoTX0LgJSHt7bhr;UwX_^5NN3}+J8*Q@Kj
zGF2};=EtKF6_vcp`L6i<<O}`3Aqg00hPu6g_nK;K+|2s)Q(&mK-D@k*cr!56f2LRk
zeeBR>fwE1ml1Kep?N+p@G_6c^xAWm0(7S_ZC|ELv{lO=Nm1~pqDus%NMBqYxoSm6{
z>*o(42wvWL4qsU6Z0TC_(3dirOoJU;ec|RQTBU|e%W{)d<=XK%K|0uWcPGo)zSTMw
z{>iNXNB(mvjr(pp^@x{qEvM?|8R@Fq@zaeKXJ==S^*{|eqP!p6k?*vNjNn@+xfbv3
zjexq6)=o}K&}i6Tf4Pqxh>#!1rQRf?zEqL*$@(}*1p*!8UC@pKfz&uTDNr==k&qga
zAf9WxTs1N>8a!I>ECJb``%)nd4zJ_7F0^m>b#jtAyk;_1!HDke2Q0zyN#v4dNl6K4
z$NmBR!_xFuA7N2Z5S$W2lh?1jTXjBJMk_}$F6xS|=$q)Bb96vp0HUq`9*zZ&ci{1-
z$mdu;W)snB<2q9FUB4rvpYtD)Lq-X_!|d)t=Z{9#TE6FM*-DR7uXfb|)MiEo?^QI6
zmJNa|WtZP;bDSH9>{cjx0S2ivD*yfZU+d6NCmK|S$R(e*Z~w-@nT0&3mey8mX6B+g
z{iC5t)Y`%F*2v!{ljp^(Y-}qNwOo*>&~e+XH<2IS2L6osI=Mi=al;`#ZD%#Q<vFA}
z%$t#e_6SC#hIHqia^J%TV_v&ebIrM?Lh3faGJ|=cdsiQxZBDrNzM!Jqi?Bj$sN{cj
z-F&Ed>q4l2+_MF~<;hq5HjJeHATHfLOg(FPGej#``Ojr7=K{Uk6V8+CwT>~%E3O2t
zCOH%IfiwqQ4HD8frl)mbpulJX+6~H|E=cUcWw?jV`U8ozpPwIgYcM%ABO|`g*QcQO
zlZa$^09h@N7!wj7kB3&M$r4dA&$#U^N)d^)pRP<@CP}5kD|%C&`(<m_^LToCzBVvO
z8CZE@w}=O&hM|4c?yNtva$7azY7~PYIW;W}3YK>8`au&ceV`(tI$1SHh(xq^seKu^
zZ^G;qP2`H0fV04w;d)Gi6XS}dbMkBNTRU6&T1i8B1s6y2NXN}K>9fQDWOMtnM%N+k
z>$fUw0_#6W3e>ybY_Az}ki-t#6?G}Vd!d7_94Fv2*RdTs)IN2~;|5WEpA4X+q`ZRb
zzZ>O9pF>D3jR0BFtos5k^iwj1v@FnyRu{?G+OmdINkPSvH<0bnwK-zdJUSW+iX}jK
zUDgUJRROL=ry21kYN)4txrqxI6lYMV(8Vs#rPsF|q<=7P`2uwg5Vtl#50xBnO{#kg
z`qRaZ_>C+qy1g;5egY?v!0Tk#9!7cP&||b6M;it!6G6v0RA#mQwcxWNh=#tp9f&Ue
zO8QGze(hHL9saPkkNQtoTwhPldCNVDSw`Ujao5QlGMUF!%({%ZW3pmz-hT(Q;Nk-G
zzpvE?iI$aCQ$3!#3K(X0?exC4v_uLj(?VJrhBf3U+&tp`|4UE6&;?aeMr`lJ`6<R7
z7Cm9)yd@JT#Hpm?|89hXu}4bZSxuAy0V^V+q~yd#c2UK{+S(c!d!IN14fK;Kz=~H5
zr$6vP8_ONH>mh{$bAayw*Dc#Qx{9ozn~pJ6iiXf@PQ#T|Z%_K{VcHWyJ6!K0-|Mbz
zrLgtU<35WkD6ly6d98KqLuoKIVq~P!V@r_14z3(J)dIx*-B9@HE}J19bV#eVA}n)T
zDEyoNi9+O_zh1quz~grFgLdvwrUKRNBLW^LM5CD?nRJJ=t$03{lJTW5-d5056+3Nd
z7iv}u0ULz0s{&bCPS1>q+#<S9aQ80p>p~iWhb9a#mhm-+{j*KYo|_9j1W{k9!_U3w
z2;DSxfC>&9px5MD#WeWNsEv)~fsVYNo_{9r0lTxgnez0RFp2X>B)_aLh92t|sGc)A
zTBAp$B6N6os9EnC0r@t2pjCw24>ANa)pCIWLf6@wkhaq~OwGeX(kAV_Nc9ZYZcEgZ
z^2Y7k*iR$qls?6?jhDWyxxe(g=Q|iq9Dcv7_n(acUf3DDR&_@RMu}o%Q04z3^D(Tp
zx+}0O|L)Hy;Ms&*r)CtF7qFiw{Mi`7V*BeOY^QHPIeUdC;%~L9AVX7{_BAAkg`UIo
z%K85?G6q6=VyX3v5DBMw8%SND5!BUJuU;LGE)}wB*E(7t1N(7Kt`QJ0qcm;9E2}?F
zzNMg`KtLZK*+M%w&o|i&yBa~yEfvRU16!&ta0sl0Mms%@GOY2-mw!Wxu$*mj7|iUm
zQ^Nywguu$rtA0YOio_YY)pyA&tP+?b|6nGmyJc2WiJ({b7E}9AEGX)6zHCx$%S?=R
z=$V>|!_&IMWqChKCdm+b1}z`BBF}a^-8Y~p(X~cMrJ;Af4g<KNJB9Rk5AC0(5eP6k
z%w<wyJqvI-A<&xw@BWhzbYnG?>0a16SjYwb1Zwpd33jknO7rMg`Gem>O8$g;Y8ib4
z^R7-Ql1T0zGb#RG4bKJ69A+s@?#4vN+)<fk_?=+z?w^bll<6_(17aS(!epqsYRC6f
zql35#+w$8n1$rTInd`QnE~+U{)Hq<3sMJh$j*Z<zH=EpUr%syQ+|-_pu6FQ8R|2uO
zuiJ74<@K&!Z!E_ek`ql0B?*#yd!C;}IFG;Wq2%Vii+J=X4f7N<aFyS8l0@ZG!SVU7
zumB=3){>ItO?6!Tz|1Zi$#RmKnjs2Cw(#(j6yhh(|9k#w0i?W~Zj>h0H@gV1p3@$>
z@Fcxkt$nOWM|^oYb@?MQH9$k;7=)gx)wYyemg7&eIwu)=bRVND3H>bA$AE$^DoBBf
zii)myZbB&FDFbN|642yw657YvGig>e!j9QFIOra(gqFyNGmzk4*(_avc1)ej4R}vs
zLvXttnrpYe3kmrL8e&TTYY1iUGGTHHJW)dhrCTBKJPve5J@M^%)Q~073h6%4{BDjP
zi{=KFXByODtv`s{Kg#;mP@5oCt&b}14!Yq|yB(|e-|_SN_VXvY$Ko6`sw;N0{F=ZJ
zy?*`hN)gVLlJG`etFOT`G-_8N)4)OJ=D{n-(3nXJ7NFaAhKB;}&pJGMLB4hh#*)=(
zQ`47#T-z93lW*~K&#b7vC10%+B-za9|9=>>1F;7|=I6?XPM`&WhE%g#TftdbOCBsx
z>$D5HV0C&hLBp$L&?4XGwoPxeB>+YJYse#h-{SUv3_gG!2!;TE=WrDvWo{PGV|S2B
z{CAIgLMqO$`@EvxeDP<5G3y*jZu!AJYU|J@{)9Q+;Z4f2J;eQMN@>vb*Tv};u4J+R
z1KifHpt5@a89xaZ7Z*DTK0G2g>EW=&{vjfI_x^pe$2mM>xZ-B|GbAJvcp{*XV``r<
z+B|i3`a`0cH@JH>PFsmZN#F^93(9wSw)b;nBxQIFZfzmC3$<f&Q`1kN6@n>#husV5
z%wWA%M$75JMV;SjxNPa~2P>N@pM!2T3G+>diviXN5@+smm=b|0UI^rx%x<JVpLa2)
zcP=d{*^PXCyEz{(ZUQ}PN+m`k)JDPQL3BoJyYxEG;$R_N4m2|Y$$?Mr@xKq}W?y!<
zCD$a7;+vw<t=5!53aq(V{X0qT@13$-Xj2Fti2LD)5rbw`NLQfT7Z$qH<cwSW2g#?@
zA4h5&%)rTc0UjEiNy`X&;KM~aIjbX+dD``^;1Hract=#p)H-hbgc|yx7K5f7K3DPe
zek4Dm4AO|S>oYA!H(LD7X`fAgFKyd;L+MN`G6TLUT1pN1vh#VDa+-p_?OM)Kd;%Z;
zU5~u+V$3u>ME06yn()RJ1vQ&lH;?Dfy1=roF3d5>;}s*QW7#%*c~=A?gwJk&GCPF$
z0)eFmx07tE{)i0D3-}tOu)%JUk_JWu;zPHqWa!d|wp~z$AWc5P_BrPRvIhhdU*kT1
z76#7~00QTKX3*z^)l7aYA{IowcZi>VpP<mpFX$jzG8KC0HT9{vpym<28H%Ced}kUL
z@BEBj4<sO|eUH;2cXF+2)==LkCNy$+o(MBFb%e<FN(1ig6#)m5xumV$ajw>6##Te?
zKj)3F+jJ<uJzXq38oH(%$cTKv&TKN<ru!tcHCewGRtw!x^Hqy6%TJP0po#al!5oFY
zd3Rk8u)QN?CWLWNt#Nq;{pgRg73s7$NASIiJUTB#=A#5gJakt33VvFhI$imzUw_pe
zhWEHBP2j7(j=_SrK;6)wPY+WMXlUoFT^)X<+PPg^Y;~UX7u`P0%w!m^N%+c!6wVrs
za!LF%N{~C4gP@DwcK&a0SFuZ+8u#}0<|0(I+a(q;>$aP4(T^YYu&$|T$857dIzb=w
z68V3|b`-;B{&&T#TW5}((G#zR#V5Q@7xMCPYCb^nS-3BemftF<9JS9*G*}&Y44t;V
z?_Si~RMdD8zSRVfHj0Gn`x5GQwvkn5k2j(F7kjmwrQbQ&=xwqhJ9BLkj^{joBBbIX
zzi&2F*VODbT%HGmC4~s31@!s!t~@XYC<(~Gc>JpA3DI;!ftL-%!;<eF1im^(gNF^W
z-tP%p^c5PJDyoS@W(u~v{-tFMJQ{Ol`I>i^vS59|(CpjdgI;|j)25@iihrzG(!bmN
z39`X&*oU^STad!DM+SDjs*&@^GCi^uOUKofr!$JV=+E@8kn|Pck~gr2rFTi56i#$O
z-0`x#gRacO*6<cM=L23Fx6mn4z<1znj2E}D9LDoC7w<h|xBfC09%KO-Vvrb$T<<6I
zCWtDc$t?eKvqK~*_ui_+5yH&6*2u`q@wcv49zr5=yBF)z(;-J^=!DOu1SdLIahSaD
z8v&04N>fD}-PtB`w6aPt@g(le!Nkg{x__*4B+aKLkn~>{8P`>4BcAQ=2V_r`Z^uIP
zXj94jOz@EOFI{z@#!v>Q(g)?4WS6~@9)4VE2VVCt(@h4njB!5;jksCK$`J`a&HNWf
zfbQYH*NG06Ft3ouKS2<4;Z(_HViC!P)JLuJt0y!55Rq`|pFet$Lh$p}D3wL*uBTR1
zlvB7}(w>yqz#{yc_VHeY=|Ybd59z(UFF3@X1_==Gm27V7+_<*3mbbsZuUYBm>q`d>
z5yL*m#>Ns}Xp@~;j+V9eEg%rqO2PZ<Dbn3>t0*hv@Z0jW15{7_iRsd$<_6wXpUo9}
ze{J%U<W;2t+iuSS&bUbP5jp}cu7g~Sytaet+UgvudY%*WRwl-Y8m&j0>)P{3^k*Xs
zg1=!+LgEYbJ(Ldn92)unqEaz=Nawf1oc#6cSG+eDw3A)ZtbEvGiHT?=!*6P0HyT$>
z5PSJ!wp-(%w@?35Kwt#Ld!_trWrwwFqf<wmYYaNm0&`(u_5ze5QuJ5p8(){?Astg5
zgr%p2$~hijp+A+z?LOJ|r2T-@Xq`(b1uc9wz)=wt2g`lW0ksLEDA2A~hZ`rE!UT*P
z_Y#z$m_#putZoDr7DxUIF0m1EdGRpFM-M&WiL%xXQ*3W--K?QIHaxPS68O6L-_-oS
zWpzz@o^EirLWAv2Zn7uVx;;Dk!Tgz!du0mOa?M&{t*xckeCUh&TGtUoFKyLp4t-x3
zH}`0be!27xdXW@*P=f#9P1Ei{5ArSLG2OjN;=%V_l44;E!A_w)g5e>_jMxM#y%uBT
z8{0P$c)cb!W6_gCWcYkD+IC$3$+G0U*cAg%R1!rFpcN#7o^|NVmyFEJ#c@wDh#}b_
zZ{tN27UCE)N5t@dT7Z&K^Vmw)0_3q-jPk}PV=3j@r&D_DtGT^_r2AbHPC3t1qr-FG
z<>BDtgniZhv9tFPUSL~m-hJG}<0&8)rnpcNC|0!KZu=?o;w0`?zE+XP%#G?zE49iX
z*dwK)2OEzX{9m!7$Ju!Bgm0Qr2kD*|moL^2K1?P!(`zwCiE`qQiMs*5#Q-QZ8PfWe
zmu(UM%D73tTbnt`GF~d43rpWkOf)4_7K!lih_EcC;bRfw?TH^y@sBcX>2KIK-b!@c
z3VpR(V#lbYDagmqH<<DY%Yw&R_A)+DF0Lcu)A(@z+)P-yMjRKT?CFZc#b9)i5#0lB
zwJ=5MFYzQQe=EDl$TVkVnORB3JCgBdCcFq<NlWrVVFwH5KC5(~v-Lm^AF*WP+(^IB
zpq&Ei0Hu@J&3`_Iuvt1LCI;u)HQL=1UxX}?Kd0po)>d5+_zC4XnqqTSo5^El%2BNi
zoUQ><QUm)Fr|brM*F7q>oh*3Wg&UBZ+7TVM-P0H~r%op9(>O_=WcOBV_3vvt5#<E)
zvlOelvC2Pqvw%vG`<7SHU7~)lpQW?xKXtpaTlYtZVc@>khDP2i>mS4$=3$5N{;2fs
zF^;3gR}+Fe{2KzELRJEN!dH0>Nq(OJ@x*V#(~LioEiH33t4qK=k*>l`7Dlg<&t!Zy
zj5n~M+Gj|>xd_nq82a&ubR2<6*A?H2QSLL(a(w*0NFP7@BQq=7Zy08|vnP*BR0ysd
zpRTleV3$*_N~ZMomW=E-eJU>zbG7-w&$_t7S3jON<zB?>sgtj3Wp&O`)sQ(c!l6<{
zFLqU|n_lUMlSj>3<r@#;X_Nk1yG4`Tvm{)tN9a{<tcJ5H9sT$j6!#eZwKL~`a34XF
zk85I5vvg1yVN$$&&Q!S^5jT3<SZh#LQ0JV(*<1a+;$X`;HoN6wC8aE5$6H$@(sd&R
z+aTP2tH|GhZbj%#Z@Cyrt!;v`NW8Mom&g30Ydv`tIyDY)lL>4x7gXVqmqy-}h05hm
z=j@BLp8IdrOK5LylJK72vl#bEZsfp3k1M`+24f3C{sJgPp<snPHY`rWLcBltdZt4}
zN?wG4zq;C9!HDZn80=TOh`-E8PT=IRJN7W)iCsl>uCUskD(ulOfU}56_jOP8qFUOP
z%@V`)IL&^Mcd|Q-Nk{A*^z{}6)~=rPswwAbpPBkntn86#4blXw)v63|vI~dTj6aO)
z&Wu7P(#CZwXYXO4$J8_TnUVyNoF2n?%j0ZK2+ZzEw}zFxobp}#y`zBv3ypU3_WmVY
zFDK_fnmDl9NO#9FU&poFQretVJ~JelW|g9uj7W!Gxpp7tu24y)<yemB0CFJs^m?wi
zb;yP55+ldoxV5WyI-Sna&Tmkokm3dt6JNALSMQJ36_Z@WHu;7;A9eioGV}yJ8pP*1
zy^+@iU%GA&m+nK#s4XR;BNzXafCf%l<#2xH=t3uPb&_8!*U<MsjdJTbs`}Hm$DHXx
zZ=%O14p5i;1^4c8_5ZR$4%GR)VBea)aU8>^*gd5qK(@&f9+umN+s?g&w{BmW$;T<L
zg!A~GQ(sGWSh`;v7gN9A?XdaH#Z2rt&VKs5VY8VQDbwL(&l_&PAKYh(EbKant+-pE
zE4jtc4TNTACKfiOn3kDxWnNP30;VF@m%1QP5&8?ElDAlAq;h11SkN-$+tzyFv&xBF
zt0>0dCj7WE&nZvR5B|L-3cO_{t`5dU*tL#wYpxEs<421=eiUMrW5;L6G-ZeS64!b-
z#;;Wb(JSH|rmxcI21vE$+P;B-!NPrV9{ZP;$Lqf=%GHSp3DZFJ%e(}Qt7}FSt|R8@
zPxL-0*K2xDcqrj+5BI&>;~^xMY4jJ`+dJ&fM%HuWT3L;lb5c@J({ns}9?=mM)n0Gh
z5ScbNyI}Y@JGUW8t6Zt}m%9KVxmRH1?@#Y<D)g>|7=gQ!em>*XH{bWGR)g7cho9jJ
zVAWs<qfiha=OOlOkDhq3*eviSmGurD&hC*)NP`DC(Q9=ERuVkhQoHt$_@#=NB~J7k
zR(+B$ZdV6GU2ZwSgDTM2?v%qwek}ABlTxg&o!0TlAt$)w_UT{o1-7!uUCMcocCo%n
zy;M!wWp@w#a?STzY|iLZV6%_D;8S(F+Nf+pQT29X_C=S%ueu?@!NEW&E?6yfQ6j1$
z4a-f%!-kTA`rDMdKaI->x%wV9f6ozm>MewM?W-&gdx6^X(?`;#6Du?ac`8|L=lO!9
zlwzjxX})RVzYQJ{TWyr_FRrw{Xu@qytI6@iO^n7@H5HbW;kC7q_X$+=L^C^FVm}S`
zGW!M5bs8%4Uf)<p-Y2vEtW7#<wz*m!TYLFM3u7z(+>cVs!*Y2#yB|FVM03}S1UuxQ
z($LX~LwT{FO%lYXuBkB%AIr7$^Z-eZF9t6{PhOU)?jN=kgr(2$*YK9`H8Ocv)jT7{
z#4MZsA-BPIPa$1kC_Z5D(+7pzyYga{m6j{hIKK~cbMp*_m9epLj~7BV_ROj6oVB{v
ztbcJs-Hx?Hc4?F%@Wd`dnp#V?==uh#*kqIb4#H^X=x~Vsh#-=JfB!T|ysM32@R%jX
zgaqjvLt)bNoS<#qa`=h>jQeTMt8(>n$oLK<7u1GOvfrX|A``IqwZ0eRC;$_P$jP_X
zZ(`I&wDDGzm9aOsv~-wrN`#pX7i!)7hS^AL>icdvO)hw4Cs?4bzTqMz>0~0Oha*>O
zkIx^Kds{gl1y+NOo#thd;Q8xoinv5Qa~`fH?L!tPW@Fb65{}{)54JMIkpjbfTZ;Wx
zCeHV^cp~hdVewAnpVrzly)lL9d(Rq_j<9=OxI?zHv>yR^PiZ&6Ew5Q)_J8UJn~Yc4
z<f<0yT~oRC#25S>ZQB*>9AI*GcXs3eRaeTACPn4gEsT1Td+dio@u4(8$lLUaiBz}|
z|BAV$8xhpAFOwz6^vZ5kH>_kttPv$eCR&{ahW4%Q&S82W^()_=o$q&=%_&hn$^MJ}
zDo)FU@nC*2bFM>u&FFmxQBq_=0r?@B>`(*e^ZM`R<*RuKZVuln?aEFy^DRw=6bF*r
zM4^$*RhWgpgqT<#|Iv7w+g7~_OhdVOlyq=2@yKDxKaXR&+h>*6GtSb@K4^7vKP2CF
z7@|-Vw8Ddy0X-ubx7Cg|Ck$9)h)7Abs9GEs=WIlpS$(+p`LzHN8i3@1q7hCG4y7&R
zX(5HlbmJRS@HA~c|ApY5BsmOO1zNf<$-OYR;v$XtawM&Xqsjg`N>=IcKrVJ9^+P)D
z$(r-K!{!#Tu-M_kjN>9?>S#(tf+l}R!&5=Qg4nY{veRw-^D4avQ3mZo6=f^gUMQxU
zEfn&$j*mDlvap`TM%|V{6O!`6(nISv2mNIDI{NWoXIZ<@($cPuSC@hzqiew7a2W8>
z$u^y{>~ElA{cG4bO?_iiX{>gtui}cnz5Tb!)&2~BboGaW^^pNF)a`^3^PXy3BtyVP
zSL@uTPYVg-fMM7*O#sMm*7WWg`hD}2%?c|ZJIt+T2LRH)oiRRt@T;9l*|Zp2E>IIR
z8Tu4~9+*VX`z2GA22-W_$HvKI>!Tu@qWbN(afreI2t&H9xbXLv`e~nDyeOEgf5GOo
z)rsm^Ud%+lN4X!F?AXacyeutDW?GdqP73Xc0<y1E?1=5)VH#R@R)k!C>r~xQWAKwJ
zS2it>^PF8BP=h2eEBR36BT^RSeGXoW<rGq@;ruX;0?LU<t0Ud3q2u9ec9K0jXP^<2
z?CMNWbo>Bf1u<Pi>?%?|d6^VMmO&dFJUGhlR?f^PBv)*+d&o|kuO~9mUzbDj?mc&{
zWb#Y&?&_R!(^%?B8tzD_b=cJmxXGj`Yh8^%;a!tkitHwy-`ZMdMVk*lwKC(-#hwKE
zhUC@Nlffo4-`XfM`%?vRe)5b7iOIfj8G#(2fNtw&1VO?4PpljjY~;c8w!cW~03&bX
zPGmLsG(KJ^@GdDyr)LLD11W?ln&DY~eqc2-T2J=}y~wL%m?!Nj$u>Q)XP7$Pi7#}T
z{jf;bZ>o1%+n9NNFy4q!a~c`+t`-JEqhQV%2Q&$_*y?st&hBoJ;<AAH#6b-v9SSw&
zby)z_K6q-r=ZTAng|@$zfBLjh=_nh$W|CCl0CjpcZ>VULR>-A^vc^b?OG+vV3%^Tr
zU8O}8YH3h~i+MrPl5}lXm!wnzj}idu5k*Bs0N>_85cl-%-Mc3zYejI1*a5fIJYV8E
zT$mFT5ix;lI9D#s7i#pOs_l#zwQAU>TGc{0oR2ofvH*6-g9~!^@DPHaazM3cHYV$N
zVF)Anz|pLLjZ^}3U;(0L5)M;iD2)MUQ3a`Ah2z&dUC;N*3iz}h8_&3kDxHc+9ZAW6
zv`&oaCRWB<7NcIW)IITXl6t9NV3g|RMJ9e$g}T?*-RBJIHXJC_Z!KREu1cMooi||3
zaeYY%zbzKR?BoSC8GJq6`C02!1PobP<Cn$U$yBS*6KgitY?t0*^`mZUs8%Nd0f?h@
z#PQz2ffCf}LK{zupI@FO@HljoPWX{~(!lsbAHC4FIYkZSm#Z?c#=(osBBAb@L$e3u
zgM_6+l0vz~m@=9vuC@Zeb`<GRn<<l22xwlXIp>m#gQusy4v7<R8YJAduX|MG50%b<
zdhJJh!o~Wnx2j~sO4JKn7h{a}A<s_7<`nq$eW2%8N_zzcpb!bs&6{C%Q+B=lQF-T3
z3xvaNQL;OpTLH+<!Dtr!-TS@75ESAQ5cF~%?(Z`UR>RY~UU=1;p63>6L8W2$5c3-6
za#OQZGSd<)(2H7560b`SQ@Je1&{poG_*B|Guc6A$Nc!DbL(?cIDpivr_q^VEYhu;z
zxu3gW_HY01$uNLulwus}N4@1K>aP^dMpUl)Xh?fet^n0Bcm`{?l9jZ0cbkfmawS;s
zBBY_A0sU~7FJ1(aWB`|J3N_9kz^k6+-~96G8`P6|z+wJV7WWo|EZm~6OGnBKI&U3p
zjHxJOO49DmkqEd~k$IijNI?VxV49++s4twzJU=qmm&^@kCFoy8Fzal$D*>FEC~)yh
zKSq+u<NRP49k!pHpBogb5|NP=0;f2yE|&}{l2|sQAj|p#<ty0OoyP2|P_Ib6+BRde
z;nFjZlvf$+DqfWyt(wa0ws)ob@h*?uVnF)33<gv8k8Z0R->$B%_&M0GPijD$_}#R0
zrGVUMtw62RpuHn3t1p|Z+$>iuL-&fC8$ApEbL`pKa=x>4Ppt;IhQg`rR@bA!&#f;y
z=C!M$O5fk(GWkAFK<xEE?W)*Dp5hR`5?x+NQc*v{t@lYX{kz-Yqg9vQS0$YlDmkq(
z)tDQyJKnr$-kbjI(}<v<2{HNiBo^zdt(`@0&)%BR@SMEs@9&@A*x2~6i<1tJGIWCO
zhr*d0`Ym|JCnwrAmw?|hiMgJvWEE9Ym}=KJZselZB({!-m^cF}X|{j+_8b6P`D)t*
z+s)dQOz_XoU28cLYT;@`Z@&fijfI7b2hV+%p{~jxJl1R|3zi41{jRLi14C}q0-{!p
zK`5FTf6B+lr&y%TFBSiUW!R-x;MBZu3=<P`ut-OcQLRK6j#58(A?zmo=}iK4@?Rkc
zM6(7up2s5chhexWDJd!zUdI7bL8QDq=w)s)FJ6A*G>Hbx7UpIOS-9+fEITq1>#{DG
z##y%a?`ydo4EZvi%pNt5-Xlv9v62q)!tmOWPP{DDKB04~-^PyR?Ee+Cp_Hciw9rvG
zPRc*{z1jFGzIwe|u%+{^8e`(G8WNQv%VGU1Xan(N2+D&}<h(T7x`<PGN~!gMKLseB
zh~?M9PEJnzJKqUo!KOztY2~8V<<lwHT@SJT=LGzU=@-<J(IHC_IZ;FR$iUnY&umgc
z)s-LUxq<t2%Va!C$CJBv?DSwQi`?r>4szBRt#zU}8`e;H!^AU%{G*jt%wp;__Bowb
z>#k3|wrKRm;BJj#9*j?dZO*#n3_6-1CTwqi=U1cljH<QT935ATyRnW;HCqo|d~3tz
z7c2*}IT_22Q@)zsm6uyP+#CpTqHZ@W)v}AP&Kq9MC^Bpto6r(GtX0rCT8}jX+`#gx
zkLz$jMj!!xf3oO-0}rRNQ~gQJY#7te1J&OfEr|G^kaJ{DeBx^-oyD?r`V$7RC{CGL
zqDAeSuxY?j=E8Z(1^q+lNg?cn#LBIa(idU&prKfFvjm3o6MH6wQjR=j?-&-d_Jg?C
zSoLWn?A-@q)lFM(fkSFY^tx~<H|`SxUI%U^_>Y?{gMf@rby+|H*AKf6N+(1OH+Kg^
zYqf&!kof;*H<||tmr6VbA0`%-!|6sPFM5aQwnti=%vXxwz?ScS=9FX7JdK&4!q}*T
z#vRH#q6Sg};|><KRCFHE@(5@LOCG<xDkiwUk%EDbb12aVi^AsSq>$uV?!b2vAnJAE
z@}}o^xbr!{6qPF$cC(kPA}@(S!Z&R0;r!g3Lbk&z(hRspaq#g`5OdiXQcCp4bK6mf
zG1u?Qhn{jSuzjQx5YX<w#2l_u%vIcv=@U(3=^gVg>6L<Oj6$pUjrucm2}#rb^p8|B
ziTP*}07NTBk18eN8nlqt;`0C%L!-$frA88AP~1qd(o)L<=Bl#pUI_(z_zwmZR2bSH
zyB)3zS#?}(JW#XX3{1EPq*w{xOi;>m-|T!=NYyqbF?F;8=Y7tPJR%+(^D!>3Ut+W?
zD<r7Q_pYb{rEHsVysJ`M4t{he&SpKi9HTbr(@`Pl6M*=nvec76DVF7Q`AS95=z2jU
zC7Y5~jlE5W=om!tz;eH(Fu6E84x@<cj$})#q+wHG-r{ABdOHF2!<O7OGjvfSZ+%Ls
z-w|gfe)A(hyPRlS1e-V&PExx$XQyd-hheE!v0!u_;O6uOMJYBWEAM~)RK+Ld+5hBS
z3C8PWK3v8Yc>yH#Ah_|fG*_O|R@zKqY~T~90QxxC;LT=wSzuxO;VOG}p4}4R0kkf3
zsSxXK`{C4i5eaxv%cSD32~kSg#@GGP_ey&c4qL+>9{k?Y)^*YeJhU`4FGJlf2amJQ
zMK+D5N(d+nbW%1Mq9PxAz$HQ<hJF3Mti64iW%a@>rEJ-+GjpKjlMDlcc;Y_Bpi$8^
zGj~e_h&xI#@J%(|^s1GcMo_Hy5D5Ixa+=71I=3AY6a3WMNOqr*2x5swda*|@f(GHL
zK4`=um>)BQgU#c(9xWPhM^4iKL9+lZ4TD%sk=XN*y^&>)z<V#lJg3Sidrwu)N+nEU
zVpg^DeBW;TymBxpdSDQ;-?~^(Laf`ZgMLy^ci3ViU3Z#q8<eV*yBrRbsNs55pSLIA
z5?4Fr0CATCB9w+ZxeU=z9<JPEK=j6q8$7o2ACSk3_r&`5dIfbD<|Z7+EW*-*1x{W<
zMTqK6LC>=);QKWg^luN5I*mUNJ6VX*v4^+_Gr?|!#l_iq-sKWRWuyx6gv1)?XXmwh
zel$h@=+Sd(>G&T&LzKB5n3D22%R-$q^)I?_(7|pzB}~$hjF(AlRk0vf>)bw<j7n_O
zp}4`lwfbjaHf+E=oA<gRt*vY0$HKKIe_V_n1sVPfpxNn7FKsOnj@~MM5aYg<<z&`9
zJQ7YsEt9yuZy?OfIPln{pTYy~w6KfYmOSdJs?qS0Ebw9-RtLp^3XJMD3`d!w;3~1-
z>*l)MStTNeba$zTVm4vvjvJ#U;Dm9<l>v@K3yh5wl+7ZRcmkRfZTZ6^dpHE(hlQnA
zRXqv3%SJWq>sZ)UPjQtcMlx5eG!s@=?dCTO>jy?RZ&HXUoqKGoX2oZTc~TLGEZLbf
zkND3i7Tety%U#{d@+h_e*r_O_F-*sv9R1ir;sBR{np)XVZ^9UM-{(ksWiEm37|BVM
zk+o(+g<nBQ<(7-GF$3WohFy2??>tVP|J3sWYEn~(ftWCyf*;C$?pDm14w@Z<H~<A~
z_BlMvT!pM2;v){@A5VeX97N;(T$OqW^y>c>)$M+gPUOoeYPe8`Uv-ssgY?!e2*jG%
za28C!e+4_IR9KPZd6ENutdjXb<X?Uld7#il|6IMwIt8qJ1_WT-C2k0V9~lOB^BoHv
zb;Ca!l(%<>9CeOnlsl}Gz6dKZA4xil@YOL#&(ZtR<VW%wC_<R><;!Ezsk5|O0R?GV
zUUeDY+58HPjieirw>NOhUd!JuICD?#T4^WYvW%Z}U1OwH@}v|LY@l!Iq)v2BRc+ln
zIvNJW8V5XQ^wdPt<Y2r5XkQtwA3An+#r{i9sKN(|2#*7l3Pdb%WFQrL|NcE-V7Z`t
zB4N>c2T^oxNJvQhA($_Sb1I=AfC9XsFi044)ys2zuaoA25k|u{4zp=Ccz#{e6V-NE
zR|&}SpZi^3s40S-0$VJ}HuE`F<ARWnPZ<Ol(v(`VB1x%UBrze8h3x(!D0-Eiy^D@#
zBeic6rE6s8S(vyz9vpuX_-%ADUVycLpn$U;xabooPggb<Wiz_d`NYK;yNx%wgkj}2
zdJk0ihdY7SS_I5ej!cq(bm5se@ObprV`vW9Z3@B;oI$^=tgJ-0O>Tfg?5a0w%i7fe
ztHI?vEN_7*%M9r@rn#}B*`FzOuREG$E_Q>WePTjOhHpn0`K&r`t62YNAKi*32go|`
z8RQRzx<JG712%@~mMo$p+3{daB1~B{9_~Bh8zU-sWfrnqBy&HV44%!ue@*&HLN-u5
zLh-@BjfH>sx#mX{6v9xqH~}3`LFkeTZlrCLnwSv*J>yD%kPsbgfoJHIgCm8Mzr7hr
z^&;qgACyUiNYa*Z%Dew-A)1|(f@{9_c(k?k1vBzjs$o}j3L0seni5<>LJj_fYC>}F
zHI1ZuykHl9x;<5Mbk{qaNL-&_bDIh|LSZJvaz!*$Rz~_I$<Mz}ZMb#*WlOi`NmS$?
zhUv33)MM>KZx?_?X(=g1Ky1~~)ujm5-LN;Y0E!BCPSgumWYeMa^z`=OzF7r=q(qe=
zEPa399s*iLQl8fhR!L86QUgTO*gAy;)^VI~!^!>pnMiDecT+?}WM#r>3I#=D@*zy4
zV`Y^;ST*R)?Ck1d7#y@k4Eu2&tgiZs;ye7yq&=mBXM$^{^xfp#;h_)*os#_2J*&3T
zKtKb?HYOBe)YM|XI4Xp;d=!%!_}tAqZ}@n52U+hwh>bo@jod&sfD`?<{hn^Q`a#P*
zEA-jj-QErXl((|7a<gtvnv#Md8#Fj*7e7IZYWI~&W3BGjxAAijGXg<H??kzObkv~7
z7;wC|G{$z{o3%~%@p8fq0*SDTtu5532UCUY(hExJwUnvL&lL3n`_Eilo#)36o1b9;
z9#XV*ME2vz4fi%xT!HfWQY9teqCd8IYx7=}e-TRf&4NEQA-;Jt6_$A=Co56D-t_?b
zqGp01`Cw~G2LvsU0Cd^2J|W*#=<NhiF?TE1q_cja1#H?VyE-@;ZoXKuq7&m~A1`?4
zJu_k%YwUW8)Sc)`o-S?p2P#CV;o5UjDiJuW<zCuXZ{%+HoSk!+t^3=u-P!*c)^M;O
z=%yGV@?dsg0A)2PRKQQ1vsi9s(x3JSNAb7GMHHsO=8WZW$OXkVRLx)jG<Fs0aw-El
zsFwQWb1D5azsHGJ+{5Evzvf3;1_l%?El0k^^E&k;&LBR&DmpXzlDIsk0Z`ql!0B!N
z<6-dx$(j(Ov7dwNfjXtnxV*SjW<H9oIt}2TnD1v}o;OveNzqGrU8q&xo!H+mQCej?
z;<5-7h%@>TXIN}}Qdn=wKlh^G!;_2_RNl()uy#aIR8&InE%c*CG_tI#+kbZ)@qMX+
zs|?#NUpbHeIR?!f+F>I-h>k_&=0qW5Q;M86QNz4PxdtyqI0{ss5qs1{gd`UF*jyXE
z=^zSmP9hmu?BjWi`m>~Gw)Ee#J`y-?q>e_=n=PM)qr*}-Fp38VuL)HS#Ro_M@lsWy
zQ0HgBUD0jK&hDb2_ueE=d5mNC=-*)xVj3@>4nJAZ1#sVkE$8seu1hsLm7>0$(xtLJ
z^DbfwfLQXpTI2R-lid{S<(nS{4iuZ{Z^T_>Al2&>r!LAhd4ukEbLjsZ?vyy^EgQx?
znk;}iDn!RDzYQdU=r*P{B18JUErzSxFWGvI^#`Ou0gB&a>^_M}DguCztCXMI!eK^a
z+G6P`2}~lYvJ!~!g6t~i4O{{hMald+etsp!^e}Z)!6WAI_6zL4elMl0<o#RQ0;lVx
z+-4gWaU<qTvb<ukAQes%$DE#p69teU(DACp!2nXfB*r{@&0@sdjxO|iJ>~c7lLzfo
z8&APt^Q4nvF#rA$0zo8sp<wc7*dP^%1qMc6Pp@#|@Be-+JAyPF{ht<q0qjL*1pQJr
z&wsx~y#Ia^|G)32QvrGX^L?YpjqTr#4;t=%IQ?hM5s2F&jky2)G1DjS+y8y5N%Vil
zQbprr1V0`o1ST{3yKursnGlqz(!%t=N_0Bf5Z{$b{0?OhOdsIY;*A&Jgf^!(tC!h?
zOE-Vlw$WK=e0ormQct$wkflDHx0e2gH1Vc~A$aTAKVX#V_1K9(9^1v_a{>q}XC0Dz
z<yWU}$U|ldGabqJOy*;SfvB?_H>-LPa>3(wK=vhQa{nHs^_9Qs5VDdFdlD3;c8W}N
zoZof0xvW+nTI}l%_~x!Jxrx-m+3DuCR^2im&xHWhbZq9)$;QEDuH1pCbnb7k4pD`}
zsfOjP7ZM$os?FSx3p$9Ja@XiX{@rBlsuZb0Yn3p%U(RNVakoafY(Ja~<`vBK84d0&
z?C``{PhN}oGjkOoSi<J%J_5m^uL@ZXFA7#q&L=%_4iAxw+!$2hi{mH-9}?7XPP8hT
z<R4*TTKlugA?O`x{Uz{}!1yB8A$N8uYwPqd#82a^j_Cy6RS%n`4by+_XDih@t_62%
z_!rCne%16UJkiFlRNWf?ijjEPxkVi9GwdHvZf?4!4M>HhA1!q|8C(4R%VhHh@RYH|
zJKF|z#vI#PziAzDSgOu@q66}ObjU5nou0CM{CGY8pWlD|nG#lW`e|(txr(}BCXW(s
zcvNk1+j-SpyjAtm8hAFG<WJ@s|2({>YO|Z`n3_t4hMDpT3JF@Gq17?nJxnQsUufM3
z^BDX^IP*~LsK%Ilwae3^%GU<%>hG4m<&-W_h?RF>|J3Tu5P;qQZC$<QA4|2zr-Qc6
z54v4PME!;s^15b^t@&&B-M&H$_peWQ-WxUuSLqt4{}mg`XNVcOUn$vH)pB~-se=ni
zC=n46O5>Eo(an0V_6(B)9!)bo^lwguTh%B^?lsvh|56DWhmw3{gB#nksbG(Z7i3C4
z`xoJAZyq+(X|F;-ASpvn!vx<gex~A4^O9Y6Zwmj6{W=rBy&X5qk`ZTS8`8%*bLW=}
z7^W~$Jm3DkXSyIY#$c9hxv_mK6is}-eH#yxH*j&|t$io|jVi8R9!-7CTQZzb9$0(C
zEJ~_%jac>bNm1SP-*(TwB`+3cPsQE5>B@8<$4?<FaMl%B6%c?Y7G`vVv*Y;)G$x7I
z)3kQtf{Hn|wj$P463UURY^;E|HRDaLX{x)rN~<<)8K-Avppi7P>%Y9oGop4#wl1Mo
zxn1Gz!QLLdlGk(S<lNC6NX+x5Bp}HeVV!&x)kGn7?@^k;meG7%YR}6s676-uKU--E
zy_McH(J9Jy);JI)MyGG|8TpU*e%$*wddou@MW-xVs?I^G{O5Y3(fGabK#D*o=ISxA
z`0!}w<Ye8Be08W?v0&<tIT8P}qhGil%65l{pR5VL%RDd5`mxl@>ntp|!J9sf@ZxuK
zy41`5Z_g2<_na<VBPWZ^OY@fOt0N0}2j3>Kt_V2IgwByvJ6D4R%@jP!?>IiWrJwP8
z)Qu>xV^3MCs!a%`+#Jiy!_dsvi(AYp^Y7=~$67NpS1u!qfbBd>0iSs*_<rFeR~d8l
z+uv5uTf)}K5d$Hsj9)$H`2OmZm5YyCE2s6?>`RTW9CGBOv`m#1q8pOnLsOY3%(~lO
zGU5iy=8innVP@`jzEGR+kbp)Azg-c8CIT!ZQ5x@PS^w9Vf&Gi|;}g9;Jw|@A!i?5$
zcIA7VG5f+(ldcPbKEyXwk@L8J7KrI17|Bm+aNST{_fjSF<xA$MJu*EFP5Tt&AC59q
zFt!d+A1usZV$|pu8<=00|1)ZTILO{NXIExY&>u%?q0U`P^m{F2MZdI@{QeGjej2bQ
zf4;2cU|s)w{l9Vn{NFeJZ#Dn#8vp;e1^;(f{C{>=EWq`Ae+6hVbnFAl*g^nhXdXOx
zw$Krw07*}S^=?kUAGD27{?8%&_BqCzi<VXl-Ga!$4M1cxtOA%oHZ%Yr;dRW17=e)P
zkHd>Vw9>I*6n)i|1>G&0lttwzya0weDlRS#E@d<Z0Rgd?MumAFWUxA1o*(hnYW}xU
zUo>LWe1@l{bO6W>12&=vg15VYqIdDGy12UL0xFBfXedlJ$EzX7)7G~9znAGYuo7--
zOQ0Kghyl|N(Xq8H$N%}=8r5{UdI5JUq)kfa|8I=<)4KA%x`FnV>@x)#su>$Q&9z_y
zp8=X7EO~IoF)m$3`Lz_bP{o0lz>z<6c^<dxQH>m7I$f&#m=wOI)w9Om>$5hxaCyRc
zuh(yFB4hYU9*U}R%`o<_!uP{hUWPd{%PnfHoi}Wha1mxOGFR1cA`Au`ZfrwNAmHv|
zVd+3m>y(~@)L<V{J9WyX#Kk|&&drg^C$TwaW^#kW&>9~@YnNG?_tYdbCUj=~rmT4S
zj_XJN5{6qDY79M0SQ$@6U294up3*jLU3(ceVjekS?&x@zM)dA`=gXw2cPvKjZEd>D
z50eAv_v{EDgaaM31v)|pV8`_5my}Y8|DG<wk7j_)*@80;8h)iimPs~xp&>o<O%Eh2
z#H&|YCTyJpR2R}`{23B>)5IBVEHzb}VY+laadCJ(Oy94lmXwsOG(IR8sZ{~NgSfba
zLdNcb5_MTg&HuyHR{&MjeqR#;(kjv=A|TyeBGTR6-5}i!N=kP(mqxl<IxiiU?(Y5$
zzCZu-IfLVja5?8W&w2J*Yp=a`omuINAw^3+O3kV^)e$db{nnO*Rci2gu80tD<Vt?^
z11^jT0j}7!RDEwcKMv;cfq>`ZfnLjvrY-ZyfNjfQL-&L(<kmK=TC7T~^noyt`pFn_
z3$-kvNnCdM#1Hfq-nN0y8CNaV#(!~UKx}X_C8?{S%Yj5ZP+I*P@U+zcYATR2`0A@O
zre0&72l$IN7i@qU8~`_!e%~D6tlD@Nx4F4_@J<{x#Zp{CDh9a8#c*Mh1O@Xv6XU&e
zq0q~tRM!wGNkFzl))|()4??||?h>48FbaKFHs<voic8Q*;o&iK7&|9nVw1KRxf5G?
zsF5_5SWXbeU0P(>sg-^q!U<^ogVAB3gANsp+n2O2m)-z5vX1TV+SmwzNXD1Y7#SP>
znH*Gr&raP}2jJXDa=i9~mgka$lmy^!YvNfPJXo+0Wzt@M^fosy2B0;W7%=)SZ_h?F
zreCob#sn%th1?rf3&|PCgq7A>Etnt7RyHR>J8^h$$v;vwX9*TaOTdej4u}yBNGP>@
zBF>4`jgQC`-CwKOgpe$qPEfnm-xJ<#6j5yKb}}y!u5LT-vtB$>ic5|DY7Un#K#~VR
z6=j=TT{E#M%%|m+Tq54t^-v9KR7+=s6XgN!oz;Vh{{RW{XM5=5%~fl_RB!<df8wyj
zBC_SZ+umVx^xQQfBQFr8h0X1=5U=t0&VKNo1{FkC8WPvaQ<UUQH(9x3O6#@n=$ceE
ziffZW`=FJ>|M>Af6gq3dyI36BwfSIKPR@--A{n3T{NOV$(<W(JL^$~7{+N31e-Lhv
zQejg4WitbbEufANBovC}2I#Vg&d}VcQXJ(?Pxo$4_1wTX)dIO?AVCT*mPRd_!4hYF
zZ#sO`;`Wdu^PNtwIsKu=DmeZ2KmwO%G52{ofd~}kT<`ufQVvgDDtT5G-X=%;-a3A(
zL|x%K+<zo5LD3jySz`4FuwqP=s4)Xp0bg6}ipg^wuhKuW^#@~D<9N{;&@=cYy3xy)
zpW%`&9?vk<s~j~>vD62EK2WANvFJuMbB6A<wNs3R5Q;h~iiJT(GXIO4DVg`)t6xts
zk^UZNn%YTf*(89-XEAL!>I6(5DThZRNVHi1+)03tY)+<x2Z_E)e?22Zts=I`lUra>
zH0Iu~$RH*&9jnb>qGId;;$Em}v39b4^Z;6JM}#I=BSkkVfA&Bigvld{i~YTkx}3qS
zgAvSsaP#r+v;2}IYixX?Jz`u<32fMv=d!-R9^cNp)#JE11X#i91oY1#5d!j7#D|=3
z%a;0NILhkfkD83UO^8ttsof~h1H|BWgEy&-Hd+5u5jJv$h9IcO$diESAx#Tj)|h6w
zne`qL|CNUR#0K8Trqzi@sm*3N8_Ut0JGXX%W^Ix45IOQ!XLIhnidy-NI#qJ6@KeY#
zak*As1W0L#D}wsn-*t(_HTn)u4|T6<#RkYWL(A_=LJ~aFSTABe5poi4`V5G-krt3v
zqto$OH)$TMpXB|gibFjG`e1*dL&T1Lu}{G89vstrZi-|^M7+A*gc%`E9p!t?GPg3<
zbFAgy(fsK7-{xilF!~odzV!4tgk9x(pdkLJYi4lrsa4ZwG_Nly9k_S1InNtc_l1@J
z)RNCJ05RHFJ2aEe%*c43Kxs6KdnQ~U5ncQW0TA9N%J=g6aMSd>O|b6^-&yu?{rP)8
znUbUsxL*Yt$*gfSY-TFao`hEY+Etkf=lN)b#TOQG1+^Y6ExAi#SHnKF^SOuHMh^Cq
z;o<n4WncdOxNQEL6PP~=Z7J4&_*7l;chb63@-9!ATwMw^Ql=v)OJdJ2PGm=&4s#_z
z7uyZl&cDgE#eeU_8%tH2DEzmdH++G+Oov~Z76Vf$4W;`wU3!djC3(pYRP6oaYf&7T
znW=<}*QZCrOY~%*mKSHiZuOs{G20Ron@H@q=BQx$?=q%eE+ZzPKi63VCFT<@IZ>Q+
z0d|@H%82NozG9Z9PcKSq9S#*u3m}^1kT@^WrLjwt6ON=0Ysg}yTpY<!atZP-)G5qN
zmxh$f{{2MA*~?<?FKgJ0?>u=LN!0#18PeX(=(fm+HHtA#kLE*7Xljx})+Q`K+Jp&d
zo0Z>Mryy1d<*LbR!Oq>jlBi6Y;iDX<Y4P|npP0#|^!0K$^jYO0@V`iyAl!}&DPDFr
ztOt}mT8(HSw_Tr$J3TWQd8c3Q4Thn=X@uPj6D|~TADUY(&%OJK3?_R=C>Q$y>(@n}
zMtPBE7yLw);qGeML|Dwk?(fC$9OkM7dl%D!-^^lC=F?*nKW(V~th8+){}t4s=z5ms
z;GjM)0l~(TWskAOLRH70HNA*=e=X1PG`d`__<-ydr#j)|5Ajh%PaD88_;Sn&gF&1P
zZrjBKIE9$UGE*=07?Iu@pMb;A%qdo}<C7AL7|_dlXxSb&`cqCyO3-zq!Z9C{<kRx9
zbQ#g#C9H=5kU890`=QoU+q%@<R{7?gRhMw-QnkvK0VU@(0^AKq<=*K|CShc6zA+Er
zx)>l@-eyVi7WVJ;!b@Ng9eV9zz&vE3<I;^O!8U~qD&_;s)h?q)=5MgJx?fDcfPx4D
zD5<P&tG4z(os<To1gJ^k;$ljox~_0HH*fZ|g@9bnMei7io59W-qr~Bmh{y!6xq_+e
zuV!AU5xW0q^j}6Lr0-y_0O*8DXebU&lyrTQU4pEFUsJUJXD;yPGc2$rGGEPFvOU^g
zBw|@DmRN~IeCFa4PzEVw2?&_071X+|jA;|%vBu(8XWEO2OMylBpY!1?W+<8Hq>FZO
zzq8|ELiTfj)YE>^6W}8hvNuUkqv|0i-vxBF-yVI^LaE#Z=h)Gh?;+*MBqV#&wKYeU
zE6ToJUL$mWl@eV9U`0P0?_VFyNa&1?0hI{$OLg}*h>s0cO)u3bw0l$2-FkxHQY#Az
zMYTJ?hVORaU-S|@jnqn278CT?sL6P>kbxFyo#82prTRjEY>(?`HrFWzj;BNH?|0iS
zXt=wn6B=Miv0rMlKIVHoxkkEmaov4BYLysWvS-J?e_8dvvaOH4e)ZhNR%G?%ymNJx
zP2mx(W9f$)Dhj2GSd{2BraUdNiIvKE$BTNm*0bVtT-$_Y+|vqR2Np~pOTIN3E$UJj
z{mjNlc^~AonI|nEFX&A<CMZ4{oz0wEx}9gl$c2fKV{M>4Ztsjf))a|iZ+$9VyFY8U
z#JY-_#)VCqw3tg2mx~XZhqoNv))Z?PC7mmdzZTa>1z@7&(iamDkZDkQtSBC#`>Pts
zivHDT)1M9`Uh1cCdBUs0i%A?opBo-65QDy;7nPN>d{9)!py(2zxL6%+qXOJB{9&%E
zjf?ekz(1vgdGarYBq@h9+*)g=W}gRyp^xOrWQv4*9K4lw3xHQ5mwfbD4OqhDs{_;R
z_s51n4ghR5+Lk+YfwH^xVfSaUe4w}0+V(M>(&a#_w<xRG9wOl~R#8g2m-IegYHkp?
zGgY4ZWtU}|E~aV$<uF1zwA#g^SWk3u=f4hf$mgm*KNB8lk9^pj)}7<Crcz=t1FS4=
zCRfexKy|+>)Jx6z_mU25>x$Y~YQ&CguUDGGCo0S*ZtkK}1eoKW@^B@dXp;OSqU^Uv
zLC*o~Rb5LRO4V(|&uGriK!fNDW7r~@A0reL`(@9qW!<r{(UdqZ7Xx|oR~W1ODdwg(
ze{t~T8tf)-mEo7636^kc@kL3uc6$sa#`L<fJjGN;dc1B@1KzDsG$IO?Uk^xjS^cyE
z6FID(#ALkrq~g6x7c{1&Zmi=!8m#9o_~iwIjJ!={-&n(&DM}S-xe|tlP2AgWW1A$s
z{9Y&oASyv2q%b|3kQL;K@luNo!yv5)5bDewp~cPVk84yLbGNq3517~N`NOptdY`%s
z9~j{v?UxP6^XlXo8JZr<FcYflm|=NcmAokTXN=-G=9oN-E#1Q(c(nlWZ8Es+;i|1B
z^6w#%M+B$|9KQf0pQ>b0)rO;)w?l}E61Rw|USZubgcWKKLcRGkTgB>b7L?34f$A%?
z-~C~?^&q@+t4w>ne(fov<vt>!_y&gKX-Vq1g<@(<fJ^i(G7{Do0onILmCpxS@88PG
zDZYB&9Rd@Q{rQ6~w9uQpYc!I;i)Ht!xyMIFiO+MEwE^VU_GrO%2hB;YM%M&K%g);{
zbr&mAQ#+To3a@WwRkitN;mS~Rv#UasH0oH7Uj46A$JbQWcM#04KZG4VaRvgz$7*cV
zD?$$mUAexR*o24=&N_X{KLvV#-JjL5n&*^fT;diM?5!e?s84)9r%Pf&1=SrZ;y+rV
zgk_A468o&w9aq>3|0^^f!9)bc$$lf{O4ln%TA1j18c{Fgd&k6laJ>IfR>Nj?*?zk?
z&{hQn9v0}?b0u@|QPzE>rs90DM)5KmgNd$me)NBzWXBi-1<VO&$xi)fa44@q9K#=1
zzDL>wnJ5faS5+@&vn*<HCtttik54xC*1`5o&HiXRT@0z^4cn$wG5<T!_N8uIDCg}H
zHf=z?tRd$6ihb1*H<p(CAO>%5E;bwC^RZvTgcMDgzl6L?Ny*6qAJ*P|^!MAD%V#Am
z+X>FOTDxB=%+C*^S9$KRS*VRkP4jQ6W*(1P<>A%m0J7@U#q|6#GBtAYGE0li<{ZjM
zu1dO44{4c!O}pRGdRU8P=w(1#wv}f<bimIGm$k3|mexexA#imK(Co{TOa{>V;tQ%V
z3+i>=ZTFvxKu2<*Bp#?q*ZBagO+xn_7?GIqH20?@Z*b0S<1@T}mUPOMRLaSi6#L5;
z<iPa4&#NPuKj{Pv!*q5g)q0!>o?vYxL)LLipN@LV7HMDBU0`C{MUC-OMrM0xvi5XL
z5358iG{+y-8MQe`Cc}#W0U;RvJ`4#-5npVR;&~kEAr5ws8Djc#mr$d=YwQgU0ZK$K
z%<ulqg6msi?6)|7vbsMx9^K84M~y(4VM6`>8Dg2*uLa5W^H)6F-uQkmuBoYio%+A>
z_6KZHAC(`Ws`|=U&XS(NK*%fhK5_Q5=A*)&5`b+`swn><_*0CB_gaBZ^@@3=_2DD$
z!tYx1)Wz(NcRaT=r@OttTZA?SM);`N(FJ{`q-;LCs_)cIkkbVlX}faie{-=*fp_}@
zf3*JE`0R4NhUx~1s(pT0BCU&xH&><e=&OSe*H-zhzs<9HU*q*kmh%Fe@%`a!Gp;Wm
z-v$AV-$%mt`?`CKKUm|5Uppvo)6L?t;QcA{14iZQz!QIJm%%+Tr~cs(s1;bx#XVl&
z*;MUDNB2Ib9Iu_2+qoki2FI!o{ndR)W^>RNVw)zfvHC>K$r|QoB97M@$Op(L*2?_L
zC<pU-QO5rp5b<WsvWewso4VV3EF#eW#ntg0XP^)jhpRWn&MGn<MtEdgXM)(LgCWSk
zwzmFi=qTg&FW#ZOoYAS0<$dY-B~|YpQXxm8I0Ubx$%$Y`i~MX97$2hhn^%;p`AK}e
zO{7KKedDZlPG=PkeugX1m!bH$e<YbGwH9oFe(Luxk1i4iWj;@ip7&JC*pFuVUS1t$
zf|;_tb@=W#*>66*t=#~9N44TvZT-Bnrf`ji>(kY9VpN4B&F2UO4=><~%6_pQ-W2-t
z4E0&SpsXb**w|BaI7^>Hu>|I*_C}7j#7S_JD4e4Sd7<?2{`78Ur87hl=dG~@?Xo^$
z$%rs$oS|fNGn`-S@Ad}%GU*A#bbR|&^%Ey|uwJ>8T*p$%S6yj}9W@zS?94cXW-uC&
zwRoI#-_o(9$>1s}YJ`HkoWKf2hC;^@n)b=A5yPi$2U~1Kr@ffr)k7!9YI`jy<07A9
z@*#p7R+xupG13v286?j%acltiwl5$#Xh<Zt0KYeI;-C*D_B;iw8~3@KUPW<makS4{
zVlEQCg|Fr+9MV#l5%othZk-`I3ilJ|eTYYF2N?)2I5uNrh?K@N&S;iu4{??;6_sAf
z5FLDOAMfy;li`^!Ja78tXgVGH_VK=Genf{tW5pNxh6hpHF$~xhWC~5H|Gm^CPspFz
zsk?=Gbr~3rT5t_FvD+^@s519dbqufH<3XkwTyiPFT!eLPoZ)8!RMkx=?v6}CMn-#4
z$8`(7<uOJv2V1ZT3w}94M$gu3+fpgC3&%9~*X;uF8Hs~8$_!(jTshWg43(G{7r=N_
z+a6~6M5|>MjznD@VoVnNtPxH$EbSa_QpI00P<%v0zisBbt2m!ThCI(0d6qPl#6oxN
ziqp&TBI4WKE1{Zg7eslqv_>ar=p}8Iyy=0m<cd#($|h=c=MxS$yrIB`Ctu>-DjAab
z->r4ek-QvOOd#h2!^dUMV&Q#^;l0uM(oxzh+XmtyPEL>C`bI6oR}=&$UB0kX44jp5
zmysTI*M4?#P&{`&f8O8!^8&1jC`+Fm;N0FRVmc7w!KmD5)Kx3P(u?@lk&<n0Hn$m?
zwMcp2nUI<_xKc4ZWuijQbAQaNW)IBE8@l%lzr5G!)&wS0vp*6N{8m@ZK&B^F&f_0q
zgtDkwofDMJ+8!85b6H<*8twO>|1`d;N8@%22TaLsCPqJxR%EN^++=0ohUSy%9v}B5
z)=Bx|;uDvF`ftl_$Gvh1tp78lbs9QfPf`3NkGN$jd+FxeK&`CDkT~*-fd7~plLP-9
z%qpAl*JT1E=n=y$Y*oc~YnB5;^s;2MsWrALxmIZ|T+?(<tcUl+dsb}Dn^ZN7bOTjG
zOU!0I2-4a)5>hie@ej903-uh_9m^l<P0)vU(GWti=5Z=mGGW@<j5^lum1=Ej(mQyY
z&ij5BZQI{%gTq_gsva(9P8{t0k0*5(iEc<xqX(6w%w)YOVms=J7@W{chPqc4Lcum|
zV$TtTBgfGHdujm{$uw2u-9ot~Rnax>SB25go{yHtg+=(k&pzy`vfX1u*sFzZIVM|j
zxu{xcTIk`zDx?BVz{3d#V_L0|HN<tR<3VY94>lRAixZ{Lp1Hl`FE=z%Km;{6HwU`n
zaU32tuj#W9alGP_9_#A6vYU7LnoAx0F6M{KXr_>A;i0EiK-tce?@gc%qsQ64B{v=r
zd!~qGeXqnH`*rCdM{Aw?*UQ6`q}H4=%&oSIvmlCvts(sfY@JH|bF7<>VvpnI?UMqu
z`A!4JTy;9EK6OWJ3Hcoz+g2n0fQTi<%eMR|5~EO9Qml+zJQv;KUQEYlTN1leY3B&j
z?ADULR>g!VYJF(7&9Pj;lID4ecy(z|KpTvnks8VSkf!LvQ<*v(968nR!t?8wXt*|d
z+N0u^VInP$4|#sPc6JGxKG+|+M-!xohjg7S2^88M(^2oQ6cm+1%Dx2)=XA{V|J*rY
z5Z3X;q+jHFCsN82`~B<wpV#ryUU9$`&^$VmS~wk%b5e&kuf2oud^E+w>x3LF)nA5q
zIhamgc*wjo`urPC3ICoQpJ>@ttX^>~tf>9kBju%U8__7D=tZ=xZS6fy3bP-kr>v|d
zy?-AYTO6L*%i&D-OLb(rOJx6*%jEgll(2zjPWxT?cwVlBgM^pIVca=yc&I$E-yFiN
z`|+<McvlqO{)1idqKz?cwtp?_7S?T*lCj_!{03jqNi(JHR3B-mhhJR?V?(92`u1ew
zeP83i-&_Q*{Op}Vtcl^}-Me(VA5F_Mn6tCd5G<I<;TP$<(6KKoi~85{#$JTUC6FVO
z=^XiH^%%8ezA7BR0ES!J>jxX(65EIq>MZ&yT3Z?#Y=~0{Y41M=-Qug%jTTjTf9H%V
zxOScA@S^Br)YS=K)EEYjPBzM;OFS?A#YY(UGOkLP@cz)@!v3K#7>CK<PTNuaU0wc!
zV_k3WuL@|(*&TWu=<__cEYuX8D;H**E*h!K^GDc~=|MY|%XvDu@!HU;4_$lPd;liX
zQKg|l0kfbAQt%V%YzC>Fm!+T4uITe?_El*l*dDr))0)HvhCa6CzS5CFPkx19uT@tT
zMlhC<Zl%7!I}+a|lGtd<n1V)_U*C+<VpqvPAtI6-7{CAobn<1#;GT>eiGJX@-0WtA
zKU7+<i>b}MkB?v5-c{4c)|>Kqa`<WO+FC&Q-*bP{iFVeeM>66UQcRUk`{aA=ttsr9
z$2e91?U7nIwhK&h`$R$-b39#I;HvuBX2qGF6yd_AVuejE3P3G#c9}TE7v^u7H4f=u
zGE;s+B9uWr3rlI)RZJY}Fjj|!T}@4Ie1KHF>XgT1R@dU~G<UAycItzO2&xOW-shsM
z?BWn$oAR)^(_w6T4;RSrcbq(V;vdw>+&cltvOWDPj<pEyLbJIjjf!Wjg~s-wG;Q)~
zCRo0@?)3%y!u|<fA+NI*ElT_FgBdRH2~t8}N&!oD1l!4M`Y}00<=$k%Ce>;KY(k(D
z>fFXp7If4~H}guf-R7G)o*JJS#0b0-=j_b-RKJ^7S3MLbPmvN0St{tu?mN))!b0l2
z(%+rfx+q#Y((7+xo(l!Q0(3J9T4n31=2SM`*SySL>H`)AM;m$LIVD8fZ9B5^@8~_a
zBEj5F=?@|%v^U&QI*Q_6mIUtl>n_{(WQ!B!=*N66&tz)bYX5)0{zfV>IJ`bhX;bxR
z8iwSCis}o3$xKUdLzPvB<wl0g(7bapVsP7IA>wC<-9_j3$?+F#@oZ|Ux3#J(1GYCR
z+g1nIw^r)R1-CnXg5^{FGF<$fv*l>{f>S&3TrGkN^<{+s_}tU2`vgA4yZ*o={r&(M
z#SQj`v*UzeXB#2PnySJCs)|a_o{whq1AS?toK0(wLYK`JK%ZF_uA9)8N3(18bxpF6
zChoZBoFu)CVIjwxKmC7NY;N!NW2SLcc^&i3DnljQ&#{fH>WhLh1*`K}kQs>s!26dd
zo>oYpI(&auD;_=7AL?PFGcKuau-)+;{0%4e<OuH~CZC>yZ%=Cw7pIyEdM4Fd3@DtN
zFVTeZjEaR%fXmxQSS_O{mve)&><`YveW<lUz`r7fq$FtTnr19xMV@b$@uIj}F)4w$
zY{!e&ao<3d@<_uEBRf{KKq(u`y4D9E>d-<WmzvbhuTo6uSK(&jf$0ac&IJm}a)=~!
zdkEotPjB|VpW)>xx99Vy#&?{Br>NCA6R*hNjMgkQrXTPAufcIZ#YEmczb^2WJVd`M
zbJc4`I;y=4pl@o&7|x2h?uXF%1p6Z?OIaW&EM%Vo<)ow3u7=YG@D>+*Y?g)ds=u<U
zTuP0gT18<Xrzyc;4Q8i#*y|#gmNsJn@xaS<D4Ic^X7kR6$6fIBL(W}B*<G9o|E(sR
zb8uk0Ul{`zLH43_{Ex9r80ETvPVBVR<<C%js&~wzo_7s%Pl68~Ktz$-w$!V#>Ql(|
z?=gdtY~CMnOW}1oe^Ya+#I$2s=xPqNRm92Xbv_24g6`#-{Q_QB1LIHhMC$2w3;~!c
z`@05Vape`3lkc*Z;=aQo)2-tZugh4rS?N+supUN1A>^o#;ifqBj=S^n!s7iA5@{!7
zaD;)&Z{iF|Skg5{r_QR*H1&{?agk${6?5@$6?7DckW$=!S%f>z7h!AlVQBC5dQOMX
z;2rOM{oZW?4c>N}Y0nmInA<un0FE)T$~t~9yte0j#1m@lv|43oLk@eLQ;{1>QZB8`
zFZnM-)fc|-Jz;$YoO<_p2YrP}#jTBtD=LFB5?)Wnp`+Jc(nC7u6d5IVV9RrhCPzKp
z8M<G}(8Qlq+e~L8cPmPL$U8b7Y)l_nEf?6WI+fD61;U}1fB2r^?f1^<yh2v|_5Da8
zGRY>{HkE?!PPN6Qf|6&MKokGQ?fX=F+S?OxYGB!~n5o%qSepiA6v&Gq)8%RvbbXc!
zT;?bz(~O_TSGNv^FzQf1b-T0_dsF~8qUbBdM}X|VpVbNQS01eIm`Sp_1Gr0-^{95-
z=AD;2=B!mARX{n0q!}vvD}yoodas9XN)cUQxh0m?-u8H`ZoE8mnpzYI@MTfeSVy6L
z+v>p0<D6f)cjpzPEad^Gr!ReTHkMoy+nZ(6N!!{Q_<Hg;#!gV|2yBErLpSuTOl_my
zvcUKi=Qd`oN6`y;mPL#_zo)Hw)77Z_6H{gfR@RqCApgDO!SdfKbCd%};2Wu0jS+sE
z%}Q1kwvL&Rmgl0aO5zkA3!Zjb+pmY{Xtxw|7Q9xsRkg2=kBYYexOTEu0_5e;s;Nji
zpVkgGy3sLodckvZ2Xxf=lASY?3UE|9JSF1zs`?txk~~PP@YZeSP=fUA&!$g_bt_89
zCa6e$QF`&=cAat1w$7h$ARS4>iB@hryz7Qv9+o&%(~RbDj93{s;hl<->#gZVVq<4A
zzz>IwZQwOyJY6C3A+ZXYm>H-cnIIeaFR;L}9n)#cKmVqpa&<pN-|IQ2Eu{wf#ugr&
zMQAH3D%m~XR}ZmeaCnozk&)Tje@3aSvOb1<z`+!@Q^Un*)zq!s?W#5g&WNlk#|$%Q
zH03FVh%4Eo=5|0Q!ADVlo(X$iqZS{eR@}P(LOocwm|bS!WrrAp?P2dk{(o%;OMU^u
zQus!=&`MEwaB}fGlxg#0IDh}oG?YeHXPCw*K?pb1d*S^7H@6*omM#wbmIEq|%A<x*
zXqa80!H_{ifMwhs#U<>FH!Arne0O%#8w_SXwRKbqE#>{K5~o#ILXnv&hj`rA(M>9=
zEC|4E)r+%G<Gxu!{60WKsll1`m%<<6V>NQ%2&n(1pxWKZKb>p|>>ihO{B_ol?+UmL
z$9-KJIaEGYAtO-UTkV4Px)!a;8tyV@ArdzaGb{UPBQf@2bASB(I~L^NT!~l8_j-~;
z&hdOag-(V9ZrGI4kPY5kmSR9_Vx&J_*cUr;?d;{iEKLLNO3pCUjqBvrB(@2+kfk~o
z;5#SsNY&Mh1x~s}LbxaChx%0=G%~Uz%k0U0wdNz6%e6{lRFjP4i9U3<+iXzL>Q9WB
z)pr&AtM%#^bYr-^s%@EsnN_8{glac^>JcW!Z2!N+SoJ6QP51{!>QJpdeI;^hVti~c
zOkB&IJJ#USQn!V%c{2<EclxI}nH%;GuWtZ4i&=PK7eL4N<^Tmv{ZFa9Z`Y7Gl)8}c
z$>AV9pl#*#D1=13Cbb)fx#NE%5Q$cRtgkcBBR0B#)*YRF{LFt)2F}WVF~a5NIw0qB
zIf#P4&p4fAs$u*rI=q4<tsbX#r}i=9@eJRf9ACAde|5tU7*&xK$nEhRcPJ!yrAkrP
zL!lMvuJ&4f6N>6`FZGa4ajq$osfvH+H5V_BosSVd`{__=-4t#?b8$%LgnS&yy)K55
zu>Kw+=7$X1Z?r@+JqUiOO6?p4NLN@zZP3XLT>;8(j$pjEWaYGl^o_Bk1pW5IYWBW~
zg=mxxx$!CeMsT<kXlwfA+E%*7h{5XZz5H8^_sU!FSL1T6Ex1#P<irtakli}pGltwT
z+2gTKT?r=JKj9Gh9y$yq*b45!OMkpUqP@zSy(pZ71T|r#FtUJDlQ`h1&q3L8ejui6
zpI-3e|A6FLCkbxzR~A1<_b_CP%d;9<*vo*=?62+hygTixkdg5%iEwfLH11BSvbla#
z(Vj*O)(7|xvc!ZEXAplCNp*DDIPL9CQQzG4u81<6v&_lm1*9E2<K&Q<{w?nNLh-Mb
z)IOWIRO=TrG(Tv20BOXFa2}&cH(jVh%LkxDpz!5Z<<WhX1W@~3Yrs1BiguI5s2MP8
z2F_yvH5munPJ&K_48K((FmJblQ3lSLkd?}%xfuJmL-2GuWWRe$FqpEmP$Pe}D~}%(
zj;0OSEZk%8o_(ur1xMJT$l#%c_V5dSrJ!S;0CUl5y=Z{OV*)X`u_)8gnHVacPF>T4
zpha6oSfCY&$N98&e=3dRn*Zs%yua>wzGY^1B!ZYe^J;NMJCxa@%}a1SY;qiVk?;>K
zm3$r(9kj)Ds-5@3UN`H!)g9h{eRhE!?<b=0!1?iklUu{;Tf^K;6#Ned{|j;XYWvq`
zr)DnL;XzrRZp^=BY|cVL=KyY^u1(j7$GiIU1kk$JA7Ai^R@hf-smme2S7v5z?D5RA
zl%zS8t56RMB3iykit^$Vh(f*8u?-(HzPLoEX;iBo*)%h|cw%FLa*orn579P=o(pv7
z$Vp|ICRbD?MXI?vZTp(rAd3g}&2C6mX5#Nf9(AFBjH|1odX9CDvVvpg4__WBrvx;M
z2FloQw#OEiF23qVnU^sZ6*(@84-F{a2REv!9M8R9=j{J8$bRdTv$PVmmzq0W{X<3F
zx@_Kb&HT(Wl+*mf5ayaVXL9P-g010${o{|nqK$Y8qMXvX*Lf+BOaHez-93UZFFURC
zox!h($=<#hnaw=D_$@4TP-I&I_Kd@!<ud^uWPi7#OVHXfw$+@OhWGG+Yr?#}>#gU{
z-LT>LFW)_9cJ{tp_Py(En7(oiS<KF!oS#}<f4zi<wg?N!+e+5a)Z8we_o->wK8)qt
z)8Ht(;;<^B^`ufkupOInw@HXi!0$5)Gmy8J8W?E(1k((cmq#^b0}vyZQLtCR|2;*B
zn2(C4ET9So$Hb_$nGgeu3sA9l8~cKx6%6o~JHmw-X_3SkWWE$kR>)GXu+KqY(r3oJ
zy;Bc_@PYY*L*y9*-*sH*BtLT~KhHxNHQRWg1-sYs@4(~fXDM+fHnL63SGPA~Y*Zl>
zLFjo_v5uj(4P2ZGN~QjtdyUZDsZ_ISP`hz^i*B67cb!_Eib01(4)6d_k(GhB%Z#iH
z6Y%|Mp{NeqghWzAa8|5{ItgW>WSkXlvQ&lNYoWNXP1y+>Ha8InMkRFX3R@3ItUt~n
z?Uc5d#96;_aZB{z9_tI9O4W2kl{CWBC{{#Z4RSlS(vgvguByVnT&(%|+q$E2V9)$%
zxhZ1_80=zblrwJn<26#Qh64*M-Tsfz7{WepEVx@2g+yhfO;+<LJ`G%cVxW1^KS>Om
z9PuOs*>71Fa)0T4zLniWASGQ?x5TQ&ecsFKJlUkRV9(JMRP!2H3E!^3u!WE&$a($4
z(l=#8OG?Y(;5kI9l=Dqx$YCK-LnOT*X#5Q(frIpxbrDrsL?7S751Z*Hp_;P%k$dms
zh0%hx2l32EcIM}624_vrJfTPUI~PD^fwwa<A7_&dji40`f^}q8v8b^PZX(SpZ$oPD
zO@M~%8aIZ0IUiFfk2+n<S89JSykVDt`yqAe>?gO%!1B#VWDn(+D&Skk(0h7vOxCee
z6TvFgDv0RQ?&WI~t}8rOPHf{>4~O?OCX3xU7~Ei8u9Mnp*j5h+s$!?pt~sw;4@Ou&
zWkE(|ymdMslLa_xp%d)EXKHzESTjsus<V1P64AaSAoELVJiul=UvcMM8Zx;zHk^)y
z&zw)}EWAFss`B(Q5jd}q)t{UIH|(4CErwlxJ=_-hAKL4NlYwboYaSe;ht#sK=&$<<
z?hz+9frR3-HI;--e5K*Ov2UNx9JM^uxYIx5yJ0XBxxS(3Lte)snym>%s<on&g>K)q
zJ4rZw>&A{ybE7Uof8q?EM#My|XF_s7E$_Y3ilJ?Z*g+>ND+RO?IdO94Z$oel0f>U;
zeJ3>qHfHFRY~OpmX5m|~t)x~?>chvQ^eZA_8m*C_q~q1acGkv96`8GvW_eok=>=>)
zQQ6Qg=z#h!#QcsjHymHZP(Q_kATA!bl0SWkT2I7(&lB$~m4)})!CxPgv9WD;bxEym
zqX0^|ISLiW{2=zcdy$WIoHlrENiGxSGK0KxKr(RxzKSZVImp7zku_<U^Qh6{cBNzv
zM_&T$h-Ozb+53@<3v&lvyWt!yMhZ|^-eg2UYar`p25yp#7(%fX%{N`65*b$YL(-Sj
zYVBgU`gjjtpWu~`U-G&qdffZsBb=Y-4_8p4AB9(Wc61U3%36J3`urz5d92NHsTW8+
z6cqlJq_v(|kfGJY-YLP`Q0z$mD(-~ExL#m1q|p*2w60Tl%`p<(!P|J=H?p)xuz&$6
zZ4q9=H+vZ_{x|H7)FP8|k<+Om;wEW6K1dX&o7y|O<QBO-EXVdG1IUs^hy3((K!+na
zx^sY?T<1-sB!Y{(oz}H=$#Nw|@1dM8F9PZ!QCStg6qt^X5W0W!L@gRgRXQ}#HqO)S
zUI}*O^zGDfLg@l7-z_RIK>GKUk4Q3@_>0>jRGV>!irB#qPT}*Wqs`d|*ZQVu;tdb$
zsn>k!l}KH8nwZVu)Jg)=2q~Rl=;GR=BqGyntZ(-h^+=t3PMKGgPun6(=HFBQw12sC
zv@G`$b=iD<K&RNz-1JfJTj6fs{Bf91023x;N$Ir=k!)zP+%xNUTiXvjAHKw~+|XgB
zm!)`GL=jJ9_WG0uBk&RB`#%~Wpc<Za3RQ5_JlTVP9m2iQ2&s-CCB3_5oikL;Smt59
z#-Mus<I5QJX2bH4!xe1mHP|^m9;I=2#W-hZdu<KKSqf3AWq!?%d&Z*df1o@-Sb6wX
z0$8`e$<VuAy$xVzNUtsqzELAePpKZ&Sks5VPOutE3P)gMhFj{r4UXj_!KfdQqhWjs
zEib%fMeaIWg;;c5>l>4m40jCo1d=3NTR({XZ%$$W2latwxX|kb%!CB{!Q2@oTjwuO
z9<!BC5V}I2-S7?^zP(pcOE(tR+C+~hq?m~fP|aSWTIQtH@<suzlhQ^c$r_1N1E|_I
zu<O~1x>Pl(x;i>XP0J4Cn+@^5qWb^L6cj^_90eGv{cjJxh@he%`_83(=!_eBfVty+
zOr2xQ_VcSR9yH2iFN9cIpeMQ~4TzS~>lO(Z-H+_%GTBU0XO;v3!gATJ{PRb9n?+`2
zE8yusS6nZ2CB8mql}KsoS~yg!sB4CgNErY0DREo{Ow6OnW|TdDLyLU%K8e>WMqhlj
zzyWHD_gS)KSR1G%Qi}dqNq|+L#xGro#>a)$j=_!&VqajLok=LLZon<RWT4j8qPD>C
zL0@N7+;er<k^1&3nXWApZ0>qm8@1Qd_p=tf_Q*)MCT{2KxEXF|FbsH|W~|c_c~v~?
znkU5J)`!73XbwT!>g^KaJA^N#hWbnCEz_f?G^Xzey8YR;BeR-?<Gd=bEGm1UnAZhN
zAafeQI<{p07~9vWzeizGmF-)ENQz>StUeJ?U@6VpwQLycCuthha|%$cC3h1>jjUsM
zC0(gMJb6vfAdP0ODD2|ftSV1mK<7^-I8gmm^cEFJy@I*Y+zcOY<Ehrc@1n|1l(|p7
z8v;1k`04kM$2pXNtM41^#QBl4z~)6lU44t1H+YckO|x28A;-9R+lj(<H;wZ{3p3e{
zyDSo2AHTJC1jCv+(@{~tk&kKU&MO>OUZE(@DanNdh6hGa@T%7p*A*#)HmkOV!J)VP
z732$NcH$1j-GP<D7bjl%)#frTW*46Ul2aJA!r7<vW$c|BtcM2y+vX$<O8}yQaEjH_
zUgG#Sc(`JDoEUw@H}mrQ-yxyEM7q?z=ir>AGn70`%*Fd<lnZ#+Ad}=SCFQ9HLY&t!
zECs^l+soI+^4e-fx2-1QxlMhA^GnYg9os+S5nL@$PV`_7sB*1hKBu=0J76{cQ|OUn
zUb*zRM#aVri%=_P*^PJJC{vwJ;`xr8<hosWcB&wc=|HJ9UrS{av00<~kFb?|;ip4N
zZcxxx>ajp>HCybT(;XhsGVkC<M{vo?yS$X`(u9l~LrH4aiyA?v$jZxgIPXt`ut)0@
zYn)0~vjY=yIt03|5>){vwr_QJGBjsyAep<_v83==jvLS*3#i2R67kau2OZZYMMK|3
z!jW?MJY~Trr8Xq&vK{sO_KLVa7O8B<bOGh%e&oW<c@7_YG+!`yighB>oPU(Jte0h5
zmyyR6XqLz}os>4)&bz(7+HppSae8?Xpp?Y0`a?Z;BsEMt(dB)-3FL5fWK==BYh}>G
zxKJ5nbf1cD>ul#M*mH6N%_?I*U=FGN=E9AIa`NX_3Xc2Im+GdWTURn%*W0lb9isw`
z{UF_)brExEG<%h@DzE424`Y2Pn=ur{B}c@QV05^oxGiV#BA*2GE*rHQclS!mvv19N
zYK5}@f_<RaOR&GOUUCEII!f19n$c+<1!8{F!wx_)a3TmdTSQc+EUynpoKnQJBp|fb
zpioB*3xB2iw8xKU=Tw{2Xj(as^A>)YEq=e1{7v@a=h8w*g4Kc)TT+W9Tw^kxQhHZ>
zJI9xL5YhxLN^|87cZ}3K=4#QA5#>$hpx}A7h#X<?0UV$9r9{iH*PJ=e^e&+yZiIP@
zo8^;l3qV8(J}djym55_@Yo;QhANxBSC{-|x;qo25ZtwL+*hh^R)j2Piucv;uYzcyC
z8D!QyAi~^P(uOL+{H0QgMfE_2(Mud;g7#IsL7m}neegqfpriX$bYh&*)ycv);i}s%
zT)<DF$olhm1>{%kz1?}r#44@NY56D6S%0;ax1z;o=Z7tuCxbuQaI(5XvcHEX2iLRx
z*T)CNpwshUNT48<D3L}n45SFN`01>#Pw&7Qb8N~q(RSN(i|2{I`Vk)mzV*Q~+|5oV
zR>+>Izy8<{>^;9MLb|_L1((9-lCh<iRY6AFd#H2`=6Btj-%tWs)X*X16DeG8q<@?T
z6v=pWlL@NM-edZI@OL{LNS}^`&O^<{`PMpTLj#_RM0sVaIpOkJD?B6X+MAv2I4aoc
z;Bur7^Ti4D_2&Jb{}g2AzSoh76i&&haUu#;!k^}8^vcJW*-3fNr+teoC`Ge)@wkG2
zK)l`k)MirYG(!|)GpgRGX?1bRtYcqC;`!J?lHNT;Q|!Q%N+{E|%pKVyudG&tWni(5
zc+S=3_a3ub3M31?T&^H_T?%K7<Vz7|gstwL8A@q%n?D@3vr;a7uA~vZDOQFeE3i7s
z)%De&C}<ORFOt%C#8xI3Da3-+oUuA@_o2!NKN1aJ|LR0TD4^Q$ismKb>n`Dy>!1~Z
zgp~|A+8d~@Tj{JO6@eUDyJgki1>D}-KP<rUaGKERJUt;awg^?!JxMNOWo3EXt}$O6
zk7aeVM&{+eO^7esqh_xm)u;*N{?=gkneyk&1&VsW+)bANsm01~#BizQ=YL)RZoWI8
z>jWC3t%=Rp=cJsfE~%66&f9?t)|*AE?CYz0ZV@I_{}VT|83FA%6+0LsMv={snj1Fp
z5WKz|921SNr(C=6)?ny^was?tw-I&IO~m`qSzrXp%=M1-^M)lb&911hSiWOVl$&_Q
zx?G&3J3m?b6%E+}j!V_`qND8nHYRq8hr5mDgV}jyQZOuwbKe_{0l+`vV1wy7zfcou
zU31N)HocZE8fSEG&sFj#AeG9X0Unma9PDQ^g8)(=D(FpLa2JrId_Le;rD)fXQxf{>
z2LGyQDqiiwAKY|{A|{Et4bTL91K8Y-gP-DuZQ$UeqUVmnV79}FU*0Z(SQ*8M2fPfs
ziZ&s0GDR@D!DU3+L!~|Y&m^^t_n2=Xhe9!#AIkQZ!vPSwa7h}Nqdd4sL2Pm-h<*m(
zO4a5k8~N|wKU0>s_m4vpJEPzIwi%ox4Vna5S%lt~KDJg|r*n^seCHI2j-gM4qA!fb
zD_?PEAB`aBI9_l*=__0MWMmYhqA8hK$3Fa35NE0d-`~w4CK}~mqBsih=W%RPfk6HA
zlybh57W<9q{W@^9OlRkZ8Ht2UXGl}RKH;0>(nd6Db~I#{&2TOlOw3?mAB7yPoKE>k
z<nTBZJd1#N_2Ttc*tGTBnbaVFN;&DR0;?TrS0)k@4J5B4P4ET@QeEjeG6%H6URUcA
zu+D58G|nX?!NC;_Z?gD~Fns#<SZPWC77<Q&%pkRKb{0mh_mhJyt1#!cXlQCYSuA|Z
z!*2d9=R)P)Q}37;p=M@o-rl+F#?6i4ZjTz*^D=JRo$CJ(`#XTs+K=>)zeOmGPo|o5
z6oM`s6Z~2<z=z@-NM1%Yc`Dh<r<rDjJP+QvG{nUE6&)p%nL+>7*VmD#p)LSaJqcX9
zM9p5;WN%4h^cm8``PmqF>dn@1WMnWmUbv&2x(>R*v=J#$qyq0YSO8>|T8UB|c^3ih
zDSBHTRf$d?*=U5oqp2BYg{Qa5zusVPU*zuH#KoDxhTSyf6@f<ram50v4Znwm^bd{p
zErH`0y!li3?#n2S0?nDX%Swd7ci>Ez!=uaC%ZhVu(SK=!iUc64a~6P2UteC6va+l2
za(QJFnMa}V{RmzXlFjD!q(FTGK~X{Ie3X)=LpRkV;p1svSv$RihkX%h*Brg8jkoE<
zsj19Y?8nS^S1Mrvdt@JwF<JHcs5;~G<BP%LY#6+9NnJo?B$^bh#xJB7%~wX{X<&d>
zV?P%30|m)Y^jMq6woAA1Tix%%HYgh#02&3Qnk$L7k4La*ub<wyCwUPpPYz3gONL7n
zXJ(ae&C5ol^ZjE1Zhr7H)F#!!u=$va%|dy*R5|LcIcZMzh`TcLNR5!`lE^a<4Qw53
zza%K-`7K--s?T^u9F+Kb#J})j<Y(2v@`c(N%8v1BDT|=XhE$+<PWG7BDSv)ll7C%W
zlrC7OO4nX{>AEv~lQt;e%LZkQxz!EYf73I8ZGji_Ggg#jq_y>M0hBpKw{CE5{|-`{
zW}bKbj?jhkE56Lm7A!4qyBKIMo=(g^r7lgA!8F)Svc)F&3M;(n(G;bH$U9_V#Ulk!
zkL8NX<sY5Tec95d>Md^FSG>~ib};6{+<zU<*i;m*%-<FE-p<n0A*~JC+Kx@+I6C+b
zQ-eUb&pj%?39sKAKb_<t9R<~Y5ux8XK#j5nlJzJ3#X7ry>2HzOimO2=-(o7Aa;#6o
zh~o(4tw$Y-Gun?y$erxbB;kN;(De1bcdo3rs2pwY*PweE8*xgW@)tKUI%#PZ^imBN
zJ(o5<s@cZ6KdlGquoVi!mk}0weA?w;)B0vDjZ0TfdHs71S@)wmCsPw0{%{&L+BAW6
z3I@}MSFLLI*sexoV?$b=+6(T!85j$Zy$p6wU$c)JGSnmz3>a8aIhTM=KLmsXy3Qm!
zsC&WrYNsdd+N#NAeod=o#mc8f%jVprUt6x*wQm>%4s)2>%f6-(67DWHf;l-z|Gk*s
zw4ytxM?&>H-K!0|_$!!@vGTr_RPya(J{1d_am$RQM@Op`sR$I>gj=nI2EqzeFcOCg
zcS*IlvvF(zDRLLvtzbnZ6;RL8QJ4==Yq8Tlq8ls$m2bv`*h*p_kmfQn8Dh-mbs<(~
zHbN7=I0jWw8i=o|79EE6J?8J#)mluvDxH2frdNZ5TjVmLN<k-7?qgtrfk!9`{4E`)
zg1T=HX<}*Bs7oIXMz3BA7yE59(QWzc-{@>GKJR!Uk&je3mo1Bce+K2UHN8rn@?f(g
zRWetJtZdS;V_P$pT7y#P3R9=W*=`tPH-o<Rlan}SMdg*)=2&-cV4j!Kx<e@iK3Drf
ziY!0BQp{yJL5)i?ZS97jHY?!G`yC#_1d!;J!CN4I;&^aqsmViEfK-bG*;^ZGFVnKz
zU#C&6GojB}c1iJzeut?`hym^SFxBZzf#XvzubLz_emuwXyo+tg>k&(z;oWPf7K-rz
z3~sYk6M33%!B#fvET-Ym&c_;Sq9uTbFUDIf7${NXr=b;7xF?`eo~ivE1TxHG`ae(<
zWq1;DuK(@Xiv<-iNDGcicYpV_O`ZSG(k%a(-3~2T=%J^l6FCZ(cAUfqR%9el!}=ce
zPZSgu?&X1DjfU$MT*Y076Iq4CNyNL`45CFiR%;P@Ah(%cE404)CoChL(Y1r!_<qK`
z25@Xdx}BP3BfZz4E<o}UMbPafMNhU;*H-JP_^F_1TNj5gw2yKv1V%mbQx`S=hC{Oh
zSgbhGqxvDI<<apeCQf0y=)9SJLjWa$9<1`(S4~9OWc_XNPRJo(x2GpadpYHU4)LEd
zdEDsQ8*c1~SjgHzw?vLelI7aL&Yzm4rs7P<_Rl>Vi#(<FGnBD*gh*a_o9dqiN3!_H
z3Yk+E**3;XdwSaMpdwN6f&aLNM+Z`^7*BsVZ%KdFw2B8&@kXLGAfl^aT#vd|aI-~Z
z{q@1V)jJU~0rLJAfYKnjYl`lY59cxU)-P49YL)oN+uPVbsd+TAS;@*Hy3nsFtwqlZ
zyPh=L8ZZeMKP0{eZ!y{Yp!Xkr2koLaRT^$P_3z~gumZBiBgohk1u>VpON6?{2s_pS
z@rCelde{06hx>0;E|{AnUX0o;=jli{VK#`QT{Iw-`AO`quq)F=0+;=!<6EPBw(8+F
zmm{;(Gy)|tf7kwLxHr@n47;Bdr+|Kk$Yr||bC&1Ze2=MwjD3pJTVZJVu<|C~oqfct
zbag-MHn@hRa}8U*q4_ft+yt*(s@?oH2c}~Bh8aQG7#P5)dNXluoHF9M##qD1ziBv!
zCP5b45yk;UzQ{O#@X<Ac|2($wN+Wy@GX1$!Zl-Yl%Vef*V_k7uKrfI)P;^YPFNTyF
z$Qtwv>p~g#3d(Z=9@iH|=929a$Wzrc9li30Chsfc>Yi3jLi9GofUFNo<k(bOuT<ye
zHv6r{tYl_aApxdekz{pcSgv_kV`sk)CSEu9#K!bGX0^ADOHJ<%_HzbkF3yzr8E#c8
z{-gq~0E>dEXGeG-#a`d&I|}+Ae|~Mp1|$_yv5hd}9vU5++)njbSZdBrw|spWc;2I&
zoMG!ES#g;?vB$=%6t*9Mx0pTmO^+g0T`_-ntya74nD34DUN|^z7KoJtt1@M<NK9S`
zkrS+<TAlN^EqPDx%>I<jH*9(C--mGvq=A*#)^ph$tA7oTXI250{pwu^N&Hh&zvVf7
zx@XAFi7h6F#~D<&>Mt^LC@)DBBNiVq=M^;@PJfJ`>*GOQ|1zbdQH;pclFG$jX@Pr-
zN;?hN3J8Fl>Qp_PV0j}lECcd|>!)&`jD}dK!|`}Ry&DvORSOI>+)63mY|`s>r<!eU
zfzsy8#(HZh>WbP-1na#Cfn*vz-XF857O(%*2x0;}Z;r&gu3wDe*}8qG!kM_CMGz3Y
zQg4u;m*Z(=ooa_Q8^dA`2zK@hs?17Pk6$Uw`v`Z%+F(WL$OBbRZP$kspiEh#t=5Tw
z#=uvEtS90uuQ+3(5=MJrgG*8h{;m1=kC*Ko`V$#TzI0Lm+eS%c63bb{uYj~qesUI{
z*rrNMTpBI(NgZK1f`NRZ*6!|JkirHK9MmdZL+A3@FAh3hq!eY4oTJ?F3aq!GVA9Wa
zTV=8#8Rby6p7^!RqmeCiKpZ%-b_&Dgr&UO*a}EIc#Mh+O)ELnLuDlLQg;M3g+DPMZ
zmQUYkp{_Te<FwqGORR&w*AUfif!KOQW}ev($hCRp$}*wP)7_KxJn4eXVVm)h*U>?N
z10yKycs@B~CtHjzw+yk<V=YQ7-p%*5b58*JI(|HfLHf==rwR8bA-~K2UKikBW$BbH
z@Azc=g0G4o)+)sHjEQFh;#*~^Mo!K7BpoHL&Qw55H)kxjb613{D8^ewp(z`7wz%sW
zq%-bt`2w@`k_chH<cyzeo{us#>*y#^Tva}rk9TgIHO>^uqvM}L#vdy_;Id<5V*y4h
z%rr@QALlZqU56pS#K&Q(NQp_UUsjXNt1J5^kSuR6_?KJreOILU`ts+vC2Lw4k$MrZ
zuoqycW0O?;2R=;+02=@YJ>c#tVG9bH$Lfa0)NSAbJYY=c1)D?Ot8-jZ0Bz3q%bIyr
z3>Rye1sxRB{`B!BgZ<-aOX4dJ&xSBpjp+z&j5Y;*3(j(=Kkn(+1R+eS5)^Z_JRBsf
zvCh0_e%EGw?hEP^UQ2qMMr|?FEGiq_GThh__E#;ybRqKH2~AGm(uB>usmb2*X&a~&
zzytE?y(b#+>L_%xaGhH1OK*{im5N>&{F(yKB{j(K_0I0YtFxBo#prCN^++T>T(T=3
zQz>rZH4`WW)kR{&?o7E3DhS8}4_=Uc=|zg0(Op@HBp_qv*8!`xGOB_#({Rhh&6^34
zrPWPW%lx+D+#JK@si#Jy)oVVlEJabJ0NmDVxjB&1cz^;5V936+f#>yGx2FUWQVJbw
z^wF6~+^;x6?$iZ%YZ4azl~Dq|B}r^?(^2sz<&91CCy(6;Nv#_c`Cd(l6PQU<x&(XK
zSfA}3eIBUdmNbeGGTrx;J)VYQG^ev0de&2|@UISE(iw4*uj_J%PR8D-ruBqKX}AF5
zt>ZEs6}15%#_n-%3_vaPHITpI;~JS?atHLIdR=pze<r*7-+wta<y^1G32jl6g|v*0
zYEpl`8&Oo5r$b{wx##+Y?D+!#K|+(StyUsR$W~6DqVtBdZjr^*)Nu60ePhVbQV9rm
zrb=*+`On^{IPB@lwE@=bKh#vp0t3w0sa3$<u&+C97XbXyd2?IpKvQiD9TGa;v|y8A
z<Nei~SQDjUnhk|l_jlhH;#3$OqY&&YGk<^a)W+it>9fWi0I~pQgVR<U*a=f=8^eRc
z;ndvDYm)^TpSnk@Ebd2tS~#z10|fhezAFMoH3oG7NzwSE=;M-N(CvxrYX`nt8xlGK
zQ4VQ6GfE(O8216g%(hPwGERXUk!0#K>x+q8?6U&29cihu;NS^^cI&zm$bYTI)0Cyz
zWkK|a(1F%or9}TWpBDHKG+c`EsFYA-|2rq6p>`1Na(Ouw#dqkU$L6bw^Yi%uSPz1i
zk}f4Kt+d7=058jyf_$x^r{jz{VshXc?_?V&BY`Mde^w=ZHLUUPHLCFa%i$*Xi^lhr
zg!5@5g)^umjfOT8lFMrTUwP*pPWAu)|AVB^4rP`Rk-f5M;ACd+5t6+*HVukoM6wmf
z9@%?DQpB-kldNN}Y<`cU_xtnt{`b4C-ygs0cU`|ON|$ra>-Bm*AJ50*{&+mcEhg5A
zlrd2I)2SGt#5dz}vO2aLC!c-(RsYi|MH-g?!@8;h&coYK5?Qf#L7S|n6~B4;<Wns+
zM#agoS6Yj~lsF(uylZXk)+pLc>xysfIynML3=49y*F@_4G{VE3Z3n79uU&k%=IoN&
z_N|?b!&pg=dUft9RHeIMJ2I?}%w6IYgY~27j{_1<Id7?vJPy}H7e3Q39$zLEzfnui
zpYh}8j|Z@MkdkyI(cWoHe0aF*9IU^8_6kXTJ5g3a;S5bIKUy_-T7c*a=JLXP>+b?L
zZubTK-_em_iE%Qp^y~Ve&vT*^@AaQAdC9fer0(ewJg<Dja_c&8U9w`p?srUh!Wc>V
zM!4&vrGHh=7a`Y)LnXs~^C<Nzbe?WD&;8yx_;V+;Z~QVhBm25=pF)J-J0edf>p>Qz
z=;NklVLxE#8`$XL$|F^2${Oy!P?QtHS24beLWL^bVMeh5KOf#`xN~TbsMn3=?RTED
z)|5DtC?XTWDssVEssn>sYGKQja09~yC#$Vy*l*yIn7Hlkk+|tVDYwXP`DHnvuq9|0
zRwiJ=gT0d&{?*J42Cx%zCcdl+3Ir$b8kA&-SYlLvxvL{kl~plbB0Xcvrd8L>hnl$&
zJH01%QMxq4Hu4{sTP|D)9pd;ZZi(si)w8)7D0(S4Uc2?D%%>Mq*Pol+Ie5MG$kt?d
ztT%nj!ZQAAImPSXX?@nDlwH^uYAxjae5}-t`36%+&QpBRk>Qu8NR1@Z-wbzT6CD~D
z%y`EB(}b%@jCckP2B`8%6%~WC4rt3>PhA=f^wrptaqWyYRAfzA>_ci*0<NY!Un3<$
z{9z8HbgT2!M+SN|`+eD02cI1#E38bK9vM&oROg}Pq<w*oX`99rf0%Lr9D?wD{KF^x
zIF`wodD&Mw^*=OT2+QVP`-$l{9S9qAI2*^Y_$cabmiWfbq3vx$nA?!Z8LNjGhU?;D
zBc^e%VCw5YXE>zt-&);<r9b8GbN#!}DJ0V&NSxRhJ*YxRbu0f!ClK#vGS*<!dy)S}
z9Bd*<cc2!TH=&$yS6xoT?e##|adO!8Ynai}-m*mn?p{`Ftflu76PxR<)t^*Y!57Go
z5rn-&KO@dXc?^jR#-9@=0KM=g<t{_y;rSBV+1|T0`*G=(^h1^=v{y(;=SEshYZgi$
zhQq3WhZ0=qKK2Qtf>cM@Tj}biCvxc<Y?f!})%6V8zui==up0DaEg$+!LTf?HD_pOs
zZ)D4iDj_2!{@5agS(sugU-%%S#F57QZ;A8avHQKWM4c<}&OE~?h28aX{lrB`l&`uS
zTzjhkiMb4{TDh%ET19L_#hg#s7DNi}_6n4@N;f-SE?srbg{96pPlrpl1<0IFE>WU3
znJ%SnY^^<l#oDY@PWJ<B5_+zT$EhK{8VG*6V9t!7Cm^``vxEeuz3sG1T~u;4w*)s0
z8A7C;)|`sDno1ocy-)fV^c#3=-9fq&YDyekWD~G^82k&?S@Vk2zvP;oCA^S_oqLUk
zx8fzI)j^g=LZANWr9WrQ4=jxcX6yRD(;mJ7xCtkRDYJxG*qo`>Q@?S$L-%UtkNg1b
zRfMSPx^nz{X7K#%czxXD*DMO&$qyYZMvftPY0btNJ?k(LgJmal;23clk^IQeU~G3`
z-}-K{qxtpy_*YfgB%EUoo|8%|_<we5+_GP{zqkovR>$y$)B_=Xol+zF(4`Owsn0!o
zOL&sJoTnpNbC<^_?|sYq9bh0fJ63SFrY(`yfdsaTe&QfHEWO@%luy0Q9C}_tB9U7B
z$L;~=BiNwpu-s;(WL@1dm@smoa_9Zn_c0`)2c1oTqGsc4=CVCs)0^?if2Y=TaKpR}
zCYV(H2W$vee5zZji|*Ny$=}8wP`^h=xY|xI`9pw?KUt#8pz|CLTd&VMjvTdI&CfG5
zgfKN?y7rU%-J;t&At7O&qZj7AkL>7}8)B_n-9~kvPq&%Cw!)cJFc?wSSFnl|_5fyF
zJaZ}Zj#rGC2&G?BCT!s2Q`XS7a9S(o%9okfz{M<5Gyav93+g97Up#5#!;{m(Sl=8c
zq$FxtS`i&bWhlCL8Qf9Z#%*<vMir{{j%vD4=@jEpC#8uAtmRJa5nijH&EtWu3g>^}
zza8=`doi`oU(f!5i!~mmTA{4fT_mjZkAo{$!<nuiEPm7$Hlx?cB|rQYV16#faO0a_
zORT50#21{z{`#hZF5{rB;|ZVF&OYykZU4zf4|)&ho!X)WZ>y1FsGpYz!D`$BuNJFq
z)X#z|dJI)im#Rhmxw(m#NSluTyFz_)Sy>~6mYL4?%H7O2(aYtF5x2Nk2A|sXNPBXs
zpB&F5SWy*Li`j9wo{OLMr9S%mc+QJcg19VTZA)V2aaVM*;MWkdd%l|~ucyBge4_96
zQW#BjL~Wk(iVF&iW0R;6`25j&50(l18X;P4R?ZCkE;I4=9!pnm-Q5por)dLka$m{X
z{DJx{mA5?5xJyv|63VnU+K+T<P}M@)KR;>Am$RI5n!xB+4SZUboK*aG+K)p@U>_s@
z@~e|~=+eH3lW>3S29KGHoLYTu4aLTv7qWsY_R9PoCUVBGGPM+sz2VM)zzl(x7&}R&
zM|tJijELR(H=-i#Vpp1K7kNU_y<I7}b1N=SyC^qhxBK=7C7nY1a<?tIvhSTZ8DHXL
zl+8_$e*-qGb*heiFmie4(Ni6m@1Z9A^C~qT>ugp2snW}mug)^d5IvSSU^zF;we6*J
z6b~C<aPsi{ENS;p<2ef~?Ka;-o2^yV(6~LJOg}qmrUjPgBQLVlWRiL1EePDzxmH=m
za&<9_nrCj01#Tw&PVQYBc^-|jIqa`qKe+h*En*Cvg|X6AT5;2u((ht3_QYGbiT}(W
zCg5(xZDg??J^Em7Tl3;uGGZxCJa`Q~m-WYv(-4*~=T+c+%TzP5X7>xC9M;!=f7DRK
zddmA1Zo)S_)l}WEEILF9aockI&#Cuh7?@<{9WQ(nBJ%H^<MJ;!BXp}7EWKg%-EDvQ
znUGhNtr&k`kow8o(}}jW@|Amvi4`v8y_4k?w~NM`W^fa{1_zecL7rvEEGfV;g&(jR
znD<YmKg9;UX}{}yOJ~0L%*b*ym)jI)q+QE11$UQ0p(iE7K5gpqw4M~uRhWZ&oCw(A
zr$XhuujV(`eyZd^y^t7t6pKEL|K5E;I^gq)vb~d^rJ%iw(c+RARK7O*e98YfbXdlC
zjEN0<+?T*D=;s%4{oEqIIYW9lYEaE($rw5&dRdUdJiD2ae;gJo+hO8@Yfa!BfXrQy
zep7t|3TEb|8jJJ!qb@{HIv1enZ(1r_{MyvHyO$k7Qi`1?qN}$Xa^-8O{&!)p%8l^T
z4pR0muG85RiBsz{iH}ze=XY!m#!{z#L)d=<i#Fj|+)ZZ{^xU<XtC}exC=gV+;*@#s
zklV$ia48Q(4miFC)~-G_>Tiergqh@%m8VXiJXhPFZQ(>*&QMkKqRb=N*4De4s&`R-
z5;SrACg6R^%|(tho6zXYkiPWz!0DwG52`4p+FVI9`x*6T!5Zwuy`5CT9+`eCry+;J
zO7C=Dgy|HHmoK&sTPsU~ru#GV-zG*k!_(N`;*Z)G_JXa<r%6`KLCI`bjI}S(J)*}a
z*oH;R?pq)Bc3?S*D_t-HlWuR>8{)HRunRJ8v5!otOfGwLDaKMlrqk6YX<Jw-jO3=a
zkeY@j!0}9ve*fD>7nwi@NjdabR?H7tmBV-Yt`>V^1Iqd5dD1H^oNUblO;b<X`)ppu
z1@j)+E!R$_WUU>A*4eeB0-Ol{!%LFq^!i~RuNXa4OC4bO__6_`-hXneBuC5D#id|R
zKZwIi>V8jKVPWNEDXE9|NV)kQ>V&;h#(#cU<?#*Ptu2`!x3~#qt73l4@1TBvHe0GN
z7k4t}iE+ooI0bCg$T#j8Ja)Oc+4U-C;aA8v$ET^^BrUYM*OwQf#T8A>F0J&1m)}ru
z?k{PyJn*1=@q9LbjidjJrXEMt2nUjwm^f>5ev_QJ&?!i;WlyY9IUG+VK%cp=sF(Fd
z+U~*eu|1*m<4x{tpG1l$>+xuWqHfBBt2!veW|+)tSPqsK^rG=*E?=P3sScmX@-g4Z
z>+;4IZ&-hi$*&ko*)8^C!X3+kV!$_u?-msC>9IQ6EfXKB)+Th7$@C5|&M-`W6Dz#0
zGa}&pQZv-=eq&$Z94EtxZsP3z@S8VlBqMt0D-YRgiF(pB6nx!(&~b!$NUPN_1nRcf
zv(WRYh1|JgnRXj@g_Vas8WeuJY`lV+f(JXCl=~8)a}rc_s{TQ#t@P*%ZK>RCXtua-
zuZcU{MBFT}*>)K-Vp~=)$C2H7vSo89Q7lL>LQyk6#oJvaP1Re(Wn*VYYr->NV|!}r
zi7LNuXz6d+vt<Dr3kC-tahI!a=6*^s^+3!~tsjHZ*?U7B6+4)U%tc@Cj9Ao6I{iMv
z=Ue!wP>#OHS*!64nT^BAP|J;}1NKa+4B7tRp$r?nc%M;EwZU-RYLt$;m$O;~-*gZj
zzZT6n4Xis(`#NFVG8(Bb8dT;&CNsvtz9UvyeKPy_+_SMSrwk$;c7I$A@xm%KP1qi>
zSooZqi%m0Y)JO6#x`c)qEYgn*=)5=^c%ypdqrl>z{)`Jj6gTNzURxT?{UpJ2mwoT=
z-&y<?ReL-0#)S}P%*zRW9lG%|Pw@#hzp)F-rQK{3yH?RgvTW*|R?MVpGx>|p$No-8
z|MAiNwE=z<X2TLzd4JN-atHKi374=t7!w2iyIs%UbuABmn{dWX&lo5h81v5v6I%YX
z(LB(Ic3HT#*!I@oXs4keli@{Iktd_7$6>pOmeCpU`Oj3+ncl7BDi2CGd;^2(zB|;a
zteaM=h(=+u>*5yg#((&CEkIU1?9V$s8nyR+eXxqkp9S?_mHFr!{#}!+5lIflG{fha
z^=Z)z%t{u0&8?Q{vdKaF2m2fO>E;>kZ&@rBk#m_NDy8egL}{k2DEq+3sh*zsyiWMe
zJb2{qfg=bX9d@f)UZ!L{n9taFlodIXwSP^uOEcY6{s{TlUrtGpIzOCXw-;H1>->(6
z3o9QlE3tT5vE5t{m0miGy>8KRKjM1Zj%m#BW3t*>lAe$6xI1@$O~qh4ve1$7OeS3#
zbBm2#m>skX{oqK~*Y2*@`dxHxEkAB{Ez5*-<m@d+8S9M`(Mb|??|kgj+3O;)yn2b<
z+d-w$BzQxrGEzgi(zsS`cQ<lFLgE2tIZAuBi}&pfY|Id|j_R@b^GSLMUZlOx5e;QC
z#vQ7zj`vw^J;L{}!?H5IEov=^HBE-Eh_K;jb9F2y#|*kvm#fp=*~Ha51LwK|=TyQB
zS7%+SbX!>0<R|POTIU=Jn-zZLHe?xhu<-I^YV*14yI)sUf?AB`856oy<D^Br8vPL;
zjUOF3wc%yoHTSX4V`}MU<)1?FKJv4;20BBE$0>T28q+a1%&Ri9k`*#4HiQmACpm74
z;c;R!zZx6w4lOTtHVVG!Br?qDUR#bRp0|%}r)Cc`zJhZpCxOH9FnX(-{j?@43s(m=
z!rvp^T`h9d$F!36s%Me-zGH$0BNH3{!OOMK%PknL%M>)Z&uvHecGnImothAY+nVck
z$HzG)lj0kosKs2ePz(YYy9xXy$cQ&)iFrPTlI)gL{d%}=WoCX8-8iSLXq#y5k4H>$
zCQ8Eivb_O^g;RK9oG22bkjU807p3_`Q#MTD(C5dA<73cELi{LbIm)U~lXoaI*f5;l
z<-;>tD54DWJI}1f3C`9BUJdxL5@wX$ifar+l2koOX!-1oCMxEI-sOeJyAeIr$hX+7
zsV2v@^%857h<CnaNC{fejLCGRR`T;1@KVZ-inr4GqlR}q{)Ev5!-?;>Kgq<WP!p?h
z(`{tr7Xjws-V`iPs6LrpGUUjJ6dOED3&>jM01J5RTPV-z^*-ZA{P5AzDK5t7%#@gz
z!D>Nur^6yWvB8pFxua!{GmE|5m46Q;MT|$m=Nglo>L~H{m`~ZHO10xbf8bW1pM8FJ
zFfk=VdV<m(ITOp?MPfQVZP~ESX}S0=H{!a5axovH4))^p`zK^=i?*rJrbss4Vw+`5
z<@74^=7vof_WRHcuGoKo95W%kO3E!_3;`5&w}fvAji}F?b{c;-v&6deZx08}nG+Mc
z^rk3X4y>}dlPXvx$NLN28Yde<5JX_;d)JbdQGZN6c0|FWy5_Z!uUOSKftawJf3M)*
zW<eM{_oG0Q&JEnv+*)FyDUmy>GUVa>^o&#SLADYLkEmYgn<})1Wez5-_D?|RQMDm<
zP<D4LPi3g1<s28|fq?WhmA>h7qS+8Moz?0fCpJ~%xU~@|34DXdRo`wpjiL;exuQ#5
zIR`48?rGIliVyPlZ#8mljGFi=tD#Vf<T{v~kwW#O-Ce58Oa+g$l*=A?DbYg#EnYkZ
z5gUPZoXWORQtaxdk3ZHg0PfbT`1Pwk-$ZYn&Jc#wsb(~ho2+gj8*Q6ss`9Q?)`m*f
zzFyj63HzkH`Jr@0tGsiE`=dlnRH~e;iEk$E=nra9UQBsAFyP@d5ERRPJm6r+T@23o
zpjRXEh0UUGo^mTX=D|RbriXQU&n;Q(Ncr}dq{Lk0-NA6SMR5jXkj1c8Fzo>y%vxtQ
z_4e0nyiMO&43$6@NOg_F<N-5sQb*P>v8S%~P&)KYO~o|v-$SV-pWUkuCe|N@I`rWp
z|AEs;zsIJbTnUp)QZY}8e3b?_cP!67rI^>`6@KE=1<5M&-UV=ctaRPCU+ym1Siuh}
z)?ot63cZ=}diQ1cvch-&HggBpg%yqP)~Sf@x#G&#ti1Yp>$aY94U!`fT*?;Aik96=
zV#4oYZdP??md`$Msm&;<Jeq!VRoWqIuF}tEJ~$$VS-wnHJ@;_DV6usdBFW@oi4!+t
z$?0&O1!3p?a{9)s0XmMCbBD{KjJDm8)hny7V@+b4LW%6W4DPl$#2tAXaE;suC89I0
z9Eo6zZq?7lI87Y<C`zP48Tc{Aq{TJgA(#<w-a9H<KH7apQ9iPyg3(1#(&CS@va+x!
zySkS4nyG1rjre?S<_md}I?p+;KkXRU`0x^wcAl`WpOUFLncZ{4`UXof3$=YK_glQG
zG`gi3KE4?j%?aG~Qg#>AQMLCPe>+y7*3Tj%p3M-(zo@$umP&6P9;9aKay8@r9}+k#
z8`#*vQF^xn8?SG1(y8D-d4CF(%>rPz@bGYr=!a%Y<3`zX84~iUu2uT}{{Cz0`8twR
zRAlPa;i?3j*uvmv)8+545IRWoX40p3Um86zXuY;>Li4~-<Ki08pf$NBK3m0zEym<C
zoNJEf?@z9If~~^$YB<^D%ZyiFPi&gOr|z2ZQRnv?l?{=dALoQ^iUxwF*Jn^ihFV$~
z4GnjYP451=v^4?61J-N#zdZF3qyBZ0>c3}Z()|$2V%{82+0Ly;BPmHGBz81ctM}8~
z*MEf^^zNtjMLH)-f(BG5`pMe=EX!f%Dr-!7BZlktk$B#0^h!Xpf2hd({ObTZ3`uR?
z@A+JxLm!^^sh`uY228k%u2dtCJg>9rqW#=fByb%>6u7;&*PmVO=g%hD*lZ`HOc3w~
zt1WuYLFL`=ysyL3p6#;f)Rj3VME73=_UrlowZ11|r%?Ws9k*CjV6DbmMWIxTl15F6
zFLgL$2#vGZ8TG#>c=`pr_M^zS?wjbWyZUt~FzDGRNy%z)@pV;)P6g+!&Hi3%hyVF;
zj`#cJoSQ#or^=^q{8%E<wYQnMdX*Or!#?sHTfR(rSxrtI|M>OE*WnaW%Z_-WCDo($
zgj6y2t^)~KiD?ZVzt+fiG|{VpNv7oG8$Dl6S2~Q7xv1Xydy#5`1MR*GPY_RS1iW|q
zXZUu|nw#It#*p1UksQI1CuQXHDUqE5-{;=(P<L%<cV%5t(eWp3>}aiv%L=r;Vza66
z_hbAha2MA~2|Lg3?h!!lU8~`%Q><n2of-W*hj_?vAUE3}-Xne26{{8apWD3#@9<AD
zezNu{8)V_rKI_pevoWT*^k;$2`x_R7geSAy-~K0f)l3exa*<}s{nLgN08Fxc_^B-o
zc41iFJF3-){@VzCCtv#Y%j}%18pU43o+%q*_k5gY9YXnCg5?|5(dc@A3~wO?`Gbh3
zRD$n(jgp(lK0mlGBovX_@$ytKn}5`>V*0|sKm;ShZLPzJ37qYc4J;PUOta=7FYW9U
zXcTZdC5p30w=y9uFJCfo;Qtgm*>jWdc|eN<;}<LtiNu36)Bd&xVjK<UwEq3MUhcoF
zsp$W*ra1r0nlkt=YpVLctSQvZ|3D>HRQ}5_`Tzcd@iZvLb8$>nv&WBh^iNv)S;XSp
zr?PSw;Ky@0h{zppQ?s(NMlj-t<CsuPy;sbLZ8RexA)&V3>73uLRq9$=EJ)&0r{J*7
zqH=nz?^BN7ZOhnztr%derGNI88vGh82MNI8l5$F|+~j0QaD>FC#urOCorDt{g%cvw
zmPd7$eik%hFCuz)4)N8c)YKwfmxeQek*T9kZ28h6w))|-VcRnK?#NV+&8MV&Ebg~f
zhR28GyEmTS4lrK%d=AxyKvqANw5WdUr?%_<8H33m6{7|EfQ`#(ge4r1AWDGe7rM8e
zbd&@9NMr<#UJ5{aUA+s4Y@5+)ZMQM+GDeFZ@3}qJzB9Auos#tC^4NHSPET(Izl%g7
zHKGZUX|(_dJL<Krfb0U4efC6SAnxsk4P1Y5IgsZ5pITK5mqK@O<taqQIbfhW?`~Ki
zRRA$?nUd1d;XWegU37GM5}^1ZdQy>>N~{OEegz?Vh^pL{-!)EvCdUqzWcN-XGKhG1
zc%1jQ9gr$;a6UlM&EUOWr~iXgLETJy^3Pa8{O^ZKCL$X&dUlRsok4!hA04#K<M#MW
zz|i(RypO=f0ahwkuR;ZgyDi_oDFf5hrHM>)7U+{2*48<_Q+OE!is`Drfe_0t1xjLy
zEntRqaBXTFn-=l*?dHx*k`*RVtfjS86<+VLwUi5O$;QFaj=hMteF_Gtcq^Ymg<hUh
zDc-oJe0Prwop&l6#2-~D3b0K8U_OwS@99Fzk)5xXyID^WKfihB6_99q9GMVUlJxG(
zW?D~q!1tm#l;q@8b#!9IcYa?Qal(jEihGscC&OQ1s%FC*Utg-(G63AMi$*kFGOc!@
z=__(hgB(9UgxLN{W933&>t$-{oR1&rX@nf_wMKK6Ro+9mFmiDzAMId#S%;It!#}{G
z`7K<eKrhWc%SK>?J}0q{g9hWM+Lx3>KeQ)^uwA<*DI_GMQ|nc6@zSNuK3%)HE;&wh
zK%(YyP)HIqQ&%Ir5|ikZULX$Vfk~`WjVd^%lRS`>RhE@~N!qAvWRyHuY>_f{bh!0;
zl<*o7N7G=O8??%gf{|8w1>y1-m@5&fbKU9Lo?A=Aq@<CMq79c9+)#|q;Lxv>jeyq9
zy>-bi^ja@;{q5EWCOJkj4Ep;^Dnt~wf+|ogqZn(?oY+1OSyWOIClW_HuCM_FF*8!w
zX<9y^m$2UL*w^ki@F@U+R3pL}U<kT;XZ&_;0X*4)!Ds@A6mHSPMF@d?3AWgs9?EEO
zba*gT3WFw`(FLHc?O6Qrs%_irh@>)LE8}3t51?O6O;anK=RRiVRXzaZW|Z+2jcB6e
zvu5USq@=xUM694)o|xyKC*LEOa7f&ShK8OLnR6l3LWK>T!U(Tc<|HqAWaZ9uoI4zt
z@&2Z28VjIN0Rvl^=z+c2)ZC04wL3dIyBpm~IAjb!JVtrd2@pLu8ckoJ5aQVR{JK82
z{A!hnL)($cxft_AV{x(iK<eA}lIO&?lL4j5=d+J$OBAabcEuSm&>-~6n(#6-*Vfnh
z;Wx-G08{4!Ygk+Ff?uU83|Mx(a%DsBP4&#y#)bwQK;CkDWnyh>s|>)2wa+g=Pis7R
zlCe~G<g+=~Ed$b!hx9Jo-~&a)KH%7F0)KV3SJSYU?r2*kTaHj!X2G?J@g^^?YLPjn
z?>)DPb1Qopi>Rm`s6Dt!)bx*-6trA)mP{xum|t`g8NsI(@L<eH_#edt$9M|3kqtm9
zM^}h)aw>sULT|nX`i-@e8N%xfxP&Bi9=pB)XjVSdiV+TRvNhtCBw#y?gSQlPTpVY3
zAU#k%V!JutuLMIR+T76I{!?DXH09LP)LeqwXeaO&04aVe`vIX*!Iv-hkWCD`c-x5p
z%&=;Y7YYgq>6PQG!2zmo!VrMR#M$lP;Z~j1<kt}B+C+K=hQC9A-U^}o2I$Q0_~T#&
z+F)3f7cX3x7@+01RI9KFAiYYD6h^Hi$19pA^bjWdAeEH*0eoXtG@vDUdPNG*JF&dx
zZy4vAC=Auo6%#IBxx&)Ob(x$z4f@x>ZPY!Tn1rOne&SAWHsLr4?woggreGFal+!!y
zSLyhtpCyjZLd7-?2+tVT?@z74u%H2%aJX|*ABB1%90xq22k_MQWn`j<9s@-@BB=ks
zXpRAat1%#0KlvW{Ab+u``<**YOssMY5YpDxo+|y0!EpA{`q?>D=-~!Gqs7lp#^jm=
z^~6kL<J_bv4FXJi!?tb-JtykRaI`rg@T+MGaRSH)VHd2a13v<*^4%+z39mgr7u;_|
zPp!8n;fYW2c;i+;Wk~~{1TjFx!NCDyQawJ?yi+O~t-kHQ0xTd#q|IP)Cf})(ezB7E
zB=yWNoUA4qUI^Qd`+-5ttU)?pGhF`goY+Pf<FPLaCTGuKz2rxD@(g9_%8I?s&redp
z6r37TQvRvGI)x2=cF@xp@C#^iBZ6_*@-xeT?D|<?!k%}+PaOtHPpOSAvdbh?EDb<T
z5cQF)J`$@q=MGr%2vi<IS&;PYcO%4hGVl?B+LYNPkRa^R1)jKB^AjGO^Y3pL6q3Yi
zdYnnks}@U#P>m<IlNT1OM|}1cBT~VtUOa!Ug^SOoW@dB#M6{64xxTQ$-6{Ax-lr6j
zxp18LcoDAgLZ}2SAqoxF`PM;?_g1?*f#XpJ+*Bu1v|d9ncYPq_OHZ1D$aWz{Id88R
z1NS)7ZF!VZ%%k{NBe)}OeX<_lX-kqY#KFf?Ld7k?qurWz7B2BQ@bKQ5ihiJ=pzh;S
z`^dzkH&<6s-*d49+wtL9FBAUCh5sQX-#`1Oq=NXrr#t;vg`H=^tOerW3us)nE<)mV
zV2g=+3}Rv<SVwUDe!@`XLLgX6oMsep+-+Qdd3!_QGT@E@4u6LFwldD^L5y-)8JCJl
zvvA3DY3EZAN>CI^XlLCDkBgf>&6KVXr}3=Uy$(R*p6zWQoiGE33ww<ebfIv86$Q|c
z@GmxfcvvrlMnnS?2y4+Yn2NCL!g~ShpUK)qkC74L+v5a)w)v3PYJL6z{-fwtGY6(k
zIi!(p;Td>y@-3V^R}N^LP#&xGF0tz8$d)@+{Ay2OupuH>A%MC!jMUdl*6gpfQu022
ztyBEu14QhYw9Nf|FB%CSy`eI@{<W!AGHPmT2VBHMjC<m;2Fa_SVPefiAV!Q-y0Ddu
zagn|V3}gd;3gM~yjNq_x|6f+w7w|`jam}EhAZKU^q)L*wcgypt-teTP{H-xxL%|W}
zx4(Y<!nGT~tLJL>mQ)=b9nsCU1BFdR;~E|wm5de;L1V?eD>-VnAAx(y^gTM<9U41K
ztFEqg>WfH7$c<vtwA|g8tAs}&yL2ffEbP3?Y=@L;ntW%a%Y3!%Xn(AD-Q3(<EVoGj
zbZavFesi|-E=J<0y0EaYJ)5?huLP)6VDO6E+}yZW4^apWg9YQ&*stdl0ygk9gla};
z35Nk6u*P<wgJ%js3ZSrTJ28MI7qZG&vca_sRzw<ry?!eD`-giwb|CgcHmkgL9C05E
z4jaueR^wUbx;&aIIOe4a{!s1FqmOs)+-afq{qt1UwoG?@2GbAVaQ!R=6_p71v(clf
zGzXY+cCzz{Z4>q^JUoTN-k?c(M{7KB9vBx%lDz-K(9_drIli4&Q%E#~CwbIF9CC&G
zqgveJP?;4#wLK<YP;4A*jb%H{w3GFjlnuFbD+pS^xJ7dqmRiTWW7kbHsPVvEC#ric
zvU^VL;lqcSu1qD7_EjT}RU>E)k^Ya)y)`&%@)I;V!3lVeCJaUiltd&W&bI}grO9$l
zBB9{u`4-O5`aHC*6bAhn8sTF3uq`XNCi|tOM0RFo8GwIxL7d4}vwM}A%6NuWw0t!5
zckbN!-cr@_yVKH`cVG~JUXMO)Cigxq6J8m8+6_}`)=CZvMFqM&s!v7zpW7&dtQ>(b
z69tBI;^Lphq50Kjy8h|5ILy^+@0i8J^r7d&8ShI=V{sVva0Z#*&caiWQ7J1Y1ET%R
zRi1kh#SW9w?AnE?UE=2L@i%>aec81>NkW`~2QD01+&$b|7E^t2J3(>8eI|kI54YoQ
z*@L~Yqdt(-@G0Y0ugAv4Wy4eGyH7<$1U>r(aa(U?F^{dp!6(UU4SE0GJr-{6PA8uP
z#q?6!k*w`)6jDW6I-qv@SFmHKAp~xXd;|7ORCsOCGIL<#2Rff^DNQ|x7kxR}DPSz`
zglb`E^C+lF9B!w~6w>+bKTNZDw;Lnm^f-3}S6isJdJlh3p|&Ri6_92=FnoR>-`<E>
zuGvbANlf%<_W^BX{eUOs<HsM}=}MYE+ov338UxPoGQ|K(?pBcm6BAmE2F5jK8C_tb
zF<9hC_fp4)yEDEizUml1JOX*-bco=nmTesPZB(B5P^pc2Wnz3hwG!Pq(Up2aIzy_N
z{>MczKoevuOPXvB^&EtT_MR%eDP$LwKmji31|uVCFv{(X+tTpdKphwXv7N5X@bRtd
zay`X4p83swLrvi8L4ei3uBFI?T6TWCUwQCns28LGAn{oE`Adey;}z4%*>xC^9uO?p
z;<q$d6Dfbqr6;{p6ScE8BS9bc4#%Me&!gfm=1JZf#u8Dh>+9VF=R`r)msqMt?Pl1N
zxcER9XFhI=jjG}l5YQkvNgBpzGg4Xd<FL9~6q7N0`qGUI7*u3^yuOF~#8w%3*uOU#
z2cRT`AF;m299pNwqr}n43Fjn^-6X$dB*;8bo4$O*{O#>+$GaWy03OqxHCv;eWT2G5
z!C;gi)d$+JY0&_DMTUiq3;lt7gshL>qAS&@PyZ2QEM%o34!@p{dd#zj(2D8U43(-w
zhLTPW)b1hMDg~8QuiUeNDP8l$cWe^I3<G!T`m!}SF`Gw&V9}yOWn(^`=z}lK&1`&p
z#a#mG>Ip+&6A0AX<0KQYX^29V<Knv1vy*e@e7B#;cdG%R&cOG;r3NH^MCt_w>De!@
zE`eKKqjqyeK~$b{7RKxevyrHAR?dEW5DIW&Rd63sv9Vd8M52bKD`pbcUq2C}gEUki
zZn1i(|HYti-GvJmqG+k8sc#`2EA-6%B+yWo0nV?>2NEst+9mMwgCX_%F1EO-+;N2~
zpCM#&yf?lQ6dKw;T;a4CUAKQ<#X<QG#`myDt!j`ML)w_35eLS!^|WDRx20uW_d++{
z{6b~UEG`sg*d0fn68S{Yfue&p!Li6e3M#7kfkRLrrm@rRN6;Las05rxD!x%Rlw0Gu
z^&kRr9o@%<--jzdfBx(-li=n&UXM35J>3EYl|@%iOD)Kl+!F}0d-`^A-@Jc+1@ot3
zi5WRqU?P*-gNu~g#K#eo9n?+Y;{N`=${twO)WG3N<2kDZ;W6K%g9lr7W3?ki`aJ!X
zuii2%u?h-mMiiq~AB{LP(}02sX=rSmnwinG{B3R{Pow<Sa(-V!1~j7R5S`bG6xwEM
zaY%FSk)1hPjMWvmo}>j%`XuWl`X0JtU})CJEs~r)+XrEv329fe`3`K(kfC45_s|RC
zZ1%7>+*hoyOV-R$oq_fIg1e>Mv@tsd{M06{a!}MYsIEIxCJeXGo32EM?DAZ$Nl)_G
z%^Y@xcmt|M3GJOwgV2?hmTn`%P1o>UJHGmBpM%}GC{2S-NLyVtW*>r4drWKaE2vgf
zRN&I9mlqiL;0`bjCSPCQ?Cyue<oAH0tZYZIMR&31mhHfzD`tu#S`Hj4?$yDZwJ3wn
zmZ7v_8xw(a9@2i5dK?GPew18Bekv*|jL5rzXW5H7G(l~O4Y>_F1meck)2H|t8xa?5
zB>63^WKahB9Usle=7uYZjzyU+nj}tL9mWL~1>sF}2p3$4=)A@&lAxgVXJH`4xG}(@
zEtWs(&m|g~#Q~qyCW@$-7;TYbG@t?0w!sAzSajbITl>a<Q*O{7pg?}b)Ye021DnPU
zmE{9abwX%5uWG3a1g}b3zJSf(EaYjc`$ym$m2s&Nn0+Ic>6HL{e0-=4p=-VOu((JB
z<C8Vl4l;PC@l`^Jf{d1SX)XnP3@)xPArB7^J@yvMTB)sy21FTTloEu|&!3-o5fI>E
zs|y_0G{uChj~^8p45ASTf-K}cDK&b?V{%#3^CKZm0rj{}9BkeeJ9YXQW>fW}K)aob
zzDut<C6DP##_tpMPkw!QmALmEQgv|kJphx93JXhy-b6<1@3++60?wamne7N862znm
zlo?1HSK^;t<$c_PU4-68RO8`FTN$OzbiR7vpt9|cXxxTBKy+haWzCY#!#$J7WQZW^
z@L0w}26Sp_hRb~4)ROOFNx$f4WA~@eo|U-FYaIM3tIKuGrO09=74OFkSnlp?L>phx
zsc<xlnd5HnL(5U*>Xfwb@`%#Ye;jr-4?WNM(DAGuhi2PlQn&?F!9NZR6T@p>Y|?y|
ziJ2Lf;tp0grT<wT%ZD&DU@ZVjRuNQl|H?!o|D!LbAl4@qVLsdq#KJhysdCLdTn+V|
z1xeF;c(4a)N-Cs+SyugdlzdOZPM)L9Nfh(c7=)YhnEepQ_oU-mTb!U0gxLb9_VpB*
zD}kPt!WAzmA!D9cy?*`rV4+#?yIa~H-PfmY0&^F}9Coa(?(6HTDY!BTJ^dYUlvU$S
z=25Ytb^O<F-`0ZW?SjVVcU#iKm3Ynu-hFoROc(fDe&dD{J9~Sc<X?hALpjTmpld8Z
zna7HGmVze0d91!~Jy)R}B<~u(XKK6wKVb{t_B_cQM7*{?f$xD5WJVq&_L&e?BT@n5
zJHI=hUsg^JudLtWx{q_TSiOf2!$AWq?0{50$YsL?-jzqkDOHApRbNi0;nA+<0n~Az
z*c5FM3rYh*Bu*>$KpMux#Dt8H4yDbVoqZB~{MPDAP7n$a$zKzMoTQaRSDz=Me!Y4y
z1doBMOt`3#!N+&39LIIUyO&U?07-@m<9}0)YH|kLB!luKPE>;>&rOC(^hF!`;Ofw<
zpbHBM3Mg;eCw~p4%ZHf6!UB^Z#}M`W$M4fDboBqu#6s|B;68pfxEA97y-A17BRrcU
WayL%TdvmxOMBZ1tmv_hL>Hh+cdYgd&

literal 123536
zcmc$`Wmr^i_$@r@4^U7s=u`=5lx`3N9HhHJ=|;LiP*GA^y1PU=hY*l%h90_0x`sK=
zDE`kmAKwq}`+CX6zHpfR?7i=2t#z+^jbG$tCGTL9U_&5~JJ44z6(JA|7YO9Wo12*6
zCpy=Mbip^wx348%g74tZsaUNy_;$<om4-b8Lhg_Lx)!?h+Xw=A1cAPMq3n{dIqmEc
zad_GGn?KzwetTN+Xy$dakB_DNJzGYG^>;oCRSL?=KNs41aK**HE{Mv1>ZJH_JmVgD
zo!M&5dWae(d@lrBG`>t`s6h|zSkz(!L&)uv$we&p4n4)p9$LfuvkVBNaf3*Z{OTAo
zOM;DFC;HE>B3|9!Bgkwfoy)bq$BnNQcCY_E^odEE#khI}t9TR6)uC>HFX7c;9Ucbj
z)u9Xh|C2XnbLaD^e8-m<_uSx9wm}LJZ=x(d9^NlQwhq%p&d`8v>&mV<+gIV;#-+D0
z{rn#`MPOrN8$Kg4WQ!Fb$DnZ7el8jwBtg|KhOuJHV^}%q8sr0Bu;{=?0~HbyqL95I
zS2}Jf5Yu?teiLf6=W_G*ZRJYqX#w&ZBDbFN^dXIXMZvkvr)$cy4gM}4PpaH}gu`)N
zS*{%-FUsQw7nc+G{h<hU=_c1M^4yk(OiD&ZMPEO~!_!lc9E&0{Ha54eP5`QCYx~S(
zQQ3BFZ7sZ)0w<@+l56tRpa~9Fd;k7Dh3tQFon2i@#>QFA&CQ)tu^MzbW@ct9eZzsD
zKkwR-xM)gA1zaGJOFKI|YjnFQo12zPi;J%mW8JM!w`+IIk!OcVA=gs=Q`oKfWyuwj
z%WqlLJ7@drZOJb<JZ8rX!=!Y)7ge2ET+E7LgW<e=>2v-upvRP>^KHqi2cZwYQZKEn
zbb(j;R{EBzGp*q0ZWA`z-+E=jdVI8U{_%%{fbAFhl#2y!rK<<{|7SV;zgB1eAHQ+u
z<{D(CbIk$#b>qVfXsX_!LMUAjx8LZU<H;GjyhN_J<cyspUL$NNfqXf~xHxR^S-2)*
zmOksKpsE@zn=G`oM#K_dJG8!b(<n$KTud<}+pjS_h|itrm0}YmO03m!`*Aq<z0>Cr
z3)fVku1R;Vh3+GVVAXm)C`wdciG0$b<yW26^4Qqe&hBoeCJyn4_;|+J>FqTlh}1od
z;=qfG3p_l$MR1l<b?KxK`K(1yB$ixVri*zr_!=)>yeGTsPu<bpE{W;a)beE(ih*H*
z<5e2#-x<&iuFAL+`oFiGvb_AGz}9(CxzFNzoGv#7La@dxxniYhi@jOilrF9PYOs=m
zO38cahKM#B;OP|>H9ui=<<pG4hv%@i((mf(+S%6^Lw$OF9+Qh8_VDniyFB^N*34*y
zqP^=|kMZf*nXcCxVwF!676RK-tSbWV{9ERD^ma*tC1R!P?kR>P@^`%l1;5p2QoGCZ
z-k$a1n>4NNe+9Du2YOL=JCN5cx#HD?B{mV&y23JUuSug(SqYm#I6}ImUauXuxL*ol
z&9G7i9t&YP#T4&k!eCCdNRKA1(X;Cml|FlY{FLwiYzv<*yHfVU@pxRb0oDj=C~avm
z1LtOHN=_V`RzaswL{7$yd<T6vCiz(T3wGMjf%IaJObKRwcOG8I?9LUW=7=bUhgb}!
zdpfx65S(I2=DWhe@`k)VcQ;(~)>(C6INjy5I&sZ-P|-8`TCCP$bw4>z3*|2GkPKq0
zp#9+w{`z1)5oVP{xu(JH?o6kRxg~Pi4m)vHou%S*=|I_t@HFbn-Gq+fClJqNWueJ+
zXRKE0zqN8#Fd-N0SD?)B8A>UbkK~6l){I82pPrDNVn`5}>(%$-<}6I>b*wq8Pv5U(
z5ai7L5OE!X9rzdEeXwVLmLTH<jS4(GZa5zHQyQ@mZUr}*m)Ghz@=s@PcC(+oYxb6+
zqk*L4{JU7lvNZYA+Io>-H=)NCrZh3eWmVhwvCNlo{9EixD@%Cxq7Wa(|9SB!Q|x1C
z-t?Q#UF-#q-v!uAX}MO9Py8@`@to!cq{{DV8FnoEj5_V1n+o#E%oyUMwic?HnA0>+
zcjp4;5K{<*`--sSdoyXDN`+r<&aV>Om<qCIzXnOVVl5HgmG6@@P|WlW<eN{WD!xN=
zw%!DgMyo3VjI@6f_AWeQbJ)PTuM^6D850YE^j$$mN&?PJY!yRGl|<eWt#NO>vLdOe
zdi8MSdh61+uP`CDgjWo>l+dIvzafjHhGMCDFYF~;Y$}sav%RJ*4UMbZ(9r%3fmo;f
zWz&@FtGv5j#J4f6PPgtXnPgXU7<3ie{ECh!V)4Z4BDtsCed`*e=zm_Eqnn+}cT8MR
z*FAKG+;6hu+Z3Nj{_Yu4(c>y9L?Arvf$E}p_y8hrcy%$#_l|`Q-_Pd_j}9=yOgl~{
zmA@wI<akPQ_YJ3&aQ|9!Fdf55re~-yzx5OYf=6(*eEgUCw%?Mxy`!HP<gd?8P;wXy
z<MaLaZNi%v5?ItH#hbVEaA)Wuv^av9RjCq}w2;Mfhi+*Y;`92sJKtXdp%eT)vc8rG
zwK&|~uQcD7_^7$tA)+BRkB6wLyq<3U%d&)FrTbRm1WYmesQ7Dwti}y7j?L4);na4S
zq!3)l3F*I5{q}X&w4TkWVU#&B)9fK)c$=BA%=z>K&ge&ppX=)u+ruf3Tkwg<w%=z!
ze7aKq-BUKA+vd}`$(qjV+3fV$Wfoo7nbhQJy4DrVPu}9|)hC&cz(g^r)SUnG`4RN5
zyjiFPG_7iH?>gLu_~862%Grtnjg(CL29?1tEuDQm3Z{k$=G>N<Gu*eMI3JA-fd?5Q
z@>dIqU?|IfE$*A~cp2y&OQ+>EgF&fRJu4}5dBWSOR$S~2V^b(6tvC{qf;2A4{}om8
zO`9_{Qml72is}-Wpo*s2H(I&33;z#1LFPMntay;~S1HLMMpSjqZjZtwW%g|CAdQHB
zW&Xb48@Kq5fOlZRUH>Qfm5Ljr2nEafUP7Jtn5Z;nSVPg){tSFx4>npNR(r6lmT$Ni
z42Pt2&r%ShACwRF{I5$ZlKo~YJ3g|!$u%LJ{mngZCTE%s?nl~%HXd5_$q7#pKDJwv
zC8|61wp#SGm=Nu||ElC?j=y`CHLHB^D0hAfd2**|1)jN{Wj)LK(XIorjF2YRvAdN<
zY0CeNBKiXbU-%LuAE9Q|9f|g;4WHqec}pjTm^=0Chi8OJPAfgQe+|O&KL!7d>$eP4
zvD8B1S37rqCNUaG0y(9jPeWD>o15l?upqYg|7Xv&%=Q0nai{is!8tLtxkp9`d!%AD
zBNxuG5dXXt9~Ux<c~w-M)e*POmu$!INwoC5UJO|G1mq4u!vS=En*R(o@xy;_I5lvC
zpBg&)-KEk*LyQp`-kU*NEr<h|6)MfX@s|+1mmPQ<Xqlmy#CT*R^nN9)E9M#=NE+zf
z+MX?eyY3Ct9hQ6V+yS_F{sWii9Zy5en+{mZoz|quCrUM(+}~4o7_#L{)Dxg9wnopn
zxH+8x-(0tRLWUX|+u=Xe8|`@ZnVOz@s{XYg1QLLCwK}5@%))AjAktjDhI9Ff3jt`j
zjs6GGW`?dc2L}GbZ%13|`JViv=IbBeBRduC4qZsru1$mL#i{SVcuv(lqFm3u56W#T
z*h=ffu+v&5GeXTrEBw3IA&^%8s|~^Sh?7^rMu1b^y`nI>qg^i8T`+S(Z6^2oNn3c6
zty9=gZlpncxlJqemL`X)($l&ka?%*{6D=iHcL}SWGv;#lE~zbzvyYn7yIHS^OBF-t
z-kfU#oI9*PeG_90E`-qL2f9~+`1mgwnyfDQyP5Y@j0EHjNWftglIYrd6BU0^JQW3=
z=bTDheov19hv@-}xUxlVVtlk2pwP&ec+ZZ$cYP0cplqh?zPM3g9{rgK8`;5MTat=v
zf+ssk?rjd<hKT&W0`JC_wvYxr$s-#jZDjt)N-Pcb&0@Q8!S`V!BhDv^;_&U18?{`k
zL)@Qq>^!I2tW+xBX^mS|Yo2t7jPJ1pM!tj3zng%q!`O&|-S&hL_-3q-l*Ox^Jo}=y
zqM-jrHKKz=6RIZ6BTh>R6W!s|Dp)`2=Qza>w*W;y0kcQaz+r4oNd@m8OYzod!0JcD
zHFDcoP|yTfE990;;l<AQ_0yTB2sLHxa#552AdyU01uI|*9wQBuxyW6-;r)H_nH@sB
z!eyl7rO~KxQ90&%;TMZVn>&MREgOeJfx%_2qozy7)H>fs=jh~{kE-Ji-P{Y~&kCq>
zQL|4SdwN3gc&Wh3cWEvD)UeAdYo{!(!Z=1sl>SJ=o@~s@@Rn`psEn&7sp<u-mRFGn
z6Uu|Y)}y@U?4r`Z?Xp3Zugs*B>*6AfOPVtGmw6!`o{E@ppvt(3EZylW$&yj#t8fn{
zFSpf?zt^XnzjqN|gV0=YHw1UQ)T$iY@-%OBQIoibAZFzo(`7(DLPfIjA`KL4?u-49
z7;T$7H%HwOmV?V9`v<N3q4aaR)-;rw`HNx}s2?oPc(lpgr!w4=FGY;G_pU*t{(%vN
z9UuLRa6B@d$t{f+-rcXDe7dWGJrhU${z?^H3i2sdq|vLZP2`=u0_jzGj>9P$lHxDh
zYZGxG@;|Pye)sFtrZhBsoV9;TL&)nzcm`02Pj)!tddw%X;^QjTmYwcGMAqhIM*rNl
z-LHU{po-hE5onbl%umH3(6*%kf$Y5aO9wvw%6Xb<;Ve~0g^c02MPbyfDuoM2Oc0;r
zD`;yJ%Nwyy`@GsI#aAVvK`*JD6yar_tQcagYPqDj^4b6*(t5=#KK?|vF&(%0)x*oo
zmg!<vEaQF$Q0LksngGkShTnnAlK;Dcl=sd;%oh5ekC@IC+bb?L7m<=&GVn{42z<N;
zsh5?)3fF;{Fe1J}5DGjZvLdV`az@@n&AstLz7lt6>M;f4xgIQ3*d24%kE*Q$=nuu`
z;8*^2_7tPJ$5u6KdaMk>Pj&3u9N^0u=_+4i4nsGrTw^DA7L0m4VU?Dyu{EA!rsX?M
zVragI#5b11`g6}9gN3*uiY^XyZLrht=n8UTI(aoxOq5Y|reky`pQ$6aAksi;s$AY5
zE~jE`{G(flK=7p5#c{*CGJD|$MBu|eWJ-glU=5!Hu$&oh;%Ju4N^WTi$kq~s*6Jq$
z+0~v@xmuB0%upm{g-eYAPX7;jAEPD78h17wz_pENWSU<sHv>2Xl9_-wTk}vWPo>=g
z{{h9;S`NgA&F9~_=$P1K1u-wWhMq$pMr2pFj?Vc9d*ABI=4**iUK($UlqYG7jBFKo
zz@ErXza~=huhCrct#8cw$POC}N6=cdXBh#K^*(h`*5)kVViQ5yc#A5Os0c<r09bhS
zzPoPxTr4B)#WVRosoZIxUM~~^3TC7(o1XtzGNwZg3(=5I7ib{7Qh1fx%^Jz&Hlnd~
zqX$gUioh24-&>gPXM3l<L%GTWYMEcx13;mx2nd$#qOh(<N-#?YM|dCvgVMB<FumMn
zg->w54ewOrhF~NaWXj~g;qP;Ts1=@iegbGICks`(OOZb#$0Cmw6_K^!c=uE_)U;7G
zMeb%1qHj0M76QiIycfv{wL7mt>Tdm8gH*O0%_@=*6Od8f5bQGlOy`r-N0*1?v)^iJ
zmdBT{1M+f50=-9rY-SigmO<p%t_s*k?ab?fv9qIBzQsB;PhulJpbXCaOjO5TCZb7y
zIFeO@kzKI-gW`3Oj(>Mp_bI!eIF5<**HG{8amS=<gF(9tJ<7E$_l%4(%;vgpwI<wy
zNd2SCT|UtG(Sm+tMDuneV9PH1bzR3|YO|hj;#T}S5GnkB%jZK=9(61C`r2qh4vE#$
z0%3;Wq$+5P2qZZmbShT`)wq#CO&&vyA)#z<5z=I;D57pOzPBG46{boy07+5)=bRNQ
z$$w>qvc)l5_wjM3H%6}j6ri*Ew<jX?+V5@;$I%RxTeUpMOpubdmW>u|PRG|POwx3$
zlVF}a4ol|jA9_CbvH8gYHFVspeL5*B(Mek(9oqCp!Mg9d&!EI##G!M#<?m!RDA{M;
z6TU|1v1!<RZEL(nD$rO)S^m`r>#Aed!}{NkV?C#YA3nsF9MKG>fj&KEQW;r>R|~9k
zf0^#=d8G(`YQzy2z(@g!ys`#GO3*US^PqFOv#%?Rvn+hwsZQ;;%o!a|O$BIL0rcfA
z`dK)bPUDeJJd=6YCXlLvlWhiMz@MFTviqR0s75szhRnHJ6raY3iELkb%}J)cpGckS
znC;a{j0=&v^Y^Z&j4^$C{(>^JMv{$axDN4(bB$fwIPj3w!uN><9o>&l=4B0lEGt%z
zS+@s!Wx5WgV0`O^#j5nrc&5bG=49S7jL4UAO_#g&cmUN=y{;!%@QdA!J)+!dhKM+r
z7Bu-8Ct}w{=x49+yY+3~sLB}(CGf~^Aj7&PE4S6LSX+MkQb&C$3zVo)-c&~DgV4<p
z{=ONe>=p&7FLPU%!WT8vW?P|sbcbWnPqrL+un*2pntm*0<e&SvtK^zbhkw2ap$Yj5
z9J5^6Eva${B3k$NGFH|U>3PhUcy7hx+#)&@C%UQc?v#OrGAEwv+@ckCRjhr(2s*dd
z48OVtr;g8859SU(PN;oNE=08Bsx}|-=4q$D-*!7Fd?1Srp}CT0vlSz+vo;#Hg9CC#
z74OstX)huHX_-4uN``0Hv<F8HtubwXrpvazgug-?9H4FvyGY5)&Dt0NlxTaD4CNYV
zc@Gu9CqRF<gHNm^PSw`^mGfDKgV+s`=kp|gz2Hjb<l}nkT(e{oqi0o4EDG9w)@XtD
z^ls6>dmU2c{3iTX;`>XHO{T*wR4`?a!!tA>qiMh@em`WM^jOjGaC8qRMB#<?r$M=`
zp^`!v0{}6oVtRDq!7!;X9pKECN(f|1>~96z1`e$aw?)|-a%iFZM6J?1E|@G73vRm5
z$%|J<$KltM^1}I!+PO9NYHgqHXeWk}qbYJ6h1BVBV|G9aQaKpGH;kSU34A5^A4F|r
z=G9p1?0lv4%2M$31f!ahvB2&jbTId)*fy5f$A?<g(G;YI<)1=MlqW}-73T*$rC#&W
zds}=38a!8_t2dxXd3xyiwG{QM%{&$UZn=yaG)K)_8V&EmOiZ$9Y6WrdV~Uei)Bvpk
zj!04dC#)*}`|l$(e6LgqY}lBnCs6F$cb!2i0KvcZx8NI@uf0g!tHh+II!)O;<+bjS
z;%^WO$HUhu=SjN;k>CDz%D2ymh6}8mffSV*d)`;50ebgHYEFyE^#A*1qYdZ5;TeUc
zi0}m27~_NqnCR1E7L}I{Mk8iZSv(8i9j*VpBl@N2nv&hpDynCsC2)la3QOiUxV{W`
z*hjK<m~!AdgRw;?&0qEDGbv45y6wGeKpZ4dcye})W?NHvwLZibl&@{2Eagl}jm$)@
z1?({mAs)oX?$h7$muHGA+s=C&T2CeUqG7oKuEbB^LFb=78}7~)fos<5F6d~xb&MSv
z057$Xi|8<9$@~d{xCm2l{+T<pW}&P?yWajDTCow%8$n!Z)e;1&AX&szs-b!q5~?u+
zrqRsYVxTZq`)lYYYCU|(oZ|CF%<@*+yXDml6S8f<$U^ty)j|r~5YLp$NwiaoXS}7*
z?7F@~F$BIMZezxJTp0HC@xn&oeR-E+pDg;nnk=nQ&NdB$wLJZM@ggPpLpi|`-W4xv
zD+~J9TZJtq&uFA%go+Y2r9ZZ*DrA~Fj;!Zdj1+Z=eEcBk&c_1Id}Wx)KLK)JoqHqY
ztcTBq4-ATy2(CIF`~)Biu4^@98{m;<59LlJFoB7OoaAr5yZ$<-uDmgNHz9QFJbr)X
zDh%Jy&YI^Dx90x&?JY!!?3!6ect+!_?MMMaw6J(RTh`NB&DJ!-KZ0Nez+#!l4DLjk
zd66s)ML4stoDhMYc$=!qzLVfc>UpsFSQ@x$AW@>}F)BO?wRu!5qNU*>!LOs-&OQMD
zpx|0lK(Q_`XKPdw&!>vMox~-DU_(x@{_>st3m)>jJr~<IWKKQG*oKs-p%UpD*XC5H
zi6cE}6|_c~`lC5<4tlD>NF3noBrYrLY_XmKq!$lD$i|9T5w6nl@NQugf@$;_hQF$U
z^|Bca#W-5!98TgU=RouG9s447^4$`%D5SNW3N(3S^gMUVWNFF1QWM{n07hxoU&m-`
zx<SIdvUmV@K{B_!@!aA3I~=Ik<vCzJ!P5V@KR(!xfI+we<59QpPNpZm4OI5`b%#2w
z=Z;0;0}d690JbB`yfcpqFErg_JabVYo-fdAAF`2E05s`u)AUi5IU(y<uSUt0R|73n
z+60r8qn45Q_f$_W;x@|`_<opus_64n(}FV(?mpf+*ioa+^E@UM{5DlecDI(+lSg~#
zDU1=CH~hVJ-I+K`D%cHJ7rUEJvXydgX*ud|9ZyuxQuSQ^MwPi9dI+LsZUcL0r|AG(
zz`|eHmG2$Lv2<JLxTbl2E{!2RGU!OfoihY$P}N2jt#s+EZI0K0i&6CF7anR<n8YJM
zsH2y{7<#~B4>!_*Z4%aVUSKXQ8Roj8VmMYu67#$dzr%oZGrUunai&8t+^FW9)9RxD
z_<UfpZzcl>u_JlAzVp55c23W|+8^p-Bgin~^~0Y6Di*xE-b+Q`31H_>DXZP~km0Ut
z&{K)iP5+thet6-T80E2-v(kJ6!gAHAEKOzL`de(&h=bR^MCz)i12M*0`;@w(p~%(x
zW159dE|PzA4RqL1qJz!6?<`?6<nQtmWV89E`Jv2bFsAfzyADB#h|oB_^E5sLDwuk$
zo)Bx}Bup^#ex<a=K^HZ04=K5KC#b)iN;OZa)GKCsfe<g#8dpv&@vfKy?P27Mi+RfC
zk+0$_90=X*zdM+&{lVRx(SjYJNFWhwgayPJuFZX%*#MHRs3ogIm`4J~#48gtpV9h*
z&G8o<H-Q<E1Kr71+34gCD62G%*BUv9yWiQ>F$kv=)Dpebl?Aq|%-Jk4bZ`XsGUe-7
zvz~dAK1VK+JeeU+f^-W^7=Rq5%l!uNdGc4J`q){5?j6`oBV~Q1*g*xli^$8zCFV0I
zBj_M*GEB2>PW!8^X-stA`3oTQ&1ZIiPtH-h><TO=vyrr92SZfFTUm9d9@MFQW@du&
zp(-~+kr({Q4tE^0x%&oIcm};*Q1l<Vhd>Db`Vk_$My%g~tX*S^^H4^a!&yt@D;EU!
ze&dd2sn|!wCu#)U3f!J5x8HE8kuyd>VvN5rtB!2!rBI_!7{IKhQE5KF0ce^cv{ryB
z&zN6)#4<bMq2wQjEC4uUH~unMq!QM#rK+k#+}m|+nn3T*UG_o7f%BwCUjg#eRu(a}
zeGsoS+7g8-MrFZkIJ2c9EU2McsBdaBKqi+N46%W+14Ev%R;96unbVO4%xKH^|McG?
zm9pi?Bd9<A9CdV8we21!2F_WvU>A9-W4teF%-dSmH6|{B4yde@)d@DDsr>7rmplZr
z?kDOHNWzuVzohK6ObiVVB{kVevfsWvThPr>X;Qs#VYP?jWDycdBZ8R@)gVr)7tEvq
zm=Qk37jd3o-#%2mg8^wQ?M}K2x*?xEDn=?9QnlWJIZv&~I1f?d(omqpl`EnAoDFG#
z9!w2MAvcxj9?uPgXG|R^5qcCn64v^{Pe^y#&u5ejjZBa|_5aC`W{#s`KCC3v@ZMV^
zIpUp@gsF>k2NvwZ*YBoiW7=9GT;1p6(7!Ceh1gr2qzIrHvRk%)Pbz7OHm6ZTtBDCI
zaKs(0+o~6U1*!1|lWiwEKHNAno_fWcd|nUgKCtL_C1!44a0|z7(Hzhlj7uLxTJeC$
zp<X)(L)8<jTDQ?##c^u7M|-ijd-aMR0}}@A`Ec%_z@&uXb+}@Pr2s#BgYm8GhO)8F
zatGT!+C-rrFQVcP99?_YgstVIjrwMi1|K}2U>3GW8GPWJ2d>sQaAZkirNL<>gcagr
z5;f_2Je;d^b_mxiciPfS<a3l#P*7O!PxW&<SWR6u)iwGL0-<BE&k7kjsicgmZ=?<6
zR33>xRI;3w{qd=->rEy7#=#q!&2QNsw(L2q=HugpkvM|LWjyrl&Q8Cvq^ljp7x1Mx
zH$sw|k!sCnFvJEP-iOr{9l;XOsyCu}R}svE@a`$7gufWiZI7&SFA1?q$`pJp48I&f
zskGM$sbzi=ba>mhgN^+~eoLZI&@EomX*E3xCwqPOQ}Psp3$D#KT4+t0-v)c#mjP47
zK&SdLyE$bV%lJZbW~^$C_O<UDzuJ<48#j4O-}@Qtxu{xS_?m<H`fzR>huNUs<#~xg
z`vLM8v2nfwgO`~N8FIzc#)DdtUvIcv)sTrl$qJ$^MME6Pv7408C7~{1>5EFq$SAv)
z8`*-z$b8z3r8j~HC#$GDvCw#duOvA<fVBAz8)TFkhzb$HFtO~r1)#1*HZFWLwT}ZX
z{8;M*UVz~p9V?@Fytk{&<Vt-u&4P`Z^mP8B!`Mj4N)H=}3`G?UYA9>p@$6ztvfO5g
znzQx@19Yy8dqD-m251chJUjDABavGfO!4S*F%rs-pv|4~R!Gc@<yPamO=OsZ$YFd1
zFee9KM-ZQRwe1;0tIkMzaM9>!Mv~h)y(bEJ94F*ilO>y&7b<l2s^0Bz&jCnDAG<eA
zL9CqMVWQb(<IAb}bNylQu4#KPo&)|sgX%edTgUeI4ZoP}!3OdTsprq-pX+d+Ku<ON
z9L!PJ#KFdn)hMlUn86hKmppDXk1S_7qge_JUiqp`a)D7aZQHk>`R!ufSBwhncQ4ZY
zg71Ang}1!>-PL^YFuSD)jI1K2qw_6nT$NMr7wNUbC(D2d!?{s8yhcS@Zq-{WK9d~f
zCT)=|zznTHrosdGDIksFOL?<0va+L#9T7B8gZ2=WY}v%cogG%crbm>Nlq&hEGPr^V
z_e^$x;H;ttUeW@>?Ipkz9skQ5Fqp-uuhd<^8eTu+HtOmd@n=QQ__yf#La&{udl+$@
zzY0f>Xu+sLakU<BYJUE<;5j^uHX`)m;)WEpfx;b<y@I@9%bW%NWsBCI=S$xwvCI43
zRGPIerb>-*X`#)mPwo<twu3)f^*PK}Kg%sGEaW>X`CFj!IzG<C15LIsyqTSF5V1ar
zqF^b#0Wo@2&ES$P9s2}I!eRRLtpQ_i{&#?InOpvSdMaQxM&jhI{O(Q-zN>eHt{i=h
z6BfDPcd<b4)0+PtT<0-43PojiAK<B&jmkOl%_uD*M~bU-%O?@JbyG}8j<u}^uhbnb
zM;`?fv5Ns?0$owdb(1<~?Uj0Va+>q9J;Xyw?p~C|DPL~}yV!gJ7I<c4)n?B0_b3Wi
zSKr&=UZ{(^g=a?!z(2m&$IMoetn41_9w>~mbBG|RsFK1J;IA0oyAEL~o=WIkci=16
zZ@Lau45bhhNB~2GlnaLMOBNkAsHg!;g&_-Vf%6`H`K}G!flcRHwDrC|Ve_iQ_`h9Y
z!1Xjxb6)+(F7~)%Z;f+R>A^IOhmMo|YM^fnpikU*%LpHwd<b6^meb8{arc7vv2v<J
z86%a;TOF|jZI}=~H^BnwwoaVL$2oK)xnyI8*5sq&?r^^CzTK)?w6qpK`&{j?MrE||
zUOoT3>X!Cy*irF?Cg<c-+DThIKAwgcZJ-J-aLU&j8OXKI{lakzqo^I%?Q1FA-qZmu
zp=iK>_y{v{W9-DTn?xojvqA%cgGZ7C-8qap@vLU*G_GB{W<FXFdv>^yty#{n?Z*Rw
zh`bxCHjB0UF$mftN<gVK0#7S+3BiGRds`;D^%apfeV+B-BPe=kw%bsM9g>^A+u5U?
z^bjmZfpdaxZI*<5rdE7pqedv7M_ethsIoybQxAqx7xp?l8-L<FbPx!p44+od9`~)s
zX;DNKCznS?L<_1ptmcPiS;C!@8wxPkpPHh_Bjqkr;&9cX0g*~vcNs8DrF*y}1eq=F
zlxd2Oj;4hM1O*NIV&EM7Zl_E}sasnY5wK|Ie8eF%?@xI_!e;o3#$;=(L=54v-$Mgm
zTwZ3NqH5WiN(xWzV=P+XNtA^u{>`NL4>d+yULY&SEq_c_(;A99ZM*Os&p&3!5=OA{
zklhXO7bcs5$%NynlqPD~-<F6-Gp`r{<8R>q>*Q&`jNbiZeD2+#mol3(){DHPi$b=4
zX*s_<*k50MN2;<RMMJKgE_Pm#t9XdndcW9(NP6A93pV08i=3#i4&y3L$_Gj)@*S=C
z`=&Pm0fl*4R&F`ajiQyn>G_KW(8m>fvfRfbDLnraIwEsiUvP)fYow^BD=OSU8?<)V
zw*pl?4&7?fS~&AYQlf54q+3=r>ugjBbpHBPyV#C{eR*u`U0q#mwBFEUT<W^?`&P;>
z`sT$rAN6w4(1=rdUzA|pCQ2c-sZ#)<G7u;m3a5E}(D5|cbfU}*T^X8XroTf=WfS=_
z$4gE0_Vz)LE&-J=latc82x3GbEpxLka97?5%q9q<euM-BpMOL_WrQ9>e|-80w3YDU
z<~ePGEMHMxd3=pvF~{!pN0$W!!m=6EG*S#R$?(X<1*1JDQW+uhiHhvFimlU73n_-m
zN@0or{`23+oAGm0*=Zs~R!z@+tvflw_fHc*v08{%O<SY7aFigO$#(m+7PLd-&4Y8c
zIaZSuH7iN(Ab&&+z*JaR_;4K%T5PSJ*J&meh@C)OxGD9#!4vnYm@LKK>+PNG%j|jz
zy5)m1L-O%T75^hkMRU8f1pGQ`1}0$X0<Aa%JSm6G*}Fzr*)<&P^BYfdUA%O+R=jjQ
zPhBw}DNOX-CcBteSji}7)FqO&W_dlocvMZ#Gqc4Xx68N`<eXO0HZ&_NbD<HmItK=_
z*-FmSjxhzQ`J!U=$9(U+yq2;io#Thrly!BJKtib(Chrrl))!;+U7d-I<diq}S1(c{
zvGwh5+XpU#$*f)aY_XTWo&5u7X^Y~h>48w0EM*JFl?s2$s1lK!oIJTN6+Uep(_<U~
zB6~5WRY41qN2W)SCFnMvhKAy@tKGLHZ8&d%bN%_Q(1_LMjFFH44J+%;#kg&&>l9z2
ztd*{y9FB=>|0n#?sEDY`m{7H%8IeN%V>LuPg`$hHN2+n#Pqi53I^p60-Q9i7xV^nS
z5SRp7Bt9En^|`<QHA?nde0=`Ylr}Vc#GF|PCZ@?mkhF1qNxRzhKkQq#cu_mMySsr!
zw{G6#LG6I3n72nb0U24rr#qy&2L>x%VeBNqx=4tu&7ts0LpG-0d(mRw^Zf|N5Ud1#
z3ZA<R6~ZW9s$(3~(06f3=wOLe>Bwj>uWM{}l!tBbmBD@6=wictyRSG{f(m0t_*YlY
z=uh0F(S+RO&N-g*G&7Kg!1$mRHDIW!YVND}u)=(<ySU7DS<T$}P(*Nd3Pq2$4?6<e
z^q*ZUxxRS!N}xv0KoXH{wLzGiXi-7;!;KM0?+)uO1&P5I<osny?RU3^3v7h@8~Kw*
zWsZX*J3#b|x_rGd61aCyomiu%qZuPYEm{kP-lzf76=){&{sO^ce{Ta0lGC_Id*gi&
z4D-Icj45wUHh4&f>g*K^i3br-d`imV4<G)^%*>=0x@=Y}D=jTW{h$EEDo*%NPie%Q
z8{IiXRavCYuRA-M^=@Y_EIIHEXDOlB^tR{sKpbqNy#0KwvFHxy&s0lw<#J{HYtd!{
z>&D&>MF(J$O$b<&sWP8I&z%Z}J|7$BNK7wNNNu}n!C8jm`=(L~f|N1ZhgJ5ysDM~=
z-1~j!qO0a0!oA$RBYUEH*8sVDnm<wR7?tE|9{^Yv^-SZ($yt7DRDG{kp_;5>z3g#q
z<nP_K<jDB2j5aj~1^wMtxg21_B`FknMNj~tKy5Z!ph1W{8O(f511-0JvnUm3fq!?2
z2VQIpbz$OO85xGSclIu@MkdZSNWpl##(FoRl^i#diJM=k__DV6?&bvO(a0T;tJCIQ
zTY=~?(jw`-A}|zMKSOP*1ano=oh~W_yH!<&%TXMP*!A&OnMCNi*Cqd8Qyj@@Um)Qz
zj|Vw3q_3S?xUa!_&Y{eXX(OW+J_%eTUafD4zi*nU7CC4`#<kmPLJf^Ku1l*ch<svF
zwq#_EEW00md0dJM?9325m0hRpDGh*|nhhS+v~o!W2?=CSlj8hp(}5?!l)}2<4t;{h
zE^M?wu)=He;Dvd>`a>I>_!0h6;x5kHz899vCoVpF_PsgBcf>%B$P2jlW<;nXD)dPF
z?S`Z3PBy_WVd|qF?%f3WHB-mI==sv1qU`$C9Jgu}w|*BWaZ<Uv%0O(PmHWh$fP}=*
zQt=@<nUW-2pi=a4$zl1sB{G@Qw+A$ExYeL6x%*5SxUZTV+gVTQ=@uNu$`1l-Bfad3
zHMPfI_Y`lctkwhd;}Dte3x8X(n&><eHex=4D}0d>Z_}WGJU@nmF02@&#6bwgWmlim
zdTOi%3Z7kEJh1H9ZIC&j|J>3!$|Ue~D+N>AnCTc03-R(yPY1e0Wtw)&;n-j&QdBcC
z!_S=yOw)=N35jy^DHEPVRNssdWhZi2+A_e`&8$AM)GA!s_ok~(G3sYUc;+a5@mB^`
zEhp@#^~qNGA3q5=5zJ?1^IB3+KCg~Lv1F98iaD3!d@n{E(wciIf%n~bUs!Tto-3tp
zehX<9eT|}{;L*Bak?<Xe2(fYzwk|YJIX9eE`dbSBPMy+7o=U`ksq)L0AKZ>MbLE6i
zUvf`7L}XI|)`-Ljd#-Y@H|81;RG25Wk!8AWUtVs|<PYXlu1yo@ZQ-?Uf40_&$>(X9
zjy7k3XZ$s^wVc~Sbo}=|+~oC{F}&dsEG6f!zv8)|6`qky+n`b~-pK}72JP(mHRX(Q
z*Hcx?Ct!kY=a)~Z2vV(H@p~OKaLpUOoO|gA<#cQN#Hm_orR3s#8Foo}_S1e0i&CYB
zWsXN$XTPf_{>k^H&=m$Ajvv23L_(y4F@`Yh@c0;hhkGi8)Viu^NC7}dXZgBg+&Vf&
zH?D<i(Rx9ziQIyfpVhYGf#}j@YSd!cWMr(}qQ@7*qY~4`M7fQl)O4ex5*IFkDP!OO
zA%Ex@a-8M>K+#E3%6;Js%*5gM)1*-j^Pkh+<`Won{xCa^j*^mljkXrCgDqzF63_<W
zk?W3G&-@w~;#<UMlY-e0>hx_at>;P5vNTeeG#i!?Jfx?Hj$eLI3j+gpfX*OF6b5WU
zO!TPfSm^YU2J!0-fC$xUCiEu><lO+^A0Chjn<7FS<52r6nzasDiVvI_nBT3Oaz<Hj
z$j8efN6y>c;HTB)(}hq><rV~Sw2p2(0g)6q7V+%>o8i~B1gK;u8x4iWAz?aNs;|Q5
zNo$>DCc!)zIBG>ySQU)?``t!?>cS|<wD85l!`j2=e=JszvZA4w&JM~zuE1Zmswi#b
zpXI_2adXU>S!}Q%A{vtD$&=4p1u%gdu^yAc<w42CuF*>gdS3ZFAXl^JAQ0&@EL?O>
zf~V24rSaaoyK1W3uq!Z9=~j5aiUlE1(IBQ38SYTmZ!*!QQkY(wO!Vgq823^#vXL!w
z=%K`QU(xDO^mO2Fh@rJp{}~+}wdjk-_me_=A4DgE1Qy$Al{d3dX)9{8yN;~V#MN*9
zVj|XK>U;u(Bdh)ci8I~5S;Z49Fayv8KChKdtvnF*bd38!qAdo%=gBHs?N^hTblA`1
z4zTAVo2a*n8od-!lEk%CGvB)fVqf?aD$v6??Od+Jb4J<*<J6UppW0uA|LCKxN!*$W
z>i?{cKJ9m5Ym;{#TXvV<?h_EJtqKAqB#FD-KS9G@1_A>hRto~{3`|c{IoG+=HhPgN
zpxS#Efn$dA*L?JjNQy21p5*8zB}xvX;v%r<)T&b*Oh-1_-1X#(U*JZraR=M;B?5Q*
zw#=pBhY*UfcLSVVE65bqc7Y-%9W1k5VCw7DUmC5nsTDu<8_3B(G6Kp3pZL(}bumZS
z0mz1H-%qGb%swiVNAs7!Q{j~$#ZW61kXi+mhEr~W2q}*{n6tIgsa#E?)T=)+@jM>y
z%A3qc-1b<M;WQubSn>e4x%hV==KaHDgQ5O%SR^x1=y7RlUX9vxDuYb1imj!f#)TM~
z7l-`e8^9cLaOh(4ygW+R<XYjs<83l=%C6b<K+7vh$kQN)p!?e+C<tr(>Gf9m$c^rN
zi2yOQ?nN=FF(JuwHQCX;k3(f%Ypf(k^Dd5S?(H-tSeX9#UPDI{pnH?uKGi8Xtf`88
z&c9V9yRDNo+y83BYHBo-cqR)_HJHhx&yBxm0>&UU94o+oTC&B<G4gK!DkVmH!-4N`
zD5H|&bYTU2>IEN9<#qYQbWs(s!H9HK*oYz%-K4+N5^gl+q0GV1+imW$w`MNse$7@B
z=z42(XKz4!!v9WSFW#JgqH$YF)%`KPr<^~nU*7QB>eT<@F#OtRzZ7=ypfu?G(}(=I
zO;0n<(V^Nqu>8cCgsMF0sJf%;@o2^d{n1m@SM_q(f|82WsNC6P@-&wCh$0vqqYZEL
zSeUEFMVvCIPAS$z(P66))v&m-3jB$!sWNF&FE0hp<Wr>Y_{S13rv};PnzM;o7A?hS
zLeOiWRbQd{Zn{%9N<`*dNNtfAj3BMzc%{OJ3lY;XI;fFzIkPtJ_Fsz&*#Mu@-4eHB
z1{U{0rHtU|)cVC8xvHM4dKHc!gz_k17olj~H5aC5*8%K<&PTew=x9B4BA?eQ8zIkn
z%L;+=nmHNA8s|ZbrELT{bJ)-llV`qv$rt6~1gmdU<$I6wrWu@Yf87<ka9AG9I`Z>+
z&h>uyfXtuyj7^5y#W`VYcf;9U{nIkDk*U{SAuRBz)RXl?N4p(1tN3&4Fm{S*0j1iQ
zkQi@~??tfMRhn$>TC3H3@IWUx>>~jIGw)q;B7{?eGH#fUX4QfKno~zPj<Ixq5Go|B
zm{Z!*pA}6$H7Qd_b`W9UBmcBIai0;C-B5F^p4l?@4nBzaLFion7{xxlNz>oJ=o-W#
zq-ETMl~&mmdI<8th^6pJUPV9XPdUBM2x$_^Y(Z=^8yzSGLz4B&U(y&KR9FsXfORo)
z+1t11z{9b9n2K^Ce+y-c?RK*qranD6F$aMNbnpN1QUFBGvS^a0X?YXT=|AAF(jCm~
zJ!)|Coc_JL@nh1C6V>ANk;OGp)}QO*z}Jc8>?MFdlCx%$813ZPq%^SoSpd1_`^zdA
z9;$xqCDP0ds}Z^=l{VIsxm<Kj6ml%+cL!9t_9sCn^3hrUSOPu79R11z3Uv!GHn{iR
zE7{>0J3{W_6&H2$A?k*R5K?;J_{^#R?qk}!oY>k?_DE0Si?tQOF_x`5nD|t9xT6qZ
z$QD;-w<phhxv6SIOL^9Q2pX00W&fvO#TG_gUBg=~68XT44M->vsM-$69=7jCqLVLr
zh$<<x<BH*`fvOTZ98>LahlO&X_Tqqr$9b`37LcpF5T3lSQjAalgNTvpb=_9Zs)N`D
z!iOs9!6Z(oO3TygxV_WWCp#b@M-ym)j)kW*YXNwHDz<>(%Z0I8*oYQm!#f+a{R|o#
zv~CYze;a0)z*`uczx<1pPRg<PET+poPItfVK}Fxl;5Z@=gyUGME5LH`NbcuX`Xh4u
zEE<O99EBN*UZ|;<kBNbqt>Lq{s0b(Rr$<Z!pfdvrKAON{7}|s4JFK2w3YzKFf417R
z(nen-Med0lD4p0H!kG|S(NVE>3r@4BjNzn~#W)@BlorY`m^9v4Y%mHR$_3+sgQ;5K
zeDlJ#n1mhR&^h#0hG!UiN>(M5LU&BcJhzFMlTm)#q?aYqv`JFeA@ZubIky4%V%xL?
zAgH0hxp%OTEswqt_1&t1t0-Y>z}Fe<E_5z^@ZfJ+^ylz4kY;pftl%9d2GI!+)CExz
zj-zbBv(pXdgAH6tMi4Tlv4riCG=NV-91^)CTQ}^`pp%{n#dXh!Zj;!yE1_f74W}@|
za<>7w37GkhqvAM^riuPtqB~XlW{34_k5APvSoqIR=BD`^FTR5PA+K-f7{h*sDM6Q|
zYeKWFo#HtO&(j@Z)JX!MaTVQTsEfFTwYWf<GrKxr-E*sW@M5#h04snhx=WU961aFd
z6U~LZz!?UaW`=hw*J%=Y((*V-b(}&upS`TS<Eg@p;8sM36K1w7BEu8t{Od))hfwlV
zvvcFh*I4v~;k}V8ee?MvLqj$fpyA`N(&SVFxiRJSwTF9$hidA|G@O<*3XUY@VNL2M
zj9ZDXc-)yj)w+TYioJ|lmYS*-dHV1!nw{l8{o&{o(XPHO1&}oJeZuNVKGeL68xT#M
znVDqg9?d@p93^z{5Tw<d;|G-@0?PUfj}#(Cbw)BA)hgMQIY4Tq)k$XF9#J{ZLvZ2S
z)%F`mz01EJ1I3REZIv^o3<NDwl^~<=Q((do;VmCAIji3gqh6!5Q5t-&!q;HX0GyDH
z&4ujQZxeYl>&6;l3oPpH+gup`eT>%D=LXY3^v9^cewQBCSi`;(s-PV;k*8dkgw8H3
zxDHqE#24~}>}>vc=K-h-_Gii}^3nJi5gw4LPeZoUz?AD0`K~RGv6p%~oLc;+GE0#Q
z%a@?sCs027xA0R`baFfx2W$gRu+*?}=~CGJn3!DUkW>F6Hdfkg=hoxY4-X!t9_wcM
z`KZ%fZoR#R^Z8Ml55`M{`=q{)2m&~ArJk03l^4ZH{rKTv#+~gLd9P6|LWC{%CsM1c
z{HObo(D_Vk)u%-E6}&*+>TSoMo~_>$Y!RxsGOTBz0ei=SyrY|fQPc+$u3Uv@pOqw?
z!y}>;Hftg*#N2Qbf8~6!nXz(1!KYkVGhSjLOcs&TuEHY+=UO)(f4GODzM~7iiQfF3
zyfR%v`TRcjO<`CB?R4@Wzud`bBILwjQ%@svZK=27=G=-?zuCouBC<;gSh>;w{_+j|
zxqhN-TES=KcB$W=g$8|)Ph;JA`jqOu55mNDu1}5#nD3Fz*9RY352lz|FcofD>_uNb
z7H0LQB#<*gjN;P;_)1tv*Q@%KIGFp6j`=2(@jMWfyH85->k~fCSoxfVuFE;8+ikb9
ztk=a(6fz#qD$8iLClr8(@r+}{hQ{g53l@c#qcl#Atk}v;7$dlfVcNl2e9=GG<*IWY
zY_t<iSygt7BTy^&MM~vJtYG_>bWLlI66rGzwp5+UCKY+3ao%dK426BC5C0{<*%^HE
zQv2kgd=7l?@q}O<#69%xZgC-w!^hJPNoR@dS_m3Jp$~=jTJ!TN!#Ud%&eC_L#5)F+
z9-it?JS)-__!3r4Ug+PGs!_c1>=_%Q5Z5K<`$<Au<WCZzcqGyN)EL*z)8l;DDoMrw
zQGa!x>5<KOSjr8(0ppWex25X!^Bbqt#17l+BnA=leVXWN;LN^A*8T{d&EpR_1``g%
zFC(H8Uxo0<?KatTCW|QWfJ$({a1C;qFV`#~Z#*#xByqh@=#;LQzqa>N?Y5g0_t70b
z0*meIIh>j#j$XbBJIO8S(?kv*F<$Ck;5=l1{YmJ|+%ZNkt=ZRupwHt&KB3%v|M!RK
z2w}Od_r+0TQy<f!It7z;dwkR?Q#mX2k$bKe<)`Q7MK#O8NW`Qm2d4Ig{d2Q-@GkP?
zC^G2m(P<-MhY?*&!xaDfF)&00n9N*U4{n2?4vQb+msM^y1vJMBFF$ZHW<ToP77#cs
z%wsGgBjL1hKK3cGj%aAfM=;WGIiX}M3}q4YFeTXGO$Z0co#p8A1^Ggme4*v$T)(j@
z*Zuwj<TuUQX!@dtK3mMXnmgrfK9mIYqK2le$LXG5tE#x)EPmkr)dw{}sZn)6Ls{)D
zb)6W_VP4-74qPyWF35u8D)XDA(6=`g*K8O2&nCFWvA@=7&mNP?$fUm$BX(;7h9|D(
z()X=yQ#qu-Ex?{%sPWfW<oN5$#%@uVpe*C$ial)#9d<CEpzKgmUo@j?rsB}~xZbl8
z2U;rTSTTO@qBMps7EbyQW?@!|!hu1XPTDew<)r{;>rsC$s+sif-+lBNT(bl63hHU=
zE*@Le+<roH#x^Ou`szXkb>2oTsjc7cp9sBqCneXJFe$#IjD!JQ-Y;FDero6n5&b+#
zL9%~c$=Ev-o2S}|L<?clqk^FPCH3XZ{cGA+GS{$1a-1=q{O1bSdw{L;_x}ZehoU|N
z*W>KN_(fW>tjr6JP^_%1{<HS&3b~wRr2DD5yk)e}Wfae`d=7=X^Sk#B>}=cR77dO1
z%i#9ve4GP}l3g!s^xv+1K)mmvZWC9v`&b??VbIv!8oxGS!$kb$&F0B>b}Sev8b9!s
zjS%E&>R(?kW(uAZC3KIv7D9uoMJJ1Je{aXs_LGpnXUWkQlEz{8EACIF-$V7KJntI<
zfZ?UR$EVhtJUbS*at&;#mj2;5@aRCriwun3_539Jb}%2O#F1{&>DtH}#Ap;<z}*;Q
z<oEA3InGIKiOI8z6J|dpW-?M9;oi8>B31Ci<HED@+U)oiwBDYGC~h#Ng-ToKdk_H2
z-TrTc4Ec3fz9nCX*$J#?i+d2I)~Dy_J>RbcBh+E9#EJ??2>MSR57NzK<+8ZuO!WJw
zMb1cE9Fq=u?Z=O<AoTHNrdfWkN8KI~;q|+-&bI6;H|~BVB7LE1M@^vW#4}>?vUXc+
z_4%&CI=8vP&=P_Dj5j5OMHk5s+1@{l=^lFY6LGLIA?{r_agY<V*D0fbXRR$1JUVv$
z9Fv?OvTGz?I0lz#p>1g>N_pB(JX4k3WeV@?cH++xPtu#a)4hi&N*UB`pd6j<7IBzP
zd<v<T=r@>6FE{SGk1mx+S(3Ux&?p)ylBqa1lq)Qhcy>cxUS7n>iC}q|z!{e{zc740
z{o1qAT)jE%EG}!?q2$l;F%|_)@rkh#y4GB(y*&!K3>94oxxu86j#~p!3KKWF#)cCv
zsgko$zp}5%JMg9CTw_9)B~~DZ%U|T8qY7SraeAVCfPda;`V!HRs^oy*o5?DkdvRVJ
zp@bLKR)i*>SvPs1)IWQ0m>2GMou|b6;mVDphm`h#U_cC*Be3BTVfAbpQ@}UheJGBl
zwEMx^meZ|C=UWHofn|{|v2n2Fb43hNT4!>F0PfF~tRUL^$I3+a7nhd4dz}i%l8wC-
zJz&Ph-jm0}2!2n1!>BdkK=IRnwYw!Z>C(MVXw`C42l7igB*(UhpLdGHO@$|>LAs|9
z6lg^31C26aOAL~F=k4~RBlZXAd+5^NQEc|-`tmefUfsjVx@*+7C~J9Xb|3C18x|e@
zS<^y7QIsR*M(ExKx4nx|0-;`yd)}yblmP3~z))Svj{O!=Td=FBpZ=6r{~9{_^LqLF
z=NPB1KVMGicg3|zuNw5o>Gk%z7bVB^ycM-!<SN#Ebx>#lK1`p+^-R>_E%eO`A^I;R
zWD|$%N)qlr^w;I>ey}n8mZc;s<U6?bi0?%I<KCiCcKAIvk|-$|KpPCHY{(pC1`;j{
zF)t^QFO|K@bnrxz!S5QAopdsPHeg!^_8%ee+1z2ik+va$KNh6Yy0-*&*HH|Io}UcX
zR=-FUzEJxnXZI7sI&r#l$*+s&yC<hf2&$u?uVS!FWhIUU_U+p*((Z4=VL@ev$_$*r
zwbB_1D^G3BCmt4t@zP=Q65o1>v9+`w7WEQ~Qh%c_+3x5kg^-44T6ezd{+``8q=U2d
ze}tUJ3Gs(nW_>2PaJzeL%Xd>06~4wpHxJW4BxrX9KNPlKKCw<B*Gv}Amjl+%exAqe
z@?J3mJ;RZ3?l0ALpSv?`yDbkV1CNax-SBj-$sUrtH8s`lP>Ghy)}^X;YYN?HhwHBl
z=oXWGJmm#@oZRQGpud{2>z}Q`DWUq836vuqVsZWM&|3J_(~;KmxjN`{7rPL%{!5fS
zyR!Kmh<Pvpf2dO#xnS};N`Ew_C_XT7IsVx2;XgIYrME_}WkL)l(uE8&XbiR{!U6B5
z!%Kx1mIX6%s&P^j-&oFxr1#1_b2;<El%4z%7R&qw-p9W6oYJA@d}kwXXw7BeH`mln
zA-~iw_(;7|zsEzkgH>HGgP6&7ymSzazk=MDDbrSl@`(jePq?gw42?5(&puR?d*M46
zv2$=_LAzvEvG;k;+qLWM@D`09R9Oi@3{c$pY+XaIVfRrj{l7lUwjjciY?X@1UYx_8
zFwz8oJFqajsoR17cL(0SqC;grt;aDZcH4#!H(?a1m1K{Yfxbf@k(iQ&n^1WG7Zrkv
zl{-FkhDhAa{H<f87OjoUwzRG+K}gaxvCsw&*P-S#cUzA;9o6!G`fry+rGMTb=p%DR
zd}CdarDog2ZX*Z@`nC}29v^l*?Kd*f8u+F(WNxYe_q3W9lu$|cv5H#VWm9pA;=8Re
zbL$hNVuQZ;+^<j+6(h|9c6&xZ;o4Ed>J(f@EI|pom^HovcfHPUg0^`nAhh<KFT2)C
zimJu8lRZfl<xDiOygZD+EdF!Asg7(tNQ<haLZmBXw|7zGC=NLNbm!RQN%~oJlj~BQ
zpG1it?k%qEK$kUoxQf~{26Si?Qf_Pd5~Ws=rbueL*RT3S%<nZh=|fjPD(^xuK>Dbq
z9(BGbWIrz`bYzz0&+gI0pC_*>3QjY+OB`Pqs@o2{H*6eGk+a`g4K)g2?CxrFUNRJw
zv-wXsMKK~OO!H$@^z?}dBC}eK`$3Cr>2cuW#f<hCdKLy#%JT-YPeqSHvk~gP{@Kp>
zY778hZz_0ZG@%9WO@&sVJUhl0yx&?QAjzq29wR!LX+!DB2KuJWMR&Shy=Ku;%4aA%
zzZ8T_gr0skjsIY0XFB%9xIRft0k0wMSUGngy2l$+T)}j3fu2mTHpZQT%ATJuNb6?f
z4x=!W^M#)!(0ve&9d-=yfp-pv>|`g#x9JsL(1fVrkyl<v=}-C|P3@jzD8;)C+XyCf
z9~e4Woz(EXW{n!{{~Viao%Lpj>!<eNQs}XkeXBEXjc?t9b`u&&yZIRU&C|Efsx>--
zk_7xxLmM=ERB+HVrOeE0_!p4^631ZoeF%`*)6QC-27%Ta%-R31h(e<&H=Tctd)_-r
z{mtpiPdYSI%)}30KP7RF1;nrFXL8SDRm)nN#nWakpYS4}O^8Em$Cbp%BT|XfBs>;R
zc6CIAgu^P`xd=<#qno)%XRp1v;jzwB4;jC^3%kfSPu`eR+;cL<;j*6Ladm3{If1X^
z#X|fg^`-8_Ygsokd56zBJ1g+;cnZ0b>a7Hx?_b`@^%c$4GQ73g_$kR4y0<0CI$L|L
z7E<i`-sjHHJFDmSUQ5b>0{UNMePvXYPqa6Rh(Sp=BHi63ARr*!UDDkRB8`NANOyO4
zcXxM5cQ@Pt{oi-3`+oBS%X7{=GkgDP&+|32Jq+C`pbCLU68Xy)DhIoc5zM>lcaK#k
zi~Jv2&eGnvyMUBzJUe`|mc8ioLLnq?Dej>G(^PhEx&)haM&bh8&vOkBj$gU4sB{Xo
z<=S_={<_-RRi~-5lRq14+Mq^*0#&~ul|43gfgyO9d@xn&$Trw`7qn1~i)_DOZI<ur
zDdH`9;ud&!$sT#(-aej6Ask%w(W=ecy)8@DC@$u-&G`^!`v$o^TA5wPY=ignAqHSL
z#YH8#{<CBqQ*D8hXtymzM}<X}<3ca4_waZ7*0onb^W0>*eBHsxh9rvRi<u?+)KRwo
zfz*_M_RcxJ{9)~_p(;MdJheKA&w;RofhDJ?);pLG69rV{alPR9|H=~eQir9SSb}}c
ztzvkhIPYdh7mANEaQ6PQUp3UX^4`V4fu<qk1AK-@?mHGmk@g8HG#)c(rs<W8u>v-x
zT}#)k2CQnt8wxZUdc@Yxfk9bJunQb9PWD>oVWt4TTAZy2Bd!v34}58znRvZ%oxo@^
zf*GM~myYyU9_ESvqBuYDG}1R1Tixi<(;gyn$*Mkj%n)gr#Bzd$PNEVNF^fZEH7(Wq
zb28ehUyF@}kE9})LqZLQIWY{oAHQnXXPO#`ROy8H2<uY%!hnXG-4Wx1QU+$G&P<ty
z7;65fg%SMx7rm#v$(AJt;6_+f_|q_oyn(1^+-XYdc%IQ^?z;W@VE5-oG;;B0b46-5
zrgiV(ceuI>uE208x}DD@0#Au_$`HyQE$zMULB^t>WAdglQCs4=He}#oVYGP5K>PoI
zBj94f;1EKK+hNKOi}Z+6nfcA3VOvyI0ZqQe)Z_OcGyuW??BE1!C`&?sU$Xh;ZBem(
zbE-c))$Do8M`Ok1na2u)@AVJL<-`F&K{|ZpvwuUNSiQW{hHx+|{I^tNPm!E2JZ3@4
zP0dt<zKMm39|WqMCB&3%$?d9zdxM+yZsd_Avq}2~myeqZtIPCXE0IYNWiSCiU%K;E
zIm7`?h5`*D%xq9*Fx3!4REE{j;w_du<<k0i6Jlr9IOdEkUJH^8>QK>doVoQsGYFSA
z4QJ`3>|q4dNz%E+_I5NfU&QT*IVP!eJ7;Kwg#)N81O2n1AO(3^t1|U^7pfkA00PQC
zmBWvGB4dYNwca62MZZ!$s~OAAjsn887Oo&~<doiF7n!->_NU*ETC%=DvI^yjyf6Zq
zFkHOv$)T@t6S&(8YRAmKhx%hgcfTWCVD}WsUv8Sz`7BZOnj9rKsCH$>3Sr$2H;c*V
zYx<{m_IISAyw0z=PpJUg`c&DyI9k-|@L}NKh34j^!hwo{vdHtqez)CyUfti~V&JT^
z%8QH9F&2Ly;Yw&z(SOBL=_w-=aF4-sBt~3cgAkNq=Tc`j@Aw{&=|k(;;jK54xo5c)
z+-)JdWr~0;G>!EgH>?(_9qSJV76VE)$5qx3Qnr02^HRnzT_VAR>p3*Tkr$<Rl{8b8
zwnQ?E8e)-dI;S}jk88vC-)`(VL%$@&a*L7}x=$}dr-jWg7O1vPYgZ>a9Xp7MQW~jH
z>1TG>!`{)-IpDM6V{G<qNL_21JngYIn-qd5Gng$Mqs!+w?_9LUBAKig)Skk4b6k=&
zX+(>Ir=%)U>hco6zqrLKeBPViE4A8)EO(RP;i2EsgCGG!TPV%Aj8dt)alzA0HPhZ(
z5$HThT$wQX?0-XohWf{9?eHS*Zk6l)a*g2wpJ(*h7AoV*87nT|98KLFKZ(x!xRb@s
zGo4(+nG>b@<G4koQ_M5;(@odT>l?Gx(dgCg!NK<Wep1lXWLLNw7x?wU;xYMwT=4?$
z`nQ02&{{UjZ`x!3$+H}($Mslg6c7IOBqC*LpJA0JO)imi35Sqqww_vb?f_s_u?~f-
zhOd+0lKX{%EO(}q;rDZdsy)|{m4ZkP?%P{&v+F{Bz|Bk-Vx&9kIqivz%6+DcJ4w5G
zQ}(hPVPYzF5-5^``uDa@H#cCSY9RBAp_Vu7Ar4P>ME8|sz*fd<qzjnJimj@|=|5z<
zuX2Q^I<w#m@_0veEuq`SaSs^y+UaooZIOc!#(skn=lzNcNK7S&=}9|OukuUWi38R>
z#Zk*(aaxbnmaH&X{Plbl@P-k~+sspU$6hXYONL^rjGZ>~&5IE|7#K0d@;|V*@DEb^
zn1A`&e}}z2O#w~YaJ30#aV>7KzQ{mgntljn-P)!9ZJR2amv1IZE8e7exry%1ggbUn
z{Y-ddd{09<0iaN%)+R1~c)pQ5FPlK0&^FzW-Bx@{Q9C<{QU^Vl1h8~Ev+6^Czw;|s
zbwy}$6!P+z3AL9C*0<Md2i7M>?+3)3hcCR(41)T48AH-#kZ_&uXn@{%Wjtyh$91nC
zy^^EMkXQhNdBVxIb+mh9L^y3obl|HJ7DeI9db}DANRE2c6_ax7Naga-)|)HseL^ei
z5(5q=ZUEnv)+f7C=UY!*u1O$%4>$(|mi?ihBz-D!0sox>%qfm>{dJPqA)_GHdPCF(
z3X<?}T>HVOvMJ-;t6Tg692IY{XG?$IdKCLBG0C~O<K=S(q?dDL(>0rD40#-$a2!lm
zO@Nhhd{Ed`9*ez=gQ-)t--zDQn6rj^Xm$UIjOQH`%d*$$qPmYcU*|iLvY1QCjsr>_
zsjaHE>I(opy@4uPxZx&Xs;OAu*tdr@WIIv`aa(7CDuRJ^DYsXRp)p*dMibToIepJ7
zl=A%?<kx`|igvHsry=@M`e$T+TO>~k|KRdP;L+U)OG|-9Q?iu)3{0%dJ|>opgBK8Y
zKQK^5QyFuU+1EW5_%`?W|DL+=j|#4<CEr%XSvKp|12pD_`6#5ux=(H1^Jye=gfw8Z
ze#+P@VWDdokGM$J>}`rC{K`TVO5&{zBFUul0iMb{3~bQVc?cGe-|#u8e!GhLmRf7U
zQ{yQLTD3V^(Ia+=1QB!pA@txtEn1sLz08jfD4-1)a1}ghhHtot*EyI#fhfaagO5;7
zRfx~`9pu%~Y)i&5ChZ!eu<To&ZwyO}6g}aVp)Iu>HBNZB*P*S_2+LCV(rWeCu-;0R
z88RRLfO=jmgf}EK8hHRLMv1c}-gkIcA;4L!7!ZHJW!758mCi2cpTw=l&0M0AYxe9O
zV4O5o9&mIv<WoE|U~%G&R@lE+nUy+7mCj}>GDQgx&s2bWtdf2Y<o^LvOeWLuXqt%x
zox6am-RV;<Pzi}HhboDJndr(jPyb*5J|dW}Tz&Hp9ivoYqC?wA=^VRf+(@|JVYy<4
zO=DXZ?F~Cvcg5!F%j38>2W}*byI2gF=)z_J_<hpAWck|<1)dMoyXzxL8p`&Vk|urG
z1Ohjw%)6&i3nxj^!$qqO;Mu=sEx{Q?#W@|6^A8LZb6jm*+#GWzY;<Gr1t$Rz;i|DC
z(^^=)5emG~BE#65^?}c#>sCv{Df)ADi3?0I(>1G`(?L_lLe;3qA4lcm&jyGd&%u!f
z!9N&eq^G;nbz)An8k)JdJQj)rYOD~S2`j)kiqot>T>EJ^?>kg5ivUy-qnf!J>D=)z
z3D>Vf>RA&TnO!@V<9U$Ka0Ed)Gm3blBV#>>RjEBRg$8{2cVpWnuKtw7^j^1h*s3uX
zfK$pQW5#njBUoy&yu;@uVs>cdr%q<3ZEN6dk0}usbAmv4hbWuO86lP~J$#bfxdjRi
z+8{P@lRXFcb3&3&fa-YVdr|--Jd%C!c@8Vm?c02#fVUcNCPiDZB}=`6;^Ok3X;1K5
zU&B}b2L!Bs{$qo%#l`koORa-+hszsAHNQl(3=I1GIyH1Y5#hXu3`2nYmR&CDZ5^#x
z`~2rG1hrjdH+5>T22s;BsnT*L&<vx_eKJ@EFlA-?j7e?8mE2)S6bDQ-C%a+hjr&i~
zBgNz&Q#-;-CVev<ZJ0*fuBC0}A5jjOLtl`Q|N7-jis!(aw;~HHm7F$E)er`-KMG$=
zprZqL13;lmLd)&QMmCtO)7PyN^uBNi>fsyS`Yh&8!a)E_!9fDJgydE=Nzv0e_Wblu
z;<JrPAqD8X+5Vz1j;$fK?3Y*VButr4naN~q!Cl9Y{be{hX7re5diyjxa{gxD=U4fU
zx|hkG+7C6ETqea?_Ru(VeqMFdmp&QSgW(R&xV|eVV*>UG`nnB*%;&lquyL{B?H;DN
zha6?zNu@&fmVODKDDKt5r(X>D1Ivh}|4B<4Q~HnDq_*&XzjRwhrj9>?`j|eUxi$x*
z^H%8CdgOBm{PJ9PeH=%1<f4LG*Mqvsmk?eQ)~|VCz?f;9necBb8vXN?P}9V$(7vE6
zi3<WK^vWw|dfaM997WNay*B_ndEK9=)^LT`cb|+gnZef6(v=dPUrQZZGYnxy??ir<
z@~b;VgXnjyi6HxPSfHVJGX4d~oDf4XirPE7-oSU-En^*7HGkCOu;-4D7I3bQTV@x5
zwq$*`{YVYGvfBS6+t5&#X8PS65<kP^Jpj{EhIjV8&KJ@&FmkmIETBz%++s5qM1U4A
zl}?9P?@2>Qh~K=hg@W(PfdPFE2(m@?J8ueb?%o|Vn{P#s(Csq%Yw26qm`ZLx(~frv
zX*p=pE5KGSu=ix)Vev$GD2K4wO?MB!<ke1MN^`wP#-1+?#_F6Lsyh&MTJx{o7FC|J
z<>ailv@x|9EcW(>%2Pgm3Hk@rH=$a(vP~Y8PU3l~+=u%eF*7~H%IB4qZj(Z)thPTf
zB<%FAJ9i8wP5soF{z^Xb<@TdYk7e#va^#Op_+o<&bgL2u11=Z1)D4X^%Sqt}gC`#e
z9N6>+3Kfjg88aBM{OR)@L^mboS|4&6vrDAHWa-@MB4_+*!{W6^b>JQ=1CrKh<y5*c
zV^L?xVHbNISc{&UOOQ_EPFj<ERwFB$benX$@!LBQtGtXxD_9UO_7y*>%Ue~i+InBT
zxaCeIolM)hz-hexv9+uEfFh_@!Fk%DeqCSodPA@NVR#OTmcBGy_~PN{@K)3TK3dpL
z4->DRE&w<gJ!+|s2jY|If259`5STC$3z$M$TF)NIe(mtv0Hy{J>B4bMpL$mqek4(#
z)x2o*RLM!6xv@4M3wxHDfX~7Z4Y?SYPRo<2dX&jBNW9*IzwOAV1ByiiySTAIUz-B=
z^QEEV&hF1J6qB6G8x@a8*Z?-jQM>26RXg`4r;1J527gF+JZ5Y20j_{Tl*YNHK(qV&
z4x!h}p&}&}C-`WK!DXR~Jbgf2=;+I}rt(&+D$p$!$;L3La5|{dfQ}=h_;avj3(yQP
zO6Q=bf>Iv$uUT*Z1SjUBlhFsU8Xmd*!d{iyilhw&9>~MdNQQe$O*t{A^?n+IsXQvn
z$&jOs%)4&!plWllP4OBvG?rWT1)`#q56?4IXooNt7&0z?`op+fvVn&(Rb9mKIytjP
zcCkE_$7(6IIEnO&t2+wy8CNf|LS!B54F}N2-=UO6awWnOv#j33j<tMO4K8r=>4k3W
zXz2-#{=qVDZTu5dcQ<9_GBKfp4focbYNcu;6E;muAPIvz;DUjTN3D?ee;&~X0xEwV
z7FS=7P>j9K?a3x9a|XmoR7gMns^%ltMg=SHYdU&n-2pTXgWd6jDy+`w+XA!=(dm-p
z_cT9Z?{}u?E63+Mmg8vbS0;vb8cqNbOmH{GsP`2oq+kof!Lwbyj6wSxR4r|imqJ0f
z-lKJYYXTvpzylc&yrJ)=mJ64iE1seEyHBkK>!Ksy501MjgRQ@+YJ&m5mVluC!TzP|
zBPNsR)%J3o+@cEW_4)K#3oBEHK{l5MRZ+e^s<}E}u$`8U(Xyq7wRS%MXgyV~g`pH}
zxHkB7u%zC0;~PIcLEE-yomw)=u1w*L*HMb>Tk)N_2faWUU;VSpvuT@H*r$M|0<FM(
z+Z-t;x}Z5Yf%gbR!3OM@$Y%l+)1XWcP|Q#Sr|o@aII7vR_3;aZ`AW;n2Mhz)Xl8TR
z1a+Ltz1nKkBeekl-W6FHJ9?_ge26E$W_E^RyX~8uj7;a#(bSh>9|Yg!{8_+5hNH%%
zRJDym`H}hP^t~{N8pvK;Y?-L1(yKU=%7qP3Pc4$!d=nt=Te(H=r?6hoHZZ)`aCHZ7
z3e^O>S^!y$WZ*BDQ(qrxh$b!K)|}yG+EYJ4rWwBf`cYeYX#>QVF;g0!qpAwbr}PN~
zn~qBQo*YKOj*Q1G<-@}rgtPM-aj7;a?eHUjHwVFV%uJawgT*k|ca)|z+_KxrMi(j4
z+35F|j83r8O#^2d%l4WKm1e4hT=yOGbbw{a@7)X;A0nM%9q|BqC{o=C9S@X%!iLI#
zcM^321Gb`UwqK$BwK9caIPNac8ANB!aDbw%PO2bt3y{{>OP?yn1C``QOb$5UZ_!M6
z(Z4c{?f3r<oL^hjX9!9hECtdKut|yK2hs=yq!oW^?8HQZzT;pyQCW9*&O+Gc6CQ^k
z27p2ZTmdU<?V^3*d`U@J#l{iZ`qAavlh2~TB+TrAV}D*?t8D|_J=i~6AiYlpjmco8
zwfCUlBWUk_uq0&M0WB@xYYU|u7i9oM{F;#o(iX3RCaU<#mFDiRK8yI}NaFsVRsLjV
zM=Uv(luw!;>L@i3M+vYQd0ZhmI0@Ks^OhENg~!#Drr5OQMffx{@JL8V_3|$t(6Yyj
zuvIxNERM1y6XM7n%s0O3%CbiOJ)8fWO+WUQnF52w`Q0(}ytr5eQUyCB7-fE)FMX60
zsbwN)7?^z%i9BmPS4e4@L@5-MZG9~<POvcyUvGQMc5*uLRZ}KPipm7?Cw|6b$}FoY
zRa#F?u_^rjpD~__0dd4~XK6MebikQSr_j|ap3aMus9Y(1)&s=@EGhrZ!=3hm^mIZ5
z9JhEGAv;YYz3xATiH`=ZQId$%w5$;m^}d)N3@kC%GBN78E)VO$;);2~i>C2ZL){3H
z$Mz<b`d=>>ve~2ygyrcY0rG|e4<qEM$$(G7g5s~jjl&+w!j=h2*B|*j?g1&J=rJ=x
zNv*8g<=9@;@z6#=_&~sFQ|A#eXnF}q|H-@%jg`fX5%=XX?aP-Dr8I?EY&cFZ1oVtm
zn|EB7RCi&dZ;~_38XKL*3N9+M>Ianbt2WL|S}9@5l*`IZ8m*+N1y4-GBHw!<q)SiQ
zJbS7TDOL_nwoUoTXrK_kYAg~h#CDv{R-Gw*5g{T_*o;J8&3ET|v`+|bk_}8DEyFw|
z4xxlezWKApXutyp%;MtwMNF0h{4)1Te%$*d`uRjv4*+v0&~6<JZIEoNIo{m;mj!@=
zVsd?(@*TLyFPhpLxrx9Xe1+Ir`jtPiEpbDLxg8jfvipu7*A3q&@V`^=Dnhy6E0X)g
zQrjR+SUjpq!;65HBc{Rd*yvA*TH)D?Spq1sN^TqzRBMO(8j{(tn;g0XIGyeR$=SBq
ztK&wHG3PEm%Lj37eHGkCbd?Sd5xnDy#aVA)V=6K_lRAOcve8CE<B{@<f}4Wq4d_+R
zG)H>z#%uP3p1#K^(LaL1o@Sjme2RXG4)nO)J1bJ9>|x!2(XBU!DjX59&#;Q@?KZlz
z6J;&ab;BdW+a)x<Kqu{Z>4Fui=qsL}R)=x0<#_2s{D_AbNbd7;=mnS@uY-uD*sH7n
z@x66V3<f({X71s^ag)P@ey;{Tf8{AGf}a!e-$Ecnh_q8!D~!x#VgVR8)sO_q0?9mQ
zU9JP@<~b!s+ZtpqZE2C-c^FSoiEUm-j_3t&2c#u`h5!Mcap%mXdewBNGUKd>O`w{a
z)V(fC4e01mXVZ8<Y_l>PL_Bw)ivpr}`XiRYzc+p)`(Z7EkJLY-?xa3%;VB6c|G_u!
z^<x<rqXVV1@5T;q(dClU*Z!_0`2cTi7w=$scwcLmb4Evd>$AEYreX1<UsX|fR__h3
zl!ftT`|jX<Q+vIz*kX&?eZwq-0mCs@J$#TcohIgnJ<SJv=D1&Nig|l+<fWga`T?BL
zJdz-}+6<Bc%}Qs-nHzwWZ?PAWl-OKAyCejB-)l(5D}I_!c~=hNI>aRaZQ&j@KUoTP
z(c`I{tcCsKTsU71zP6M(8gDY#Ao9?}3ILN~;NZHdt@t-HT%fBFDu0I<ud#t@f5dNU
z@D;!`8t^_Xk9P~~aUls9K%cc>(B$HnbnW(v<jH9x7ZA&uOkztcg#9|{Q#f7cgdZAt
zj^Jo0;jhD`V)GRc;MLJ=cP0h0r{mA{zsDbF?~j?PSZYr{m|6g_Y5~yTHIl{#7+{L6
zdU-yXreDQgC^g#^?`>hi1!U7eJr#L*246K;Ol$D@m1e*0ZQYJE=@{v?PSdsm9b7PZ
zJ%p+qt`USSL%-@~XqoPKiMmS6*n^8k<<tIJTi!Dl!s<_&t89%EZhzLsa9&0)Y2>Eg
z4g|W2ybF&eQg4bWMi`O~MV#muH6|n#;36yY$>w5_AD+HE!M`t$Z$w0a1q}h{6Ga?|
zM8ZPISpTr<BoE;Ga%rob$_M;}w&rPA{mPE1san~pf|#)A1irESTtbqLBqrLnj^>DK
z*!m%K1LGRTzsN-VzqnSpv}Z>Cv(G>3<(E|rEd_EtaN@Xnu)`6s1}2z(9C1{6w^jKe
z?(6`s$A|NnzJpi~rJy)x%f=h~%9`rtUQu{xnqbq$`=QFr$V}j7pZ`(6#&j+(OeTNQ
z7XM#tX-ov9uY`1xh={fkL0~9qHlm*G20nnV4I~ah|8Bg2pWK=%(tZ{z@BRh!Q3w!Z
z)7%<E=0~Ei8d(MXTWoXO+Do0QvqFPMA4oVr`S|5_Cg(Y1yzwabwG_ldH%-mq`gW+%
zm0N+#f{+s*sU=IW^`F9W>?>ocdY>;#?MjQ<+RFDRgMaI-(|>`gF%hHk7$31&3og=c
zNq3@KjQ0%n$DdT^!~oTT3HowqG`gCgb7EC`Wiyfb;?hhgm8rwI>x(i>lu!0}V}*_+
zGVlx#s$8EXXR;h0#im9B0~+k;-|9^A=@5&1F6RTQfAsVpV|J!OG)?2V?!}%v;4xOa
zX~S+*&@=(QK39j>RV!-04E%Whh^KAiS0edeU*P+t2YxPZ`pWUWZrUd`a(fqEGva5B
zgAl-0h&*7Ef3kELpEb-D+nj#=Gsj?qUR&f~jrK9UrK8`sLB=if;8zd`KOOf@mQL5M
z`C3gXMgadu|EN!>tA8NvQJ+rS6RQ918Ng)!tgHjo;mkGK7wyIJ%6Q`8Pf-_IQ}k&i
zltoCTzgrmRUrfMHIalhW0<eBxpyj!01c5>TN^DJwOYrR_A+U<0loT1Z#sx008})mw
zg7=ktql2*c^>LeL+H=30fO`D<nM`G>wDG0eaIcr|i-_uuoL}Mb-6<G4K(|j{CHP^;
zpiP}6dlAGyU#dG~Tx)5#5UIjKU2;5XI4+t0SXjJ<BL9`-Px<EyBRoLSAqGj!Jv<vi
ze8!~)IzUKwuUu%sBS{`cKRoKu$r?|0Kke`C(VVwGcs8`r4jZM&7MUv;xqfHBV#MtY
zb#;5)&#@MePY!G)B0B2l7t92POq=~y?Kw=!*L79T2pCl0Of>rQvCwj}ME)KjssBAf
zF*8lyXeQK}Kolujc-iEkh@pXk7!k(>K~3v^5fa7wPuGvQX?owFQciz3?r*>)cU;^}
zbxTFPh+FMSuuUUuA4$4n#m2H9&mH&;snIz$6A|uPfry2Xq7dAZvT3ucctb=U5d*i1
z8ezD$--r=AyUJjL8jw<?Ge1I}CH?o&y!<Cs@{KwDo;Y`D<oZQainnp;5d^II2rRe8
zl;IdnqrCt@<4^J;$c{>6l?1X54%dhKBWhD@6Sd~|5R0Y|jC)ss3rt2ZVg_SZuc`0P
z$iTM3xOeC`HMQ6a;~^6+W@Xv}cC&T;iG!VU*ta0_cT2T6y1(JZx#W5O?%Npr#G$92
z-3G~<FVEn*?Bq6O`C}ilLq|*3x8S|+=_S^qhZ`C3Jv8(+xcg?^^8o(1y~bMCX#crt
zE%G;!^p0w!#5d!cGulmiF83%sGE(;IjWQt(pdA!hn~REQ!@MqDj*$8r9L)M!^8Ifp
zl2dXQdxy?-`MTC!TsV-3galG3(&u7MzmgGXBtU-C*4_#fG*(TX0S4K#&jBGhk_iDd
z3=i>6<Smc6jUomhDy}ZANG-ek17rUc2hJhT3n57AS5e~O1f(Pl$?u01^M{`ch^@w?
zjiNkHD&W5$eEPv6|8xi0t{{rctD{ktDmS)ow{xn<0=^>YeEAcaJBlI6A$2XA-J=wo
z61;A>`VzMSO`&_BrU^;AAR7g}1O$+?o_(nJ;&UCxDS*zh2Dk!JRZh+Dz(PPkcKDTc
zNqT%@1$byjy`?DoDZ02XZYwP_Q{f@kI1{)PB96NvRz|1-$?ooX_iH)>7?`4R`^Rvi
z%`R8iZ&T;-8dp`=G9N|7>Ck9Q&lHMJavm3wAf^Zb7)yj)f;#w_&$>jiFGW^3F23p<
zNH43(;qP8UiQ~?-v@`e$2L`>TV1PB|AW@y;G+fjneUz?RdI~ankslW<i~iqPPzN$w
z)SZchHi7x`9-acS06w6;91UyTmf5^)bi09*s6Qv;$8#;8J0T5d3FUELoitxZK*Leq
zb#k#m2gy&(jq5@S^SOX-8MI+AKvciK1t><|z17R+jbN18-*E{2{S^9abvtN))Sy4?
z*23VjK#39M{d5^3W5<EE%hTJlKL|xfliGDsl_hQ72r1pG95fuO!;uibdyy!92X31h
zaddF9K03u;;lH&w1^6N%D%bZJ_{f|Rkx@}kVr~fFbp>85Rh|xnC&u#fO_2f0V`))v
zX$K!cM$Z_<B&xsSe)FyFfrg0AXDD}r)lw6wBV!aDv^_zIm`8&MR`zdPr~HpO;p5tn
zvbZmrQpVlW28jEL<R@E3`hx<93LraOg+&t3G0LiTKB-Q%N3`~KgQPnAkIdzR+W3pc
z1(|inW6*T2JvR>L?<gc~G-X$*ku1$-ut4+ySwKk7>V*{VO_iGqQ>9tlA5X(pbS^{o
zaCJ4s_3=7T8ty6wmG0JHkmKRz#5DZ~Y>Hm$6VOWXT5+q2__q3|;RNZouVjO6rVms>
zy2`TQ9kvN5HYE&~tNZXvoGF+`j0Xbb&o?r-=!&g)$td3wY@tf=I#=7B3h{j(JX{ck
zM1Po5#dY4VS1ygI)G2(Y%IWHdMdChM+nl@krW&K@-mD>JEX9-tsMUTAb&0^W{z7>+
z+wx)eR1f5N%XT<0gFGt2a_cHf<taq!QOkCaRrCu|-jl#HV6nuUyCqyNv=-T%wt?&$
z$l*TD2-rCxvD~_8foKx;@m*+VgEL016N*NEtm&_y6pHKi-`UxQ-X~AESk_BK3VH3m
zwGAgAWvEz3Xu)ZRsv`tXV{~oPbZ0B3An9BZ*&DnX3kDXJj`ruG<89>H+M1N4WLaZU
zXy`jkvG#^_DqItAfy&2WXWrXUkgRYp?}0!TXa?4T6L&qPXeM!bFbzmItY{=!t%p=6
zkuqLvb#(HeYT&C-78HGh&wwQuGKbG#KHEI&!vy*%1=mNZS6rq3I=P8NG#n9b1oxd3
z*6yFurRVJnY<m~=!EZ{EYej}f-+s02+@?0@<vqD9&kV*N6swM>+^u?jZtGMRoDCsB
zxD;f*zAX&iTMe#frEI=pyTa;pcMJVK_r)jJ-Fsu<ItP=2T1`Znk|x;Xlbaj24)?2$
zmSm}FYgL{G9|9h3VPN|xRc(KYUT~H(Y|k8>sx~9*ziIe1`C22AqoL@1B-KI1d(cO(
z7WX@ASGnBob3PivE}W=H6xnu!K$Kz7p(Wy*r_wQ!F-bVNyb@C266llD=)=i#Fzdyp
zjLV~tRj~_ZbZB2JDJ`WPG8&dYq2c{>7PE1+%T=has~Z{-0rkz>TOj`$=JB!w@Sd{i
zGDDe1q?`4jT&Me~jN-nUp2nzq7<rE@kj#5y36o&Cl+dpR5i}7s(Yi78u$G@ShWULE
zZ(~EZbf04~CY|B+>8xk@4>XWCoU46|S`LF&D%f|_Io9J1Ijc}=e0LY?>jqtk%w!+L
zR5bP0hhrZ6*^=Q)Q59Z_w*WS7<T^?mY23zN1xRFCu^p0DoXyg%;5^(ExX5LXG49ic
zE|-dHe%0#*$-1e=y-uT?^MyRq_)P{AdH{Fe+*lG?>UOQ#Gdo_aUd&QJ0B`;l(|rFq
z=kX~A5L*#k4zWOTR{1f<*fuBwoR4I&NZ#><IV!O+Z^??r69&-N8)TSA?8fSxneCrZ
zVq+c+VL2O)ZCTQ<$S6IGOd_xMdRRV{SdPC;z>hxmV&tPu4d3qE?8r_ulL1Um13Sn%
zd}imWdGQT6yvejQg=58)eALCRduZq}io*}%rcuffcNQd~nHi_fj&F^EekV32h>1v>
zWIw`MrD6x8VLUx%A5>J`1ZWx9x=9%9Al3$V0uMLcMW+<%DqQ5Ig#KPotu&i*XK5=e
zNUigLo9&uu0yc2z_Kz)~Q^{(j>!P{tkV62iZTe>X6}U3H2RRZMw+k4up^4#eAG;z-
z<>Yl76{p#cd*&h_u<x~=Yw0sbl8I+hG@0+oK3eSM`J?mLOyW|P*7wS9PL8)|&ECPz
zgiu*$%{2YQmALccZx#O(55LxTO(3V3^+Ybu{*lWbLK>P_3N(r5L$)g{7ivrjh=p#2
zni~tK0+~A9wMZg3mCL?Yz;NeoU1%N|E}c7JW!(N!x<K<YHohB>98sLggOCUT-|x1?
z9{Uw*3D6Xf4h5S_+2xD<RPMKsCZMG}8?qa7=`-1h<XA{|)ZnZYBL;3dj&Wn&?s3na
z)soZzG<MV60HiS*4H$>gpS`sMpbJG{I0F5#hk)40?rjwZK!#OUFd#Wv>UhQn0tz`-
za3T|?O$3~5rgR{~3PP5Yoo-IQ`~IvzFVDYo-g{MxdcJh*p$kG>^fxD_Q*1SII_>GT
zZhsKpdEift84G}kok$M#RHnyd2gKo>6Lih@mLBS-Kud~Xmrv?b^9e6sMBPi?nJYts
z3#!?S03W{!6e>6>_c$PdP+EQg8Gius=s+|pcd^%La);|H@#Dn7<$VzNUY7nE(L5FQ
z{KJi$>p2&krzM_v=9T~n0u9H)!fujh5)ugb(H|SR(4%LE%QhICHjZ$QnJ0tu*Do6u
z9Z#<=$O}JmVb5N0K;*w1Y%1*~e4MB$Vu(FbqYM28YreA1g)F*}0rn`!TL^*l5C}Ga
zEQ7bQ>Bm}Mo}{y(Y3xfoAWbn_54hvSI|bzpJJ;v{;>}viq&Iz=S?%<d2z*vYwlF+M
zvFZSDOU)B`+9sJoy@$`V{ssazwnA%tMT~Wv;z7Mlj>uq=3__WU5mF8VUh<ZOBjg-N
z8J}f`)Xm~+1U;$yDM~**NvoLnf5oO>-SK)^&aqMef(|;rA}@aix+9T#Beuh@x;Ych
z8LmK&<a=MV=;<o0jp1N0{qkwPMuaOw@1ZIoJfeBcy0b~-t@_$j%<`fvaz4ZYB0rGH
zUj{c^;8WT*w~zXQoQmZMZh&&?Loy$ju_&@J>dF5p3UXZ98;XU9xs-_<DkfExxc(Du
zVQw_0AJ{J;z+4%E^LV6jr##9JI$FkbfFsH<t`_Da*7r>XA1A9kNbYl(VAa1-yW(ZY
z-7J`>*ZbYgD!j8^qoq@TL&Mqtp^a1$u}t9BW}TU&$VXra(IiV}XCE-+@j1@T1`?7R
z(69un{yS3JFa9#UlrQunVb6*RCUf{4VZ`!iE3HXDdjLq(1n0imiJ6)m1_Em`pdV6G
zrVs@PVtyfT6cbBgfgzuUw~{;@rNS#dU?MZ6teQ`%PqqvAJ{K6QOflIm|3Gz4enh9m
zG_fRy>|K>53I6B%;(0HikpVqkNL=ta_{QN8*u3iUZ#sU913ROoT6j=fC1TEz*s}bv
zUenC5_fPhZsLLK9)!I#o%l`LaNGZccK}EI}>CqCC%H`S@52Gl^VYsp10q+!Cw?pZ%
z-Cz>&7s(48d1^97|NJy5;;SnG-xLgmqUz_M#tVU-#!?tZHL=wHnSy;PCq*od;{+r@
zQO*B+O<Ump*js|g4L3g^HNt5-9ItP`aV6gtKgq_GzMi1wrVa-nxFYMXIyv9sYS(D^
z)0#==f(pr8CSYERfb<fRnn?6o8p@ubYV?Cfm-s-Q#XrB~<C9<V3Fmi2XiqvrLtCd;
zXz1@i?+BdW0$79T3bc-l^|hY#aZKr-iK60y?N|!c(E@CoF_*}-dS*&CQ%Yt;13FrA
zb!1=7Y;PpX%f5$s>R98G;vk;jOF&mBTd;VnuP5D4hAqyriNs`5gP6iT=i-BelntRz
zJU-j;^XY85tg)9MX}vO(W$h2WKJOS$m1n#^Z3r7QjESjm_4p$|ViXDr0oTg^j~6<K
z6u1{I*_tYju0?cBl^8_0fP4TkvmVOdFo_h_-%wSSX90A)_Wbb$XwUpwyl%xi%KQW~
z7w8?UVE(4wa=Jm!+{}>QPM)1C6b%f0{9GspZ><Xrb$AQmoO#_|z)|A>@e|ut>VR$k
zb8y`Y$(3e(JGe8+>~T`9=_JWAG0*4NU%tUgA%w|UNmEe4l=uqm+|^&RP}UkG%$85u
z1`kwfq;@rGq~7i_Qtw{{<~H#BjKEaWjsrk}3c4X!KO|E!_eX-7$XNVj)_Ov`NP5d~
zyJuGs#aCh>O}*6&`vQ&wwoc_s^{&sNdc83cpz#ZV6s!pL>he~9mhk8n*H^y$8BYb%
zq~@04n7~U?(C>gTJ<g$Uymt{7A-ZSU@<fM!`96she(Z!iG?6HY;<|(DYY^zsGfTr}
zv=e8_+Fy<M#q~aPzeDq@yCiXgX0|?FLKWmZVh_)x=1Z-<==@%A*IB$bNi@e^8T8w8
zVW_qPBvqwspwMG~)j1V7?(Bm|2arACy&b39&snK*`~JKFqXaq!yTLn+=ZjE)9Cqf~
z+J7Iw_2}Kw)PIGi>PaEs+o%NW!TfPekTOde#te_Z6q`HKX;bG2I0gd|0lqAlyznZl
zQf0clu>+82IcL11Is6)9MzCPaG?}801XLoRZG?v7!aqe|m&eLT^d`<gOfWnYT@X}L
z=@c+^*?*!6G7)#`GcP_G^`ShDzNJFvJu3rNB3NTDXtPwu#mLhEBsa6!d6#nkhh-i`
z_5(8d(J<ch=jNgYhEiqp^&iQUi`~e*i+0cyuNT4}6%cdRl4AM%P1CtjB9GYMcW!3T
z3G>7|@g$rU;D34SbqKZ||Caj$2Sxl<AsqhY939XN3^<H%#hf`kXPSzprdn>Xzr(xn
zF9k!H+}OTe#*YN*yL_2XNBVl=;W4zy&6%}!0X8vr8@#We8i*96Y*rOHzk}GTfhDsH
z3@s+Zae~_by{+2i&=~ZR`neMb)yoqkh#Gi@7ZEvsWX5V>9n!LzYss<sO7WTSz`&3V
z_Qh-J{u>mgmU|-LNte!USdL#nD_WpjPxISQiz(Y(59Tc~RhA{&MtXp-BZ-i^O8`iL
zkOl+`#wRAtZ$g2Z;v%Uq7lZ`T5uQ~$ol{3z`k>uC6DZ^t+hJaMXIJnL@Wc7{P}xQ)
z(f!<<gwYpTFwdZAWacfA`AkkS-L2xT#!D`-QR&pz7hBWH)q-^GejxJ>&-nreSFW7N
z;*y1`USZ@j7*!wQc=i3W+|QKnuu{d()DOxv(NOXdUy*7)Yv+@bbUGj+^g?{)73O6#
z;+)~?j_vn>Fq-jI(Za@&W!SWSyzM~rYX;-<JlMA@toGz_k)8HS{nI<I4Hu3&h*W~t
zJ=rb3w&MnyUq&ra!zbsoz^tB5wGzfn`PJ7VdEj7;^ZkB$f-DKwdL+OW<MZ2Qi2dS1
zcc&#Tu;0u@`%BB*XS8gPK^HZfksS+2uN^~Mz7|J@f^7}3eXj0^dZpEUu8K<2BUQR2
z5a(2ms5-)0=_j1!ws2dP@3O{>GU_YS`Yb_!7YVp&e-aPEW9=$NZiG49e{Sw+=$IN0
z`n{%!WZU0=hh%wn=+ULxH8EN>84lxUi^tu6tyWp}R&!%p=;HOs*2Ua`LH?fB4iQ8L
z*O_?wNnL~*+d?Vy@1zlft*grqPCV6VeJj-86f6|>?+)OU=TGQ~yr)$;D{wiW^?d?e
zj|9`zGS>H;MrveutVRc{Z1#L378bsal^$evNA(j6a$k!?GmM#+xvD6TwyPO4)eWPo
zzoD+JdOI}pn%n$YYh8E$fM&;-R$8Wdo#pFDf;5G<x3kQI49lz}%<fWKdqF<hkH7w#
z`XHQOem)1)-=J#_|C`4#&^A-e)toErxoS+ILCu%q^9GjiQY3{4U?9yaz7&3T^1drU
zOtcs0;k>RNgz!gnTBon(38G+#P&d%sPbJ`PnNcL^ZkHp66eHBW=HShRC9h^kN=Nmt
zPq&ixiB~xAu#xFcU5~%|BFF{Lb;3<;wg4umN-Emtd+}(I=GO3QLLNF+qGQiy;W7th
z8o!HpE2q})f48h;Xlx<|jnA3$I=@{bS|OXs2A%1Lo;??Il%?HS+rn@<Dnt1k5{;`z
zc@WOK+{N|X`$Wox0TcQXux5nuGlCymQDJaz_^9*@t9Jus^&0P@j$0}4helf(%9T+#
zXCYoXgheFu#a$3jsE0))OIr?NtqyILc~RZa6ffseE{uWIzq#oY=CFrYl;`yd-jn;`
z{}hm#Q<Q$XSisCu!29Z!T_aj-+IeJ9kweyO`8|gM4~StyLhua6y3Gqr7ZjD-F69?a
zc#5=p7wP6bl;N>u)2(%|U<UZ#8oyN0gYPt<A4#)C!*$e>_jVL>utYUwUlfzqMmAnJ
z_-u80e#}A}>kq5Z9fJ{1+9jTHu=E-RL*c$}NG8<T&!63^K9^^nxR1tq!c6J8GJLXQ
zBN{CRET!6|dU{3uc)s#2-F4W9@3p}sRlThd3k{=FgO*38;?gcKzIOT9-Y^t}ZWf~I
zF7|X_k|6$VCsM@<3QYUenpt?_yTJ?^Ukj&6KJ55;$Gekp#o4L(q`=<F+Z%XN$}aEZ
zg`xVeA7R<w(6Byo8Gkr8T~eH?Q0^`+aZ!P$v*~!zr-!7n5%i~P+n#p&N6$R*#fT%b
z3-dQ?Xt$2$(dc1olGI`0O!h}mJ;$Ba^f<aUx6)#lk59oBeZ42RG@T?9p?%5J80#eF
z6poD7I?S)qi!)mz<F0;pY7cCEGJU&JD<=j!_XqSNp@%e57L?LyuLjnTz{}6U1$RFq
z3R&=7#%=oa7);sEHc}7Hm4Af`rmmF_-X~3R!G|R}hT7s?=&o)+`cKzJ-3GFPZ7Il4
zS8b^8zi<TW?lu*aZ!IJ({j>ehS7W8RvVo%XVjo2THPx8m6t+xWj-sa=gB~Y80%!JK
z^^>sW+N}Isz!zeIST_wIQ4`x-*2)DW%NxJONk(pGDY(oa9?kt-U?21+Lc+bXW5Ppc
zN3Ue1To7C74la_~E19P6S!r}iohYbxT@8r|TJ6R`pIIAl!M=Any;HBd>oJqEB)~W&
zGPzpi7|LG8@6|ii;Lr3q>Z!m|n20wXyKX25ywyfjHh$dydhI(wkGl#Kw#^?_bg^Q{
zYZ+EYQ_AO*F<m{Au|KivaSk%>IU9E6;<l?7cKM&&k}Rp?;d1aV%O4zNvm!}^|78JW
zqVLe>){=9erLte`-y(DmbFh|KGws#DQQA9Xs<%cpu0J15+!Yg9fg0Go@?dawtHKru
znl`C1s*~4mC!lfb`zhjPJKr<&o9y83lR+gfWi(5?7ZR2*YjH87{{1)uYm(Mmu<42S
zoM8%RE4sfDnp|jPjVoi?psQuq-FJ61LoR}}*N@A19p4J&vI7;hK7anN`f1XImNeL$
z-*d1#8_zUtn1ivo%DhurZrMjGD`X0_Tq+3T@EV;WK#@c!zBc_J-w3OEb2*_jO?w;q
zKzDpJSyXCYBaET3^Qlc-_uPyG&0<uJOA?;t@|M&0BCpk8D0Nt#KpxhJu8Kfi>~`w-
zNJQK`h!E_h76mcr0_z}?{H>!l$r=evzk!Ksz4XW7_E%EI|LtJvIJP$*>pRAG$dn3&
z%^|dPf|VE-S~<QIDkKsRZzC+apCi(dv<^igYca=4X3K&~I#+j#?(iTJ=(0ObzL74S
zz3F<;YP0hI@$P2A+F%2dI?kW)L>kek{G|ds-uGDhB=8f<_Q&AMXG*-o=|>Q|`^7uC
zin!$M#&@2n6k3|Ed~hRSUZ28|TB<*eIw@G!5l=1eps1pFO~M?8jmIy!%FGn19(0(!
z==U(7EPh)}3Vf?}aJZ^}AxgNJp%(0)9iAJ1Hn`vg7muXOaSyWedXI+o1j&z6vd<bN
z6?1t@v854pN)7T<U3ZTo3?gWBxSe2=Wv;N5Z@nspPU#lzPGc>61r1L!x9$9HBojsK
zy0U#>d|5B^?{F$uC=~&3Z||?Vy7Rk?IeA%K>w1}z8=GfNFfUvC#rfwxt-JTV7~28b
zsO|+mN@XKOh;K^Du`|a66h<VXl(EeetAjj6aQ=@%L>j7M>CU~yA{29%^5Qu<CR<z;
zd)&`E1G<CM)MUeuzq#yX5LY3J?9X!T)a@b%)r_oesg9zu3(EdDK<ug-?avXBs3)o4
z5swEejB>hS(f?_VL_b~<1+Rh!Rm13PTy0=`g{kouh~q44O{aLQSt;o4We|3(tVrD%
zn`M8%>K(|nhLm8JsEHj2->2q~w}jP5b+6OQI+|Mft;6ieTi9v9xUehxi7IFFuCe>{
zf(079DV~%vkHa)J;Kz@`9*I|$H=)nXQ9$f~5wlnC)wsZen~oJpt#q4|UTn8xb$Ilb
zjNE{_V~46zHOFxI*$_xq;dku<P*WT=@>4Vwl$b%NrQHsW>n7t5ru0~_=|}rsosIIj
zbujc8xev<;(_cx0l*X;~88^wg2Ada#+gFsjSb8H0w=~%Hx2vq~%awE4AF67P54=Ru
zbB<5W1cwc%8~a#ddhrbWdzmQm#qIeErQ`;Ri`za5^U^Vk+hvWF+Xvk{%I{q3@Q!gu
zypc63=R`HLMyY>zxVZ7FIW<>S>k)V*#qs!rnVFec+1VGs6Yk4-No+kw4VDx<b%XMm
z5tf^k0a<Bj>Fe$a@yyzEp0;=63aY9w-{fE)n#UrMi>>cyu;=bMq=f81O6M6Vm(ktx
z*rA3_&FeXigl|6mRKZ*8%c6FI<WNffQ?*dCY?Ze#y7C#ZndU)w_G*9DWbnS-238q$
z)>!Rx)<IHo+v|n<^Vd!knfJa4#+~6?PfzPD<!5rZet|f%hwB7`E(+dTl^d@ic9NG#
zStm!ZK`p~(uikLBbGr)eCzF(x{vHt_7f<uYVWRf7@_l*Lb@RwCboDc{U8a(XoGOii
zZi5d-YUr8Ff(XvEF}J6x5M!831!wgJW0@%VKDHg1x|%`j@`|#G>^~;MINSH~i;5tE
z1U(CaPqR?VQ&TP;prD|@_2dvD^-VK1`?sb;^GooXUI;7+vZaqO7whZBb)wSk+QVlF
zxd!(9iA{^cSA530LcH>b#dMv1L2Mm40t{OW4$ngxF6qPzP%707#!8SjB=GQ${^()a
zy65#=5eb?;f1l>qN#0o%QB^hB3Tlz)>rS#R@s@$<%${OI>cm<>DoPG}sdu%Cmq%~o
zcwApdqI#Fu9<}#Oc(2rnrV+;J%$HQ>sW3+}#~-YBM>e+6WEn=B<QfJk+dO|yXJlQz
z0y>Q_S1mjMu`ZC9<DZOHN+>4%4ow`U8n-=4&JbiIuU)c52TXW+I^chYxp50zc?&`%
z$ZKXc|4JvhbD`hb2nR3NBaVT6rMq-W$9h#YKbQ6t4>9sUe!MfpPg#!<^KySQJ5TKz
z1>#x0?@-ynSC^P`fz+E?5-0H<!vSBa)wT|a>GO>b;$8<!<%9A_=p#9Em+?-pI;%Zb
zanpBPK{Z@CE(&GfB_c@r=l4l%;KeH=vx@<;<S01K@W3c%YXco@;zHgOziO_`IxTL&
z!!sNU+;UoIaa)sat$5k}BkGGF=XdrY?Iy-s6}TFs<jU{6gexEJmoBfBKH*Na^&+v;
znRX1?Y_g?^uuYVG^PxICH~lE0`U2xf$l|SmN;>NX-2L}%Nj`(ho1yLP^WAwBu~gJ}
z3(FJYoloSdlrP+8bD(ByUiay-pJRW|Dd@YWkDaD-hUL&JlyWoMdvP)DDXFh3RYf5k
zpk@D(Qgr1{hs6q9qSk3#_lq+4YBt9wNns3;?CH9S@-7qRQq6U16U8go3Ap{!=+n0i
z{e_JPKf~>HW?<an^;{Em2UGl2S?j)Ug<l_bhxmf3uei=VV~69Xp^1bzu=|19is@`t
zzTqQQsXbjZTF7vg07*>Qw({1VxtymEPv`a832TKJagZ~&M^LK5+ViM>uY3on@ZqMV
z;=zaci!vB2d6)iMrNQKh%}r)EI58)&8Z+L7fob0i*MtnqkWPQaJ7%Ftw=Q3rprt)F
zEW`1mf}S<Sx8Q;B<HfbKNT~lADv89PwS^>?TMKZ%Zov-DbA$c{NQJ7chSy08W^Xw$
zM&L`40MVlIQUQT^vtbB?YXg=Fw~9BY9nC6YSvEavFE5#%$&!C3xt@&PDY@%q+PyV`
zoibSBflbc5ISc8Z#T2}5(lo!ketu;vbG0=*=xDM<F=i}ZsYmd%ma;-PSLl!7tST;3
zwG%EZc{k6XR1$s-9!WjRi^RNL)y<PrX>jFdOkErdw~OJ6m#)Imt4x?wNa|6qoAwWI
z2}qbSq}lY^gi74IOCdOqSMt~wRmzC3pqO-<{7A^i`t)|@q@rMS6E}9)s!ie0T7ECD
ztoIlBW<+qDpB_O{MxAj2FXoxd#Bkb$49Y--ym7vP`vbI%s{<}^c&O5Y^S9G6FOfAx
zmFo}(#BybR?J>O%lXVK&rv1bO9H^Y1dF4NdQ{yt4ji>)EsVlX0E5i$3pe|xTk|&lk
z)A7Vu;DKd(IWsqjQb_Dp9{!ame!)(bY=6dRsdRlPH*jU2XF6-S-G_Do&sFMmBi9{D
zd$p~;t@D!zz8&`b@!||A(Km0clsIotAg;Mfu>ouRQ3*^j8{7+`g&-#cEbBnA#z-FP
z0-1wl7>tV#?NHHqvC}`wPN1n~Bu27L^B(?C!w^9QI&b4Bd!1LXcFwk^%J8J~u2!VQ
zO}G0DLQDU$&t3`Y(+kdz?Y`Gg|Cl;{w5ZfuTxw6QUlo186YDc<kTe=Kcw{(HD9y&a
zcVc7M{7`viu3$7fQ*!JZLK__j?k@oFuYVb(P@5TpVoqEms`WS9#XI*?8gLg4cxUG)
zyjgch`i7HDlY9&=eR@m^Di@@;JS5+$SN6J?noeab2%!r*++dunRr5EDUWYe#b-=sH
zxLr!ndQGV`6u(JFvt(f@L1LJ?*y~X4{yQqH_w(P<Fy||O6xYLq<p@euUsaD-P|qJ8
zcHBjSpnNcUsTZC-cN8Li<<-ECoBW=`zZ}n2{~>InYtXNDe=2Sa-R|M=1s3<7gr(a{
z&rZ%kNQp0nHo=vUSSrry58ALE`1zlM7Y1@zVyvxtO-==>Cj*&c#Q@js&|q*z1pF;8
z&%|NNQ^R1QLa6$N)y;t;GkCwfy%I|7hEpb;lm8&rBO;$6R>56?MDM6T{TKtn)VIN7
zV8TCjhg-WpP#owHsNvlf{SN%X&ob>@dvjP+5b&yB0xbaZ!Ud7`<jhxMacTDe+szb{
zcKacLPhSU$r>*tZYEO%C#*VDh+n^$1sDwbTLlaJZ7j%{f$GlzB0vu<gPx)^NR{6-w
zEWhK*DN>3-apJsEM!9L2^=c@p>ac#wiBw^vLI2%Y=jJL<A5ZRSWZ8V2HJ$6Q&Is4~
zs0ZR$755dU!4Vs5JmZ3mEz{|)+lrfG_6?t1UqT&viI%voI9n_A%ih1wOXn+*;4{sG
z&6ZRd?&OC<lAO5Jy{7rz#E^#B-rMNGlD)$A-G?<OIO-dC_2zkt(^q_{rOEFp`~ll1
z?Gb}|pQL52?xIiE+&`m$?*1?l0l`?i!zODa29;VkII|G6WQN3At;?)cJ3(vlr=KDq
zfdhqCN(=Yg?97nkl<GTsu+~P)QCcUi-sl5t$IJq3;m`f4>Pv0<?e0N8g0()g!0QhX
z&-o8BZwTH3cI>xs1#77vqC+1oIT;~PXT|_P^V*SaVbM0vVwe;CGj+Dt1MCL-5c~2;
zEWB29ab0rC3=mYlc6x`k@LGG4?d{V>&DlZkdV+4Y<r)$XQO@#`b1AOXFydyd!-(&4
zD?TPY(!5V<AD|wG@KR7WMFZGuZD^R`a2*&oW%Cv9^ACu?`UZQO<`yYri==CJRiSOM
zSzx>0o{~AU&=3%DhXr1C<#t>_F(=>5r(#?$$zsQ;=0+^;wR8-4pe-o1{j{Vx?vzMG
zthk{GORWsY2ICYF%3=UoZL#cHUc=yq%(Q6l9ktFrdu6~?rOfuStR^yCMbDbjWLAvo
z&`jwhK1^cAnK!+qJrjdJ6Yme-6n%<4^?M>G=%Y~Mf*C>my^kMnLLaXnC|oj2Z?rF5
z3_h)T1$upwoNoimDeFGrf*7R3b_ffY#fNy;Ylmdd`tqcqFSkRHV+vuU7DX)){hN7i
zQf#;Z+rR-2lroA}jVz5MiNaJwv-e%*U+(5N`~4r<_!*H9IQ_L{m;yZ_$Eg%Cz7F%*
z&jh`z$>aF>x~RjJFC9D&@5o+nfT;}v8WKnnG<eP(sFgJ=+59*&eb`tZ`m;MCbXX_t
z!G!ljbf^Hc@2fYwcv;!hoPDeTD<M(uCukg0pX*+3WqsUmt60@dg(;0V0O`)d%)5m(
zpR|gOX@ku#ti?RtJ&deJ0@`Ph?Rj&1B^ze`8B=jw9abnV8t>fns`OI6E_qPQ1VPP+
z(<g<1lDD55fAX0bp(<10fa2ir;wEW$yuT9K=L46w($$h_K4pwNj+;oYf83wGH!uGx
zSeV=E6s%xFte}GTKtp~T$k_F+q}0{TZ6odL%|5LFf{@dF^VmtHiw8^ckb&(BpRyG7
zk2(-#%2e|5@)5y0%(0KmVrBtHR_;!{dZlMIo3`;d#5+dzXC&;mf%EGSa|#uGPO7+S
zx3wrMdMd>^@iBBPeU951#;R#U#k{nSh6f(5<H~WW?c)AYVKiJ%_rqB_|Er*HRp;uS
zk@g8^+vT-jr*Gu`sI@QGncuc59eur68vIuOL-%LVPx%u|>sz80Cvasu><B6xotV=S
z!j|}*aHd4tUAW+dzo|529pkCEy!iPBi#oJs2@U^4(^W@h{XAU|1pxt-Zj_Ri?h*lM
z>6Di4?(P(6X^`&j?(XjHj)&%5{Qll^<R8ydpWU6AJNM4aVsy*o%GK03Mn)%l_3H8!
z7Txw2Q$4sBb2U*C(;qRdI!{HQ5enGpQQBQe);QmcmkDTqsn~u(I*#f^1U!6$TusGz
zvfNL8lVPNLF~7aEhmLso*G3UHvh=hV;w&z#y6OXYKe(!BZSQGXQmg7a%Ne0>iMw8+
z_RDC`)J_j9|3el52@?M(*RM1dR;MjfOMwR)vOB@&Bg(4kJ=fJ2GoL*;oiNM*MaMhr
z?9H7JtIZN$#3__tp1wTNjm5ZIIQByK=jOIvsnw|X<E<%6r|-?rFyrY%0s^-y7#tQu
zeQC86kGBkTq!I7zUxue5%l9FzjQRF)V}K+q0t_)5UzldcMYnt4KVR;e9SpuBboJ%W
zAj>GPs|syve#LAwxHQ=REu7oL^5`c?Pi8+oPW7Hjqf+`%PjgiJ5WD;PCL7GI_wBab
zITMtBkW8eObBQg?zn{WkP=?}6LrF1c%dX16JIJ}QxaN-cS<lmq<nPfxlO!MZ-u@4q
z@?|QZg6!+cM5V7Y!+kV$J&v-IJ|Uz-@E)2E&_vAfYcx9Jd9Pd;&y_=9x!H`;65uNA
zUP>8%j{i`I)n@(qU|6@k&wQrzkvyxZ+DEd{U(tv+`y=G@`6j5d|C>6cBHeswaj$~h
zF`5C5lZ5nYR^)H;tbj=e<;SZca_wWME=i|^QPgF=Oxk901Mr1k1rI)6odg-7JLt44
zQl2_3sIDpjyFJ(A>a1$UfLT#1hW9)?YUj@3W(iXV$xe?5kHr0G^S7X-E=Y=m&iOH-
z<K;N-3H)mc)~i%Fvh(u>R={z+bCss$NXQ%X5eGto*T72?6Z{iI6SYjNbTjtyhfh@5
zw5fn8F<}~+$N!Sl+K<CTnF%H;!<{d^NuTnyG}2*%$qvt47jUIC(RvSj7#ya#=OXs!
zboI&iOfb$2PzDqA@{`)n!WQWC_>YS=5;^8aDc2vD`S|3Ll6sZ#J(p^klR<Aed_$c>
z3OM7_3-|(G>isHG>+JITdTh|y-oP7j20g!@*zKwQ%hZDulj4>;9z_GEb-Z{;XQd~b
zW^TAODL?*lSsFMr96KCge`!LkV}++bme06mJx=PGE%jGy>MA~8Cc9ZvR(H^UmuD3u
z%MBZ{8e6=)u~Efdfpz^J1R9JKZz`Y}1ag+)kSj{f+3aZ?&HP^bvh>G=73`em_JKfl
zX$tDZs;4pLy)hCEtT-_-&u^IKrf)76$m+=>fsp1|A;4Yz?6SA8YM-YIi1Im4oVms}
z^>38lU592X+rsCIe@A2HDjajHcWE5lYT}o+)Mq^<Xn)x9NN^(lwm`h?Vq9>l@tkoM
ze)r;OImc?f8fQtF;)GVhhhJUuhi~UpTNx1t3IulPx}xFyG3(hyu3gCuk{UdxqS7&7
zBHl=iJSGEb7;5d6x2SnaEP>d}+0%=Ylk({0M}=zF!*Ef{X>~+GX4xRYX{saj)dJsM
z{u3dck-T``b9Qjyh|H=RpaV~5_ffQ&c6Paro^~EKe(z)07mpYe8ZI5b@@uF=aaMG%
zDt?GSUygNmZFs;LbMN>H0N%lmdo;cC%X#sr&hIiU$%3fRSlp*b1*Qjmc*ZOVH+zfZ
zZ0NOcke2U$P49%f!*q&hjvl>L`EX&ID&m1uBJ5^H#t4cI$r%e55|hDoGf-hgcL5hO
zC!+W$d;?3J0PrD)3iVr?yQ%IT3NF5PK4Ef~e3EvwKQ(&IwY+lMxkzO)ags_nV1$gU
zaz(qq*HN~(dDLEP3-jl~Jb&OtGNI}XiS?V$cU(}e*^B<7KN<y5{^jp`2>&y9PzrWB
zV7#npsw3QSY;NVW{UW$}a5zB!JXQdpEuEDIsnG({aFv0IquyR|ok!teN{v}J?w50H
z%1xG1tPb?ABYS%e><I)kWaPHKm(M=SH8N^(QuA8oqYcZ5=x!4C2#QY%7w%d}DpZ_q
zkGB$Q!-CPAkFEMk?ft9)Gbhn_*R}|Hk)!Dkz|GD)USf@{e+2Lt4(8ZBZ{ZZq{=T&c
zsaf-QZ+MaJ-Zv@uCt_Rfq)J`L-S1Rn6blOWFhPyS>V;D<Ni)+l78>hP&?K2Vq@V*8
z;}DLB#|p_AgB!l3CVRPasOkN*;sm=;qvvGty+^Ns*S1%xIjLSwwT%xeuD43(EyV2u
z9ZmhhDzNO%sZN5azb(V?qW!PLHn3I-4ZhK4?RwZT5n>rRh5>I01nb2A0TQ%{9A??2
zJNQbK_T<jUnerWxAS>Dar52N|w$;r7$7bSq_->Fu(~>paZ-$E1gejf23!`Cdw7sR<
zS&_Mcf@w@GM~szOQ*OTBXYj`d53DbV*D=}53~;~l?qbcWip;|iR{8|-`}<4W%`jW5
z(`O1r2Y}~Zn)2D}PNK5NjQ&Nk%LDbt+Y-)awI$j8^HW0Qi^8IV&ddQaRNg^HB#)x?
zZLNK^UY<-!^hNT4?zy38Q+Jx85`%}X*!^cpnP1_Bv7QZ?)3%Fv);?X{t5mvy)srB_
z+51Tg@virESGb)$-RgyEVd+DGN>vF1*G!qyM=L>?;`KtR<XA_cO(i}*te+_DbAnSi
zGq%Nrj0?R9*H$0Kf2pWl!Rkd-y%{F-N3ESOR!@0H$g6b<$Np~#t6%K?n~FG5M(cXK
z4tR$6%Vs#XM&l3LD379=cM;ds4RCP7eg=gHK%^wQT;QSgp&>tClZLuQa!$-vk5@J0
zKep3lz@BZo68@gQ^5H2D?$28&xU!kYX>C6qw#_<$s)yRw9(L_Zl-0icj(;wllXdp>
zEe*Pj&aBHz?5XRC@Ug~%l!K8^S|_wfzM5)JCeQ!~?Bl^AV0mg76rDsRSZk=O7DB?}
zg5Eni4{h*nE&=fJVUc1tjKn&UewL^AB7k*zJ7Q!J9oF49<ck55_3rTw_du=L$DeE%
z)LN<(>Kz3cWbrP>Mhe@N+n>tfsj>#c;tR_Hk>o(lg&zBIBeOdKt1{x;KSnw+1sU)w
zT}?fxPdDkJ>_K&;R-5XZT|P!!8iW_#%oKR4)*hbY-RV<N4=!&GCgOzMZ|Iu28T}W2
zV3*LLI@R!#E1py;%+=}ct=c4)5zXgvLH$`;411Tt<r$r1J9++JwZHfO#1N=Lj-}3n
zO=;>UJJ@|i7m0x-!H$C;HjAZ5IMqMztqI4f_FJqlkLrGOe)Xkz&`{;{RAuMpqI$25
zvNV+#97e>=3~f)>oPWf!BgR(dP7F*NFmHKhW)%u{*VP0Av#z{Wx=%uN%iMJfr%ji*
z=exfx9&azj6z)&nSqeWA9rX}S<L@7B4g=Fn8(x7oU_x&;L7XofaC>9K3uZ>4d2@t(
zyoa#9o_|oReWeansn8FHXkU5*@x4Fk?BB>r+T58#VXwPt-w%Ua8MmC9Kk<Qkn)4D<
zayStftVGzHpt%ZC<8>xs*!NQ_Q3Y3LJN&!f&Xii#w!5mnIvtp^pQ*QifpeP)adYM{
zk!n6t0<q;HiNk9PQV|<p|9m`aXhb}p8&qoe0oDbLVX+~ud-!dTB4SEy%y`2<G)#2R
zbAkedK)@>cqzaiH^|T$K!<WX+V=&p~s1u=&3_!~Si3IUlOL($!(|E3>zi9BPPcJPi
zoGNO-v0FW8A`}VXG#zy?#^pS6zd-j3agG2=+IvR}#@OHMz5Rn7vy-TF%dkFnxL?Ov
z7O5!@)FZI+50=Zdwkh$u(TA2KK$j{R{S}j5`D}W(&eJ>8U0&^vt9Mi7X1jQ_$-E*!
zF9MAzaKef>l?MX>r*!YI#_CjR1M5g1zEH*u_MKPOxmW$+;B#QX-j=1LK6x?2HfuRu
zakq0iK{Rv}w$|opU?s~yF+|n2v<P3#M~8E1=xd%&r2PHgw9jhh|AGf=oOdKu<B{Vp
zbqJS3xomD^*|=;J?qqzUvv0!K(nl)p+JOhTI;lfjwZI5Rid>m_W##!Xpcl+Ra#T2d
z-RrFqCwa)C5mdLGbjgKf<5IknyvsWi{OlURup)?@`O{iVggb?B?W&6P@N@$eJ&o>9
zbAfWtG^R-3@n_U(>Hi<u^<fXqSk-w`C?fZ^6(JOAW=5S8_qxvwXyubjp-rEP2Og9@
zvaeEar?>TflWOZH|0P7jvAVY^yUWJxB;M%=I_^Qdrfe)6O(myc4=cFP1rt3$J_2%#
z<x$4*eVmLXK09q#M?-dl0wy-6+Let%nOzK6j{1J?Hsg~Um1*Y<jBRwQ-7oK7VLO6#
zTwQDdMVo!RQvZ;(QP96g@me~&#_^Vt^?h(2A$9J~UflGol;qy5(J>I;YZ)N2F%^H6
zwi<XNWQDK3tEo8*KG-1-RUGpKWNX~PaypX-YpWXyL_|7aKw7Raw$IG{)?5(EBQ32i
zzh+dr=xvR9tIpII@vJ&e%f`81Qj4=nX#lFaZ>5y(?MR7Zv^DCaCX2ORw`Cahyu?5e
z>JDOXaov1xcPQDz+{jJDV-2WcMjIR5KTG!#VTQQSgev=E1zmonUhE94^bbJc4W89)
z+H*^Z@nmYw=Ef)3^R5)QRB8H$Z{=eC|M8fN$MK^@ec7oI2jm)y^*!H(ZEJ6@h--e!
zH$Jknu_Fhx4G8@akybe5$3f0+F{G<A98iD!RjMpzo*LOb@%P=@Oq$e4&Z_*NR7<A$
z!k~xHzyC@a1ku0D{@kid+$k6?IZj<`VawFH%yQG^eOPfefA<-tGnLd@?_ivSP%!2f
zF|jb6A!pP;t`d<OUrD4v{MH|ELX3Z-ep%8+AW=1Itsf)q@ZQ4{q+GGHDC!vptHff*
zcI_#;5nYXJ@{CRy8LZ}1e4&r~kQtX~jg5C}h^y6nkP2Y_&eO8AL191Li8JgSs3+p%
zmm}D~Zb=I=ryFQVqlVhEH<7Evs@%A!$|!I(hG!SS?xRz)8EHXItmvcfM*nTY74vt!
z^x#L0Cv<UG*5JdUnc3JsACvLzpO1NK8x)yOJ~($p1f0vkkVmqaJYoe8`WrpaDyFtY
zJpo3py>+Jz0#w=<peg0&1~U=;JDqvKjd1_n5%7)E{q3;RrQt}8X{`>D-}@}?nXqeX
zY0=eNGbyZ17=AlN%rXhRfIe@H_n!&+*0vqZEw--t)$d$vYftAPHJG6a-8UtfHdvSM
zSi)ik+wQ9)Rkxu=av>)O-N`&vDJbeVl0ekER%*Y7+Brg~`jZc*x2q+9v$gzww!#RS
zo*en~7Sq@Qa#k!8)@Fr?{I%TXOfAR?p=B9^>{>ukF|+4RYL~Qx2kE@*D>Chp=$8%0
zF1H7~D}tZHM2dzJOilSn!U_kT%`#dco8>dbRz#>{q9?|SP;gTH&OFmq+{*<z{n`0Z
zz8f6%ZEP9_TMn1!lW!8~JbnhEnKgD0KHVjEtd}k@x)x9$kpYtn!2R+sgZF_VYI9s2
zluVOL=sgT@vch!1gu_oProm7k5kLV<vZ!FWK5n$5$t8JR+TCoB?ELmKnO}3wls{o6
zP6h+#o&d|&t?y*ju{m$mP1oQK8@NsDQI5Q1^o9n%nD=|3%u&K9T9vw&)Uc(&G|tjZ
z31>TxQ^5f~o4`&@GTyBv6-d5>Ai|XH66?g%zpyIdK>~gty!kGTic_e|R}Momo?qGy
z`pp?YqEWN;)9dN7LGCcztBd1_yly_z*|wXd%j&u3+=}KubF)#eZO2SR5rS5moIX>$
zW%`MRzgW2Pfp-(9(fW^1&-wlBg(6qk#-^42nVqv0QAGZg_9M-^TSpIn`OV+dEUj&9
zP<6cAIac#i=_M)E<(rYwq95*gKa9*j+``E732<I(yk1y;b)2*{60TZmSF&1U2wCij
zdd?Fp{VCVcYn~Z`j=}L9doeHP3J;%;akdIKf0{FcBu<VDd+0}wJP90FX`KCXyt5g(
zlWQ>}(|l;k{Q=qAd-f1Fh<9&_11f^nCuhP(M!Ea(n=Z%YNeJpwE3~c~1ZO)J-Rt{d
zmdsxmp@T!PIgu4$<qvVPdtF6~oVavJlU+4jexiw{EJ8k8YrzHepn>fxq<EzU6*<~5
zw!KB!xp?QQDcW8ZT>rF@?yWqSR*6cHO<Q)9ltLAqQIc5N(K6O+9ldrW;C3SZ{rS7j
zO1z~02%y5f|2SN3xOH~s4|g}>re>b`qji2XTGe7u*H3Ebi`obNq=b<Q_nYz+^Qeu;
z(*k!0=BCS_fr_0aWO<IQHJheIvi-=>7*GXDhh+bH2LDq-UrS@=$}K=<zpCAHYf8a3
zxQXscs76d^$JxC|w>mXN?@49tlCKd7&xReXF*ml7(Ydx3(r2JkQ2RxXwq1Tc51+We
zfQw7++%@2gQCW4M+~xNk=B{Zm_5Nh}?MZd+wbEm55+he0THce>dbXASTs`Mmi=tG<
zl)(&ST=%1!J8ER^nsuy&HnB19o+RKY7s-(S&U%bkbqg2G2pHLTjTu9FZ=?643*=jI
zLtB-FH}YKnCI%IEbhXK7@tTtpv!H)T()z-;$x@}zJ|EgDnbA@EOXcVUc&)BxQvtRl
z6xlEVE6dt-=uDz}A*O?G-5`WSJVDTM_JqW|Y04Ee8UAQ$yBe6ze)P|LVXCDf0L3z_
z5;m(8=vDoEgubHxzrXvPln=9xDbY?m1J_o*I!Y`$@9T~kHo&Krc7}f3O}r}31D)0y
zqI&XCCE*M+;6<kY&Vz<{y2w)%fPU-Y_Vc9xO#kinqjF_^O8d+i5BTbGP_>n~_=Fm^
zR1~7Ygg$+nym_S4{q*1|Z7!z=v9&Bb$o;->43n&|k-ZHXU{0<4VQtOYV62n*^^FH3
zc$e41t5vM+o}{AC^fZ`<bQwT3mN$*{kq#(BqZk_gd!F&I3rV@764zg@#*M~}+*4(t
zx>!!E6y=pIk9S=*&AY!!chNiNEt1xp#GtaN_JVVv$CbcSs;FH^1Jz$T+q_bm-CY(%
zP{{#4H>Gb?bGU?`C^E6!Bz)k7SH5E$@h^d2jpS^Ps9!l<HnH;(aF@`R1zHY`d9QYN
zkq7ZbI3vG_zZzD2rj!Eq@+t`$gfb!y=c~`RQ{3MZ5J<|%j4e|M#;y;<Gge0beqL-d
z`l3s2N9~(@lUoC9z`N@_ejsoQfHODFyw+t9Pzfw!LrX0k8@uU+zDQ!g3yt2Svo1I&
zc-Gp0@Y7So?*6XPnrD=ATW5VI(A$~lnX6M40n1Obm^It-9k%f@RS@w@(!$%K1*{GE
z$}hMc(KxrxZFKN>Y7NmV$u#}=OMf+Rla(GRjh9uvsKt#r@fVJ+?O=NqMq-lsvIEx3
zzv(j^l(0FN;JN%w1^G}-ixLzY)Y{N0$LG#8;0V@2CzlX~G0ul+GyM5(tb4>SArcfs
zPR@?{hvU<OR=&|MOzI$VCovN5B|OEWH>#Y9a0Kd1(fH-lB~SGdSTW?SXCxNmp4pXW
z{CrM<GApkD+iNOhr{7|nQEHhd6vj;L)c<a?3rgx|m7{#l_n{whw4<sY4JWc~LpT4;
z9y&Fzlv&|+y=8K}z4$;uk<0S0F<=c@`%tt07leFd^Ay_WsVZ_@l{mUMIQyPEaymbd
zP~x^S^I=9E)Je!2@={g?nW4gsnaSBBKh=R3eyqiEOio{@7^9cGR5Nr@#GMN}2s7U&
z3f9pl&!S?TtCypwmcdV4jeYq$V~rZ;2slh(cenVsALMLRw@~!+;SotLL4b~i%BjS`
zu`?DLhfb@&pIuDH9)`js9-?5)bLK);G1%i<4OC;G{rjoBQZTdziy&jcK;snd?WWZ@
zzW}Gz(9d6yiOQ?zVj1IB&v5_xy9V-8nKyU}5gzPfY8QuiV^=Ah<w?I0dhXLVr-kZn
zl`GmEn<!RBYc~jLZ+KN(`M?oOH(Fa;dkqf{O+-x{GF7+<)5$3xis=AuK8Hs_BIV_6
zzOV;pGu!p<N=daJq4yc<<luz+U1FW=8NfDh;0Qz{FH}%Z7%?L!4@LheRDGw43FpM+
zO2?htgz0Jtj~y8M83WwhVA<JPYg8^D3QIhb6J7Okay7FdhH{9i%mp_(=$AST{O;uY
z0lgqA#9pc4&bd}tXI5c3`8ei;gg==JIf^jFhYf|RUukP|XMrBWUz-~{Z#gl+%%R9<
zrc%{On9Al<Kce4faVw4;CT&hc$BY{}>I32zhid-bwp_54EG=(SW4`0jLIGd??cEI+
z>5@JtnatIDw%HF_Us1cm3A0N|-eO{6dU<<;ht70{9NFxS`QBdcXXFvPE*eM9ng|t7
zD{_5bxH4*J;nW{W99uJM_d}X3*GIYBlP|T8{ufkj4NPrnV562HRb*gS@fxF*T~vOc
zYp}$Ni=xQ<K&l`tk<pIJt22lc^$m^uQQ3@aPF>qzkWd7>#P)-S&QIIXS|;)hY3#5*
z63>xnu`{WmjY{IfLVij&l6H+%qQaz$OFk5spKUoglmHw@^XAqTgVTk<RI!Hpplfe$
zF9|t$x&APL-f%Kf%P{ZLK#e6jT@_WwVR{#0L-WHG;2bymW5q20O=_@bsl=%i3Z<Ba
zavGm}RV~haFzgdx@{PvR1%_?6r5lH%?$gt(;Uhsx+67KkS##}j5al6fqBM-)Y=8I0
zj{Km)i%Jq!=3rn4?9ESTvw_%Xn`kR5yCKFN<R2QN<3~Z6WHSNj4d5ZTBnQIQb-S{q
ze?|C*M$$;~m@(~)MRh&QDNP*T1dWyMGFybwLTc<zauTL<ve+FstanX*TOHx$o!aG&
zy!I8sgE3E9s{-7+ASo%S&E7<yY_{a$f!tnVOABA{RQ>zvW*oW`aBFYfg<ZZvp%KS_
zbCqejyL!XC2DkuDW|90==%4a_r5Nq={hgrW2Rs;-EQ-#>(WcL1a0QbVT$oV$%L%HV
z&tBIqh^~G$yL^E?V8MV=so<@es(6W-&+lkBYw~1OSyr>$tS6qGB1U$_#{}5#uCCcI
z0;y%O>{I)+%BlkYQ(o+#i-RXEROhYYY_{qX4&ouJwPJysW%VyRaq1UtZlA6QB+<KQ
z4o#guViM#C?FVb4CiV*J8;6#gPu{978vzx+rsOHg>ga`OKEck}`6T04m2ePeZCxHN
zr!zZ$BuRF66S@xsydyrjycEoz)CRKPr%yzrq`n!#;b$OpXzS=$N0wu(b>*GDC~j`%
zrKj&pN5%=w@)~`%*|*irsYMcjD!8!ov=qK`)fv&V#VkfO-H06ZUtCo!M)<0rhql_6
z#AhOqc{`|{-0K6++*2n{Xjv~EUVITKv8>AY#30O$T-}Gs0{wH#D%msjXt1=bnA5yN
z8zot9*7`guy&(zq2=tr%0J*{ev#95zIL|@vUjs$h1btd^qr(pX2o+Ui`jdi6j_bqS
z;wZr$Im(?w<i$mz4&+4JU*<m7UuIGQ4bMs2+?IxVk^Hf_A<8elc=v}cri#hhR4MGB
z=<^pr<K<tMxx160CA`KSNf&bdG5s_2*r+`%rK<Hx%)*?R$D!ul#neSH0jE#=^@k(z
z6xS=|-NSZ&6b;7s@8I&Sl@$+fZ<yBB)&=`zS5^+e4`y{Z27R`b(iuY0OGAT&TLu5k
zoSx@P!StheV4Kcvtw0r(zMw@>nm|@JNJT?&Er|Y@*c>Xe<IMaLqe@!uKAFx#KfLu+
zjfn#ZpWwh+He9$+di?6~NRz2u3Ns(0BSfYFb}n>J+PYVMyp~(~o^4e&vxc(=5u@5$
z-#gVixJ)xXLxJMr>e6F3RMXAaV2f|9RH+I7(z6rMXp`jU$zcvk)S8)?u+j{Uy7^1j
zh3M-rm7nhTKUQvFOKNBq4n^FYXoCqKmEyHjkPAb@Vgxv;7%ki-a^`wJ|5+8@ysdIg
zb)M8-@bao-YV91x58L3=m7MhIz--$R1G4>yji$jM8R$^4f9u`KXJSGPh0kG=GsW6T
z{PAN+Sy@|OUy#qs&E+m&cJ!umOkQ4ICDqm0VzIO?H=D7U#$w9psi_{MVlkhsHom#I
zxRg{@W*UxXi6?W#3-}`|;{nI|>t}xqz1afW^INN`x~8XN`MqIJClxiKx9+7ZEiE&<
zZ7z03RR(2?r^}5dDZ4^(g@JD^1uvbQjki0Tqc<EQnJQ40%{?bGQUH;OTq;@3$r-$a
z8UNo%ueF%<cL(0?^O~RIL0}1g)$;VErP3?D_>tvAixF!D-m&kC#9{Zv7gabtnJb4a
z^#Xj8C>GQ^1}Z{LORmJ;?rcHVKg;uJTc`zI12s%B%_o&*0gD?%adU4IDcPnyt4>6$
zcQz#@t;obu;)A6%9@AC)w7H+z+J*{~%n{yv_Lp)a2rbHRPw;-^F!Xz&)Xqo7W$(xO
z7ym-e1){6oQT>jd9gD6cLS942UH;%B)OegYWsH*)(=<*y{CIm!tt(#^ZdnS?69OF>
ze-sI;wQr(vE`so(1OH3mW4*~w(;tn=5WRBjM)h#@;%;4rQq9&Ef0&KCfW=FK?E;0Z
zcx)CPm>MqcG&D3M6WOG-26&(DQKh7$2;8o<6pGYDtgILW1qHtf3p1K8adFu0b(Cne
zf1++s$IWUkRByyVK|v7;CjeE8kD}o7fR`m9B`tM(a7uQ6bOgg8iYZbb&*O|3hyVU1
z*_$pNet;|wgV*-moNrT5P<$XK_a_z(D*->s4+*!c#S?~#OF!Li!oic3h{$8v{R!Vg
zB$6y5CPrbr{y9kc)NHa%TFaX)?7z}^h{A4+8OhTW6)-5g_)$3UMYXbbXwox{kw98+
z%?%3%j+*-HkLU7xEs+-rc2jRsF4-t6Gs!dTOz&4%0VT2OMu;UH=<8)`bZ!2b$&zSw
z#}UAv@`Y7@w*}F$KsJM;95mBP41#8a%#@?g<_P8Ivs~7Wwe3lJwN}TRC=`U$4bF3C
zAB(Y-T;3^=X|0=dm!J^Js(f4ES3x@bdd`neREed|`6^bFdj9h;l)0dQIL{@g4;XIX
zF((1;*)g*Q^UA7B=U<e7pbH?B5y{a)t#v$;p-envdt+my=m+QT-@p6t&KK>M>1}qt
zgQE1l#c8l}JJZF4WRbgdkl}bn|J8PXMXc<+yoBsolPveAM-arPMcvZ%<Z4?IqN3!s
z+><zLwYx&FK^?zzvzu$v?H(0CE+X36+O_l1kdAl%mQ+5oljMEn<7YNwEt%9!AR|^#
z#*INuP+n0hNm;7jl=GQIZ9(i(m)(#2KN2@!BU#f$@y^p1eW!49#r2<3>F#+-@KMxh
z#*Hy|d>^h(1c11M3I$^URlg__9SfA^b|No5n#L2?4x-r1!zITi{pU<i@Gb6tuwEB&
z01Z9Din3`T%~aSfdwZi*ly_NCs?|#=liXZ+2))Iad^I~5)crYVUNP3Znk;;EV)i;Z
zXqAyL)6M^FK(Bvqx29{}Sk;=-Kugh^%xF;u7J^nDu9Hb!rn&^)d0^jWw#Q9(X1j;K
ze!W$9+(d6{Z_k&1`>>M{K4QS+a(7q_7DkI9{u3=N88vl51(J_?@pfr9LH*GcJUl$t
zX$LA)O-)S+7<2L-6;(_h4if{z>iR_M_UaI^rOEY9smf%Q*>Wg_ci4KETk-bX6|yL-
zt4jhZ=;Or6-oV1_?(Ui_)<u`<bSvDl8Gjt!*w9T+Plqy?%tP1G(%RVC(q@Pk)lsXl
z@FaA<kC-Y{RpbzsX3V<}BL4ijq>W+)3fLXj)BQO{H*-S2!|1;eRhGd)APDO*hMY)a
z9OfaotM2Ryorc*A9dV=$&=$bd@iM>G=e3_|sCxZy+*`teKF3bQ7`A9D%1>SS5?Q}B
zQ0=pKWH@fdHkbEvrB#w!E~+sdBpG)%gW5+C{lnGBtgjc0vf2!x#JA&^={M}v<Uwr*
zE<fojiR$wXyM>S-H!b#-iAuPB_Ty)WW3Jg@EzB+6mF7<&TFUPWBt5TaMUOS?@plG)
z$-L2020#c3WZxn2chu?<dqU8UdLy$cjBcZ11+LEUAP;TV-;Qb;{h^kY)_CSAQN(|<
zqU}ybf(g@?ZUk{XdpQ+`Ua<(QUMnI^z@On~A6nD(8sdc$H>3WL?$$s&)b;hX1!U3Q
z`uT7AhyCjQzOm&(KtMpg)2)iK@@SK*GcqBMxIZ#Jo72T7Tow}|9-d?mkNl^?Bpz1|
zsG_2xL{58T%-DQP<;v314@^u^g9)sR1|vi)W(z&4=8c_k^g2a2kUpy7sHMk)^5Hn(
z`{W_UOgIY*3p?}G=7#U87>qs+yI1*M1E*#<Q?klT{{jXE{>__>Re!=Fo3M1f35Q_V
z=QRDbWo2AacQo#zT0Uk@{rsi_I>e_wLIruZ<6M;|b-mqn)X+w>-tSfEiFC+VyNxs#
z-f-vAVxSvc_{J>K@nr5A?fp4UtvucLYO3iShgWyGTRLg}Wv&BZ`^VeMb76a})(QG=
zIkL!AqG7M}TJDuH8@tH$qdO-gw^nyH+ghf(Pk_#Kw6k`E4vJ?5cpZGY-JE@4e<Xa#
z;=g&jo`3#%{f&zwft!@yhBjymJY8wS#J;;Mmsvpn5`c4cvz=lM4n6Ffnz~B~8JYLN
zm~?vg*QeJvH>3HAbikERhCpU?!lP?76C|_O=gpgG9vs)ZLKPQ_r}vc`9O$9I;XY*+
z9Crs-O*GtWB|w2tq9IRBLrbf-JC+GmVLBf}A)6%^Xa93TE*Tmc+JLF=rLQ*}%9KG^
z`j(s{>~S);vk?QzKe}UHVX9LsRYBghYYQFe@G(rWo6UiPai<Wj_AVN(#)9Zi#W?&%
z*3XPlBa0KO)?x{mdll9k#s@o4SxmJ(p#WOpt3AX7r*keKt&?gF{lgeGd!pPJpvR73
z;{hJ<#(3M16cQS;z=jDECIuC|Mzs_3>9;8up0?*Wi@MJ|l*ax0n^-G0U=|gKy=b}{
z!0-DDyVBv<7>AXcGxOqsy_PN-8Trm)GQ+2?PT-G1xHXa{a4=UHOru&Ag+XF}I477P
z6slBhh7Vb=mWnSW6Oa48+Enx!1%=Mk8cJ09Ml@wkKiSa_Q{7=Tuc=beOykC{Zq$If
z@pvp#Y-XOS82Z^2x%1lY$;mM<95-d1Ra8}N0cgL`3P-5BP-E$GyPM5*f71S(f&#C%
zu()W<WG>AtPkUl_FjGo){$Cj7jvqwR-SW(+I%~9l8dQek;}<f!JkKKq1l5N|mG@)&
zb5}2Iqtj+Qno9oOcK(2hJu8;x&l-CHoo#fbCl5WO5=4s*D8BE;)KA*UQ$oTSJcWi6
z!nYgpF^&GHg?{#FB{^@)MH~f8^c4qD%zz2V1p(mP=nToSKVG`u=6z~4VF81z+W`z2
zYH&P<!eY>UZD?p%Q>Xrf%XqmzO=nLC;v@L|;C!(;UmeIp9;_ceHL9CS-;Gg;iax4W
zFp`pHUf7LiNzjC8<^JhI)dbJ}e3q2U^-3L1wcFa-M31fQ?e+I2a>Wzb-kF;p?~&8y
zeXV;${IEwI+51gg4l6fsySKN_$2)_H;Bs#qbWYePFrSdAEiixzywh#TUvF_G6r2#I
zSnPoOH=6TP|KK3W%+5*v`KF%TbvQ6q;%`k36sEoTy*HQ;ld2r*=43t+Yq8bAKlXdQ
zr@yylbQa%y8TKXXv4;>b9ZX*&x_e-T@l~W7&nr-@Py0yl(%x9e0<x7_^fi_$8Ua_B
z6X6^&$~P|0+VMQ9v(zM!QsX;A!T?E12stBdH0W9a(9<GsJ4hqJ4}`?Lh>1bjU-82K
zU)G3K3bAO*FC{edj8;VxH!c*^$?S~NFL1CBq@@GnFkBvt<(v-BLRX{*T{p2y%Bpka
z{^jyDi$L?lVK(Xj=;QSKT<7*;7b{dKhW1CFtQ5EW7E4S=W3u~@O<|aO5x@L+xqaH!
zSwlKlW0$7~2(?<xuj^|E5Yv>J-ME)sA^79jQu^Q%!0|(Q8eVc07PNFj!h52vFS$TY
z_XtS+_%VG3m$C~(V&bbr2-%;!KnfvOoxNeX$hbIdrmnad?c-6BD-35$bZBX`s3lI-
zWnLj^Vq5!n6Wi9v+iwmA8141!v}&EVdANY_kS!K}XC`|ZHk+$3%#=!(;(e&-s(SZw
zwKY2VcgvsU=9QJO$&F3jh0B`EN}9uZ-mVf%o|gV<N1Csn7O21-vL<RAk&6kK5ivHG
zMctbS)?V|M+B<Q%FMY3SAka7~2c<s>R4>r=!w<jYH#(C!LtsO9m@VFTC&wGeopI3f
z_)VVkuJ<g3qM>grUDkv)q9!?Ww?2PGVN-bPo#P6a_?d9d>W^uX%?3G&k4G6^_FGVs
z%s_^Ib942oAn7oVsXzcmKZq8MPlleUgo&%^`Xk?;zXf>+jOQ~?{k)0uHa_<i^(ly~
zRA%*Igx|Ms*J4D9$&I=Vk)N89d?HIT<)f;tO%M3@R~HRf^T-i09XECQblE2C$<B=^
zb@N(l5$2<^e;aRDoNu2BuIPMCnPNgm+qXEuG=+Z2|Mul-7QIealq7W(K#f+Y4X3B4
z*SEJZoMFI)uXP4P*-KDv9l7m$Zvc}WYjn2v)^2Q^W9HzP+pzCTpd5Po-M`6XG!Y;a
zibJp20@d{N=&U!C*c(M5i^Zt-X0hJB_+$ZrxqSnEiLU87z2cxguUAFj-{xdY@6Xn3
z;KmF-ysgZ886Tf`3(akLhR`)Og)R@OsYN141P}F=V}9P(o4)c+b_>>Dz801H$m4;0
ztUjDLtX>zrVXh4WUA1BY-Kp}8lHZxZU)I0MYHvgV2&4o1PpTfFKW-x7CbI4q#N85E
z{B-VEM0iW&`En2^6uo4S<0)%?Nf{lWD-0xbY2>?Uw+T0c36(mNbK#QNKX(rxvn3n2
zHXJmc22u+%qP~^M@)g0ZAy5SL=-v6<96TD+w`<)0BJXZegb{Xi7YUZoULDtvj&jwJ
zfp>mYVF`sIR`N?Eu=(84U>%{h;)!e7?$6g^5rLgEt0&R!tHbx-&U|08CB72a$0gWU
zrG~n%gH<rRl+A~fUzVfiAc+0#Tjk?4Fe0ve_i3c<@*@+ub>N&^P|jiYYsYj%m@IDZ
zwy}Vh`bd3Zy?TeL=K;L55AD7gwKk!96#O5&-=HTB6F*on%r~nIMhJ3ZbDRs}RGG@`
z&NL+@h`}!rD~~igp|{xAScc-*Q{3S&AB`oB8>4*Nw`Kgu?ySSxl++2A<VEz_%OX<9
zI^QF`<m3ot-(026h5{hfO&)@G@1*M%>TLHp3C_>XD1WFoxzx4>%1RBZM#et^W9K>_
zU?eNtV3tpere|baKdfHX>x-g*GMOpy3`8ZL>S^*h3k(S$dTr|^nZ4O<s55g1A?!gH
zRk?Eak8#aTt!S**s;?tS$@x(pc<s#LPMg#l9jxyFnGaP`!DiPA{jUFPieh>a(M|C7
zzWR-sUw7OMFG|Q^Fvu))hhg8-3Nx5qRdFar{P~U5-Ccb7f#QRL=-1Fozt5PzVozad
zOR+FCh|Lu_G$e5btj&j3PNgIBDhTC8-f2i;_vL!4wx}StH);j=Gg(f!ZyIY#y#G@v
zLVb55P^1=;I$dBffcCTC?rlvCgqlj@20lpvE;(k5d2f$eBIHWrRIua|^h0y}L1JlG
zRF3hRq#GIEb+{|U&5`@Uk`LGf=f8eOOh%`9{n+gp++^C7zA6~byl|>=D-9w>wEx!q
zIfo$R;$<v@3Vn`E2{TfR+WN&D%-=4x^>}cTN8f1GE&Wt#RL;v7o+p?aL-sK^jA2%v
zFv*|Q4K-@75#RT4|L$FrUOt(*pnr+7E4O+*DE5kA;s0v^?tHDZZ`74yU%D>#HC;uE
zP<lSZ$Ei~EOSIkDn+LXy;SDCgaTr2eKr&$eJ1S=RV(_jqSNA9I4g3xcE{`|cnf+;L
zX`4lm{o>}ct9kR-**k@d(UY}cx)`_RY^fAZg6fJ2-RqOp0+q_3TI+2FK;eKBqHw4b
zElLXb6`jkyNjzS+L=CqaU(59_dFRBlILe_&a;e`531Ui0O11Di^O3uNd4}z##wwB|
zaWGdn{*y30X!MT%t<lnNW!YJ9NM_g^-|=X{(DH0P(X%Mdjg>udG8u7(;Fl8+ukA?c
zN#M3`&?!!SJ5K*&*Qe!X>i<1E@!|aSzWyLwCiZ=bdndXX0``TSZ&=u8?wAf8dKb1#
zDjh}1&bGN4o<YD>cwBV#W94~iyCpCsRrq|}S$TPUE7m)G##G{y`#91I|G*lnbp4=v
zhSNE+HY#d1JGQUPd&4?J!1Y31;Y5V-Brg@+qwqzSgMdBewFVQMeSr`ox3?Q1kLir(
zQa506l-SLxq81L1UPL~$lzAm*Ar-E)XO(axHPsK)N(ftC<qC(h_7_)OnQR7YRAZ>!
zLbwWQkTV+kepHJ)X&%Xc8XYw|Adar`W4pMsqo_5`{-uqSo4V-l8y^LAwDyKjIMU&Y
zGGF7#R~m6Cj}iXK>q0H2nMq9ah-VGf{8rR5HoKJBW$q12W@>`9Z|E^?$51=`vgD?e
z=2(p^)Kc0bNn(KSW$(#X&_{mRl1P<0I&P>&7*+MNTBF3KB&)rA6)&#(m0J@6A#<Ud
z#ul@PMPr;0KFmq(7uz!6Du@<_Ic#=X03Eb5k_Pkm_{hf2o+d{b(W?vQ1e~3p2L=VL
zt%vh&0lJ5B(d=ZU6+LV|Po4^cR;@=ToZBxn6iKBprl;rIGgAbZCulwg2R0rad(7}>
z${HLh1dpS|`tvDO^Jn@GKoq&#CvSy(QaZYDkb;yMPt&3yyow}SRIZ_Sy|qy&R!3^N
zK3%_oEW6*Vc%z&^Al(ZKc#Y~(0EX)iCvSrP-v%(bqbqCAS0XEbk(D(TBo^;+aQZ3?
z$9HB*iD_wRdvqB@MMVR`!nQVJHJa8N5FqM>!#%%1Z+YFq-#`a%%GW3vjKWprTD`1F
zO6()%XF~{)_0{Fp#^O&D^`*9_t1CmyCK^x1NWzp86^R|<EREH#br~pjr6=cOP0>#*
z8DCeClkVYgHmS&#dn&X~(J4gO(M#8bx~Pl?fe%8o_JkyOym@r7|KLb5i(W)fy82oF
ziR(oX#)*VxXYb-XbCq8)Osjh9XlH%EtZkOjJU1Uk_$_Rb5Mol|Uu$(~D~MfqjR@fv
zs5?D#GGf}ywaIMcNEJfrCRpd=BRX|j?i2rt&>!5wb&VXUZIh97Bqi(n%$jDf8ZYG%
z7aEhw*<EJwTncNg*n%FK?`1EsjAybS7!e7Hueg6?zH|`;zW^x@&ip!kdQa(Yic9n?
zxKBfoEaUi&_*X*z>a;_yyny_!=Z)g(Hw?)oGaIMX2n5CBGf{Uut{*FWWYprns*3Yq
z;!B7KYJI(Uj*(*BG!Q_@X@fb~8O;!nXYd8vJ)X)BQ&3RwnUPU|EK*BP&#$qO8|a74
z#I<+#=fkL2N^WBLAfh~9D8z0>QdLz2whSBtz{mqvgZIm6kXETAb>Ps|#YL&cLTeXn
z`V%+EDV0=IK1TM6c3WFnr2{EaJc%Q^qobp~3n1SAGFG?QvY#(uVYRiiq@7%lk&$f<
z=c2)T8|$)%bS+;bvRQfo)_HYnOTp=phK9xvFgU_tc+f3FLq%g9b>A67L{GZ9x~xvu
zIysxnQZnf{IBtu&&D{irh06iH;?$eIGoP&|dX2)<DA}D(xD1^-zLQc~JYAtOuR2%z
z?&{163Lp|85z8Cb7@s=xVHDias2nXNR|CaC^*RJ>mywqRn~6%b;R95;%&5A|e>Ftc
zF8(~oytrK?`&q4<^G%o>ONG!jBB-(ak=mZQP_+H%nDyIo!@9-;eEJ^Ms;S{wK+<x7
zK=Gu!`Q6>jY^Z;)$iPc->)i!;R>i6ISFknS>ce$`=<YomOB0c;8oYF_=Qn~%R!9@k
ziTf{T1{z+96{VFhZRggaQko`}Sa+!yu9adoq$xAcQ!Djm25HoWc%N5&kBe7qJTz{Z
z)S>S<2>Ugc(6Cr@JDt<ad0~_JnK4|T*|gYV5|##+{-j)=KifZCTT?!L0y*ttEzb4Y
z#{$W=wyxjVZt>sEE{`N)uBNz{A<P4X)S5T8*BlB)2{P|O)m0#oSR?H^<J~Kx3-xQJ
z|4{4)CXfhk4JIhAa-fn+ZrmL=(?KjbIXQtp@vOGLaW_tJK4J%nuV-XrWNYb*hKH+p
zYK^9bt|KE>mLapnI(qvfwSvMzDz)Lo#Uvn7$k!!<-833~i-n~$r=CXr^O7OA$GEsr
zPn-?-JOA`yd@|Zww4tRB*HyS$BUndnNd_}cpHH({o^ePW1lMrwldp3cNpV0qE{b<3
zUb2zq3{lwDiLUEKaIvK}08)<IBIB^i667(*(eF*$_q@G5-K~V3+IVvlmF)v~j*+|D
zwvPH1;uhb@{4Pr(uE^7_+;X?I0(FpM>F;Z;n6lbbX9-DwdB<QRUllmnUb=@~c^w*(
z2zF&qQg^l0-S6$TJeoRM#e^Wqx(SqT;FyScw-U%OG<#-7HB9WSihjWB@e1wzmEAfu
zn7mr9k?Hom^C7c*ZTv6G+auB3>EaI!)M?F}d4m*RP1wY{x2cMR3Jtp-XJ}|b&VH*@
zV`&$#9t^HrmvQ?@U<HV)Ke842*$I(FmhL_X8f<sBFXsYawDOWbbQj0)LMCPTA^t?4
z0nMni_;^6pTBE&CWYW!j67giKmN%=!>fYqcE~`Dl#GxR0;%;_Y<(BUo3JJOxYM!aM
z-{1FWqSgd=R&V%W=5g(4|LrwTWJ<~~02*v|2UvjBTWj-0?C9)_NJt0<*8{k@xxIvg
zV+6?vEHYj|hR&w%M~3(i)@7H-l9JEhfOG?W;hTg6DG7;Bc8c4dSPeI9&iWu#ZO@6e
z5A5vmNu2g&+MNi<1YF<L)bK<z-?Or^0>$YyA|jAS34DBfqBw^@Lfn}V&eI1-_?Z$d
z7&sIH5-KW0a&q!7U%u=tH}e9;s!+ihR<@HOw%>6xTG8RCc6+AAl9)=dNOw4ydjqig
zInu-&b_YE`g#~YDHfU4{-usHO<lzCeHJlRO-0W`Z|M7Jjx1Gr{8Pq*zs1O<Beb$R7
z$&5hWN>81;M>P1QJGp}<YZ;V@up*hx{Is@3%~NJ=yX2zFm*xXSxaG%w^*B5MNiXKu
z(-#w3Gan+Dj-4T3aEvK>$ZYtVokLjZ8z8Aad?#_R#leqPl+DyRHteXiF7=o~wqQ!g
z2(}-!RqLcddvWKxwxyVIs8K{_x6FRMnRh`ND?(_oVQ98kMi)71WT5XPf#a3ry!liR
z#e><V_9mg{Fv~)5Ch^O~(0JvamB~D2wK;o9j+i5PlUW%I6w->4j1v>u>7kC*#jfJE
zNpF9fw#ImJkg7;5q<zY*K1!tesZJcf5HclYo4u{-U22(oPuM!o{MxKT43|f9dH@L+
z2h>B^e#t@IMPh~QH}r)$RrHR~`j=-t6Iv69IeI|b1+cXye0F9A&~0$fu*3%*Rh+}_
z+qc)&TLW4{iR`f9Ad}&LdVJUgD=%F-GdoLfcc9c2hOfG53Y5o@{rsu@-WfLF-LdJl
zUlI#pDa4UZP-nbv08qDSKM)NMFGP}j(ad3M0DI%CEME(>n@mkj0eL`fR~8^v0`b@k
z$rmf{M#gh2_uww-=$9xmblw{YTj@4byj{~=;Jb|eE+rzQhOqZB+6K{v=2I`-N2(7J
zd3`plX6Y;{oGnXleiN_vi=X^CSXkVd{NwVWwZ`%Yl|mX`#<#LLbQOJy)%P3x(zOm|
z8bTytJ+om;&+^iS(^oucRVk|*K9bsAWl%(+y}lKY5sUb!MFW387!+(iNo5%+nt}+c
zV6{B4fDxrvovd0(a;L&Udeiw<m%K9kvrqtc;vgwIL(|wxIuBt@0U<i41RvShz&X3$
zHQty#5<}^r`4QJ%II&^TU@U{w^I?8%Vq|co#AFfA!Q|H3#x;{S;t7x3Zdti(ZJeTK
zpna?r!x@Df$KFhmZ<)oxmx;xJ_c^TuVcxrc2LtJeG`3$1wE@KIt~YeA+XRTHxZ$C2
z$=CBDs0;9=KbeYlqaN`(=CY4@CBVNgE5i6C^TkaiOga!#C+^<CjPShqPP8|M5-};H
zfpkx(;KuOh@iu=XhO^EKctun-VwmA1{Svz5)+Qe*$F^%MkNbqkQ|!A2#z@a{c2bef
z%3FSs-3>1@A-8WY{C*I}zd2@c%Y@eoRq)RH>Kk2&US`A!Qxs=SY+n~f$Txk8uIPdI
zzu)d2-yqmNnXf3p{TD>>&+)l>PgDDubgPfH=ee3e{Yp;G_3s3BE`%f`>GD0-DOd?2
z8^gpI2OiQdAmJh@)>#s_@~RSek@Nx!F8J6^>Z0M@XsuYh9gTmdcO&8_@6UU8j7gps
z$%|-$Ba+-<f4bOw@53v82M_i-hnCLQ2n?tt6<`#OjC@q6kbFSBuq>(&`GT088S5P4
z-&ibOrMkVYxM&D<V=hT^K##lR+wyZJk?r+NO5XT4s*28ga5p2=4d`I=V%U3?l}*IL
zh2;-Vj;+=cqrz?o?T(Pq^tSk8a1ujv)+A)*Yf|7s^Or**)P`gZLz#;%Lz73f<+ja-
zj@!l^Z@M_Rb@Qa>l;!iCsy|buoV()_>rG?Jq!&s5iuaFSc%#Y+*YE{r9DEzJ>s;!0
zg+oW9ZL_}V(NJSKU-hoa4>T+h=kALJqZZ7e>4`6>O{gt%jzi41p<dipq)Fdp36X0f
zj#;xqrs-*qI0W`Ozk(K?hy<#?#uIbl9!npXoJn5RIwr;C2>@#c|8icO4oSTAQQ0J%
zi5d5=jzFtK#7wDo4pxv~^G$-fW=3I_H)!Kl;O!AsVT(p|T|^@MREi}dW?_2Z*;aN&
z@>@N7dxo$cRLtrjl%JV%YtlUg4oD1rWMvQD!Oxzh*TL69g&n(!PmWfYqA)GAxU!5T
zI~AqybP`(7cH_YGxTSIFbE$DdBm~djx{J5-vqM)a+lq2w*gH$zgO!01`?lswkF>Y<
zyWZ3Bv0CGqwD`buMsaT<!jDsgzt;@)zkXz7Ae7_StInOX=Fgw9IGHYdPPpXxOD_bl
zS2V*EGRU``VA-b`*`B89@o?D2_YXf*xD*Iv_KK35vqsq|M19>9h?uJ3X<$uWCZNH=
z;SoaW0qjusv@u)#?oOfTPv41;0`jRJ!Kzoxd<~;r5^S%M*tvTQP;I8m=a((mT26Ud
zJMlWq2{t1cFjpfE6_UEDv?(jjv^E=m2TcC;Q#1MN19H%2`|oTnpnUaG1j{{HG*2{j
z|M;rZA7eayc)mwh4>lClSu=u39m1dWMykgDzOJ%G{DZJPo7kVUrJ)vH>G^RegSmY9
z%>0-Q%u=^8^Ns@!Vp0yLx-aI6udCoTL$|d9k%u<gu9y%YdMCr={ehfloh#VGiYkNi
z6^c~IbTi1DW2@Lw?P27?Fv4XKs-FxNE{~QVtPpFbL+F_pJqIsElmp-|+o*W$QC=|m
zV$+81MYyqJ1AOrv=J<=KqDZeme2Cq!PdrPHk<nZ}x+>2P=z7K#cJq?Y2ljh~)6$#g
z6hO`?b!x*~3-1<fbiS(>8*az2lSzM{`W=znT2qXYU6}DX`7go?)N85G|D_uBq{1rO
zbBcnvcE6=lLN0(a|2DAGf`<?nlatHa6$v*3mL##EBX4@^`>$jw*Ot#SFm-I6;guJh
zBH<VJ9Eq>?BLjKeB9{H7^jGwU$BYcNDUZqNNE4#h-9Mfl$bXM(O?4VOu7{deo&Kaj
z{3D}TwwEKm>(BJ|*yF7Fg`@dPJu8?L;pct%?hi6p*hYhWa;{YI<Q05wCniw0r2_Va
z0S^!F<orAXl+f7hj6e%_*E!xL61~ovB*}q!^fV0&<r>;-7dzc&s+-gKt9Bpj6EIj)
z<2;(yy@C@UQ-c{*rBH!07cZOt==*be6&)t;8{=dNEjFWX;z$HK`CNq&5|#R6V1BoD
zswXMPQSMFqUJtyP;}q3s$7#{0|7yzH8Dv?GUAaOtfVQcK=EsBI717B#S}f@!yx2PG
zfiRjmovI|ZAHiKxLR*heBQhMr=2Z8~l|evsq=G(GqNN%8tt?+76rIZ)H&>ygQ^O^j
z@zS-R)6k_Qj4I3$4vi>OD8aC%26aLHR9i+C=*AlREBqM~^0lQ44NH%2jHBswRaRO>
z<=@F)lITTwl%Mo^%a?!SSpRT(XqbN4(8mThWwo<M?V2>?OGj&Ebm=DNOUiY+D2A?Z
zRz-+UOX~1cLq<wUgY=E%F&WB<`%nl{NA7Bx{}E(OF<cOuKiZWx(`^ofyIhNseoX8O
zsZO6;@TW*Kk_d;p>@lF^D0)>BWqZ=~^}gAoGFAK9CwK^BH;b8|k1Pt_a{NKk=(Msq
zrIqugB)D5s9?ekv=g`uS86CasH~XS<CXF(>9nX5nZ%(@j1*J6j9a?&I8IoNesCaM)
zhte{OSJzO3biB**(fVbgA_tNkrC?4jYNSfOt4zw}v~iEDhjz17@4nu%6_G(VCC|<?
zFPLg2r~m#p^RRrC%~5e--0?0pKXb!O$2jhtQf1I;7R!mOSR7t&bi4GywG)l&60Z8u
z9&OX50b>I$NL74?Y1?84&E@;$J0Hboeuel}{ysv5jj8{$pvDgfZY9W-3&_%fMl1JR
zWoy;nEO$e{<+%yuh=i)gd~sLZIqFOU+W9|oWAK=g$)_nyUNO!5@e?Q0*@4has-;f8
z<#GH}`jjltOa!em^Ck7iBrENNXfbh8a?v`h=B9z_>Ylsgk58obs<Z827`9lG-%R^K
z%*hn{(35S*EkBQqTd8?%zH{{{@k+!LICQTVwO75fCKLE9Wf2bgTR@%@rt^O^T?JH^
z&C{hjlx{?n?oJU<ke2Qe>F#a;X{9@)ySqE3ySuxaZ}I(q=it%9Pdz;Q%<j(Ix%Uof
zUbjDc4poj<^tz!$-=bmD3Vum{xEMnu#uY04swcDAlMeHd6hE!SblP;q(V;e28hFv0
zT<SRGRe~zVAF6^(r+hin2RVWgq=2a{n=Fj=@BVppL>9ShIhiGaTW_~3G+%9gd{YN%
zks|e)Z2SEw6}2VR+1CUF;vfm#+Nz;R1T-|?@NjflS=nqf*;hY!vh^FZ_--FubAj6Z
z%$&<O7G2r6VId0dDSp(qy>Q>|m@`7lw|l}YO(?XYV7qr+J2cut;_2b-fpp27zcazj
z8-lmzxuhGyHyi2L#VG`~$r%z364<7xYz11p2=|2ZGrip9MgexNp0Ia~ydf@-TCaX1
zlYW+{5mfFPB19C`ISIj~Nxq>o(s30^*?FtTEED#l;Irbqj3lIF;14&{WUU=vPnEBe
z)vjKAHQ&qG(kIFe#?TS-H;xtJobkO<EYXWEg#BE*Kq~LPObAIp+(?SrQ&CW@TQBay
z(22V`T?Ly=^Q9gNOmKQ@Gk|EN+#VoRpsc!YK-Lt7#d7s@lf#y$3k&DLK_DuFd&7yD
ztsSM7qacD_DB+V@tdsrgd?S2)pvb3X`s(?C6XdsYRc4A24K@2KFpM0Z3Z}|0&!cPZ
zu42esw(-ARCqv6f<k{5F^erY5sI)(4iY+?0H>9j3wVS|D|Cd=UNKnX}94$2^f?98F
zkLD8MU@Qyy_wV0WxVW}w8)7fKC|zB$(a}*rCxnI&6%|Dz=J_NhhPXYP$|!|aFlpj)
zbr2}7?-L&hE0{Yj5_TW86yEYtR+`1rb26EncIgjQ&Ov8<K2zd&;)m5?8KFM0uxzRA
zr^c&Qx9@9LqwqpH&(#<1o-nzWaz5u3!DIC6`m*~gsR~ZqP|}+#Q<;$D9v-N+1wjSM
z&yTl^)YL3v^5W8pu>lw&PJgC%ZykrUv5QSCvNrOxwR)<`0-}suxuL4F#mizhO<ubZ
zKx=os9pBr19vz3dFxGUIWWf*jU<oo<Ez0!Bz#I}Uf<EgPqcSX5oh|^OoLOLW>m=$T
z<NW9Bk<Sajb9mg(sCkRsZFElWg~Uru`^AzpX}r)gR__PWCNa-G8P<mpT3PxQf4U##
z-vxjk)yNmIrTEyazNdb<Lf<N}F>7tQ)`BWh8MkMfEO+JW$4W6Zr=<ovxptG&3UBYr
zuef<!5BFC{%2f{4>$}>^txFwZdtJN?P*?h1)gh*t@PE!%7m=B+X>9!Pg$)`%c2qx2
zhL^#Cg2u|Ok0DRsj4O5Ty`v?;8sw5vCuvJ>bO!Tgv64%u7DL9Pup-QyASq{%Ei(6A
z4+At5_WyP!Z2VFuadwp&Ju$Dt1^|X%$lspNkGOS>2b@tn$zdlaR#ykJ!d6yHl+sBb
z)4!%E6&TN!zX$IhqENjij7-=+jz!XU_tIsq$`pu)mY<p3oNjA2r|b!(Cmdkx1GGwV
za&rMbfQ*I2e5_+WwtxB1WP>B&;SviBQ-1Y<VRVRw$2xT8i&WHuZ+Um2s7GB#ciDP3
zk>CV){eG_tbPW#vjGwdw?!sC8ozdd@buA5&hfy65uH~usQiDeb06YGY;D(U)yNZqM
zS4CaaFL9a}7w4tiYO_-#-j_|2W!`1Z!}i|9pfd5NT8{suIqLp4;+RJ&SJ+s8M4zaE
zICdH~RUhrQao6C?AW;Zf?#gO-bO~yINp_81+M>oP`-<(dUA+QL*@EfaSzc4j`|YPt
z>5OQLeYA3DQrbK@Ze?g|KakH{xBw{T<@Qb1-Z>w+o*8URX!m)oE&t1Uo!gK+U<&Q4
z4E6hU7DcNM=*8I3Iux84*Au#SMB@!O8O17{{=#OrcHCWcoFCp9e<OG2PLMdnD_(XY
z&a;0L`T9=5@y`83!4EUlo2_iFCYi!J5-KKuASM%rdYA7GOB!|2$-QSg`U5;D3RDG*
zKg6PI{qoQ$_@yO!86{_L;*T`6DMo4o0l?;$rh22=JwDz_GMgG|J_feZlu|DW-wuG`
zDm>0CN2XCJ@Lx}FVcjr?Ib^twK4;_{yKvpRzzJN%(I~DlATU1=Xwv({4gqK}G4}bP
z_r!^m>2ufT9s98sSy1Lz5rux84+=t(jZ4%kW=flGwpzZKdG10YWu&9aEd5$P-!?Ra
zmQ`-Czz%f+sL9qBJF-*7n(FiEAUhM5O!=$U;6M*_BJe8!cgkZD{AOU#4fqy_2ncW9
zzV*STRbn!kpc?%9H=flHZFO}O;sv4{dglbr8Eyb#iKe>vGK8ZT9=0=bm%3x;#iB*H
z?5+^+&(86JR81!Z@1EqcX<y`XjE;+xE8O|r5F=N;5vAc2={s@;*y07};8G#uFm0z-
znp(Tsn=sOAv$VOv;+Y;%{uGXa;v?1NIUAqQ#I}o->OHfPv1Nt`MB4*Z&<5w*H=%zU
zp&2YzrI_lW15T^ty<rD!(I<xBK}nHtPEzuXz@dZGdrsWyl*}DThNQkGJ$^!ES8(5F
z?bAM%6;M)ez*=49^<;J|+uU3q0vkV4KyAqU(EFVJHDP_a2sYf49Y^B6K6wU_WFat{
zU)|d3NomM}M`)-ErB=nNhX-d$JmCU0d!H6M)8Vb6)#aB;1=!`bzLQR}FM+DsrsZp6
zz0!kGb4&`Y-IFP@$dnlO$AH0G9ZD<uXY|qan*Ge^Na}?g!TLu%i?n&cG^=kHZE}ZM
zD<(|(M32ZC{_TJp*U^>-4}o(iftq(f#0D}Vmcgdpp@Oxx!Ao%oI1)>&w_sM7L~INL
zJPeFNeRIom(Z(R)Vd1lzU_pSk1molF)8oAqUOMP(Cuao1GfLj2Y*P+xh?~LW1uFOc
zmF}5ZLY#J4VnZ|X)zGSW(#19L2f0)6I{b+J0|ZM+W`y<o_=idjy27ZLDcdx`uNfwk
zg0_(s?|d62naEOWu#<X!yBI6kM}GB8xzmT*`-6cotodHxJ-od&B1iF6?TyRI!McT;
zhF3-`Ewq4Z3?ymz+gmq2?!I^|LtKzL1p_ST5~`6jzi%QxWBojQ<n&gwSLW~P!4`R@
zwwQ=~X7}hkc8jz(o7r?`wQwPh1iUFg_Jm4m;3IJoKPd?j02c!)bNaG^?N7dMQpIT<
zo@slcO2`={e$u|<q%HuM5d;hV)1vRmG07(}t=|_s;sc<L)!E<jO7w&V1uG$`GAi|>
z`>Odyooy9_M1_s#^TM5`q3YJUB86}ed)<$P4Z7be|DD*mzLF_qKoT`v)K*~CaDRy1
zTw`gfrzc;Z6cO<)Vg{glExm(-C*}k`mgB$X&_3rvClS`IFZAAQ815GBRa4wxI_aSZ
zQ;@RM$gPbR=bbVuv`bou_^ybsAy_D^-@A;=)box!)me#)XD(GC*}~#%O_t;rvwmk;
zNMLI%5`)eNpc!g9O{!rRq)jA@B*NLoQ7B(q1V#3ZLw;jg8Ur=IBL@YSV>Ob(?)#m}
z`UV#Y-0jKM<9ZF|(S&~hmyqW7GnkUP+oEP+Fm;lPOM!3SQLwFu^Ss8=mDknPbsUBg
zBqDqlg~^r*shhc>q;%>89dlhnH)SyC{@C9)tc{nsS=Ps?dV4<`oB9D7T__JnN0XSZ
z$)!$p-E9)*>V9u`mdo+=g&Hwp4ukFC#rre4@=@bTIatd`IJP5<pMLg>yY5cBgxm8&
z$px=T@yPj|QvRO{u;XpU9)mWeFa1+l%{bA#-E@P_V;b>4#OX80i>~%2g8{w;4Vt-d
zDXvF#NYl%JmPb6lq@ZAD+I;~4^h7T4mo1^Yxp7EA->KB1cES*gWD&A!@}L6FI1KXQ
z()$Kf4W3ujmw)MfUof;#&#09i-nX>msyEtUoZ6zXA%!~=%LEzXhxibRx`?7UM=QPU
z#9+(C)`cwy368X0T{x{X6XkOqHLW?FUw35*omAvGZ3ex=@)ik?RGw||Od?3wA=tgJ
zs(Tf`P$*r~{2vJ96c^9e#%r-qI|g$s!PxW(q~XS+e?G_WcXxI&84Sja9c#N^PL6E?
z7(R+2z|Sw9NAl%C$Tmk82tlomG_)<DXdDJN<dmH5lk0Y{Kx}QPRuz~Bv#K>_Y;Ts@
z97G=MATD6V5lgXPVuBoo^g_=0#uXv8m8HPZT;a7oaK+?yRS+%RIGroZy0L{+*;AhW
z30@qUlCDt$g8Kc0<>)KI>3&+L64B1Gy-Dtktna?@C_)Eh6S4_~QVhP5*hvGzu5^j4
z`)D>6=dy4QmhXxJ99ij_309W*f38bhw|YzNtQ*{XgwG^t0Ko2mTdj5M$L|I#7oz1K
zGvNKa`D@6}$wZ|(uB6IJnafj4{sp7YS8@Oa9L`oyKHgpG4aCrciW2lQ@BlUd=%;^l
zJNyAUfPsMlDD3#iICNBt419b^OHHn^9?yW18N+~%&(b-e`vUu*Qv+ZgC%k%nXHB#8
z?d$GaJdTl|Pb0#?qVs`${Xcj_3#LkK7R*2f9(k7&cIp3gq)&MZ7!jWs2AldXS*;&<
zHds^Cd0=suT{gokb&E1pooL_IBzXx$>c)AyyT@w6r**YnMHeFAUMJ~Kn%a)!vyIFm
zY5_}AsIngZH`>qz7^B7F-^TBQ`|4w)OKSGJ@B6xwq01b~vrepVG#Gw#Pfe0$enH-k
zfHY!G(~K`r;hkNZuJXGex}n5|i}vKFwOc2ERa_<b7wy!>HIIt(WD1Fm!x>e?J}19e
zoPCr*{|X3J;7|!V!-#nyRLZ|2O2p7nJ0^nuV43ka#o(ZYqGzSWVuE6U668p=xw?D+
zHG+LRb-*q7AFK$Y)%-+?AV20QEHNWyp6KlISU}_`9g991ve-HLx0+l<^Tom^%9oZZ
zjcrChv14Yw!P6bcq0Q_}iN$uCLLpF)bH+DAU#{^wW{7NIO~x*H=tMSM>Kbc)*wC9v
z47b*u7rT%sk%(?vQkq0Q`$+MS7+91DeVDSVx|(l{i>Z3Q#VBWUo#lV>9XfZ%7F!*b
zZ~4_6S34#1!Sk(UlKTRp^B;-JZ-tOZ61MeCcU0;E+K9`V{+7^fe+aNgTvlQWer05L
zwK~*tp3^x_l30FV;77-y_{b}Il53#7mb`1l2SyAk3#Al3+P)lL;4A8H>-{c`fw6HP
zkgv=%yYv2I3E^U3v;*N#2Z*7v^7`{sAE1w5F`HomZ8rzewSVwNi2nZ8$D0kx%=s9B
zFz;HEi`+(qH|80W=b(vGLy?H_gJ*5*={!lL8JB2K*X83TDCOt|b%IDMZ9yUvuen@Y
zVr9?y3}@$DCxB(L+5w!{@?3tst~;504*@6X3TInzMoBone_`fk2$|oG;nXdkk5g<i
z>h_c0Z3brW`vbo@wR7IABB(2>h-aYzaP)I1pyZoiR>7RV%S)S?7u(d3m*lpEl2)Th
z(29IcU>^C5?L-$Vvml>j<J1h4yzw9x6TtUdSlF;9`?nMbK7kV|TM87BqYaLhVWa}m
z78VSkx<8n&2?s0#>(zEx8yg!g`#mhM35AtHRTE*s$<u~47k(s`_<He;1gd390Da!r
zccIPEz13p{r+vPST;qF5bfHun{xv@ykrKo05c{#SA0l#=b9+pYZO^yfRY{Sw*T1n-
zl~(uomtB$!nuOn!;ZY>{d9r4``|do!qY{m9tl{;Io}UtpB28u=>G~yrFov}tJqf{Z
z1eYG@kvT|zr!;D%3d^aj{^=I@4ICu@=LZ@fFCF7f#G<Y;&Hdt!l&$^450NTr3nlfD
z=IW9*^`daNp~HBgj$4C1qIh(*FsMW}5bdcJ?pV}UO*GXp0OWHAw`~7aUO(zrQ{1pD
z4Gl#r4E`ET70x!zB)`Osf;7eYsEml-_EGPWYNF1TzDhF8YO&m8YJp!AjF@uUuEtC}
z+;kA+4!1SaVmEs3;tgy%N}ufXm#`3cZPT5jZ!>Suwg8H{Sib+sj8AfA-?v{=k%Cii
zrqJ5#*36I1dM1Zi?rn8W@B}d870Qp2q4&n*9;1KGdmq>Lcjx-_%L{U2%mrYaSe%s0
ze3#85PyR#1L^-beZQVQf8!VgPxe}2oTgMH*s^c=yoCa4~>kY!r28?D>+ptIiMEOT(
z#4b5G2S=W$ol;vzoz7uQqgM;L;`hw*z0P;joHGxr7(<1;2&A8dS=Dz5s!Ah^f2*NW
zzn7al8nWyCO2!$>x^upbKr)smGVJ@`A+6Im=jllMvaPB<8BmZPg!Y^q67>$@n||SP
z7TZ%)4f8jDL`*$<w`Bfsq#Nb))gIST^b4{sgFmIPrA5mD4|BtiLfi^0wKtc}94p)5
zY!G01yaxoMPMoL)k4O^wU3OzCkpI!33TtM*J<`Ea8AvbEt3uC+YJyVZP2-8^iq06l
zu`eRT&T}{l3(<T+>xn(oT|W4WMLg)^tEyj2%NaWk2xxZ10uZK@^CpL)7U?Uej6EZT
zjtC9SnldhS5fdbTh#=k7UOE%CcatSk9{-DhIWmk!@L27_CA#XK-NjA;8@Z+c^cOz6
zb+VDqd>8D_Px;c7s#IpWOmOx#qJ1}^3l6g^zddQ98|~1-HgV~@taqR%d(n7}j@(|N
zjtoK=06oTX86}aaDZUzHJyU-nhktdmAzS(dmxxAp^ooBGE|~iG3z{m#?XarUms}OF
zWQti4wSGmE5O-f>|NVK?NnGFB#c$Xr{UzrOE4qZM=>dO|5s2w0el;cDGFSNc5HC@+
z@j_+V;^bz@!D7}-P04ox{J3>~P`se4HFq;g#|M++Z;{RW5$s~{mEIi)^Y6BW=>6`*
zcHNy(H?dxk`6+=(Jg?9IU0yKv^X+ckm`_EWHFRGl^Tc$pEOv~Ju6$MC9Jsyt)Nj{j
z`=@Ikmvze5>&!zQDv8#6@*%gBaM%tfgR+~{CU-o(+psnYKc1ppmW)3BclIJ&ONYL9
z0-y$vMChJx{Y~O?jsU{Si@P8LyI&fOPG#9b7~n_y!Sh+Zkb;q((Ec|P=>HWGng5QD
zZx&LKFMNCX?d|L|tJRrJS5Y2i5YLG3gpLb;m-FY1oXt0^tQ|v(>HUbk@X12X>-G6|
z-RKT_A#(@H1*J?a<Tx1%c*5yyg5@gz<z27#ei1nLkYHj8hm-Js5f{%7Tzxs-R+|;0
zl7~<AT?3b_6FMuDkDLbQ`f#2}=n?Vf;sJXwWG+0vG|n@detj*a0?Po%Q^vcyQoKZk
zcJ5clucgX4+3$ELm5yH8iZ;cD3#W0xCGdaS)nD_lpimwMTwmGnUK6^TL`r1#4WVR>
z31$<RQmSukRb4-n7Hq$vfA`W24Pq17<$hU*m!*LJ(YIXoqNvQ4B1A&6kScybWQ~k+
zYyT>O&z+2R?Ogz)KGeN}Q$9kQ?h%YD-c$D(Lt&Z3>c)Ll_VnuYKKf{_#G7yrp8mA&
zn<=DE-=7y+K-xWqUbwkp4qI*0JK0+G4fH9w%f6B3Op|0zXOh{uq=zGhPgMrlBF&!7
zBE97xiMeZoa#HmvfZg$`#9&#A;N}vD??84OIf^IwA>#q%x7nebw@bV?oHu#D>cnAZ
z&*7FfUY9mGeP{j<mQp-Z&;5%u5qj3dift`{WB{&}S3xp@>c5cSQvr!yGQX>Xu}ux2
z*#b2$1bTSBOg7oe$LIur6^0t=x!ztKsYBrld2|v7f1@VRByHJ0-Re!J;u1Ol|2AG=
z!`A;ViA)CC-~4Lw<goX8EtE$qu)FtP;qjIr)aI7C1*6sAb8K|G>?^m=W}Y@pzy=vC
zRb7{K7|f1%9XogyYxvvwJ@x@M(;vtS8(7?6US&xd@1ne&srR<uN2M{OHP--jtkm8%
zH@VeRERv+HckuhpwSIdrYTHz6jf-a%1ZYni+_x!-7*J%WIw4Rh9r$_5ts2X8z}5dk
zk43gN+3`=*b{~iPACikuH8?l{U93@8Zs+_rf&FdW%Mr>qalyk+n7#1jXiC04Ddt2;
z&*&j_gW*eGZL`IRD+0<@8i4;d8K1t0D+onjRh?C(W^ELYp#Zz7@kC+M0N%(3h%JMW
zrPs~okdXwwLfwA!lrtAgZ^wx$1F<r<%QxzdGhdnlDtJ&G1PjRvu(>5an3qmmX%R}_
zGp`{HXx%{C|3<VpYzd#KodFaX+}#eJ_tPkZ8<Gq}{q0K~foa~vl7wlHJRg<`L7!ZS
zLdZY=4%+CF>6ivWQ)>^NW&FAh=ijMv)IT_&*EV&C@@&ULwoWmk(@B*O=KPx^a%nZg
zPlC&txw;y(A2LAELx_nfU%ybUFCmZ4`0}=|qdQtG*W0HE`xaFN<-XOhAmKO8A&9u9
zGbojT6n)||Hf6Om-MsKc*tiaZZ%Kl1oW$n+Z%FsW)kp&!iuH+v$4}<`s?i<!&My4>
zOHa`YnUcFIA1surOca!nZzx{*9DHg1)w(<$GwD`nF9|T9NQz|jRFsDCf*NaaL^ld4
zr+L${Up%8*?OoHp)iMEs={VV4AH~=yG{&J@x;t+977|@BSPD%;yZ*gMp+z}`KWmW!
zF{4fwDq0D*(EC@UD;f*fW41Q6Nl}AOw%fPgZpG(hoPs@$J-&2q*3O}+zCtkoMRshg
z@@}~XFqb$1Pwiv1Of}eV2(B$Tb^q{0#TL&UD&VO5<?0dzbi7rcTzH7*TeQgOTjmg;
z4)beLtHb0j3rL>smPse)tNj60@y*fKs0m)D@>Dytg8p`tr)eyrLGFs9H+{cP*>R1C
zzaOu5jvQ-6&?vSv!t6(6elpZ&*yu>y1~u1?ye^EuQdzfNQNCz4T1Q1R<EQP8%=2Gb
ze&SluXa-vT_F^hLDy(ip1#x{+PgDQ0;2%6xWc0~LmNKi6?+2(DWSMz#x%X4?I;#*F
zBLr^QSt!G{Z)SeG;Y`Jgpj{nsIR}N6$b3J&LRzvF@B&>XAJAU{!ucxVAf0!SEd%bL
zM<@{sG#QfojW0RH0;e?HGObkZ)NqkWbi?y|XsuI7R(yERMHkbt%4e>6JM?@X45RO;
zkBg|tSBU6V%5uMcD^un>p%2c#rEgrJuS&{gfVXuL*_(D*nWm=*Id~eWG-UE^Aqdvc
zg|jVbPuyPl!eR6jxjP9Ri@Z87aWduHtMHmH6Z7$dOTKg!%U7zu&U>-br%1*q%&;d9
zKpc-MH(=Bzo90(!eA13TlMobIz)!y25?B%LU4r`Wgg&CBR-Ls=<%*Xti{05!O=V>a
zS(}Dz+C82|P&TC<{LWJN7aOLoj;T(V>NR87T1DxQQKeeS^A@ki?SAUqhaavY8KVHa
zH|0fXgOU8uH&;;7+z((@#8x1v#^*4L%X9p96khuQPDhqsPsonwu|6!>6;|>f+0!Sp
zc+BEtRoAa-P7_To(Zw~aoRJu#fU0RWO9FLQ&CF*&4v?AOR%a0trZtfjtI(JBzRt>l
zi?%pLYQCTjuyoz4Tp>%gx+?0~LdS_DuqXZy0%clvA?TICjb*ap!kj5KSqj*!7fD=k
z4#O9@@}Q<M(>QRt3hz63FdzoR%Fv^5(~}<dy#9t1ITM+P^azY8PK;TAso~ws8XfEt
zp7Q;>$X1)vM@^j1B8SDkNEm*R7r;T=3oegn7@*&<p>RZaKeHH}ZpwMSbx>D%V*F(~
z4d<H5M35<^kJhQFsc9CF6(JZ~9$0|DGJDDIB}^uq6TYzr8kTrfaNiJoZE|h63WETR
z6_u@yUF*IvAzsn-D_gz$sNQ_3JtcVz=}sRl#;PzTieaPo0XnXL_bnFMY*Bu5M5#p<
z>a4#m3REv|rCeY#MiK<SlMoI57_a-4{m0`MIw(T6LN&*YIr9yZC6AUezb1@dKfZ3P
zK9uJPbB6&OmM>I3>$6r=$ls7kE6e_Dt`j?hl?iz`l={8!f(rf!=v_PH<;HUM{^erv
z@S#*D3sBH6PcK*Rmvli8EXDETCO6aY7sI4mFga680C6$^Ww+0V1-5}AI|qOqgK^Bh
zpwR<p5Wf`9WQ0K(CPf_aZ=cu#W1caefGle<aR~G*HUn{ocy!aC$*~W6)S>e-?h+6R
zpY<26JkU3@mbixQL`cvVcpUR)Cwh8rPLG6H>hJ}9d0w&QLaRml1FbL;%h}gCwQt);
zKgWjh0g%c89__A?GYRl{Gv2VKK2(}j($H5Qq0Q8Bzrk#B_?IT7XH4$iT%a5KERvG!
zdPo1cqb|V}@=>wlh|r5&pRrnbBi!J?Y#$drW>QtCTHz0ijAI453P7IlLg>(L@dzR2
zjRCM+Y{Ma-#KwWffqJzWFegJ45D<8=(g32Ju!{2{^rmoPUXfS^P4RhFCl{A*1`hj=
z<G`89`C?R5qGcL1tuCjv2a|bYf#CYjq{%MOK`=I5U;rv>lb*U>*ykdp7gH$Exb6Vt
z{5p?E9w3|l5fE^`m1sEzv>d)eyB)0?h+3qisl0cNpp$xjw_l>{?Amnm<I_ua1p$~v
zl2H1_KA$4x^7+2v#<G@0X`v&rcFeOz9f<}h=1k&qhW!=~AAh$kO^|Wn!nXm+Q6Rgv
zzOX4e{3<C-(b3(7Mi^@%P4_R`X@}lcu|q*L-nF|j`4|A_U-cyt($W!Oa~IXK!eIHy
zHDzVer0R#CpY=+sH?N1s)q(0z3jaU|K5K7pZ!NhhvK)i?*+$=B68CvnYN~)|%K7&2
zc^9s>)k-Vm9Kdz%W)ACki<E+u3Yzp96Tyr#P{iv4=B{H%lleRcKs!VLwm941$T&6S
zxKw4-TPB&n76`Pki)9{eZhQbn(`oa8l}!DRoF|{NWWV%-z-aU())1gT{`LhKpW+Lh
zfw<mGJx|naX}l1c2{D$@kz3j)c=(Ey9IhKG0*aJ?=g%EUv{rP43eRF224a?HSmNg<
zemT>7KLtnzhAgK5J!zQE?anl&=>|G+Nxc&=`KkM<bEllD8Lq~J{1GYKdZW5at*baa
z=>8q8FzG{c4nhf3PH8PAvg>?fxnwvt&bv<6V^@ds+*T_P-rnAyL_~m?Vr172ygW>N
z{Ju=F$o@uF5AeppGDQNHtd}mhtn6#hF-=M$0s$8wp7j=3y*)g*L4eLEXoE6J-B3_a
z{F&aDXaUq~6`m_Vm+uDBnv;{0m6O#D06el0paK%@i*?U>PdF=(iA53xmy|F9K=vIs
zH-43xrV9$NXK3y2_6M#?v2k&D@+oKV{^0O>PK?L>cMSZl%0QixuEwIk`q&I-7J3uu
zD(FJKHCL-Odv8-t#>g(JR+5|%HKQ2|Q-G2FQGJWU><b(`EZ-rb@19s>;pZ(#9rdp7
zxq$BXu-r5AxCmFiHgHtu2{=C+TQ)+eM}jrVq(yRBS%CKFAccFDk0p8wX)|BP`t1+z
z|F2`Pj2LX1UqP_!`ixrK_{dV&OrhCXP*7<_z2jjafDPeVK|}A@NIOk|#(1fbd)e*$
zujc&5_BN~a+9wSS4K9xdE~T0-Q%lQXfKmTYD$oZO3Uun#A>gK}yEsiO!tOUM`xOz<
z{&0IPq4%P(bT0&O<BO=p(a{k&QhWjCIzVD2LOjqJh<SV$4|Fl%0#97P9RwZ7Z4H+i
zo%Jp5O96eTMzg^on?&xV_Bj#&-5Gv03{b@Y)_=2we`v;(80I&~7Zh0|3!unDg#s}5
zENR|Z3R#5O@2bpp_E3_OfPb6W=C2&@S|RG}3{}vSKll<5o+vgC4eESlf|&fYxXZ{b
z%%<~@3tt?Ek-zu}vp$jBSjJ>!Fk#=*0quOWi<i6Rma4<BMI7|;vBNY4df`E15a8^B
z^++1@`PS_gn33pt?<dgD>J(+^Up5HH-hZ*2XQE#v7tw{149xapl)oY8yiTETvK*4Y
ze*Ido#$qwQ@-H}oMtAKR&xd#&-#J=dcVfygc|15h0bvj?Mq~7Boh>DV4FBZ>F}P%t
zl9Qd+Lpg7b8jf#(DCFhOAv`fbm6ppSUciarCyXWjr(n33mzzsQCd?~=EK%a!-AZ!q
znI)7>@LZZ^6B)IrO$x(;ZzdCfXb9*)gz#~sD5UIuq6=wxp-~U!2FQ;KtK$YIwoHJ2
zU<Mfy?&Yfr${IV5JZ(R+JR}JfE*t--1#=xlZT5g;wbqe4xj!}D?{8+mATIZLl;*Nu
zSZ4rHOjkw=nZt4&QcrArpAma)<b?W7xT$|Kr`PAP>a6MiVo%UXhkJe&>X)j5*@X&c
zk5ySqjd`;_8gWov9XBS0WSZr2^UDSC;|Gjg{n>VJttvGy(8SQu(e*#n$D4(QgcK_A
zy?*@~;^KI@Wx?L%`UtfZD5yd{#3V8U&P%*fk(#clsk#da7zPXR>GNmBeLNDJw{8DW
zDfUd+9tb{R0%04(!3MQQ$YPwf<5P%>1h&eBNk9)hG2Tb#X_|=8{mOsC2hm8A$6;iK
z%9<d|sN5TUcg^9N$qx4Hre(%-g?0b&JJD{5z6Djmk@|=|6+>I<(R&keI|6#s<p;lz
z+6^R5z$*~|-L?}M8lt{@r-yPgY47z}KnNT$_;kHz`LDhK>8-LGU#5e>_83bD0i{gV
znu#=nbanN0r3-Fm&`fPD*NYPsIH&PA%<xbNIp_`r5)%?!raAih#_mm$r32K87|!>%
zU`%RcGL^9L@yF`y?CCvymJ%Z~vvP6>-RXc+2{XM~<q9w{c@G-%K0Z(x;j6m3XE><(
z)HF12`T3jd{F{Np*mCe95;oT+Q%kjQF6+e(bvaeLkg>Orhl~~m2xYnp7q-td#ym@?
zeL6W_vRY%;(RFeNq@RS#E)X59Ep+_~8rV-a_R4}s+7Kyjl*~!rTrMRaP*w^3?ComG
zGxRIap797mo|hKle=liDK`gI`dgaB2{k{P6c6)s@M9s-DyeJVsI`}tzMpM#B@Z)-`
z9oetcx;T&<#&oZ+2R97(o-WUO={t`d{sqtvc+FIfqFbj^IgTDJ1GxYWxbO;k29hdi
z(bQ`Ha9tE&^@YQzH3-^$KY?I#rOo#N+~F_wY%j@ETwEOQnTfQtG{lR_7i3iNY%z&_
z^NE2=<=+hrp;y*y9FN!ch$JK>QGgXk$Jkf|aETlTgt(-nq+bJ@pveTRJ#ld0sx4d*
zCRAWcv`P#?#SQ}tt7~p91w_I?fY&iRd>nS^akC1aEuGvyktaWD#I_kifzbY3;kdji
z8$Ao%)503bvK#vD2L#zGH7GI#Y9u-GKXNeonzDkQVfLRsjVLNBFZy})2E7Tw{`C4a
z4CHSum|gyl1n+<RTDq@XCglwLBxgtg(|*e*Lt<gD1lpUd!V%o(OXo;r7W~`QM3me#
zkPnqQCI0$|fL2(mp>iqBLtZH6gG)}9g~vlvVeB}lF2BhDC}>3Y*5NiGvFg*!wd&m>
zu+K)Q1lad23qrCE*j?Wv1BuT)l&Vva6>y9T3Igd{hf~g7hpl#XE@@Z%7p5Z{v<VPl
z;hY|mSQ5EQ5?ZiX?3kYU27_S97p|9Jpj@g7T>l#=vlcU6A?&Jl=XBl^U2q`rg^kvK
z^!u7Gw0MDc+2e7gfs1|iqzNK2vL3KDA>wri0jZ%56Q0v?(=|=V{!$Y!1dtEZxS%ET
z(wce!|Ea2eX5dx?pJqCpKYg0}x9(t}4$})=;NcfAEnGc33<Ww|n&`7F3Ucy}v2o*x
zy{uMs1%*y8NN7Ax3nIXen+&6;qS66eslgQfL~wymi}C>5uur~Rj{}hU>wPgNEz<pk
z)6l9X6NqunI7*s+P54$^$kmzc&kU_U^iKLd>;EnneOT>~9+4c_%&`8hnfDNRUeQ9_
zmd4s@MVdwI!<}HPn}++J-KXmAugr>AvET+ILFti!#m4oS+!bB@n3KHno*aV5oW-8p
zpJGO4?4S`z!k>+0#hKmOMmLj+*z&G}cl6|jc_z-MDHuAO&e^-P4E0{L?mx2|>j@fc
zOQ{=P2u6$9&mOyq8}_HOW%5K);$($VrXp8|dWTr_ELJNYKgRF66oD9rEVbK}1D<Sq
zSmTSDw}~GZ8B50Qf!s@jgqNH9w*DTN+r|rcet;ks3CX>*rsa(-Co(gq^%OV<7CaL?
zPlvf$NR42jfa;jv>n!tdt_r!T{!BQ!No8f&qzU{;)&;EiQk_SE3TkgI?3C{oJ(gf8
zsubo3i2f?woi$??4N50t!V^Z2$>)STXR`8*3f)RAiq;*_I85h>hhsMs*o~I+^3zMy
z+dao6S8+*e8X{^Jl)&e{sSa9yP~6%1z*lq2Umvm?To024+wpLI(#`Eu*L=qM|6BlC
zBpeN|Cj6OPwuRFhos{Lf4}#D<s=xkCpZrmaErrMOlb6@eiiSNjaGz5?v&6k+D%2X+
zS^u8IjEw+?@0!je`n~5#!b^>1czd?_?%lg6?M(|dg23=_Ly%rWpb+y^wtqM}IvPx1
zNA_aStS`0M6t}Up9XMvBp;<eatpMK-2C22;3@05OikD^cZFXbPUe&|HgA=GqAa^DU
zaDho<rx6>lM*~OFAK)0YW5)xm+l;4+*P?Ch`D3-3Tp*c%=o_@$w3_kMU0irxJi&m-
ztIC7OQdG%O+;OA#H6UOie*}83*?N0glPZEx5&2zl=jmn$hogl)+LFd^hfyyE+Jwb#
z$YUcO-e#2+&_Dhf*Fs%SY~vF6Ip_y!Sgk)7vzMiQIB5NYD&Z||CW}g{%p9yAs>by6
zJ98B946D-!!fdIt+PNy~=6!1Txy9n~#4oD$e(!G`clFL<)@}bi?K}*8cdG$aM?DM1
z@M!=M70z7Hl)G=)Xacvd>c6QO3ukO5?6k{nq0NjBzV^)9_;-&c&3Mc|98yRWkL1mm
zn2uLupPCE>D0EIGhQ}#u*Fz=l2tbpu@wA&zafuWQOTl>Y0C3LUv$IE|f4K1qAoZy2
zf2;w*yVZN3cTf8H6Jg2yTJz->$isoD61(YC)NrpR=yEeNGryQWf;Xgr%L;^%AmLPX
zV*o{R1`sVDEHyQZIiMrsGJFRHJBr;6i>mH$#~@u*ZRWRJYNRndim`ThvduaIo<Him
z#4lFRf*U!p9wtXxVr{Y+K3|nV@SS@zHKFoNVzf7vFdk`Pk~7lh?CH(swF&B}j%D@q
z$&@^B!QxLQ$tQ<rVcDOv+L+?0{BEj%ZtHMmI@y%vJ6Z0jdw%Sxz0dkxNjUL2=-)V=
zS5Tymjt-t&u_T9cHajL5nk_0rov4YbxV9+;&bK?EBIUsnTro-gg#u`!#(gYHvlXPg
z>gq8=PJTK50h}sFz|$2UoREfVoq<5{{b$t33b;h|{vxB(td9b21W8H7sb=YNd2zgs
z41f&!QqBz}^Jaj*W^J7gw4g7yA}E$wj7HLs?qNakX1V-<#AyYRkB^TD4+R9OB!&Df
zUXjwki$P$M3p=>vI{Nw$!8d@D<6qzz%?&C7zqypUZBTd3RGSlm6$V=frb;aoxbDjE
zox*_N--tI%DO)37yisw7vsAU!|C)cJub?PtUQ>3<>NhdbZ-{~j7MxRjXcFa{L(;p;
z2gR>>CunBk^WN#84OjH&-4yIw{#*NgpgmuKKzM&oWp_~x7ankAqhaUF=x+|HFSxE4
zqDgy(>JF0fRd#?kU@?|p3>wkIiaxw|n7Cj=ke=}XNlxOgW|f}Fiqr2}V#3;gdM2ld
zTO8=EOx8>$*W*Gc<n%E6>Jr3cEg-{4hH5!2eGDZmjl3kF{|%M1u>af0H+$P=wpF4v
zkQ6O&=KdXPSyV)k`)Nsw|Moai=G7R@O80x!mxTdx*u9D2D>d`^>X)<&Soo<OgQ!Y9
zqdhJ7qf~(ZLRt|GFSggP9qIx_BB-ue?Dr<bk6Wrw=xEAaysv>{(dPWs)s_8RO>ON=
z|Ln`gs?uRAH(wei(kx4CKX}+bex9}=T1eh`>Rm)}Ifcw!-NS+~v{Y6gV9)M+%ariZ
zTrmAeoD8B^ingV4uqqE8m{N;dS1E3G&CU_$<aVz^{Q3%`xfpf$jU3R^i;S((SMHLN
z%%e!!(b;_=0tBwb=-ZZ5gOekt9^CfD+E33PPH!zckHS7FJ3TH_Jf(a82;CQ*E_3`-
z&Fnd4cNJ}S#W34h_gAz}qRC^bQdGi{XK_0CL!^JYeZ`71_CH%rRO7>gIBpLj`J;x|
zOL(->(^j%?#qT1L;sb+y$vw1EF3=A<bDP}}%!y}4c9C>n7vMTAE-s3Ri$k^QFhvcI
z6y+-li;0QdI(+=t;UNcH%h#?jJ*oVUQ=Y!;#v=Da86|^B7t*&Ms$itIzQ#ol9|v<-
z8SQdBJ#_`M9pfmNYlIe|(l@|b?|x-V8t~P@J&)x31^cmJ>5#UmLGc((;{!oXNJVC=
z>l;k3)(^xqO}wQt@K^M+Fz~Fa9@<=Su;*)UGVBEX4b*m43M;LsSg2svdo^YvkksW@
zKWN8cK=jzAb$(g#^4~RAie6g~@^IFpHsl~tvM5y+Mt3waDXDDz!sy+rQJmD+8Xbvy
zvwr?JB;o$u-ww6uT`r1@z(wTg^Lt7{rX7@ix<wt_Ecl8FIGc-Cu`*k?3RiS<t?8CE
ztOu6?h_0pD{7F?WiNL4zt)qj^3}5)uofScLbeuID{Yxmz)DP||6$Yqjl^Ot8D{y$I
zV)X#thfjpDgM*SOZf9bW6qrLpL!eY2-0OL{yG?n@zi|>iU(!=#h*5_~o2Ap)TusM7
z!9~hFJnXS_#UkDAS-xI)ep>$#TJ&_S6mVi8?(JwFaSOYY%lC7z3xw8Ro~rx@u8#UL
z!%;YuR2030gz4<`R)|D7!K^=zze>deD(f@&)s#hp&<r|oPfJ-|OSUGboEX_=5qD@)
z(rgpz=&`t^BDA~{)_|O$^5sEk@;LqBAL0H!c~j;oyt1}H+%ypJudWqlR<A2PrWFNl
zbDd=ODOR)dJD=aLBX{hyOVo%(wF}pCVn2|Z?4OQ(zJ7?_E-Y{E4*XpD_V=G<IaSVc
zQ=n*MI;gT9s}}rsNg>F3PvWW)vKmO<UV-|`gy?xUCpE@0oK#>bzYdt4zJ2?4Wd9Nm
z1r7y{_CgTMmm8(hSRn1{t`QGzK)-4KGbB#y-GA3*qA)Uz`2o@Ibu&0VJZ`r+J4pkJ
zuJ9H~>g9K*>v=WftJIWTFzM<MS~bT-((?cO_0|3&;R9#qv+?bQtn8=<QY8UmmWO9~
z896*=DH?5bW_ILy*NQ1VYJDbu^|JaziPuO|%o?{<40=7G%`kmrN{l)zWw0&7k2CqG
zL{ql1IeG*4^l<}L+`6WLkPgw$)xDVU$j2s(jDd+4FB?W@gz)w@nz7a?Z7;=QQ<7q;
zrA(6gh-19HCs|i6I3KsjG~3U!(qEOhb=Ln;+g$fJVwsNT-yDvY;L^K<ItE52qbAWF
zE{u&QP_whM{oXXjqov@20Js^vS3m$fxHzg!7(nx$jD;lv<ln07#UKY!Pn5OXxbGvS
zmbw#V_uGaVFpO3j5P@ZcX@3;|ecd2--ey#2Q(NTAFKig1*zC+mfPrM1nu&O`1C{hQ
z!`MoqKCG!?ZWVmkA!3R=fU$BNG6$&UHCQU{*si$na8DU`ewaKchxv3#`A|z{@Q$vN
zM|e^PIu=&Wq*~7m-Z+;fRoTl+&)!ixv?E>EONv2x>~fHL4=Y_CL*U2@v~CnscZvxp
zrykn+QlGyXAb6f^>&a#mTQD_{&*aT=II<)53{P7Kycl$X*+AxPd$3x-6hWy_1(NFU
zH7KHQ09XtF#!WqdI`#l?UOWALM-VpcUL%#d{n^=B97v^w{ZVv*iI7q-EjXs8YHnL-
zN3ml*{=(#w?|C|=S(#NDnfndKICF?TC>)-Q$hz-b&b-Mg*+c%l>26UFT@>K?xh)%u
zh+os`CeoN4N4RO8LvN!nP+a$B5l`R0DBWMt;F99zj>XN2BwDS@_ndP5NZY;b>bk3i
z#_u#Bp^f;6J$0api0#C|wN&<X)gjldvB988b@@cxr?BTrW$L}=(P9U`C)34!Xy1gd
zzkUxi4;Q3p7BEhgt#_nGTWX=p1vT1_PKoVmo3WJ9uTkpb{5x>Kf_z1X)$;mHE*ev#
z^+iHrJ&-ab_v^|buy6)YgZ1q-;2G)yFA1f;x=#<cGjnq+&S&5Cvwwqyl-;|LX`v;Q
zsV;1MFDzxIz?AnMqQ_%KbN%FGnlyVjvg3^mx}`|lU_~h>G*XE%@pq^pTq@Y<p76FO
zNj_*dr?rONTK4loMKv;V7nVEUoRJxmD3ZEJnfC4|0bD`aNF}KK2;DL3n<FHLuq(bw
zD;mK^UPZ!|H>~0E`&YDGBd1SQ{kuF*YPHkwn!1QRN{<#=D6~AaW({N6dTy*%thUBJ
zf%G3U)<$Z0nf7Qc;q3u;)h2>v6%AegwSX<kwfB}9191?=u7;2Ee`BS8yb28iX9z7T
zctn46(jm~G7^+zYvuuIQZa0(R5RqC<@%-L6J2x=v1+~^jQOy#KCGJa5L1c<(r@w1a
zF~%SH`}g4^|6*H5%-Z1FF~)c6*gNW*U!z6-j>NAUEmfha8<xC1H=+S`lmWEs3{8_3
ztKC;KI?`qH2MD+(76N$A9G<MnTW{VIv@^Z&2HKwjM9}9FjN6W7vduOZ;^}n?i6Fy#
z-tJ&CmVf+{3sJ@^L)|sP%OAZ|zlC(|ZQu9FlD92c{)k;5EK8E&PfhCAlz{Z;YvN*T
zZP%SO@zGJUHT0g5e8+%&|4-(;>1IFFt%yRA;9NU6V*{5QlCp`_<47g`J5NWACe*!g
z(>@2|ynw*KB+;j{^Alh+__XP?0-QXCUVUxl>%fw&{Z~+ngOX`?u8Nl5<${S;1T?vT
z^-$i$2l?Iizg?PA*4DmNoSPZB_%?3|)&!PsPZ+WN5haP8k3UV&$`{FajcJ2ur-yZ`
zzB_zfY!S_fq@+&sW)RgO(|W6><HDJzv0HNv4KeG|OK0fOA!Ns#veu0&=HY@!QUntO
zhLgRC_{d^1VqFVm3gibbXMvK$O3Uty0j;eyZzJ#zqL5vOBa-91nQIV2lTj^-+>C9N
znTP`%bDV29sy;}!2abI!d+%rcRZWj;{+H9<Tb|_N{;u^sbuAMPfTr|AS(7Nz*(s?n
zF02@FAW|}pG#ZZBJ5r|9=bG77os{kGAuYe${hN-?&{X&Q>C^k0NM-Yj?B;SC&wsqe
zt})Kkj3!K6o`KOY?P4=N9*Y0jb(7G~?mR8zM*V|>gBPRiBDJdg3lB=loHHCWVEBeZ
zYzj=8u&}XD!#y854)C^zlGnAL@06?bf`Wp2|Ndpa$3rd3oGI_AV;<YLDAsQOqpLZu
zB*Mn9iyIk@ddze6u}IQ5<RX~b{;qD$!3cK_?GVaqRHrj-T(6)gE}Y#%nscAvOGUAF
zKPLrdeLIF3@xe+$yMks|WTeKNz|2^WbXF&m7S$Wx_n4ySh68Ky)aT;$<+@$dLJo4>
zm*IJRnOG6T>@k1c;7(+xXnUplB9=E%zbL<LA@58D4-LVzgNm1LG)_(sP~Z5>NLSd=
zUeD<zom0RR{~2NQX5kl<rIZ~{vSJW5crd;Dt3$~|4pdG~WW!z%c1BT>{$m0bMa-s^
zY$DE*wn9c=V6hSv)6}P|_NmeGF3EvqJeAY3jv1>8G097-57{(0)vWQ7O8jA?KGR7r
zhdZe8tm~+e8mH36cb$L%ED$i2I<-}UE)jt4)OOt|A_HYU+`<6=B_d*CdLC)v@~Wy6
zxe1ZZABauT;UXD=Ma8l9@LlxnI&4w6QfwRq6Zz^VmpUxJtsbD5=cQkHa`^*Ah#hPf
z7rS*=MKroOoE^xBin&}k21EK1y$vO@E+O-j;;f^KJ5LZ!@*am3BLfFS*M5S{1%3mo
zC8Vw56Iny3<=qT}y7Q?aUDy7o$(6kJ&WVc980Uin#e0A8jPA*K;^=JMr;@sRXKF$v
zH~QMUdgVJct1Accd-3`)!`UUDb9(m!O*a~Wuv~4hgM8b(Kq^cV?9%vCJz1-S_IT{g
zGp?U>j&PA|hOD|%w0~{h<i*qTT%ai^L5$cqhxgmMyG54vkw$yj6QGVkPfOInfdfGB
zIM<($5fF67GR2~oC}Eena~i=2sy2~vqbCUj);PQDGpLf#?yc;*j6x0!8O`qlg`{<M
zO!aHT)h0m<?;<_O;^T;OW;pICU1W&*ljZ&O_H|}MhSX&nyftqgBSWodM5);l?rA$<
zl6oQ29@7}A?&@4_jfm6@9>27g)e2ZDKysJ_4N?B~_z(#8`3%k}rh960zy}$$j_#+e
zTjxXEF}Yc;cxXrG1X8NzSLOCM4rboT)vH>*BpzB^E?n|J3utrM|JH7c#+{xG*;sMc
zG~KudN8J1Rdsz9$RKK26WDDw+0Ya~(v>qKM9aB?U*roZTr8LIl4qC8YS|!36qdKPK
zC4Sw7+1Jlt@Wdvp7KoICL(7zw3Uf2FOvlDop&A)^20*Zhk~S7!qaPc!wYkX*h7>`3
z0FJ^mu29Hiokd$aL5pT?u61&07R7$|Zt1bSY?D4Bu&{7oSkL0_4o6NoNUOowtn;w8
zZFPQavU6dXyQQhFr)L@bp$uVYQ&5!EkX3X`NSwJE&EW>NrqH9Xuk)#-rd)q29Au>k
zR$i~0#nXqUwVcDwF0DAw9kS#b*+Yp*pCm7;G5Id~6g1pGGg|Fs)n{63N-2p`OpKo-
zC(Mn-8wE8B<1`oZ*i7coP&PaF+)G)BkFMDnzs8`2eFrZ~=B0Kb9<%t0`=^>jd1!F~
zsswRH_G-p20ay5CZ>`~^<8%B;mw&YuGirp?sE^{mM>s!i*YZke!32$Au{+yQnd);p
zH6&AcUb#CxC90jReDl*5`fKZVsp2aG0fwg+37l3}%}sYPj(Ahi<1Iu#7}v?()*e+j
zNm1QH0Ze6r>j9)Z<?c%arLrC8MLM;^myasR=27k1GLCr1=oNZ(HfK=U;o&QJlO^jb
zODE>+8g7fm0=3Ap^U|$z@0@Z~7k$6)==@cj;2>!4UNnAux1W2z3`YgKbX?r})p8_3
zwo=OjRy^bf|6EN!{&ZVMLk0UcY(L(ysGg!Kd}YW(>3|ux1Q1o8u~YWT+W$TCS;@>W
zA&(!AE9Fj}zqH^&Hh~mo444j50n``bEx;@i!7DOS{??quRT*IvPh-I|45X@p)I?7?
z&>^P{&9(Q+%~v23MLBt~ZHbo0U35U^;6F+283hT?z@oB7o~xH7=87Ey0WYCQBX9a#
zIy0{fO_!!C``}^j;6jmcS<#l-%^N%y*I&-u(Ta?2`GXHlhzpNvU$WjjUFz$1TWjt}
zooKOP5gH4;L%B}WxpF!$owCE5Qru0y7fh0CED3JHFXcew6?MInX(;K}y&6JWV7LE)
z@!W*hJ3f8L75Mq21sMV*<>?KLDc#J=RMMBD7>hv4yM{-<$slt(<kcT$X8e$cyW`{N
z_T}KBG*e=O)w$f2XpUDKp8%3Tq+M0Bbk>MJYjlW*&n!l*;9-xi^lqxS)KDmXw~h`{
zfeEB{rDX+o-|&2;hf9v#O5D!H2@xFf+#?)Joahmth-y!IxQER6H16H^M>jLFK7VaO
zT;ft*!mJj#g}YJ4N*aQF^XN7vC0|K$dW;CN#vhuyOq^)qCb!-_M|N@8$gRCU#lVSN
z<ZweAj!Kd1N%@89$Mn#oM_K(jEYmvO#_le^HI?}hQr@D+ze_^s=X|@l><XA3(7iL5
zO`h5Pj^=ZR?gx8x&|=wEgDkyY*JfHYsBq#1m<MkvaXOZBOGv=<kdT?jBrhP)g|PMy
zDq^EJ9eMa1?`j<~OBdhz7XDz%V(*4vH7MXQ0_6b2!Gn?ct<MM4E+eLL*HxFzL_w)2
zzdnN`zt6a2^2%M(6f-Gs^3v_Vz|X<PR8;OyiIv5E8DWg8eNv;w&AcHaRgpO8P}t_I
z>L^Wqe%|;NhnX|jNzcJ@P!+I)w^Y=<p37#^?RLWk;09X$vQ&Q9mUpyWXZ=`sI>u0q
zP6QFbh|M+d!UiGi&!7ImT_C%a%h-PzVqX)?+7dIg<JMbp#Q%K&rr{$cu8p*z&E^Jd
zf}^=!)q@4BuNH})<9>Ew@-2IJ#w^oiR5^rN93fkoJd&GnmlYrU8Zsov*74Z4#e_sI
z3Ynkcr9`7pZCxf~cDLXJr89ddA~BMG%D8%->#6RXjjEi<Rip>g4x-3CY1+N^koK|f
z8<~Zj<Mj<Fl1pk-g9#`nYpA@L?}5&)@y?h%(Lyow<QMAEWyck<gh{=6;%nMMRcgAS
zjt%dmsfbLR5q!a3K#|%@C5tm1+z*Y(Rm*+aT1aUm@UC#GXSAN@Y^J8<WWS|ld#0bi
zbOp-obqP_9AC89E8wvF7^HmkJose91le-qSh=vqmvLgKd;{DZhS#In7UO|bP{wPaw
zageHj(paOj?aRoM5`e1wB<%mb_OQe%KAP9ipkH?9)V`}Rbj)u%kL?7>6Mi6d-=<sW
zerWoqCb`ur?-c{K7N@kHsqlTe?h8U(W;-#3H*LgZW9^B>W%8!-PU6~hq6hlhId{vo
zq%V%s@UzF(u-ivfx+667tnkw-wk*vKZFP=ZMRBs5*Utj;KQHXy39+qPM3-teOxwDq
zlLMDq`e*<4%+!3xRF>B8lA@jtKewhdGZVP`fF0HpESww;E|Y=BV*!!t)5qDzQ<t@?
zPv2&eJr=omDd~xik42uhk2*5^YE-xXO>6<X{eh0=1dNGkSqiQI8>Cn{N%@nt*6T{E
zuWIxvZ$DCh)u%L2A*)c2Os<HrM$U$KRYMsuqnXny`31VQ!jID&-XCM8Mlbo>R?eo*
z$S&E4*sil#V2u9o=;h~D!n_CevHsIxLVj~y6IuYPZJ(`(zy7`(b%h~y@TT)~e6$Na
zJnGGid|X_3aVd;Cd;_5a8oOK`i1N^|)x}<kzZ6ej=;8y7LF=OSrha|88CHt78iMxX
z(sHT}dtJai3^Yg9!aE~u*(3iXxq%wn%5v!k&f;8JS#)sG|K9Foq$4VT#*k?S84j`U
zr`zMX#r3cZ6vsWUIL5Y$U(JC*BG%T-5FkA==%V)IOD?)5!e%?MRXF6%QxqQvpJRBZ
zd{sX%)KZ7kI1Tc!jK`7EvuPiYbU9Q{4Nne<70@;R@8Eo8P`aaOXkwCIb@oyu!=oKZ
z_ltyK7yqFOU0qo<b1qa*>N9$Kdht18s53G;4vhx;$Rp(j7Jg-#I!*dfhM2EEg}iAq
zFuL2yY-M|A*tWcVzqpdIB%`76=zq`Bj)8;iw79TP7##otlc&^i?2o-|CRN@mB-H-*
z$6NJZA-oz|Vx!9&E~BupDxM4IiM0NYy|;|2x{ba?H?|;P0g8kuoeBcd3IYO3v*}O)
z>F!1Z6r{V6knZkMk?!v9?!Ie_?|<C$;eNT_&Kbusa3F5>e(JZL6?4s5=~CVuDO~^D
zwyZ<9_Tj44`zM81%x#P2&}8v@6xW{j*8Df&Ds|Hl(l$TAE%L_SuDlu&PfTYPND3}d
zFyQIs7xoH?qAo)?8wOIiEC#6j*+Xe+Shfo`cL*ADX?rW*ywHAY#8GNS8Sdj=-r9Pg
z+aS^;WAmGP5}V-)wW`lPUp59xX)vb94gSAo=5Y@5ZyDY0{MDgb9yg52H{*;mLY?e|
z*M?Xy?e04s=uTr}95E9&UvNpOX%qbS(fAjs)<HLnj(U1Dy79NitHf(s-gnrH=&^lp
zNB3PX$<Y-l{#w|%D$g=;QKNTNIeJ|NZ~w`qXf3qIc0pMb3f0Fq_aEI@-6mxv^E4!i
zY6|mm*ZoFpXNavabwo4EBY2f@LXw1g<wX1TysqZWrFCq~pBZ5kk*zY7&F`MoV1Gqi
z!d`p#Yx&<UOzx6;reBBAq@IjFIp$R`VP+)iC9g^TVU8iL<>bSFFxvCP?oO%F`(SkN
znwCfG>~?&^rM9cJC(L)nB6KHOEVAjZPsNF(hn-$FI91uVyq<0nBDyzL6V3lXh$re&
zgw(^~Bm<$@g?Ag!JXrY*T^jR<035d~<?DzgO^1wfi&@JH`h+|*Dp8m4&|$0zKo7e_
zCU(9_jq{#nKUcW;mRqF~RNs*E&|QVwUn#uwPJ15Viu>mYm0piGUlaqIQ8%Dspm4_f
z{UJv|{dfrvr=vxq0lS=Ar*|%fpkH6^au(g&Nv(Qi$m@2SEA4E}^qbAaH`kL(iWLPe
zf73$`P^5mUqVnmb?9L+{9j|cPu>%fjFZ@%>Ti)t@+zq@Jk2&yfdup*h3EXul>vkBk
z>DG`YzF<@60+%w4fHhrDZ}bhh+djWr-6qa#5^6Ad|K81U8;tnh;K&A?VMaD!OL5Qs
z@lZxNRo(rsP?`3Umiqmb6XJcXaozOO`Z;y71ZJ~F-q4I1jB=Zb%GwZkBnEYfdc%)K
zBVv_yPeq^6iX@keTYN{H-nxD@cU^0trQau70^PUBuWoFG^hlT;3f$<0=FGpbOH2H~
z!HVZuPO;EZi^2L_b^2u9YrV}>SOilUvrR&eyW9xhdv1KB@P7OpmwDG+J{f%GDemaF
zW|RJq>sgV`BCb7a(t4k>80yM8)EJG=j(j?3^LsEG{tA-v4m$vY-~R4b?1BAVlg%!Z
zaV+f6(JS0*7kGA@49jXG)$Yw*H7JZcok4gCmcQNJy@MHL6Ax<1-ceD6vRNt>8K%Xz
z@XPsK;^QAfj^fxhpkbKY6-=5fx@|Ubp#n#5>oNLC?xlU^%c~pCDy!p3m#;FbFODk_
z9Y>>;b=6#=E4{kWUf(u3zB=5VWK<V~r$JFDt~t{ilxj8>KR|Hu4x5XZP;O;i#Rs#t
zwILeQkiKqRgH!J{GCH2fpwU!a^^hQ{!FA;ROkU2#BMbFq!8WzEWi<i+o%)V0S0S2@
zxh9K`ZhObLH&A@9O6sgHu3)n2r)KiK<2j*5ejJW|pR8YJ<sS2%5WAbD?!44_tnMW(
zA&RET#Tk*mCDh({OGW$QN+dOwNlqML<HP^30BFRJxkC40aW}j}d9XoE^MiJ}##-!-
ziE5jhP^BNB?QnD)igdA9d0p_Vz|oPTSAQjD$MNjr&~Oj*ieQbAQ>q3T8JSMDeq2lJ
z`{2^vKk86*g^?l2A3x$69l^stDYJyBovru7wS~AX!<K*E(Adb)`OMXiUFC}?Y&rP~
zx7<WTQ==>yxMlM^E8L|#CajAQW)=l`MGt%!Vp^ROhxY9^1UxJf>oT{g>n#a-y<TR0
z`%TmS?S9v}tJ(djLF@falL%;Bva;^Wau~HsR|U9SiN<nb5q-iDT)<RS%hZln<r~aO
zu1%62TQfFLq}9QzzywAQ=b83YeO=6($IL8#$r<$!Ylja9(h6^Sz!ik*<yg7mcKccE
z9ShMxH_@V+gjW>Q52BQ5F%mAYm%Mm?E9L1|W?U;YSd2NpHdR!SUt-7`RbYenFmok?
z%U}X$UZm`$^0h_cPZ=&18V4f?h(W-I5Cym4>xgyud=y0krDAcdyeFVH&?*95X~Oe2
zbmz_fI+U*-<CSdO%qbJ&Grg+eJRR;))`#DHNm97o@qSyDA;ZVc#xT~m5&4QekW`28
zX-EJzXlim2pZzlU8`9;VNJQBEjU#Dzs6Ayk5M0DSOa_i7m!RL2KJp@!gyB;|-%Dod
ztAzSFIBuG#ij@Su{d)^MBMA5!wbe&G)K+jIiT$o+6X*2Rp|01@9il^(leBCtlVzmR
zMI9WtiHL}_3KSR_8MAXz%GHZN^<lMQPX)BBszFK8c;p0%x0=Agpf{NRHBC0F6I$;8
zHHXUm`y_pw36KO1WE9PEN($&rXrL&#QF2+lyceoY4{4x8@>}gmHb{A*)Cy6A%-}lc
z59Ut*EyX*i*_$o*Z<H1TOAe@6xqy)9#GK{HXCElsbDwNWu^5gjachDm1xk?Qvzy9U
zf5aPdc=Kwg5Dyd?5?^s~9<8w5+U^o#(rmuhy;!Pxq9xiq$f?4raDKqQJV+_0bFB=I
zUL$`flGNO3o|jQ7W~lvbD5vM2pU)bQ=C`4)BrMVUd45(?<z@l;Z&nxioL2_V?f5Xt
z0V(&5fXyF}6gg1#Iow-QT_}cNHoj{il+)V8b?7bmoviXrbZFg7GqMG(K)_d_)Ue*I
zjaC5d*i>n#O$8B1^#V*Fj749;^$1i1Va#Y3uVe!X27M3Y96eA{s*o=4?CQ#*c_`a+
z)~oewvr~itggrl@7pa1nAS9J@K(Y|zBSGqyt|}`b;WP9=0PIEnT?|Ov<ymiO+HB9)
zWGa^#nVFTZTFE?R)qLwC%U8_y;0OgP*!lUZUU}d(eWsP5!q?6st3j&bJp`xSB_j4X
zJswN+U{&XBe~+UC%BPVPEyoun&)tW2IB-^J9@ulyVstkVK1=198hO#iwMT#ectiU3
zDPYH88=O2nF}Q4z<u(QYly8n;^%);mJKdkcM#sQ-ith!0!mWXiF%puJwSYeR)6|r<
zxeuU1RMBL7u_p~=sh}2%5*7hEi{+8B5cp+){6LNSx<I^NT-+o0M;(f&>w796hbO>Z
zaB1b7z`Vm5A>-h95a00{UI~Z}6#Tz5EDzrK10+eG=??~;9PPmdK>-us)+k|UkPU2p
z?87O!M38<fFSS06<WJyZSeGK8%0UAd&zt((ps-T*$Fw3&%tGStrLUCqJ6URDxOd<F
z(kfetRC4Zccv3_BjQV-7mqPq+2GZD?)_*=0n*4R`MyS!@N?tK4L{qgzzirxqgFDt<
z@Dy4y+yN90)Q>3S<#RR}&Ja*RyNFU-K~g$F8f;^;iD7DH<`*b_8XFl=R<-r?1Obsb
zzDrGYUK@84BpEG7md6;OkT%i8P*t8OXFK0cf&gHr(svve0E<IXO&t_Q3%Ki-m-SR^
zSC|o~M++K58rSkbbQho($9q6;lr08wuQ4FVVNKr>aW=4;39CoIZFi5Dm>BR>A^<{v
z`SK<6>^rYe87>vfmzv-)elE^{>|j@%;!X8MHAU(-JQsNa>C0JKv2opNQ;8a@`V4$F
zE<P6avpkgV55K}@qp+*z-kY%-f`&S4V8qk29_E>8rbgh*G6w(t)0S*gsB52{lk*PH
zjR-Ix3#CVM&0!a-K}keqVi%^N_6$Gl=?4MV$<+cPyY-g{R1i6{G3IczBJ3HYabhYK
z%_;AhmplqTN4xj*{VUf=19PZZUqPvZ(A<6B1Z^;zwHOSDZ0MGKfctlULr+q9OlNP9
zOEI5dwA{UY`~6NR4D4rtjTpZ0@x;AQ$CqIRL!<Xc^)R@D?AWuQdA>Ffac|`A=vOzU
zlA96OFJdU%NBgJ$S%|(4ASx;S)1^p6E4vWCIBqjVVZwPZ9e_IUY^Jk;cp4}Tzb3`2
z$L1{B12=|%^=q@u@iecqSRme{U=seB%4q?eCNvfH05(EtRm?@(dIO>42G}jrdP2+b
zwKdf_*A{OgE_CsUVI4`N91*vjRnP~w8~1Ag{`CjyjRtbfsYF?BY8K9bu}kk1Xb4XA
ziOYfIHqOA}t2Np*BjI_B$E!SXf~BVF$@9i;Wuoq;)e3E7>#u<!ufmJ$eU7!~_v;s(
zl7k@nbsW`eT*EEc;DAlPUyjW0=g`dvP$C3dd<*)Gun3;N5Ee$~=H^DAl;03A>ZwwD
zus?{zFN(-N8N%kCPEZY(u_;SRqQIk98ZYQm`2d0&dZXp{5rF@Jj)<RdXqBQrzLk`c
zLZhaQMFwndn<2tM4R<Cq_ws0GXXn2OJ*2~jLo;k~TU1Hhqb#~`U9)Tu$FiwEZ6d>B
z{GABYmA)o1^r*)D8@$5xW!b?87LGqOoV_(&T3<&v)U+c)#?qvze~86kRj10NAv74(
z<!#*>H7jkbC^4EcluB*~G)S-XC5k{JX-ejZ6M(Yd8At>n>JRnaIYtw_C~|oK+yk@4
zeV~TS>N&8swnlTrwb6}FZ&yGDX1qZnk*@o!k7ipsqtR$7*9G|bZ{gv6++l#@9#?sC
zDx{5_Sg#P;7vD63OC)pn?4+{Tf-Cx8+_RI!U-U4tLV3ap>*sU?-p@<N$?0t!8_@>6
zBAMR5eTcDNyn1yBfzq=9i#TlFrySJ!d<8JS>vNCvc7)P;ZsX$q1_(7sUfhFzBme@f
zkJ_&F=$sjh1lF4a619L^bFQ^fCP3rFmL|{s$O2Ic64`1`ANUW@;e&W-Mi>OW*Npk*
z(LKK&aor|he$qU2{9<A>n>4zAB&h>$A-iKxfkEi}!F?44nVwY}i%V6fN49cSw`Q-n
zye0f7XCUU%Rq{-GByil;wpS#&U}A1gaBp>l&1Cu-01i>wK8W(oMw%dRVnFF5U~Olh
z)EoCUf`ye@1{#R2y)ZK~%lM5=n411BSF;5lI=DpYy$N-PKZ1gglZgY=QThrj|FpJh
zLdo1I*Gf^pTd>)U^!!{PGb4+gzQ4j&9$7rSI##gUeRI~M^0naPf+w;`?B7KdZ7_Az
z^{hBCmvZxXOD(;QT`b9TqHZfz(zs30kuO8L-(Oxn!VQ}0a$&Q_+a~8#OUlZabbDzL
z(0*In7(~<1(n42Nm7M$qTH_eu=iq_K&MA#Lp4kU8sEUFcV^fj)C8|S<6!c~r0w`o7
z6g5gtkH*ff#KOKE!!vyM?p=hi<2X!x1Nu{tS`*`QbM>Gsf&Ky_k>}Q%q)>DUXm}+p
zeYe7PPavu577>wXnZ=4ukf8lfz1Ny;5LZH`fS*k$k*g$J{$0)I&U+<7^1=1D&riMF
zs}H($;MY!VJzevwV#8rHUw70GEMQH2|E%WsT{*Ms+X8OQ_e7Q)YFOzPr~_N@Pe9!k
z5&Goa1rWZG@pjzlkrjITcE$ypdmJTXYcdSg=*3787>;52hCtBX)ytP<*vg6-&%kp5
z%rC0n@*_B7LEcm}1R?VOE3Oo1A4wC+KN{=zz$QInLd6inNMUr3Ak3>Y$LO?P5|OIv
z*0ks6>GQAQaBD{6UvI+m4ba_l$2UG?@B1_I$$pf+66O^6zTfvoTl6O%`dG&BVnEhz
z+ndO_%&LOQdkOD`DO?sk-m)pus8+`Lc`bOlMd5OXP67~UXpa}f1|G1Q8kej0pJ|Cl
znI6D=<v_6#6k|xAJh|y5;8Iq`fr6lb9xe34O-@fsFIfRn7If*2K>3zjH1ah#Z186J
zdISDR(Ka5?9;w{N_%f|07o0p$o@xQwG$Rue-n1SQ5F^gBJ1{}%&*bHqMzyAKzh?=2
zz~UmC)D<K)l__c-BY--G!6us~GAxYd!B=U#Cc1wXh?Y^=JGG&589?3#O4&x67V$z$
zKk_R$?q<Ch3wJp`yCl>z1C8z!vK6hFm#d8ChgR+(Rb6{zgE*q}WB(uyd4^c`-LEhL
z>P@SIE7_Sba~N90<mT+5w^8KPt5*jX^b2kQWx&b_=x*J5EYMwor$6vWLuLW&2Z*$3
zWK5;>=%6}i0L7FM?85d`*QOkww*tosoK_BtWwi!>G7$9p&b#IlzIz9b+M?%2C!gF~
z`m>FteIAnUtqL|%de-BXFF2fS%VN<Fxm6}_Uf7yFOHZ)wW4Ym=aRaRaQZZuZXLR`6
zvNFNF-9x_Z=x2dURYc^Pb5xA<<N+b!E>U@p=_^++ZR?1+lf1jMyMUm^=$*S6=iDRW
z45BO{K+I|cmf-Lj1DMRvmZA!t%(q;BzBa<ev|hl<(BAX@OH+M?^_`i`EPTKy);DT<
z{t;_JQ)J%W`xI@^E{&nUoUk;mM8Dkda6*n??ARWQi{i$k;}~q#{R`Ot2A8T7Vbg9q
zDQ+W{KJrM7ci1ES)Vo8S5uQU8mEo0+_^a^f@@JfGiSrz<Uj(G8V0;H+9p!_Ay@Jlk
z4=ZLcwl5yi)VzW_5``hEPkCC4t|LBsFDp+aok0C5XS;(<*w|A1&OZ5#Q7w!KZh&14
za`35Qhj!ntoWIaD<m8Fz9;`p-P6dAPx7j(Wf44)NM4pYaB^c9Vpan;4FZKC=>&o=5
z_Vea@`f5IGF4*Kt#>U1Fnt|i@2COC!RR$Jxp=91OyY^Cy@-w~H710TrQ*E7i7~knq
zL-uq#<=rhe01X~*cYeHQb_Ej1OrK)ED)(?C%RK3Rz`c5-r2r&j8iM7`*DA>V4P$aw
z4X|#&gV|IB51Hck>?2vw*s=rgJ^(9j;o=I2ihBS1)Od5#=~2tm%=GA^n0wD9+{bw8
zbKfCpL)Cs(5&s%Z4FC1=EedA$4U)`If&6eBTbNOCygX5XHLh3ZoK<8p^V4;3j`F|H
zEqn)`3$3M%MSQ%x)P$xhygC|b-8g{e006xFIS5@cYITVpGVtyfE;e(f<?B+W{=nh=
zjuuQ-tveUnUOmF#MQUZ`qvLa{jKY7v?rD218j+#O9%+R3L`pD1?@Y~nWDoEB7+mX^
zm$&yTXbNEnF@!>Oh+$ghf|wR#SYy>5^k-I`08!p7bPwlS2*c}NC9XPbpefmp4D7dB
z-m5CPp&KwRDt68EB+rXi=Q=@~^H>*&qR*f}o&6e*a6kA@-yApRP8xrYoxJ=on_3Q6
zET=PuW8`BBsylr@h-QDG_6#-c4odlo>ZR#v=)vtUFFHCnFka{+L!hIhH+oj};X3TC
zjTxSwHXZW^V=+z~gE=2Y+ad?dF7;B-s5&M}-hvG5l?qP$HY7LmZ)Q~63HxOh>UsT_
z63F;Z4$<qTw{yxwyH+O3ntlG;-nfn&8)sHZ$}cD%%)`^OL#F`o+088&&x(nWu@R#j
z^zCPkQhShV_fRA3{zH>Z^#Sh<bm2yXMkN*FTYrsAc!L-|=G?&078)w!GuD|@SPFHO
zw+1@r7qO3@BS(BHk%Kf<Oic7hNh{oc^E6h3qg@2Oj3@#dC-~&!lM}RJaD4YciVx9W
zpo<5+j)-wEO*2P8R@pSTfLfn|+2s|1K4Z&H^Wi9HV7{nkXlRH`FVF9E9pcoZ6H7<o
zT>dU2{1q42lhh1BPqEz*v?b3BfjdVx7R{)EzSNuX0iwZg_{^!tyAV?iqTc&e2Bh}`
zZkxM$!lq1U5rl1|gns)GZh1Ej-9oexhxRT$EjCh((o0)2hmARchJyZP+0nDF6b$|`
z(Yn^6|0YFZfR2o*8Ol<v+FZ%1JoEs^{KELuR9%E&WkcCor3l1Q2P+yNA7!;a;hquY
z4Kmpc>d%Zdr``eVFa+bM?OnGl?N2t_vQ7h|d-mj*zi6_xFKSR3$k2}(#7e$=|1=cA
z)F*dwJhgLZw5Mpd!p_qK{rP$eD4VhK3U3E3ts3GF`}^A3wwu;ITqbCtA;9>z7+O|H
zhV|02GTG8U_|+V1<$MrQTQ4Mr{Ap`rsImuu3awl=!s~Y}5#&H6L>lI9%W+K6KuLv$
z_4&{X4%YQtm-mlJas2QC>N;@B*mkGZw)ck+Zmk6FUKKd>MR-M*0SyMF(*Pq{toNJW
zqXG_Uur$$bwx=m)xDhLsxYONAqim!XH_-awio|MM;uOTjLx0VFN31i<YS*#A)pm6M
z0m<IZGoF8#($6<Be*>sAxm5MLaN=Z2QuKQmVM0QJxTNH#tgNho6HM2ewFCfC$UdBD
zYsU$u<S(jMc)`k=eb98bNsmzPO-(+dAK%#O!;NnK-;RfzqAK*50ma2LX7})a+_mZ>
zui0jqz09VHpfH8r24N`g_h2!f7v$NFM?^Jtjfz>n>0Y5DJY~FDERLYb^aY;A7?sm+
z38o?Uj~(NKPR{L<mfG4SuAX+PYsO8<AppdJ&d0YAjWexK-eiEQCN28xl1nSXhuOV~
zHS4g`npJT>g6GDTTTy(8$>ar^B>L$bjjY%lU0>puN!%{`W5ljQ>DStf6r&ghGqX%Z
zMErGJ)k}AV51Ngn!xt6j4A0g-CKfMgVi*><wq^!lOQU5F0tOn4_NdRhwWKOYMZX$L
zs$B|>M+MoC)fu0Aj2ZQE#A|W1{r;DsAi!eU9<3BvZ?&+ZzLw01Nj@j1fgMJQ($KAn
zUUw2qqgqvIYQmbm!smM@#OHU;ic8N<bea|c>gL|xq}2Gm%+cX1sdN<ew)*1+)UC6s
zf0&&OIqA69ud=s{iaA~u^xCw^NCkuj*T_2y!EC8K=1B^AVy`s&^FBuy7Jg(8M)}!6
zn<J`w5%slrJ`A49{x=ZI4bNPUKUplytH2-C#@<DNU__C;3MZsg_)$bYHy)aq5&xld
z#uMdJH)!F&x&Ka-HoejV`v0bQqLBJ$9|LOs1P|~~kChMOUm!>&P)ylvUDrc|zR~lN
zyG$-Mvim<~UM*b(RjP^blQ0~w?$nQ!s0pCcWi{Q)Dqk6Y$pb|h_sWrXI|TfX6D3h~
zhg-X9Jb&72n*7!sZbUYt)v2PVXp?0hQgb@oDEu)oIraOq;i<E%+K#^0<Zd(xGr5E|
zTE|#ZIp^Ajb1<|`+-?GtN{i=zW@*x`j4J%~q8U3vPhZ2_hUC@B7oOW95*E(6p$beH
z2hLD$Sj)Uwu=5=7-9eXAUS(+M>06Cg`9Q+uL1CLB-)W{<e-IWH{(6p-=9iE2ffo!@
z#wWP0IZnM(vKxj9T*b=gTw@;}fO<P^PSig!+~<sjWcLO(<})VW^OFJRGX?59VPRo2
zLy!m3uYLn)hJm1|xgf7qX1u1<o^GP!0?2rNf5Jh0b&s7soXkL#8b_J9)YJFK#&|NR
z0xAKb$9{6#A?Qu+R{85cR;tBIzpc8cCOIj%SDM6a?~Jsx<Pz8x*UfD>o;v<_vl(~g
zC09Rq@W69{!T#{Z6}|%th%j`_XOtg)6L<`%VQ^*w#~^9=M5OZ;?-%NsAL>*ODlLMq
zZsoYs@K66x7jc&D4vUKC0{1O$?(<7EMu|`<Cn#}4N%dLb?1@6_+XkfOs*U3Q;cm_m
zQ}lmxRs6(%d31byd~+8)XaU6Hq9!5qZ1D$=El&7~@`Ae<3sQ^AAU^e;)HTx-X+S?r
zjL)h|+~fPIqMf6r-Q?SOMtmh|H23-cnE23$_kIPylCcgv19$G;?QTzjb4~Sety840
zOC*(8?m^;HfQ*$Bs9d8l-Zj2haovQLwz*MP@t;d?z=qp0&gF2tZ%&8L%f|<KPz1=0
zUO=@gR#FF{gAPS`c{%i+<&a{+1L_C34~;1^`r$QQhgeQ7CEA}Ib@wS4WS2-OKD*it
z9t7YATuU;IjY&Fi7M(+k6HXf@HvjXUc)M9W6`vgk3B*$|th>3nDcn|p<lB$8Rcg-)
zS?*z}^H!ooQlG7-2$V5ZYsJAHO6y=SxZ~9d4`SXRGpQr#_cM@pwiGPAZD@o?qw!xo
zKCV)}g1-(p(qF%ML;Cb-$KZFc#d;Q^3cRaJ(Cxh6(HtZ8OF%9f@dB=C`KKHtv|SFb
z??J;@HV5JRtriTEM_0Wyr&ibammCN)BF=kI!FRU#&~an5asS{zo+AOuPQOMNyh%+?
zUf$p#<&|ZfmvqO)ARl*}BPsZ7agk4ZfX~LwGr~P8Ao#Rxq#j^$)s7z~;zi9P5~CAc
z*o+awC!MdK6(Y~JUi@A2FIV0%;(z6~cY18g>VJ>;Ez*WjLGWX_-SPE>!5D)Rx2a{F
z8W|(r;Y$X8IErg)c_I4*?F$HC06r8TKwcxEHrm^Jzpv<%ecGb2F8Zu|07<po`%~@H
zuMjm|{MJ!}Z&rr9CloZV2zU06UVu8|BLgwg<1K>YG(xNX-trH^Ps?#GQ~iQ+_sjV|
zqdPI_|Lij`U2W7C%FwEAU-`Px5#&8ke#oS!w{p$V2pya2;7TUZDC=_(lJvSoI{Zdx
zxdGk={v^>TNZ}`dyb~xuxv}UEO5cMd?uiei5E)Hp-)!4p$e!Wye!({b*U+KuhAgg#
z?FxaN&2^L1gMm9?7-wHT7YM-#f9kMdTD)I8R{YsDnXEvd#f4b(ca{_)vcWYa352ii
zy$xE6Wvw|>54FH=#k&bb?#CZ_(5M&a=>DE81ATtigXXZO-fT?7kDhYluk~8Cn+_AW
zRh4=^U5lD>X2mw$W8%5ly=N>AOjN%FInTteee^jSZBL1y<K!d&&pK13Vvy2%#COh=
zCN<=KIX^!?@WFiJvfb@|UKkte1f=j6pMX{)6fGo(sjic}Dyhl(R+AQl{b1|>>qJa;
zB1C<D5xs_fTpQyT`D^-5d;>-~2*emxi?vpVG^fHJiqgs1M2JJauBm98JP=_kSas^O
znDDeJuDf`Fp!(O9+#RFu2rNzNp%;4_7mMdIHaf%4yPk$OJ<zwTx5Au%Mzei34M1*3
zW#2m+f7ka`eI0#XRISZ@eq)>#>+D%Tzx~+yY~aiC-I_MDHA};b8iw3aqOuM%b|0Qw
z@I|*}B&&=zcoFkdmKSqCoB622kv>YKv_q@hU>W!pnu;t}hp8^Bko?5F#j-0pxV`ps
z)u+!?XT?DnOTET9ed)LL<-<<HO|H-0tDXu{hIm}p-Z2As_1oz<RjJq7+1c`lrB#5&
z+35z;+kn;Ufx$Ad%M~7v_=3j;Ztm5(s`B3?rdW)BZan_V%7J?Fl}73p3|~yfrcv9V
zPTsvYL(_YL+LwoThY}Jjt*gB2>U@Xy%<cVuXq)af&UO0xICRwXlqHN!c@lnPw53dy
zeX!%8=ewA%Yc+v!8G%43HBT7LPsy%!jjrg*q-~Hb@@(Q+UyZc&^A-~?wi(?^)Df`g
zO<faQ7@0HZ;!w>%j=0g$i}yHfDG{w_s#d80XLKHE3Y6d#velfShgO{>Lga>31hX#E
z%o>pTHo$MEtb7|0NcoUfuCLrADbsEONvKe!NU=*Gqob5dJBoXmzgxU6vqd^yqPVQ-
z5;ZAfY1r!N>~)Pu;jWr0OX`_@Dh;xae^`I2zTDACl_<j9yuN8_83sdlYD!n$^+Yzu
z&{wrzS>!QkwXPdsDp+Qf3~R_4JU)sz++Gpf*-W?`B$u*8r&OG+(6Z8c^x+k~syB;9
zV_;kpxGT||oUL50Al%w91ENo*r38rQ40%0LUS&m+a~w8lbIH2q3+v8{Px3h?=u<CO
z;iKu;>*v$kt_$zjE()02s>LcO4}=-v!C-nk>L%JP15|uTm2Hj$H?gkK?A;?YwBmPu
z_cq_UMRhROz%NaNGwIv68}b>*%U7;MoBE^z@xvdTg2mILHJ+tI)17qvf!xG&WSXR;
zB&fZScoDEkdOctb?$PN{bv%(Vi)_*C5+t@y05bt@zL;{US;x^D<L%di|6u{DvY<-$
z7OE%6G#QA8XjmMyO=H*$_4LvOyzmpD!g_FUkRleteF&u(;fffIg_<&;#qz>u76#4~
zUmqVId{%>W$SY|<vh0_4OG`@+kQwc?^I@j|M;7S8OR1|TK)#b3in%G>Y~N~UXsp`a
z^eYw_D4PvPH~yfDjy?<Y_4OTJG(mvQw?VYGL4VeCJVlUSD0$37p36PgWllsTa`&xD
z3?8}j8FP438uZ;t{7uo+)fDH;&fF6#b?2bxhS>-wtR})q*w|_eNjO->PqH(Z^73ej
z=9!!OK%9lFPsb+1T3_c_QmTFkH+^QV83+gCKhyHnY>i9_qLEF7lSoFi2Sg5(tgNzi
z?%3xm+tSn|WnhpAT@I-*!ANsO2Zu^hDlRT%)G9(D)x$us%Auk+OEs&F=S=D1RV--(
z1B0(Itql6BwbsCRk_crxycRMXI}b<Hp$0ahXONL(^q~^6C{u>s=3Y>%Z7n^mcD@KT
z4}bEv;R!%y(*tijUq}&c71zJvj!oPPrS<R7Da1SV+p7y|eu|TJ0^^%bdtgEMLn_J}
z@v1>#xVRV}ArNTx{G|v*YmYb22lb{u^cH1FA^dOY4dYdORFc+%cW5}pEOdUHp4vsT
zv?_ehzw_&8tx;X_#M{e!z(qFIx@z}V_~u2_BBr|S&^QE;@>}39ciR>n5@v=pOiMf)
zb={ti_tBYc4hw&7{QfJNA)xUbPIhE3U%Eue&R#k^<#X>TFO+G!ijAkFA#I(9`(YHy
zlRy?0qL8ok9nM0E_13J(a{`4tO|Q(VQ=5vtQCpl!RnTg$B68S&=S9c?jH5l0kVJws
zZYHp`2tpw_67^ruZXaS@j_MkNWPmRm?CFL8X07&zsE=k~5QDa{?m#Lj5A(aw+2Z%(
zZNv{$Rf$=r+jl?@sh|*!OQ(>)v0BXc;rF)=UlP7n9Wq@~dTc8f7Z)5nJc^X@EQBkK
znTT$WTHb&My`X>q>Uq90naz=}*QrEzVwJgkV9_v1VvZS%e<bXTOT{I|9+_CaLp8`|
z=wi`26iZ*U*ISKe;5Q&y8Iq@r!J^Zer%Fvj#>Xh6v%pYr!&gTxfuQVe$m0yL4!=<Q
z@ZI%S%;gCsZsljO$GVqgTBEL_)?8X{xWnE*A>1!23cr4+>c_F}E9!-5v;mV`^_@*)
zert`+3R_ESTU*`3?S*hg4UgmFR)aJafKUi^e(>S47!6J)LkS%JcoQbMEVY}HlN0{*
zgBvZbC%6lVv9XS$dny(K8VWh@T@~_^_kWlcjaWp$ag*ML#EIdE{ocqSWb=1`sX+Ue
z2Ttj1xNhS!BybLPc#HvUL^6<e@E6I`;xdD$0jH40ViZ-CT`q;1IVS$I*HveyWi~sD
ziDEn_(!hVhur`u*TWQp#2p)0%kfu(PiDsNCAS<&pXG7emjSk6@_py)rMcnE)BGsw9
zlW;#Ou}|Uk9<#5T!0Wn25e62T;!{62$=IY``y#oq;S-gwH1;A}d(%A3D-&NYPpsk;
zBxG?26BqiOGDw!Dc$<CXT4<nLC==~$YDv7p&0cI1LXK^%eBJf~%@PS}Wf9Wi!tcvT
zWs>(+tE{iOMKRC1PA)8@2ohPp@)uqfJ8I=vd+iUB-O=b?JfIkVTQQklLAu@0>gShR
zA6?5_!&wa-kfX9Ha2A!m_F=<x{qq5n!RS+<_OfQyvQNphMSdca3F>LoDbw}ClyJXc
z-Dgag-wg8#7ujl{h`Fm=xznqttc-<l31u}XEj-InNnIhc@BZ;{e1|#2F1v@Xltd}K
z8>usyRJXJ{Ur#~a^v*KUGFrbXv*)--Q&@$l(=kFwj31|2{&qlu0ZHf2LdlqLVK0$R
z>^dGK(!}Q18YOC))iZV{=(}PKEu!W&>BH%tp4uJLA4k)=MkL|5mAJnD@S(tXN&p^;
z*Kfn0rL^}$^h39C2%KoI{pqWYHznIDpg2eKEQz!Scou#xL86sZ2tq>*A{ugX0e>?v
zUBPAEK(o07Q<8=bvzfIuhO{2)v>t_gNgGH9&k*IM8d+C!SgotUF%|Jg-5r5}fuFt#
z%duCQ%rwRk5s{FPEcRu^q&1<bA#8CCJHoT@gwdiqhcXfbDm6HsHjCKgCElILP<GVs
zJX{Nj+*;rMe7EKPmc~n?BRi>-k6Ht>8=Jn-`j3=x;}RZUKZGGPS`eY8KK?2BzDhQa
zNaNQLQLliQkoavW_vx7xnkuK<Jy-J}sAEw-dc+HeB&6Lj56zdPw-Mw*HnU+LsW}UK
zbYK;r2UF<mQYP7TjI{kmt-O-yUy5WI<lHf}2527*Vv)E&qMEBvWRL;oOg4hm5Gf!a
z&>>*p8c`Khay_!6m~(t*YpV;Wl^y~MgRo=(`P*&COGg!KjX>3q*<|`po&qDPKt3}X
zi0QP!wv2Vc?0~G{5nv5|)Of&tbDoMzcD=8O?9S|;VEIE$FA?(NxM5nwn#C~@k+T|^
zM*#^*5wB}-^&`rV^+iO~V+W`yI|+CF5Fr-o>}aWCXBHjKBx}smNYytRL%;0X${6;>
zW-%p+R0!B{0!i5jyyI{xaO6vDX8XygZde2ai1gbX=jZ0XtGKDtolKHqw*#CnijWo!
zDk>^+INlKaRbNj7w;)_mSngBlCpF(1XV9{w+xg<NcWR-q$qdz?ybGhi%C`h&fc|jr
zV~qU|fGSA~P$5!_A3r_1TeTG0-7K~!@f)Y$tan&-@qX$zDeXZvh95uqO!D*wx2GCE
z^%NKNrS)@GYA)=S`X}u0MA{ZS`MZ_1+9Nq3#~1m`^_F6Uy+>3gj0;;FRsGGhV`?&~
zGltO>Ln}{qhx`)&5i$?Rj>UnzB%R*$h0Oz~cF_odh2u}DZ;s&^o7qCXf;fp(2s};G
zur_%h$tPU^>Y_eO*5O!%0-c7gt~|<hdZijY^`P79!kv&CHix3$i1l1lA`z4NpPO*j
z#jrWoUuXoO#d-{#^!@&XaiW8h?w^tD0N5|Z;*y^c*vA!o)&nJyeQH#H@+P;oGfBck
z+oqm7wyC0~@?vHyUygY{I%&qJzu?flemXnkY--n?wY4KB!;z9U?z@e$wIG?aJgP--
zakwN8atR3#{LI7U6-dejz>y$ERltXl2@yN&fA8$_Tw}{~Jad32HeGU-K8q%^nAd+B
zHb4jJJsRvu>)i6R!M6<^p%AGo75P9$p_)brphSJFV|MZ2-w{4HdI1%UK7TTieA78m
z4NXm0HF=%kVq=Y9M68GBka5B|uPsGHUko>oLH&ejTUTp?^(Q*Xgsjdxl7Yt5EhP|J
z!5@0l7P`kWNiUgx8K}k<Fhx`M#YHB>M--Tu)I9oSOYA6%n`S%o1-DT99*;vgR8T1w
zVvM?a0JM<w(&b8nW**ArAFOh$BEWKAwkUi5{(U#NLBe50k3PA$>@?D3z4nibkB2)a
zJ?`xw_>Sh_&{iBSrhP0hnjb7kOZZf81{YB7(29Qo;NcbkHbz(YJ?a))&bN$ZaDnkz
z^pb%ESLMqB;@<1ShEHDiWa6ha*ED872HQ0Y`-o9lM}O|>>oYSU5=zworfzCF5`G6}
zhU3gs)ZCmV%ggO|hL23Qx4A;JhK~1wf^2w4RwGH%#KadU*a+C~F)21B-H<X6Ep@wL
ziwxN%G@0<A|Ha&=^jPQ+>ic9umLujNc6N5u`jx<LB|yk-)()eA;+T1_J|OV=AnnBb
zuJ)3(mGq)Q2cz}r!5r$yp<HjMEYod-qP#_AkShIqn?i0S6&}i8-ciqu;{?9llhEE#
zxki7I3e<0JJn<NMBiT&ztv22hvYGr0s=Ju%<`G`bF|=Au(y;2=5I@b0&A-(j6(Qgs
zxc1!k4Qh5Tg2Y*KOH@uAF=QU6NAG5xiZa!|((cO2q9HTrldqrAOj;N}ls~k$Q>8l`
z6Q7w)So(3K+8*UozWSrFZFX(V5ovYs=dq9Uq8$|d1}p8$0G5&k1Q$(uENPHFL^-$l
zaKrGwdiDD7V*r!o=K6X&$}#|eX#v0}G8HyE5ajj5zP!k_UVRai5i@l9GL_BB8`L`?
zPDuoA%2aqM{FeAgsX5q~C;Fr1^4MJ4z7N*Cbc@w5uK|?Qd?}*{^;^FTbcc&`D$h=j
z;%3?rM&xG-FWGapk^|Lfv>e4Hw5=t*l~2t-i9`wd2l`5Fu%VOPUCzFL)55#z!e0k>
zIL_Ep6#s06-F~M=N@4S;$m3w+@~`9lPqf_5o7Zp1s#t{QN{aoNpp-+ChNF)8fJ>#L
zBGc4#=3@jP(87F_i;do;6oJZC(wh%f*47JvRBXQ^#woE}So^jYq6+cZk~UUD;2@i!
z)X3>8Srz*ZR8qQ)fmCx`qW+v*TsyFif%t(TNOy_n6`w2t(6ZQ{^Wp16rgZz41Ah_v
z4)CO=m!%P@-`Vva;41Zto9MbSa@adWSzwApC#@f1rnI)0c9Ih|NQ#f(Z(?p(8qogT
z;lyiL85#8|2MWY*hT<Nq5yrc{9coZ-lUe*a(kSdDC9V}}bEBAFtbC(_J3J?&(X~vb
z0jjF`02-Cg=zYKPK%Dt`!6#N48W8~F(jOK<jfKd5TcQAf(`A-xy+A-46pYNi-a%J1
z90yNd!&uCVOVvE#drPqz0S-a46tDRnCrB#LM)s9|y0X}ERLHO~>!uvhv^YmMbU6LV
zw=k~4Y7VctE8WocD=sE#z~YI_TEhyK-d235FD!RH`7rIx&E22@yO)KDjNgFT^lj*q
zmk?7(xm>}7cEc3FFw&`%%RwYiMyd_Ai!q|bAoDJnvCic+cpL)#{76#N>MQ#D<to7C
zDuwuh*<koN+!SxX+X*5a&?_{dqq}NkoZQ7Cxx+3X<euxnJ7_W#N}cf*2`i~HTE+L(
zJv1YF(PU^wGMeu&_tQyDfNA1F4x@GGUHgwQ9$sjuVH;rtG?1!Ry{H|z-M?aiGEYDv
z;Q#BrUxT_ogu|Joi_}0lumEjuz2KYAt};rXCtVnY00>9<6o8#~@0wZgG&(ss<=MOg
zP-<7$0zz7lJm8SW{zJ{o*Q<#qVKW?O32$xkVzb1?>J5Dd6b`W$SI;bi>^>c%f12KO
zYp}*zDBzM5sAL}<mSTS39<pYK8Wtio8G<cc2;uSBOcKIb^r^A2v3rsw_+U4Cfp_FP
z3fp8;gl<Ybu3Y6{-!TK7wn0{J0Cv?e15X|s6QPq%>dTNHizod%@BrCpy5}c;vp2!k
z6h0duC!0+@@onnfJrrrRE*A5RakE>GeM3aPP;t`9K#`EUEg{wvpwdsJFDkQwyYp<l
zB%%pP{C7}SAWr762aLdAg>4baqJ!pgOEi}v1j~V6ab0D?rO?(o;srhFj}8xmt-ghX
zgm7}A{{VcEr6xO~N3KmUlTORl*Pn)~fbVeK$?Q?mw2PEUB9(ihpJHM$K|yg`e5U<Q
zCgaXeuf+}pm(lW~adYijw1pDF&dhCpmr;VUZN>8KaZM_6{(|sV^SxpMjgwO|!%?Ul
z7VPA$q;T)&1ZSj==7cv}H1ps}Vv<5jo`sny?{u!Dj+95wRx$IjRYm(caFH-?1mGVC
z*F;MyRGK{(8T6xPBVbhI>M`9mUp!c6zMN*#lsqzSfS#!;!o#&w;@=k&*qijapdtZ@
zCi#UwcL1YyrX}SvNmfE{pTN?f?3#Z_m6!(CrN09~%m~P8BV>e=6m3&}Rf`_lZpIAm
zE-8?$agnXGJYTw374EFV78_8?V;1yj@iS5@`igS6*3WWToUJ-Z{m2SQlw(-<zJGEm
zk^G{<QtpIos=<)}PNmrSn+Lf%;k^i&dN7&gAU2ZhAcobE;zAK<w;jlMr)!#UIw|&x
zc#n=kk`j%D$cI8J>ch-RkIJ;H&dC6rqN8NVDS!2iLh52s%pPq*DlrC6jMZCz-}c3f
zk$m;h{*({i&frPW@bEArA%)Od<xysu@C7w9@&KJ@=(1h{6G7ZEFXXbzdF_HiA}CT~
z9iMZ2zJU6Dr_@#rw6&|mcH{n2*R1|NSO`R{p0KJg#`&lb$(@l3H4%^*N@f4|6U5B4
zboGO~cc=g7CqkW=Q6tJ|Y6SUO|GmcL(fIl9A`t(7zuh#mL_-)!7(Wm63(egAtPVc=
zK_9xms@ID!&Nr)>H~xOqy_;wZ2qgqWngs@n?^h0W10;PYqRk^5GhQ&Sbg3vClqucP
ze>5U}9)&)6bU_7?{HpMq-pRq?GyTW{UaYeC@y6H0RLyDQn?ezt*gtgYwlxNy*oP^o
z-pNR-AEl2`bKJxEYbX&nT%7O_)IyCpQil%vmwv5rlONa$v-bI2>GtQPJ7a$$t6(O4
zxEfk`Qi{xpz+NnD6G+vR{%ni>H@ARrt0C7#h+jA9%zd^~iH%-6vg_Q5N0^|vbQjx_
zjOx#$k(>1P120QyUkM4`-d?F#qnT1)bY&nopCvlXQu>o>^qB(1UcHd=r$l4A#8^f<
zwc0+~if*MvDXEnDfM-h^ZKUq;o1W(j@5Wz^fuOe_)r=`Ts}xPu4vN-&WF6k<kSo;b
zh%d5ZZ)VgZx}uU_(z$qk`hN?eUcU=5J_h};Ojhik0bpxi@AZX{W*jeL;~}s79ZBi~
zSV5!j<;zMd%7q0F6INvNCNG*}1o6_b-_8?*R@O|Kn}1&@O>?OnG0tnufknf>QKa^P
zIqeQZ4!Ka2%S<+hnc0*>NYDAQq=ul|S={qi&l?B~3tL!aq^4zM?HeJJzK?~Y_dX~l
zguy)uGlO2!qO4-xM8#oL{)|3tx?k!t-lOs;2FhU(|3DK?zcqQ*Y!39wQ_fwyqS`rU
z>2>^?t%*tGz9|VrP~<`JRqUxTOe`;F%+dVS+lpxVYD~YgbBrCGF{xtcu|$A9q$x{a
zY-(x>aA}Ef*70&G`UKQifl*N_sNpd9sB)jS7_^0qsF*uktE*T(ILJ*qZgY=J@5;hJ
z;+1dkR9tvY!Q=^pn9LC*aUaPnM?app+9L)ufn_2uaBADGKPM;;DZUrEc5)VmmfE5T
z7-#d-o1rl=V$~SZrkBcbrk9q|+S|n~LnPjDf2~}l|K4cy=X-iHt$aGtI{YBGICQ{7
zzp;tVs>`6jra_M-#P*vndjUFqq)AOg#Vf^*P;PF!HWe*4Ha34!1_lOJ*$bA9fF-e^
z59Q+0D3BXbx;0G2z#~>j5K(S#Cs$+Z*VNcbt?;X<O$$1J8oS>6{py-rl@$Nlx_8RM
z#FVEMlm$ukphLpWrX~?0BD48RsZ5o1Xw|MdELaT1*<5c;c73LYx-fo!L$@?}_;<6#
zm2a(IaKgCk=1y5IbCHT(a4<?`!`XZ8by+9Qy<4e+ElVYtFV0_?aT8wYX<nRTTjQGL
zgS7L_^11tsd73c=eP`{jNkYhAz+o?zn^Ux<DmK8>*3aE;`nrjoJsr5+Cv<S>C}rs6
zqLS+Q=O(j&(W?H{j1(Jra_V3m%9Q|D_rY}y9&3%bXAUvyjN0vrdyO>JN&m|s4~m5g
zwsh6Y7JXOF@uTWXE!$4y;dQE#&IFgv{t>u!Fa91*(6k>?jn(;U5uGHgOUn?5<jg_r
zk~gRCUWRiWGHJAk_~J$fvR}qImTgKKkCkH4mP*a1mv78zn*7icTeK#XM#E87WQC~a
zU-@-%3-(kPRz0}RAcXYb@UXN5zgSTd84S5W+UW9%%---~_y9^m9%n@CZT=mt48b!S
zWg6}SzNJb<+M#fD?YAa5Q=Aoai&k?pi_DV>kZUhrcRgP&KCrY-!n1d#lTO9PlDXrr
zE{umOvsdo;c1}u7XAj}r#7!ThTS_x73f|l<EH0D&yA5hj;OR7S1~<zCZfprIg|f1{
z=lk<Ey8-|2cL>D2yRZ@GfBtv=HZZ9SF0=9TT?mCRT*x0BoCq3cj+wbRJT5#=y69Ym
zbc4)mtAQag42q)Oc6Rpa{tY8DsHywQ)%<r8{xzuQ%W(x|TmMh}c4lGq!Z;Dbs*jQ7
zbs0B4EDa6KhQ%2&vAltinQAY+5|){nSs{M_MySvJ&4Pi?`y9D`ei#u>FToO=ugw4b
z_W!QK|L1u?SyWkJQPEzg6y<{e^h$pYiZB<jwUU5>Vunh6f!;vp2-$yUsYyD*iItgI
z2C~1FIZJnL-_C~y5hx6cgW^gLuoyyCV?_Mi$wMHDkp<TQI3G}Ixt7Uoe@ID9EnHJm
z16d7UD8rVTFFgb0k_`Dg0YH$utQJpUS%&=gG>fLOa9LfvaACWdsV!NzFEdpl-mO_P
z?+tkXocC_1{#2~jV)!R4bS2U7@Qk(e{5QO<)VI~P+>mSXg;XzAVq#)R$$wVVXq5lX
z`ryiLT$755ib49nPy?Y9Ke#>46%)l_DeMoKISNp5G9}UlFDN=%Wl|r@J%0Y|quXzR
zDc?GjGh}hZqbfxd;u$gZjiHLqE=FnDLuJjHNU6K^c#(sElbGJZH4DR{jE@Mfh4FeM
zb{2*&>4r+&AUHXW_4f_QQx#N<H1!dwNuH0!3kLM?AuUL%-JL8k$K?QBkEl9-oAnz3
z0XJxkQ<jY3u+q>hDf>)K(Wfbu*)k#%o0>f{U0ila`j&jT%4Gg=Ij6?VOi9@dX*9Nh
zfmM1mnQJ^04~0rfN*Y^Q%q{OHzhaxQI#V6|^whGUq2V3FquB7YU0rMl>qf1nFi}By
z=~AYYCobJEU<=Ow6u&A^saZ05j@p=CKYu3M?ycI4J6~-yQxO;UWd}u-#T95@f<!~E
z9Bn4`&Zn($Q`e6<Lo1BdR)3~fd2c*q;89JpSnll1P{gs_43x~vy^Bkbp8UlKb9q?E
z>v8Rf@_W|-Lhm?k%%Bas*G<ixpD-Wb<XOf{`unRF4dD>s$60vcYrmNDM&H=8-1uhX
zcCu|+Q}>gm?pNmg<6!bTZSJ0teExL?b9<DLI~*6=uCA+0P8wD2ok3NezltfeWtMtp
zaC~Bd<W5x`qDP?s(~1zw9rA5lbgGp~s1qtB#w9b0mk(M+i(O(obmqg*Cv&jBPf1Dn
zCx}c{{9sfx{}C+3Lxj+)m(rvEFAtCw*E+mql|HGlv2b9!O4Xrjs7sTQge>vrhv-8+
zkJ3V|%)Ye>MSJ%As%jkewqRW;6@!}ktPEVwW_lZL0uF28thXmBs;v=O!{z=w_!_(G
z?Q|8k+5lR?XD7`PH4!aZQ2Ws82_<!h4%3>hF50wTi?bf%qSoN#&P@Ml8WJIoi}X}L
zT3ASki5NO$RnBW@YCzUr3b#$?XcynVdCvMQCM+xk_?m_Mv$70PUZYL{GNt;y(~?Y6
z%nR3vRY-PsgC>dBeI_?Nrt0fuK15x<IaxY$e({V#i+AKzW%-$<1ANskE6}m~g(YGd
zZyajxac2yah{$qom9G+7xpEU7d?~+C-Z_G!$)Nh(cFbD!{=fiwwu1SGG|HW?=~;))
zXpBN0Z`<1a>jZb2qE8MDCf>Iwl`7p4^lGoTm}O+S@b{2aBT|*CZ8eLBay|%hfpinS
zqU%+qZ!F6rNi@+8q7C=&FA~a&O&#q@Mkp|Q5L$lNBDYDDWo_v%5fo!k`oVA-nPp$^
zvsBcl>Q}TgskFhO-KyjjzJEq4^tjBdJ=CBBeTS;Dan8G!E^x5z$DVBc{EE!d_P8za
z;(SBE;TmC)^1T1KH(4W{rjY@8HdK@zOe-G;Eo@8B%0N1NInwA6mAs3pjSP$ie<0Dt
zq2j>pcHIas5qW@t_P`sOwMzb#S{<E**$kWgIv-E&TC1mFMf3vk6!wo8^5na&sAgCm
z?$SOTrn{?Eu$UMfqO;oKa#yWjaV9HCXK*qRFZ=!9l~8-tLq1M-!!m?kITb}H4gF)E
z%bWtDkJWR%qd8L)6<E54>}?yBeslMlUp(IFYG1m$x|U>!OWVDs+b3g^C?ldN>P=*|
zc5#3Z>twC562JRdYV+JKY(-Y)1ImB|LGDAjuiDx{vVTjWh$rPbU2zkTTU98{S1TBT
z)?j9U$VmE}0Fptv%Hae<24C&9A~r$yo+oaan~x9qjb*9T4E4qRqa~Nqkfq)jhpsKD
z%1)UMW%}?a#k0MpEDx@r<fRCf-jxi+z@y`|czhkXvfjh@TPzh!_CJ1)I2w{kU;jJP
zR2QARPkV1*vEe~YJhgo4phQ5KWq77QDr8fxD|i{1?Y<hTDWQXhlu`r&{@IhOO7bc}
zyF15Kq(Wya#zL<3R+YB0VWjqDYJSV*n&<l`DII}RzO&fV1=-GF<tyU)$AvZ-q}s2y
z$Z?43NVSc&$Oj1=B~4|sB#x4Q)2C)Gvfm}n(!YwGqOFxZ*qbUpVZ3;GKTS8vD8t*U
z>xcbOPma*Q&|lyT!pT#xK&b|!n$5zUugr%~D%qz9|LRpX=3X-MS~4M9@86|y`P7DO
z9UuG?=D!#v)l=IoTf`9E`8(I~F^Glk1_TCYmLJ=>9v!fAAy>LS{3Ih|;|jhN`C2%b
z6pzco!9QW@>sL{q^AtfU1qAcI_HC?%{25KQB8h0~f@y>{w{}fzXdkcd(AzDWL=n+E
z?jsl4+}kj=`J3KGh;xC1NBO(HDfOnKBcWKEI5OIlWbJfXT1YYqpIJ#lAxL&_`(!vK
za&lYlss~Y-cvAiOCE0Tajlt>mj^2<!;jtYZn_YK`@#(OmxPNz=@&i0+nVkY39a=|X
zLpqMzOWo9?cJ6#zSOmcXb$lZV1!nJD|4uQMon-v7c-+enWon^A$ES+aoB~QVZVrPL
zVk)?vSIkzYQ#(_iF?jOJ@(a>%Fo>xd4oFNGYpx8(OUcVG4itLv9-k=7ojVE1qOe+l
z^TDGP-4D})Px~Fk;p0-<?=jGxD9T$~+w7O>-ZitS%l544D!LIwT9ISoC90_zbaX5o
z@s@#3kxa;Dw>Lsb;lbZ=C99y`?9Y%+PXzWbg;42S2u}gU(-Nh*GX~AP%8^pOf={2e
zmor#p1-{B4>&Ycbk31ZN|1@{f@tpivE<a|BiO?-tsw~+ud&tHwYw3kN6D6d{k^KAC
zYGx3q37c=ta53ww?+7=xcFah9)av;Dc#eyr&7cp8)4fv|!G9*Tdz<lMWLsjZF~s$I
z7$-X1d$?0`kOW>mIAd*{Ir-aL(WZ<{B7H6U@}_%4;sNZp4mTIhl$?amkF+1`>GPx_
z;&Ro;nN`TI@&lzhgJPvSNn`oPf1b_^tb8O4i42HO8L~*73QV!l?P0&XzpvFR!jNnm
zfiy@Rt~f5Q*jZX)NYnM4Rj@t&JZQJ;by{YYXhA`a!?F9q#A8&h1ew8oq~M87D6{0|
zcW6JXi;j8Ci0^NRg{XebIid`y$d)d>UiZhRVjXnh+Rf7p+g+>nsJq~@wS3PL%GYQr
zNHTRz9aTwTiTR;C{ReOHnR)3RbX%J4tl|*ej&oi6*5#?{5bxu7%D<MbkY$O?kR^0<
z?7`R$5t7uiaq~|2A>`joM1=3qB4e{ub95Jafkr3Kr%wVo3CaMDqn)1d)sTfIYmDu!
z#;O9nCZxWH7-eaNfSOb0Dt^Y$(XIsSj5=f3u$+F+gnq%|%EX@TPs~2;O+!bm*0HpL
zm2r|^KhC;xzsrx5mhN_DmYyv->!@T!KS(=&-URrnF=Crxv6v?hbx~hIhD9_aWc(@j
z!%_)vr@v5D<sC8oas#6)t!}j!zL25@ncfuSjRuyKkDf_$s~{~y&XZ<0JRq0s8p_3a
z7iPZhd($%H{d0H(@Ov{MW%c*mykLF-Au65|02|#Zj<%a%bOvoigko>}M+{_NX(hA2
zV(u*nvFIVAQ~cv1)BY<#Q**IyfJe7CcZS_!F_1CObXuE*roBzJqa|E>dVHxT$)3Y9
zM4l|_jZjZkTA$X-W&uJEA6XNHQr?RfyR$9yG3PslWX)INz$#dV{I$MD7f?ydTpY!s
zrLYF6e4YB%_eJ`njxQWD6|;(dw1|Us#J3!K0>9XRDTCxHEa&9fJ{3gW-ZL?y)mh>&
zdB)s4=D9VQ@U1=So-7%W7#=nSdyafdsR9`hNt@EqV~0}i(*wMj=~>(|`-5Kn!`&=b
zFkih>tfR3M2Bm&pC#bckj)kW)*8e#HR>kfJw#~s(ihlwwGq1Vr(k#mVz`Sr#-QSFs
zi%$fbxMy$3QhTcMr1bCrkJ;uArp=bK?Tj`{*&bGL$M-uMrRvjKumC}%m|AyJ{vW2k
z0xHU`?HWZ<T0uZsq+42AMM1h7hR&h8Q6!|LyQFjIZjkP-A*H(;{=@VB@AH0Z=~{wo
zX6`xXioN%>@B0v{v&-6ju5P_`iwaA~N9@kMI&S{xUS^OTLP24?SjM~|$b~VXjyb=+
zwgrcdHr^8OohcHBNdKzJ&i1W<W<nF#mGa}$P|nWebd@SAsS-yL`L8+jIxQMFPoMRx
zlxhez-e_^;D^J2#{~4F|_&<ox-FguM!x`MRtI)$S*ITAZO>_008yGH{jz{B`N?o+@
z7ss)_Bv*Eu`M2)xuBpLxk5vr{*vndIIP$T`7Lr~F8qAuR;u8EGEZT#KZ+WNKOMn)$
z^$aOT`A03mXkq?5q(aIPrk3q!G@t}lSis#aG9k|q7R&ECzTRyV6_!*;82@Fn;jTep
zU5oMcr-p`c3JMi2_P~)qex~6r=Ro%RW*vLa9c0F9?RvE_*m~C1KN*Izw6ymaA(iK9
zr%I3=0H^VT!{KFKi;-f7!T_H-BUaJc+%NKsP1}a!s0Wkff5cDMFxh}BM9~`r>*UMy
zDD$#h*9IOnmr2P#t#HrdDzy_an8lUFkRoe>#MN@BbNtbD3SQ1~9=QOws!&H`R?v<a
zzq<*ZyJy?@!waCVU*)9uLSRe8vvjolngC&ap~c6?%lQWmeWU+Jk$ed3(G(A+vhEBM
zE{#kF9gbc%3xG@eYY){fom{rYEt^JLu&L=vM)KA(VcA<wy1xmKo|<+nLSOByDz4f3
zv0MJGx;l##8qvGIR&c)GnpKAf-@4>IAgTm)&e7h6+S~$YQ$JP2{Jv)m5@*O*vR;Xc
z(N&*V;XnO*QSCu5*WK?=EG#czfKN77xQsD-N{^oDRGg^TiA*0Iom`EP(xtWLbz4+z
z->M3Vg?F59XUdT7+k_r|&34#&MUqvJNE94*<C8C5P>PZKK5fwRUj92osbQf>IS!ym
z!w6AQ%nkDV=>klndyTQ{*AK}dbNY{R3rMS^dc8vU$w{Ur9pAu45wQVB0*C9R*y`G9
zO<q0S?j#5FVHfQC32L)M(@}z#bXalt_eNtF><ye1XLeTUK`Gn*erB%F)|vfDZwwP{
zXtywdzSnScZXBwi@lVf9Eqz#M)LlLs1;6E$@e&8daZI$FgZC{E$d+PLksK>)x#7A6
zo|1~a<fVHu#pYF3aBsuoQwe2u(>Ki0WxCRo5O#BdJB>PsT#Md%&_~I13Q^Aj&2>A;
zhcEIC^<g@rv7xQ;ad6H-ry`LgA3V0{JGQk{sKFDTTi@B}bR%og8{nTr#)t-0@os9R
zh^DZI-ff=ydJ2bseVM6fe(>RqwDtcPTu{rSy2MPJFnZokdFF4bAIM&N{|!thH5C)~
zrN}jWs*aYJdr&fK%P_H7611pqnpe-?e^jQDNRWMC&HDut^6*aY(4-yl5Ss0mBO-9h
zj4WG3MvsqiUJ~R1+jgF;lX=WQ#IMBPxu|$1=fmk@V&cjTY89{G^>Mx@J517;E4OeO
zK|uNe?^u{D7!UJ~II2(^&wvmIld!6o6LK(2s*v5?g=hRVQ}?gx%rr7?^MCTkK)-9x
zie_$58qkF3sR1%>-fZnYgnOHDZH2q+AMn44ME_t$f@*EbikDBWFE?d`eM^szR(DN`
z4&M8OsvmqgN_?m`-?qMe!+ZDMuPsfrS83zNw4ll9fF^Ja{|Pa{nf}lQs}Xznt27wk
zIqS(``91l5nLg>OgxsMjMVX%;52ydg<RD+J;pvu=Q!=W-{o#Af(>c!y!850l>oVPI
z40n!wX25D|%e1uIS+8W{`qwcBwkGK=eqImA`+u1PtU_|OQ9VMJ(`5HBT<~h1sUXYO
z(Kz3ArXqaR?rY<=!icJ#lA}FzV(d)Fui{wYw|s-X`Q_zIOd^A*DWoYmzbAiW%Wr%n
z?ZRnSOhUnx`qOUf`tG0q`_+nt4i9U0Ppd%0F^S4mH>F$&PbmLrMsRkCeU*@$W*`z`
za-G&jz`u#@ezGzc&$DLo1y~M?oTW0e@@8}HjqeY=AjKZIfma*_Fe<VZl!H<_0djd3
z>q9kc!fgAx^PE$}`|Rr?qEu<&P`gWaho}eIy1;aiPcvl>$VSW<vNHHku=IBCxD`E#
zd49JSn7MOXHa4nPe+`^>7mq17_8739WjD=H`}pzdddiy8(14zWSKV>oQixER;%oR8
z(nY?szB|Wr0B@*>CB-L4$UDoq?=nb|s=ksI&gNLcn&kRF37t%)wH;NejTb8R_PD>x
z`>F1FK5}!)a%H~o5fsQ6l+dyp<AM?Mnmqhh8)KyGK^c~vXKIc12_CbKL%HA1CmWto
zh~DxzSSgx31oSsC2kZLdr+I1M8*bF%b<hFq0wQgzqL*3z-WUBZs0qBBR7?<NrdxGX
z3=*Y%>f~W%X~DF#2A4>a<Rp=hob8{(J$>D>u_>^-*;_0)yfU+1Or_t2fl<@)u0biq
zawhAcXaO!ZbadX?n;uYMzQMZ8Vm?<V9nN0<MZRHb1&ad}oz_94u`O-7?rbYF(-Z1x
z6W|i$As2Dargo0Dp}aL^h>3~5XB{VSd+NyFU=_Ev$Dax!-`+o`P}$X9e!uJL>Rh>t
z%tX>G5u#Em*Be44zn{xyr=E{1R<&wsb>91Tdv&$dRm&be+yo=J()Lm!<}Fg}4A~y3
zeThS`e#^lUo{|5Xo&<}qKbXIEJIb<gl$x;8SRn=jXX_8R4Zjdpx%11fJ7c>}Qw>@1
zAM!&dG%luz5ZMi~nBZi&SM00kFe34c@uemg^h0_VR-2~iL;2RSAjzn@I+3w}B=31Z
z>y>t=oadcK%F67gbEqTZBGN7F$;<ENXHT{LI4OPW<XX1IkP)voViJ&OyL!c;4to=f
zu%^v1cl(MoSbb5j>v^{+=XM|z=~ZDgqyM!2h$LgYX@Bu7dyYvYF#bL8cXMv5gG5_C
z_g)#0j`6kbG2ym0?0MqP7#sC+t|DLTiG_=f0OrZDUrFYYyPnz8*FPHBKT9FXJZGJ6
z>p1m+kDt%Zbzwd{qZ5xo)&b&C(|KWiF3>M%rGi_giY+X#Jo7lecmkcABSXaTPo~jE
z1rD^wy-{eyHc**{l3L^(!oB!=Q}dre{`jLu%-Zh&Q9o}yPL`segsRjj#lK4pEil+X
zhlW3Nzy^3#wwbeY?P&#9RFwRYu(nT;r%r(!8{4=*LG;=@-`5H^_Z|hkCf`$z^{j0J
z7T1l=Rr*r^<a_NP2I>zn4Uv`URh~p2GcO#3TcrJJ8vng-Q7wi@DAKcJm9xGgopgNB
zRs7E9>9g)7D`Wi_43|2~K~{4Cqj2^t^>WCTb1qmmubZ1Oy3&tvrUgt?!31Q0u$sRQ
zOI>PWmMMZm)@&z91k<iPlc+^qj@HtS&d-VpSo1Tu_DHh;wr^;n64o-Gh7{Qfs?~ke
zvY0Odx$Ux9#7md`;UT*|WWHH^incarzg^)pYRG|(KH;Z)!esqrqb6u$4l>DZl95g=
zZMvMOCB|>j`~8PnrF{Ye3CY!+Z+_2h*gimW+pEcN3eiwE_L4+Jr4SGVOnw=DJeA{%
zhO(<XlV+@&z0hZLW1Z@os;VxI9XoQNNfl;3y3#mi4je^}=Qfj<W>v<T^Qp<86Q5j?
z9)qn4b5<`@W4~jJ<L6^{@3he5y)_~h7Z+TA&?gqm;AwXY{c?sD$xN0DN-&JN?Z=85
z_UyUXX_lRQl$a5sn74N^4;cpd^g16kJUYK~jd_ACFuE78)gZhS`6Pn0|BJzp*h~gs
z;TbV58fvEb-nJ~FF1ts)QlT8mKn!nRNxr=a^oo8$Jd<N>CF5i2o+0o0<>Gqtvz$$w
z+9!o9Y|(C>xNtg4X=#2>)k&1#m&?0~<(HEQqggR7a~WD|V&RZIdy_b&y}Q|_CSkmU
z5Xs-=i@VmKHkSWAkgd2nf!-Rm$Q8aP$WOfSszk>SE!o_BP?SzD3~Nbl%)i?l9do>^
zEn924taaW0MyLtrP!nu8`u*6=jostarQ&jju%V7exK4_a_WtDD1yrX(<ZSl6{R-2=
zP30^7Hr(;&XhWnc5cyEfLbw~|!*(OA{-J_?`>wzC*LbNz@~BG9PII&aha}q|SjHX&
zQ(i)&zLMyDjzcgZ4zJ;AC2@Ghta??>0pEPGA-egn0``RY*t0Uv5zIkB^`(-;)3NLn
z@o*$|)6{o@3n-zs>iKIf4#PH-*_mKYp|$4evjYS<wG#U})2pkE0O2s4WNMz@mGQnv
zP|)UPJo%)6NKVfe{SwGCYS)c-pMG>U{;uG4gv#|%Nd`<)<i0T;?CYmTai#2@>=lQh
z)rqJ3%`Fy%=LhB22}4giNjIjuwH_$&$xS$pi$^y~oFvs#XIENoUXe4lyJB1MeTd7-
zolXE6s;$=W!*~CNLJneGe=;l`=-TY5vid$s`^mYEKU@O*11AAt4R~4)obM)aabR7>
zxaVqyvcdnw7h9xr?h7uCy&q7kNPknjN31%SYi)XcnWxw<V=fq%)8x3rP`PZq{avW1
zYS!NQww5)fbW>CH)Ivtu6)U2`dZ96pX)!8ehO9;Ha>;e3*f9liZxJ7DG5>q44~^1b
z9h{im-K#4L)REc>X-j_l-XWC2=%S+d@%;R0J*@{K&F*@M(y!k1QDk`~DmJ>Pcv1U{
z@8m@DK$4x_;GQYGc-iyk!`DNoCI#i}O-Z(t=tHOG<yTsI-H9C$0C(6q;j_QFU0u@u
zs`_49?yBZGN2&7f37Tk;>rQ2g#Ml;b9r)^<U<|_}-4go;L}~<xRQBUP>vNjEWcFrQ
znc&Bht>QYkCd$L-q=pyIAy<Ff!4S#ix9n}T<1w?(S=M_+N0gRvGtu^U>Bu=T%BYH$
zt9kQ!v%O90uZG^uF^5eHpN}37i!a>5@vek^TcgOys`t&HMa=DEc=p!XC->locSgX&
zpm+K%W(O2$fbz%Z6rts}Lpn)c#lpvRWj`DLWHO(Yo78gfN79^-R2nRGrx*6d-Lddw
z=)W8<_R)J4;(W$O_8`mm{(>)Anc`S3@sZuFst@yk1$>~Uprvc6awZUAhzxfgsdMJQ
zo)r<~_!TwO_N_d<zNBU0kQC9pn9>|6buO806#}=xh{%s>a7up;Ue-z<`8&PqqPK+^
zU#XvqK@=(v`Xx~~p9IzA^~J`B6649-ZdHy5be(H7V7g41Et+Yx{*=(};&XlY$SBUv
zem0VhPVH}b5%k@0q2c}nZvMeMV%}50t;%h#F*1Q6#=^0S;$g&r4xOABg7BX#@LJLz
zpAY}hKMhHW2NiRz(`;zz;z_2yK27^VQ_hz*$p)94wtJ&}ru6jMTBMMBBdR?n3bGKS
z=f4*B_GTU{id~PXZ)JXk)7XE<wG6DQdlvjF)!K2gyX#s93-TsHlofg8>}FysX3qU@
zj|86U0K9YX{G+t|izojaA^+r~@smTC&d2I!J*zeX*RK45To!5NzaI}3wA#I6%aqG~
z%f$bq`ULHc$L1BOGM<3*ku}xnrlL_3BmW$jnJwGd5Uiq5QXySUF7e0zdh&sqv70zo
zAs{mj-zPIuLt^kv^~+POP=#RZ4fQ6;B`K|%9`Y17DI2A;!Y2IgjH$SISsBR3I1MRd
zJ2ZgJu#c4r)+Uu9`fe~1`?J{wNvQ7V2zUIicGM)^4M)}wF4oA(wl-l1aP`T#Z$(L=
zc&xdRdZM&6<E9PU2YHwG(8-cy<Z4S!^Q9j+`Lq$embd*QUC-Hs5KF~-X&4v{?tR2}
znyma0ESW1AV*RQDZGWueCdGFoG~EuIv5{5X8B#<n>FP*9RJmW4SbuKsI6Up>N8870
zj~V2Xq!0C-6p-ra(Y?9BL=N?DsEAjk3tKd^5H`v5dRvp2Kuha(HObRBT^{EgD$uGv
zO2y@r=$TIM?_Ynoy}wgubvUsgfyRDLQ&kXDyz#pAx88*t9I4}UX8^A(<w|<O8TY>%
za>4Zi8S>FsH<!CcLgV7)I&7g?hHf=3WH)xCC#A@u=(UphV?`OVb#1q|p57uhGy?u6
zq@Eq9Vd<Bq1Ea4uruh2%reZ%PJ)Xw52{Gr=i=9zx8V{45=Jqz!Us~N!*hUg>csx^l
zJ@-^e1TxCoqMjvFys5^H0F!vFWpv}HUdeJkX^=B!n%LLxRAT!^@5bF&2c{}pfP{sL
zGL3`XiHX<h-_NYwBa(73a#7;OZ&W$Z*c5yF3W`a*^ug`z<{!#xmDp{2DQJ2TqBhjz
zyF_n8X+?3WV;w?>uShW$RdSfeJcsVqiiRl<i}Ta+)1&rYRYxqPOgWaTA3i(0FwwQA
zH9W^=oepTHhpoeXuMjfTBvpiQ)r&-qk29aYU0F6c9wXxPI3OXfLz;AjTr+9>4U-ng
zPR)ok#Ko<DJ@$S0PT*o@X0BGm`ts>cP{_(Ps?H9#<9F;Ba@E@j!l1qRiuo0y`d1xO
zn7DGXs{Xm(9}PN~4ths>*x6XuMzIgl((7rNJ#O-btbt41XVmKcJ|uJIdG&JS{=gTi
zE}gVg!rYjZQBNcdrH5?HeV6r^cbZmNs1+Hmxio9S@5RJxn{wGg6KQ*UaeZ5bKbE`l
zBf0U-bFcsW8TxP!_k4uf%KuAQqpz83e(mOz@JU^*mGhZIBjH>qgv6pMybfWaLs-IW
zv$KeHWs1`T&4G&Qrr7-1Ak$^z071iGC_Z}g?+CI;`bhNTbONXS4HHQ%z0SEjeAlYM
zWDhL=&#k0OvU7d?mIv<<1T~&hH)K3f=H;y|A4v+jvtPG>+#Ar+Rm2<dR;+Kal{uxW
z8>uN#I({renHil4xVPW>FeM+#+T_S3(^i#Gb$A66w?ErAgZ_vJ=D<OrBu(bYp1k}U
z9d*1qjf&8haQL;*dN7puY4Z)XQO;PpQh`|ET7zD^>q+a>uzd%nyAjFtfL6N-mEQEv
zkBPC|KKPFcEh!xz>0==$X(nFeSYj6SI|sdNgecIx3QB4E86his8b6sf^sb{%{p&|j
zTA0J;=P@=m=L&J^o4oM4)EW^1&WZY0uew&VjgAbf-URga#1$WqJEZVx!K?=ss$06w
z1yX;PhPTZ3*Kn@02<%QqXe+PtDt(?!h(G!|t3;H*5{HxSt81#=JY4+zs^2{yvT{Rc
zE{NUw7gI1Tqdw8m^$GUj2^s_qrM-@>1~uOGuccAjwkGdO4p&@eq^IFpa;-V%cvkFu
z%O9n*{w<+HKfWHIe8QH^AMd-_p|{A&U>)m>KpdrBG%!~*s!J0{PE6!@N^Q%gv?JKX
z%<H%vL^?+M>pL-`e>eT&rMn+tecWWYZOjxQVQ6*)_F!;PgReQ^;M&RV5QdM~pA+*s
z+Gu4?<HQJ2S_ZycXP^7*n3PF;S8k?cJkOh_90w*%$qe{oc(y`9N#;TNtn-^BeOoMT
z2p1N|oDPZH`DOiCKF7Zhn*2mc_qKvvQao9gfNJ_5F952s>FmH%W6aGFibQabll5Gz
znqH^kxME0O&r)z3H32(Db7%H|H7o7YIE5Z;p=aAcn&ns0A>KZEU<F%5sNRvM+9xHH
za>vJ&>a@}%8!zBSsr46Et;?#Y<&{b58sd%$T%l!({W%x9ItG^+J0W{K(f*<1w*dUo
z{NfFGiWZx|ZFp=sdy+ZmK1dCROkz<1A~Th`z3s#A6B0Jf$RA}@n6xa3Q0wUEcWvh6
z*Eg8e3~=5HbA`*to?a^FjGIP>616^SsMRT00k{Gb`bMImgg-y7Qf27kj+U}1jEuI;
z$p4#rZ}>Dmdnvz&?;7}JyTkAKNJ|~66ihOadQ}ag%IviuR_^E4`qlg{`XYDocOZ?9
zxFBug10fZ%ld)y%Z*Y2Qm5=hiGk;<_*ej8LO-N8`G0z+O8h_KQ%zeqtjDg{2Uysg3
zEn@~=%-FWuSw{z($EegRIvn}yQyn`lIBZSuY7;L<fzA-Ad@YJ8X^TU+LbUM4xyeX*
zta-<>aU7p8CEA0Y>~5?Gv7p{96QcL3XH{alG_XyP!@55xn1+cS#&I`D-ua9E)mI8E
zkD6C4^v^=ap_DMOLc=Bh7XkwA7`dOl<T957o4CdZwA3Ii*`Bax+M7Q=qSIxPMSQ)m
z_$ug9&6Tdfw4Mu#*$7Ih`gE)zFqz9=hKRVukI`J<Q23B%DB|MI`=L=hdg~0A$>B#f
zUYbdl&Ql0z1Tv;+>1E{f!st)6yVnZssS5AS=OLAe%cVB+8xwJx=1>cSoZfpLdu<18
zgqkm37*uN3Ncv^|&5>%n|E#?q^#1W;nMXr0=S0%`@hm7Co?!)4(nAz^!4h7?RvFho
z%qx%>8d%7f&!m+(8_+au&4!y9eS~4UXlBYIbU)wstaDIX*SI|e*cD_@?|Ihs4Rs~D
z92c}LbXzaW(Ri5eBS@Ki<<)YB2uQVLy4YMXSUpxS0{+^#JS{ia<+a=*B4_9ErAPKx
zd5wRhdY^1xBrv`fY>{+WUir2z^_Su^43l7x$ZqArwxMd%v%h6>!REh+sM>kUN-qK>
ztdZ+Zk)3>gwIc1%*-jp2yeBUu%OdW$UuZADe0#u#5Ly8ZPRd1Y(g;Xh&$HhNl>DIB
z%g+-Q`RTrb@baY3nIQ<xl2TSMS8UPrcm^h(t3b1{zuA(ZWWO+2hcA}NKr;30n|b!2
zjK}PnSyiBv<jP>1NC_8*flkUEI?1!e+hk2EPA<Kb!w3ygi7|q7ncR!(E*1n_%q{u#
zEqQa{OJ}yePE;Sc?m)%!rKUw24Zhhy@1ed{FGp9q;r!4q?k?JTx_x<wem@7hjF<6I
z{(o0`^ZYzuN?-Ct{D)75R52F9D|&kUG@hgl*UpsH<n(WgfJ4hfU`Tca=grVy%rDYL
zosMiR8&=kUZgmG8OH(mVX-B>S29`)MQChmS&vGdKK3@>Zt!(vn>nulx2dn07k8Euz
zb#}#E8jg4-!^7iTH{S0P^M1KsBe|GS(=*WjY-oZXGs5<~zw7Y_y0C+$Ve-qfOQY=_
z&%-Q6gg!Lm;D%fi;^7(mUVT0gua>0T-IjG{@1O_G1T4xC;nm5B;I`Ih1N;I>cxS|G
z=La+nb3q23{VHlRvm=FZ^h|4pk+iQUUcIt@Pv-wSFC)7|qQgH<K1YUbv%!U5T`Fm9
z+<(P1NS-Q2UPjJK;_Xs|2fuTj>pQlp{Z%(LBt0&R&hELOyMDNJRp|#W@5p7{5bCU7
zav7=>@QIe>aXsH9D6~Jm3mVzAN};d6yt$*GSCFJgNejWi7~8IW^n{0ZWMT9NIkM<Y
zqY>#KH21P@_XHo*K4}NzABx2JjZi%c<F1`%2RFMd1^_S8B#+lPbd^Zod&_11u1m<D
zC%pc?W?Rn1ncSe5FmQ`!#+{oxeIwENjYr&QtYe}*wsW>Zr~6W-GDV<Qd<DuZ^!9)C
zj_;qGYYuzY-|@N4z2QXdyINsp1MxiPghogpb$%*>2o#!hxQBs3bh*Xm;G3uiGCMIS
zNOz?ofn%BH>hh?sYl%-l|Ans$g{K)qpTEW#>UL<-Vt-~+M0ir${L}@3H4-bnrh0|`
zxVmix6-5O%i8VerWpf@qd2(0y%H{YhEaG$fcp_eNnz)W*&~e+)qx#7*)V4|URTruG
z!uve{xDW08K>Usr*eT|YzmJczS?f_0>*;0TVyPj&tLvXd*w~v#<Hg3_k&t*VV2hFf
z6FFY_WA#aHq<rEY-NAkh;9{o?^L%wefAlkpfeAtuxB4?n57*sH-epBP9DHZzSc2C>
zuo@A>A_Yluw}$uOzuG&nsVk_QF#UX$|8VS)1RtC(r^_30+tCI&W=j`ZyTON7Bw3}T
zLk(33@WbR)erlvOe;xE*7wk?e=1sA54?a0I1j~)#<Lmzf_Fr7rD%jq1uhrmXMM_~{
zqEu8_US_T!h?VM-gbqMxhQ7gM+~QM7$@yV;9c!(tvQ1#X^S}^F4z6cDa$neZd1qXl
zm)Eoxa-0<tMyj~vug<8Ns=e(+LaoUuedb!<#`AK&92-dq)AS1COf{tY@X<f3$Wr{e
zLz9<O9h!lKyn2LE*cnmAPACHR2-RR@c<yS7gs@$tVPvWC`0&yvmJom0NUu%2<@?3d
zoHHhP!JR{;K4o@zd_-WAdKqtsUq}1cYmvtfh_gM~<8Nft%j7soHEU$L3Ib(R+s8yZ
z)+O&?EbF+6?sH#7bFCT3*ibnZP0YtrT?S_qLL-vZ*;>#g&)t+mBjkFiAi3P}a`?gS
zxM-PjZe|6@gMG^`a;&@c4Fuy0f*)<N*5WITf4fp}s$u`jch>*mJCC0#CWNXcyymK9
zp3M`mGcQ*n<P{f3vfCNj;2!(a(VEDeF3c{fW#f~T3<J-(C`yue+cGotv)*PGd4B7&
zjrqB;X+pGwxqMS_Zr;AEeM(g?EyI)Mj?b>Nh_3STNKO-$e+nf}%fPAID=PM_YAG&*
zj`;#9Gn^s$0@C@yAqE!uKff!eZI@|kC?S25Bu{yIaSx4&jXO>GdaxD~sjd44YFFBH
zRf{ubEA-|6Q|A{?mRhipAxmk$mGW!iBijL{3@<CCrJ-#Z?jRslR`z{j-}<N0I)8Kz
z$1g9cayrw1Bg7yw@(AuD!<1<hrl=N>rVx<t!wpi?xxgJz6Sc#5)!&<K-5qQ5(i)g_
zQ32-JLKoW`EZ|*1|L+{0iB}>asekXmUdTtF#MInA^xQeqYiFdCpc_23N*AMp8Y@NX
ztCA`{U<^(ws-Wx;3(oSwLN#(JVD*nx8R>r#7Zf7wp68@~8c4hNT(tT(Ojgbd5TGL5
zwG%7H%(6f>x$Yz%uK%nb)DWi1n3QRd=$p6KrrK+}zrKd@!9jJUv%7$xQ>~XycZhF+
zcA6W+%Ziyr*7^I(RU8_HJhc>aLR3^Nwz4i$FWVkE#;2*7b*?{V6cv<g7;ae8SXd~G
z49UX2TG`sAMx>Cs&pgvMGoIS0fun{@fTuY`%P_#&Tfg3OD+$`pB%A+lotG%E>RW|A
zW-68KZLkPHuDL)Tfd&%}l^~J|a7Gjrs>3^{$NRHQg>nx13e-J$lM;JI>oBnrv}DKy
z*vSLoEQj@Oi!_M!d@9_gG-p{<fZ#G28IVVPJ+QK=w>*Yn)_iI?UbyXw|KXRbKnEof
z#I=soa@_4`nF+8=PG~#1bKf3^{iwhJs)o0b;Sw}oG?G+WLZ_wsg%ox6#7t3AI^zEM
zR_kj|EsEw<hW_na+zPX?wOyNQepB=h5|Brz(!Ry3F6>csQ4<bKxNG$Ub79=XvwFI3
zgo+ew7^(5}goHG9cTC&QEl8VZghYOJncdLuEh#2U*q`LrSf}JC5zF~|dwab-qNgLM
z*aGK2n?rip)>oE4IY}@q#i^nlo{hOC{>-zk*GQ}&Id`xod&?Q}Gl+qz@?_LCQP-Jr
z*!d|yjcythq4rffUY{P6p5C*aLVL=x%v{<s5ww*;6Zzt*(lJ+$bFZp=)Ztu{&-#YL
zZ4Ih$bW_>eK1EN+f7A7aw;xCe6if|ah=9DnmA8DDGQxAKawBgT7vqjuF)JrkU3t(@
zkc5M9r}caG#J030J}wuEz?7uINnf4w*5;7!Un#L3z>_kav*94wgl0_$LGe3Q{pwIW
z{<#Dozl4BrO-udt<Iq1*LTpuA<889w-9YKf9{{0I<d+9Q9+{W0j(TS5%kz4!ApMwk
ziV3%oT&5qI4qFm{Tue`En*cl^N*j~6Cu{hTj7{)5+>*<3^@2-lqNZj1CsWG<m7cZi
zc-K0auzpT#Xg<b)5U;Mv+B4|qAaypC5HBYq+cLhR0j|)Nn?4~|-GObYjyqo%Q|}sG
za(A|<up&nFHNhPIH)1ZmOScb}9<{Fr>O`c|WO9pHI1Q^|c$x_{i$Ft)`}EQ6BiXZM
zEbfTJ9(56^BlRf*&LGk+HnK0j*l@SiJu%VjzieOp$amg<?roqSCL{ARJDHyBxQG6I
zSftkzr}MB)HNE=8B!5Y1BWXLn(b)-DeLH;2xCps9LdefJ7@uG+jxk%RY**MU9jkse
zzXKnaDC6{El-#!zz>CU-AcK!7doGqL)q+fqYUKR=()AilKkf6lhNz}^egV|FgB2gN
z!;RTFVlHvi4`tyu#=5M2%MmK|Ca9h}@WFV;`J7We67RH+N!(T}kw8Zxj9P>{$RBeP
zD4jaL{5Kd#JTOHZBrXO6#0RyT=c>N;^iVYqCo?}g$!eOo$}qn)l+jYsZ0x!2ls-NV
z(bSNZGVJt2B?(Z@+AU<jpR2(J@=XT3lmnnM-8WIJRZt_y?uaNfX^gAxsWm@!sAEX7
zzM}-6^v;}P{!a6lhx^_+?U?E0^s$7>0U02R158qG+^LR*$QhJBB5z)j{CP$_U<C#q
z^+tRH{Wwi}S?QdI+Y!_UD`!b&)h-eA(={NhS9Em-vfW)-GbK^=!{bdFL$6QuEhYFY
zNw{0SSATXXP!n+&r@yqNmLB5?sr!M*k2qsRcZ8xhVYy=U2@9;-{@GTC_{Z;RT2Rky
z*LPv6-1YW<pMbU7C^O_wR*LsFH$VSM6*1h`k{r21B!F4dEG_hG%w?~ZLBKIgEDD9t
z*0umj8~6^|^#Qj)tvAMqdCaA7tCX2Dxi|>W1N7LTe7}iU7I@uJk91Qp)e+tP$@7i|
zy2I%gKwD!7{n^^p{LPxb78;B{Q-hr)EUeMrZE)THy5s7a(}@W1?{y<4gm;A``CXpy
z3O~HYd*lkW{s`ikqmP3{YFzcM+|}f{fBrD=Sw~03z}^|ofAa9Se40!kP~Ku&k4JT~
z)jiuAeeCK)VO!_WF)VQ{=Aem%f^tc}#bemf5wtaNO429ye@GocCA{Gcd)n_(){cIy
z&vHtf*y>l-ZXJ<o;dKN^a<l;B_Qt8oo2+PeE)7}Tox?+O#<#BDZ_hBgm?5Q#hEK)5
zE;HJCk9)G0Ws6-9eO74}|32Qwy*3@`_wkW+Wyw;v&Wy6$yruOfzJ%Hd3)pd>AaUPp
zP$cCe2J5pX?~WzQvt=(EL4`my;@5K4Tz^dx5^{oS7AjXr^_DFG#D{lZe9uWGBUDmF
z?AuZ!Br%>NN=E?FzKSP90c(fF(IY;FF6i&Ibs9dhdQXNd!X}R)!1^8NK-&eX1dGOl
zA>DcpWZ=({Sdsqv(A>1Y$*rk0iG}KWOyuIX%P1n)rp(-O34xn&Zmt4o$OqSR%Ftj}
z$14Q}3CWu9q}~`ZJOVx`3ukT^F&{>ApkC)#?aX+0hi^<raOS7nzd-$w<vd`^mdkfO
zY?c+H$9brg78qx0D;aXMmJ)GKn&*V5%fSgj;5-H)fTddd+PyJTVl&y!7BgiaH+R$k
zWdRMN2Sk{$<en1LP^Sgf_N>(H?+YNY-CysC0Ii10DdhZ<EwzYY=xgDU`tX=D(zLXN
z0F<bJ`(0iMAkO4tcM953()^p!-ZKE$9yBv(R7B@oJh^hNEz>=b=|kJDN}U4=A#p_K
zvYACk&+RjyFYl2ImV3}eiBWu$pfk9w4m&xs%2(z^K2WN9v*xfHL;y7RKQdHxG4h2~
zvipPCa%tM$FuW#o(x2A+>#IOn1B+YC!LkNu>VnzKZ?<{({G*|{+1-fF4T)0tKu_1t
zrVe!KoV<ZNb7Fe&Ui#H<R4*Hxmub8-x53ddbYVbM=328e^CReDrtA37Sz)}ihk*rP
zYrM&<gDow88Ds@Pqqp1OXf3HLO6=Fy=cY?q*Yz24Gp?@Wk8l&8>)8JtcL|Nql5H7d
ztg{^dNwOe+oi}4@DVZRF>d1BWhw7l`dpeD2uBsbXrdmItpLFqxA;Ub=K>hg!d(z>)
zAP>5u$g^;iz}RYM>YCM=?IqK)y)E>4UIzl?gErJboz!$bAyH>;X}6l!AT6bZuGuo}
z16@Md8qCET2}dx&(qn6R2&wwi&ztXOUb2y9)<lr2TeKexghk|R#r;&yQK+xJrvZh<
z7(Vx-M!p>8h4=3!kx0F5q33(yAE$a#?oMs!i&XHM`$?P2Sa*F=LgrIJv`x%r*Iwlb
z-Y{!s{>yQ$obSqBh?}GJ9^LR+#+}imN{q9}?p3#98_7}W;_qi#_3ft_`UOG3DxA9S
zMvxls_RaPBm%qm=7NIH?eo++vR9s!3*c3waRD1K_<J4rSsG7M1I@$x20y&On@!6J~
zJ9A=+d%+E}nU~*dJ87-~9)swzVy-8tV?wKCIi&%e+kwTTD&@0;cXu(vQtxMiZwOp!
zCW7Zal=Yl~d*MzX4d2I>vCuN)05L8TxX+v5Tw*CnY_xeQu15|SOPM3---;id-LRY?
zXA&d1k&QAJMCx!kW2GpG6y;7fBXf&^XrJTaiehaO6$-%Esi|OVNM2Qe%ryF~S;RL^
zEJRRSwo0ZO1YU7Cy-=nc?Yid$E}Nd-Kvvc$blgRXV%$Ue5z4B~yns!3r^Aj=rV^Xc
z$yY;klAN-B%vpzD(FvW^xQ%_!jSJb75~ydY3?fswx~Hda%JqzX%r2l}OemSDzQYuk
zmyb@wG&qXZq`<}9G_z&nN|7b3Ea7{L;z}>hsa)~D?YrX!%-reC!&D6b&%}`3?FO}D
zi}V}!jh*H8DDTfq8coCPY-X`@zjZCBvbBNQI#+aMUQZG6HFi>5{@mtIBcdcXf5QXU
zMS!4e{Y=XBhqz0ak9vU;8W|Z=wms)u$g*HL9r7tRqbb`D;%|)r)K0@E<(~?bj+bVd
z$~%Bm5gr-k6YS5z?0_)_B_fXtI(K=3cyK(RS<$h}lg~YQ8!D0Elf;N4M#)V2UTkq6
zda*I43eax2ippm>otH-PqnB_0K5pr~U+&g?Mibs^T~@w$gB>3{F7@Thcs!9-$^9FI
z{nL%s3DNWtPR`|Q@ul)RdrD9T2kxq$fbuTN%tXdO9%!Hw7NpS>bqD3LA?FHx5X(K6
zUjjsCL&1>fEdi$wR<EtIy{h|Deq^;S<Az1-q1C3xdsebEBl)}3wu^>#Rm_1Y0St%Q
zi*C!+YvbKZkT2>uR>rCVO|})A&}WFu3u+p5X_+c3+t={kUBLR$Ze3BEd>~YS6_hVy
zf$QLY5z{p%)M;|MEe3Y)%iFMndGgVVC7*S>Gni7oYNMkd!m+Dk{$yf#u0@^A!pDYd
zV^d-`_xGo}lNZ&s#q>NO-!k)^Z6+UaYK5+uN-^P6W_~tUF9xN+&|*OxtlRteUlKsi
zPrTeCyn&xAq)PtVKb_vSQ$br1Ws?vR#`4Z42e;ooqN+Sv*L9)d6&e`MRP3dS*KE73
zHh*TS{j-X?l85tEe(yU{<~}s^?yg?)rs^kz{1FS1OUwPn2vl6lB%+j9R23)dE2HZd
z&Yeuc(CKOuIEy6u6m8)tsmN&KQ=sHoZOFw=tcY`3KGn{ERSm?~+2il-2d168L&lHm
z!a)f_+q)rXaAJ!e2|%eYDBRfEl4u`gQ*k<C(DK!ILk0U5DwVX7E<e*706@IqGo*vv
zMGpi!2kT{U$?DQ6I-g+49LtVL<0ovnQGyoHxEgLpvbafTI+A}mJZfJJNXL{<`q_gx
z`2o?4dpSf9_qi3Mm7613kS5vs#X_ts&)~vBAwJmxvkp+A+$wa)w~OYr;flM0#jy#o
z2NUhq^_(V;l?-RN0`h|aCpz>j`kP7Wu^ig)?0s}>UrGDSHvV;W^+9I>Yy>>9jOV&B
zE?mscl;BawS$Sm@bg159a$xv7$Pu8PQgAX7lkzM1ETEe6CD`6|T3))e8AhAU_W*tO
zf^92)LWQ1<ZA~h!t9PYS*W&jACilN$Y3mP4z(x_RImljuv<Vpq3Es*EN?le{dq!4l
z^gPqt5Jh!u4F^*T;CJZ}f!Nd$a?fiVf^|M7CFr`KXUXN##PpVMuw?l(n5SQDDcOhq
zmY1s_ksj!2B$wK^jyIWE$2IY!<maQpy|_0L)&kSOa{F>>kSo2`hjwS@bt_J%S?X}O
zMlO>nOOvwl{Qh_pDYN&Xxbc(RFMqnC;uULr<!A@i4EZJ+(Jjw6T6d$|>uPFjSokBb
zW7172oO)Mnz+C;XxU{`6@2t2k^C@V_?Kcqk6fyRof5t1hKyjMrUTaRSw!sNsAm*K%
z8T$^Kb9jUI0Mf;~BFb?~q4%ir*z=xt1kAytZH=gS4;;ZeNTL)KwWO*vqb8mF2d|^9
zvCO*g?N^;QQAeZV%m?FaWtSi-?*v?e{_dE8$sG&(FCBA3sdLUkCxcd}+VVV4bokbO
zV26Nhs`Fma`d@2C+RZf41Fx)%j^CX|M>847U}*ClG3JwKU*FrRoafKz=>gltAF06>
z^5XFiNBehs?yW#?oO8s9iu#L#x1wV(y;xi1Xv;ELfFd#23k}HKWaQ@}{jhI?U1eLy
zlbz5_i^<SJ7!&sBTVM@$aGlViZW?!rS2$K7xC?FT4^IM@na?^Lq+Bs8>u;GN_qO~3
zFR#2y_B)cA*aSpX4QYwIu*KlPE<hI0oM>r&#dvLLk$hsqpirc2v5W~z@f7LXonv5(
zC4Sw!)BMdPctwq4@?_4}J*1$oFLo}QE8_aZLZ;4_m5pWXv>V&W{m7eY6Lxj|X|@7N
z;|_!0>gwqXl;63>O*HNQNDla~;OJ!)`tu3QY5||%^_vJWT4fs3U@#@bib@g!!ktGY
zXL)hepRc{bW~*Y92vowtF8yC}<-$~dN!z$kVN`NfbH5*RT5S}N$b($MWN1#~J9qO2
zyFU#o+W#?F%~)OgEkkErQPwsyhbD@My~(NQ58J8%2QH0suoSG;_T=qB^<YT^{cccR
zWj&#`yvwn>xpCzhxlzB&=#28366=Uyd0CE-T<iYz+4iB0kl2o#<a|+bG37Z-aP(ec
z>s!{Nqjzj9E#m`9u)#*r@v_wI8JwQg44kMpVaDO7N)<uJ_mfgh`)~f0e?A&Bge~Ii
z@qz(YU=m~QH214sddqR4ep>o<pIblf@Lv)h&VPu_DU6kRB%!0>U)*c;WTIY4(9k?d
zRYlw~&KonWOyuid5D6){U#0}=vL$hZXyyruaj*A0zq9U!c3;r8vh-ws3(2<&u(QIa
z+!dz<3%&{n$Z>HuwR&8<p9Usql$Vw^n!AcLom?(t9^blckEb+5;b8e^<VAiv3Yokl
z%)4^iVPLEvsd~qeS^nG0+pASQpX@YDR>Mmpw?3+?Apz7Y9hMSFyHIw6H;?P<O^J>j
zC(T|>zFKKl4qjQ2J0ABGA%vpovtpqJ$mEU(tD$ST3-s^op6;zcc!&rFz|dyOea|fz
zi>QTB{auzT*IEz(WSh;QYU?;?u<4jYbFf=a)ro>PedT8_bM;DWXfQ|mO?@2k6ow~n
zg1YDAI`7wi`_7AaxqiwppjMw4k0Nnc(U86Tri?+9eu>#TWy9|arom>C63-p%4O3&D
z5hoI{u>J~D(`a3`jj_5qQ=U*Ve_bx9G}<UK%Xvz04*Uyt&UXe`9w)S5lp>s)lkp>h
z9I=59yu72Q>zCfk80X7QE#uR6vk78;2+SjYJDudGxbb3CwLFN_mvPfw8FF4~7vuO^
zvZ@c%bM(sJ=tB2axQL2sc&;Zrd9NR66iK)al@taB$GrpaC;a!UK-39H{Ittqy1%9k
zj;N69_NUmbu-%4wj&p1hr#5JONKL^+0MZK~cSx4IxxI{t{q_jO%IEb}>=><YC23>;
z!*hg$?oSgVTQRV#PHY*t=ZtNs500KZ;?RP{h*8)<;Q==RC`=g!8~hkI_xxMKmG<dZ
zCZ<3YLvz07_pQGrfnpG66=!$@9F2Bxk_x7nz%KLN^A-shm|E{kvQ<i*=}jwY2RweL
zUFp4vu%;nH1~!&3je+|!;;a=?=;hs~*I+VmF{Ls(+ZH@eVgdezouJk1h}Zi3#vPq`
zPw*3<YJl#xA&b7oe?3(5WrY*@)xYTfiMu}zXh1AkBFdaZf-ROhVQwy0d^-552tk%h
zz2Yszn<ITsr_JfH#OZN%Ld6%$*y5+M9f=neZ0Ke!+TVtKeLc-lJ-{-_)Ts*YQu>b<
zfP|-&ZNW|CiveYlMxs-`>x&-^d@O7?_o3ebm8PH!-XHKQ*)>E9^|*S9XF2z`Y+8Ct
zgN!>(854voQ<jkpEq{a*laz~QjNsef2*$h9M7h!7uc~_c291_m!QsfObXiD9_0x<$
zRd>}^jG(q68ZOqz+c5awz+b30_J3)}lwqa2ZI|UH^BP9?LOhPTyJ_SMsC9iGm@yj$
zsIv=#`X|#LIoU5e6|xKllXTyq#;7=7qFCCYwzj7(<@<~`j}(7J$U50g=UDUC5pp{8
zOW_WSY^_MtP}{6oK03GR=igH4H8pmnr%&896CLbd*{7d$u6MunaI2o(i*;fZ6Dwh7
zOONt9+ZN0c;gS*-E@5W~XX-zhU8Jw5js^O0ECD`CvEJ=A;sR8pwe`VW@X%erptO$P
z3x9S0N56Ug*7ye}VcaHPh~ergec!D;z7Y(v9$Eu4WrWK1Pqcao>>|DgnQWOBql$|w
zs^EYDI(z*@1zTvjiND`?QvP>r)Ib$%=U~cpriV}e6=_Fexe)DzQZtn!@o(-rrM9C$
zmc9d(h#E35KolTO!9b2mh^^-!h2N{SB0L1|KMncIWXw54FxeV2T9$QWt<`TjE!djg
z1&mED_l?!O?>C4nJLAgu?pAf_a8Lxc7Xl;LvfoWCoP*bWlX#n!?aDS0Oxkh=$T=sc
zA(+8}S-!B*y<BdjHGduA!9-XD&2~w^fd%cCKMVq{$lv}94ou*yE0G!X)%zsY*hGOa
z_{_)u161M3uX|ou9veU`({MJ#O}};j31q^l+5L$4h<w{1Sh6z#@RdXpvLuR4SmNye
zjxYi}Y^1j4nTJJ@nx3%Zc0hVCNB5UM03a&BtT{CrS}x1&B!bYKp)>l0EW$`gXj^kH
z0k!58ksdzaeFE%c4@?jFVE1@%f`Ds(coV_RX^Y`0&J36PXr@Hk;;mlxBPWl(9JQ5X
zBAzGesHhXdn&jnMBH_OAjyq_pC{ICT`PcXZD#czgwDJ1Egaol>0lpusv$1qQ{@g`u
z@YpleA14t%SR$nQ;{yP8A`D4c%!M`O+EVV%iDEx;I_^RS#N?r9<xH13<*UmC|7H4M
zOD{gZ`%+?zt!!g=sbjUkY+k_S@i#6iX|s6j|FOa+{Gg{Fd;s2(k+iLGH*s?E5I!FN
zkC0mBNUybulgT=V<<Kx81W7NWPy5L{vlRM<tjEX9z^%YQ%Hgf-#XFG?*?eIsjls|Q
z61-O65=vY*kUWatv~#Nbj+6Z_n9LuCBw5HjF3Bhj2cvsgIaSeA7$)Ll-i;v7+sRCg
zYU>E5evS9<-G5JbkHHIqf>gLHo{Y{`QgBem0FPl78}gd5UVl2`HR3^6B|ilo4%cGP
zB4F=;%i{y@OkUO6R?zL{pY<>VRB3N38UazG9r)Wa8z+l>9fbKL0m*eLXK^ied|<6f
zj&%~?cW?Z-@=BpAuH2d0CNq3ZN(vCJ2JenUS#m4eV(QGC*tpqq;J?`74*{;Sng+k>
z=!n3I8ty=n1X*Z7un+gKA)gm8zw|_F>PJU%ZX|l3HuTNpK;NsV9!`|gqMN$HUd2SQ
z>9qI(siSWn9t>3JB4zXqfY%cfxt}_UqfU_Www$XwZUm9e)sRbZPAJeRq8%UVu1A)<
zj~#D7nH=tm)RgaF1^R!K7!H#|<q2%%nC{={`t$SfhdIb+6X&OOgQi$!o<T$s+enQ}
zf<WH^#=u7$2Ls@CXT5cpijcJ`H|&X{zUekDIsrA%#}|ar9xwJUc!;DGG-T!Z_#+dp
zH~D>jt8-W*4kVxFHpGYHKv#Nwu}%f%gt{h%JxefwQG8OIl;#n*D_{<Wx79B>l*qQ<
zMt!|--uN)X&11#0g_0nhJOMVlBIXb}5fl_IcU1rx{BzRB(<y@}KHztBviAeQ?cH&h
zS}(o=;X?kiGGDY$xk|vLgmG7pSE3OE1}<nfw%C+WQ70CY(T=2Xd2AkShuFYm1+>IK
zDc^&&CG1nN#*Q_-A<tc;qP4%Ek95(|ydh`SGD=tpY-$I(^sO~=(0p;yT&YdcAp)3l
zj8^R}?bCC;`z6MGb-HdJy#+jG#=t=Q1qcWYlV*!HN)AaYU=}C9?VcQiLG<SXm0sJk
zeO;|kQsg~9OHNBNTKWXw=d6b}?2XV@*_OOBYV(!_)DyQ5A`Y?M6%->1td=nJc$7S$
zR6xM?s*R`Uco%r!ltS@+L27Q+I41_wkEA;P&*$<<9t#X?zsFYD;6$N3z2$?0(w(kM
z3TAf!KnmfKmqQK>9w&%MOzNL$6DelnSV|(%JRvaBK-F?_uD0gq_SrT1WY>X#iLva5
zmn=jVRm{K;j?n}&ko>ENvdW<~Dow|r3*1;dH;ukP&1dNEdw&eq1xVj6q{L|y7l>OA
z9qUhPeakxFhAR;s@rr{fw+#@Z0tLF&xlr)~I|r(YYHxx3$>X{CHXtL6sibM7DU|Y3
zCC`K30XqfQj$}f%@bmm0oUzsONi7|Ta|@tuCePV_d%0a{!@yD_{8%Yd$(D@qhe*|~
z&e)atD+Pt2>4i{4Ew{WfPf#tDx4Fn3HImMd66)zS*1ZtU7QEAkj}rn&4K4TYWnP4*
zXK$QzssXwC!MBplQ~m&FQ!4kzxRbNXu&<jgUV~vJxD9>SVUrFwr7puh%G^Sg8#tcc
zXm6TU!h>35{@KLgf!X|`Uf<7emp}((Z8X9JV}#8{8)6s{U&m=;J;8a7M4I+&|La<q
zMklbd#^MtmpSxO))+L>xAM9}fRV-Xm9mLILffn-?#7IkC#7G2K=ct-8yff?0CkW2}
zTm>UAY>cDRr_XGkPIPok#a=elJvz9oMR(~___tfWi1!7?CZ{)*$vlY7Y~OcJ#FoEN
zwXjy$qk5n>^l>FrmM#ZRnzrwtar>zi8oDGzt87o%iQ*w2NN6Jf+CJrR#enUy$%M>e
zH1>Sv(uvLD?(Omss*}t4N3?U2tcD0gFgjPDi2`~b7+38pB@Eo%5nXM`Gi>F2&r`^q
z!neUi%#8$s;rO;20+KUVXYWh^=q%x-rH57p4ZbB&arj)y3<g`BDJ7?4_|YPtAH+gT
zdT}@N!V*^2w`Ju)y>Zi#3Z~&1V9;}OF>>5Tz*qe7c&2{QUEtmo5mgDc3T9G%*x^7%
zPpkMr9FmpwS25BSOwNMxC~|=)f7Z2hbI@vQf98HSV?T0yw6QrKL_J%T8R|zkxXa%M
z*gnZ@38r3L(3unluw2S!$^4R$`D`~%rwM*nRrl=JW00x+FU~`ln;U&n$w_y6`~g;J
z?hgSEs#<I8?~hcN=XVJ>4@v%An9jc!9sQN!mzqK=XmAROtE_Xx#8davo#j@J`RDd*
zY@CMhddCaXHY$P3W6K8vBXi8#gVM^@&JF@pa-k7%G5@}vta7<WAJ%d27~D@OXRe(l
zT@zh*o!Sjh#KiE|R(xQE74x@zqb0=j%vw6vn=j@QS#vuMTSw#`EU8f4zKn@Gps$0?
zh44Bh3Z>})DO8Z1Bk1XWiKO>EF5r<QKmX=Uc0F^R)O1viA)8MI?JI7yqW-Ps?bgW)
z%fXJ|m2?wNR@OYoA^%J{m2I1rTZ0ap;O^8HT|>j*BP3r|d|+p}xSBTFfERgm&uoi!
z|9yoEX65*mR2}wQIX+MNxo>e@;^f0*CAJrojp`NX4BF}){YAy~2EzBbT~B#LeZElO
z;KGNqBZAR^+h;A`$VhMKoASKn`9yg)pP&Uhv)es3Xx-$noY+xE^CLqOQ?Uh%-HX$x
z7^{iWl0;7Gygkj+hLf@io%<7HqMzd0<E#0=Yq;p9)<JBpHZ={V*$L~sj?MiepN_d)
z)q$Fdg^wrwQ51P%tyWpKZf4s0G<0Fg;R@~HehXmF0W-|chO2V{VPX^y;%U;RJYnLi
zriM_iY>DMVvL6>F*HV?q7-XzW$rpguvOG)zcVD&xAixIln&LF)i@PBFs{V?7&-KA0
zdX|-y*x*21^GP<H>G*P+D@TC_e{mb0I)diN7Tyi~jss^2B@bYuad(!(bpBu6#w!)`
zAC~VU<|CKc+R~>0Q%<*BH&0hxjZyyHFkvVr<)E0TnBGys!ujBP8%rb_GqrEP)`LkW
zH-jOs3{wpDOTxkN_RoEQy&jxp_#~(&oht4}hZl;XxTtgOxmoO!b8uD7o4aXha$j6h
zBLEE?>m{Cu5=7u`Z_n=Fw5Y89v%6|+?3`LylDdL~iUeAT$!0Y#&rnMzp>mz?xEa#d
z-r>aT(EfWET2!j&9p>R#hFs>`hHCGVa4awov95U6HebG$rH;iusHv;-GFOo>ZM<`e
z1dczO&Px{+kmLRN?FzOGT@<+MAtNcYY99Hpj(d8lSbG8uw=e!9z?vr9HLs$gm(7gC
z*<4+d5{CGVX}QM-tM2k0KSu1FRhGP^E({8)zagL>vvdzJBj6g?b<bG)JNA<H|DckN
zwrghhC9V8&4e-1_Rcwm*Jg(RBH073tN@@^U9HXg?5^$;A^so<U3c6FxH3_frfzpgx
zTW&f<W$8#?m?MyqRqYTDm2(d!=oy_xmNv>39!zI)h{i>G4g{=bjNq;HiHiKyd0<xs
zCZlE9(Eg4*n5XKOHdvAZU1n01a>Ld(XLO;@(@I?2bGO8-k;uU+lW}*Z!kn5K&C~Z;
zl84-RBqD=@VgG7bZ`h*&k~&WU)ey_y8z7glR7Z5kq7tJLYFRuz<0Fk7vqxX)nOfj-
z_n%|L3aQG<Up=GGN1}F22p3o6_OfFnwC7L|7WK~>%!8g-uDfl|M6uVxw$=x7lUK>S
zudGW7@9H`K9BaCWq|9`PGF#k~$=s^mz}!Mi(-m@`*VqK}uMhse_TDQhs;y}oZBz^h
zvQ?5G;1&=R$r(w4fK8Sx5=C;7(Bvd;K|nyV<Q$tMIin&XAi2pjh~x~C^IwhoeZTMW
zALrtXb8#5WfYH!v%{6Dus;9!N2U1v(KMvg?ZEI^^l%-gM@O8TMrK5O-R`I`H_dmY4
zVy{>1d>kp2dGFTV-3LctYlV@Xfe4|<xXg@v=Hgt$HFa}Vx%ZMIs^2YMc0Ptl^`4e<
z<sw#EWzF7R;&$juRvzj|t)SAYJgO4PJ$R-=-tD!=Em|(lh?l%P30Z>Y=CF#P+6ph!
zuc#XvYlZvy9BZk2)OTcLa^(jdqp|j_t#3Qpnuq!!w*UTwmJ#-XuV7s6-WBtSWBmpy
z8WJ$?`)SQF4R%Bva3a&tbG(F@?(A8J6n%csoAvVrohQYAOE*kq<W)AG{6(2vt@?Rp
zvD<Yg<vg!iS(1RMcmId0IzAKS{2pqZ_Ads0N^SLOZmUP(f+Daq9qFcunHXtxvm20q
zgM?eMTU@_9f1}^P`(y<O<v2<<t7Iz8VLJ=voTp?M8hsu>_=9JH@Xdsd4`R@f%U32+
zX4SKN>@AdXaTQyT=Wi81bE%wM{w3#ZOxEqP=Ne+q!Wa#88c;vnIG!Ifc66u$fm{x=
ziRK@>OR~><dgfMUTFKA4+V6(?Kk7(CemCsZeBj8>bymyp)~nSD>#Rnu&hcS6#s4Az
zQ_9m!>8da%kvA=yZ@8LPrUvbS;dTD#*x=Atc#4EnsXUp|BmKM8)pmjckrL}j*P!qx
z%B^fES1BmJf9P*01a~l_*_ZH){zS`P7}xFf`S}giTvZ{#q7$z>NHY_&mbn$3mb=W8
zv$9uRx1bPccWP4r0lCA$h~Dc6rwbR(I}*}^H#0r)#C7XQit6v7n9E|zJCARNvZEL5
zHz!(3_T|YmRr?%gyw;vVnP1k)qRq!6xr<PxyObyxOc$c?D<fG9O80T$crg=csOd6<
zs(x*c84gJ3-9aaNKeVaC1>tEo|3?iRTAaF)n^xn&aN14g9&=3H!7Uk7qjF>nLxxbp
zDmwy>?F2nxLD>?$;1qFxPKE$D@W}COcncH;WI53$mnwfQNt5c=BKYO$YPlwMh>f)s
z;&_w9>dRMo^m+G@=cS-eqhJ4V%2ceoz~%1D;_7|}wl1+^B7+bY`t$NEwP<6|sL&Y?
z4+=`m9?cJ>-QAXEg!9R+83guuDvu!Fh6JSS9DNJQ^7DCCA}>w8EG;(%7;0JRSn_%D
z5H&3brM{ETEuf~knY;2xG)60b#-EM+I@xtuCq4~@_V_OdMdAdj`1W>9!RLSJ5kt;a
zyk#>)bk^1Gw@2*9|LB%44RJUp?ML|=TKyOLpT-8AXO;vWy-|Kz=xEl&@D5v<FqQgT
z=qB~v;XI?mH^r6SK>s>7$Z&og3>sOi`=|`T#U<ejkRp7{OzcGV!?jd*d4)&f?AfS>
zbxH_8)Y|TzL(>)Z*DSApC85EVlfwqwVH0w)nEZGW;PdMm0DdC^{7JqQ`rBMJy33eP
z<bRXz&Wtk9i<wEj8AE5`V<GmrF%y9#QU89U#P-2}o!w%e0#rHvd~{K^@wRPIG58rL
zSzKd1n23zRx_5x4W3W(pL9MYWU|n;gvBd2P6jN$$@3%pI3{$h*@VUR&WYBJb8yq6b
zc=l@#diN6=&N<iHU(j9b@du!!d%Qm2_Tb6?0HTXHKt!Qv!pO@O?+1Vp9GUY+RsQxl
z^JU-A81v(LyVonsCy*_@EX$?Bdbr7=oU_ub2}#oCQH}ZGTn6%9uK9X%>czq_u?)k$
zwIPhd`8@1*y(a^<AIpr|z7QAj@(mnl$7xmc=WFf{w;_IvdAt{JyAqm7XRE^$T{8FL
zB3-{tMslhTh|->76veLoG$o?CswzxNTdtvuYIob6X*yuoOZ#mLv;IPqj{D0)!aL_V
z50@>gLgT;q8Zxd7l%Ck*1o*n^7Orei#DaINE#u;ppeLd_D!i>S=)vw19J`Q{ALKu&
zbImWO?aRNc_ro)Q^NmkSoYQacY2h^60A-Sj`nN^DU@laf^}PrgGSLTd2A?0gCQOP(
zg51GjQanx0Y&HNkL@S05SFN2#H0{@vnVRWt$GHal7r&ecTq~=7wfgO(Ij7GY$gW&A
z+WBHQjapwP29HG?I0?_)9kp8-COTiYcO`?A6yabs^W~zb&7qZ55eLVelJs{)P&E0d
z`C7OJ!aD?gdDe56j4E)SQK#+Rl+eNQPM)yCIfUplDyf5muFox38SPbmw=q|S#x<{e
z*nQ?dlT3DO8G9lEwL|jm{qmDg9+c^IDB#l6^nrqc{cDZDSs~#yNcbt;VJOS~XA*gL
z$={z2ih?E1-WYFK6HTNM*?g#YR4?1!lCeiu@)A<C!|~0#3rmt@30)WtM@~M&(j;O?
zIl}3aEV-0hFwx-Z-$Xu^N>QFLHv8<mPKRW>YXr3xP$Kpw|FWoaSJqZxhHB2Pev`R~
zUIiWkD#^d?9&}~IT!zGCgsS?rXua?CgXe-7?xZQB*KJV-8;3iW*0#;NH2(zDqLj3r
zym}M*ccVw`-0XaCTn4^5PdW*yJvT>;FvmJv;5i)lffHIrkaCvy`bAs?hiZKiy<$K?
zb_nqX#XcMC>x6TVhc@E|Q26w%ldYjE-QLn%CL=5$_eAst1*QHA(@k#^dZEVz8P{AS
z@iUDxA8#I?yLfUQ+7D-=dC&21ak+gexk^*oQ_M+7M8uJoVHv-+dW(uvd3L?0M_EQz
z7O$wZ^y}c@uBRMYR|F{NfB~u*8cx$}s-AmTDgDT&;?*OcoiWJux~z=Q?k;}kAQ4nw
zc={oh(`$yVR?xO~%#fOZX!2+p4T%G(e2zo~1zcqj7>?a88}TR=&9CH(ylYT3SrInn
z1o<kt<pDqZ662qf3wB8+yDv4NzCzKf?jI;4UdLtO0l>4Le7)f)-Q1^T_Zi&b=o*Kq
zrjA!|&Ph2%wP&d&Bb)x3U3p$%*>Oo1H!Jqt%n>Pi&}%|Dp98G>j|3rq@^8W4Fq#ch
z&tx30(B>ZnMjjB~)?h0A2Jy(>yYC;?iW&+QI6ArpZ7A-{>n+AqJST;^h{=SN&#pUQ
zRi!p)qb6p4iX9G<%JVk3J*0PjGXjNVO7eexS@psllJr?)zkYJz8Vv#DB=37o@<i1i
z6L`^zwCYKRgvKU7oa}~UR1{QItS!xjg6%5kHY8f2_zpG0U(+)_t{zH0@9jmlb$sq<
z4S;YV;nLu%V9)x;^e6AUDu<pu{k^xu_>|q^y!4kw0f<oLR8=uW$ixnfT>UxS)Ro*p
zq_uo#LlPcKAjTMd;)N20IGp@-DX7B#237&gQmDGNPfQ9PRA+BGSV+P74~jAM^Kf=G
zE$B7%Y%k~IwssFD-mwF|#i`Zil@RjKsJOO7d-BSdFmWvf8nVngmBS<5Svm;pNU8NE
zF$ZkK!^7W~-PeDpz+q#Vxps;$IH9rb!Y|O>-5LEbLGbyLPS=tX^}eLE(*lBKIO^=g
zEE`r_l9QeHb+D`GZYJY@K>b?MY4v-P*gd0f$#_2~xPK;wOfN39o_u>J9u!;L!};!A
z5bb_nb@c~IDvnfHcDm}hhTN6$`^<VOV~tPR?omNiBg9S$ob-Z?Gb{E<E{z@F4mL9S
z5mn7jYR1OmqBGximTWEhW*STPe?4rw$8@gLE10r%{$#jjbuN41pP<Rk1p089TlA04
ze6AE@>cPM3gGVMQ85?xlhpD!HA(wectG~J*|KN!dxvr+<*B~ae@;Us6p$ev{e_iak
zc1`P6x(TQKxF~zcSCU)H#+>7w0s<x<_ON=1RX6T%DXtt^D<xk^YQe0(Bb1bWr0Oby
zYZ0-gPEuQ`U$uun?DEq)I9^WUVT!~PBC1?ei78L6-uQy)Czm(J>W>pz*s?dZO$cr_
z&77;1>^2lBkvXPKNJ!&N_;lI;T<9u3TgPH^g6LvU=W&|N^>U_)<lQ*eMs`e^yoLNt
zhq@JtuJN=GZ%J%_Y4;G_xt+|UT#Qg=I<OBiAmzVU%(m=!FI>aUEg5<$?%&-C^e<Qw
za(SabEfSv5m3*u3GGSh>Gj6y)BQ0lX&)7sJ#^gLpbbE4i{QE<_R)wzbd=IMKa8op4
zWKmJ)!Td}=3kv!2egEokzJFmu1}mIbwn84{1rtRqgd{vbhSeW$iDk?TIc`?=6c_hF
z1B61o5Yn+MudLiXuhQQ?4W>!Sa=m7Q)!v^r=%}TVUQ1O)N-Oj1wwOdX(5rQ)Y;DS|
z(olAnN1kyHq)l1ml+Q<sq?%|H4h&29p|;Ycxp!<r%m>wkS!Aw66dW0Gm|rE$V9CkR
z3U_~0mlmgLnpnYS11%_l;P>7n9!-i())n)=Zsr*^(5U`uQ8A~cQBQFh8>^KM<F4IN
zQ@e56Yo_)ZjN%{Yf=ju-IUWh4;443TT|Nr*fhf~wsiTqfaRze>#`0yqwx4ZkYqL!Y
z5I7vPsg&6|rQ)}|D!$>H!e~u<7wI}x)7Q%Epu2g~B0OYnT$J&iP-)@{wNQ30-I=n)
zhNPrp!C&fCaw=X=*I&nHx^&wgnskcf3ps1Dq^5G`>S^<&pEkc=75^M~jmCLaXBnx`
zO~k_5o5hFCR;oFw9ia%LSsapAInc|8_P-UvC&kbpEm!pZ%d0DuGa=Q*C}u0<gUo}d
z{K{;*g@N81lzUABoD6l&0}1i^M7oP9alZG;&91YEzI?1$pp_Pw<6J=_TyXOAcRu}m
z*+~MEsGi8<(*Z<Eehz$oZF@=WFqhT=mRf|uJ;9xDGPT^R=$Q$Tzg0m*IOA3a`DOH7
z1|<#=-Z&Ar8KfkSWwxTH0!BWG1>+w-x?x>$`04Mhy*U-?&X(@4R0X~8jeb=swoEd$
zT8QuErIiA`oh*9Qv1O()oWC!(V!uUjd+q4zaIM-Kk@9b1K0Kl9K})xv!?7xP3kLg;
zH<3gZdo7|e77mh5zGKI^=<6t)A_<Yb4cbf^_#*X}RkJd$PH3$!h++f<wZ@l3vw68h
zFeRdSy}^cz0o48<CYdm)=|_~?LmHHa^U*S$?Y&99SNb17KjO*TCu?{Z9xFDh%<00~
zyt%g*j}D@R%^yrk$Um(>y7efMQaNQNizPAMG5b{8$+L~lXm#Y@T&m=7bzGtd48!#U
z6vH-}&`qe%IheOVtD>?Y?rNPlg3ew+Avxh#DKih5T$s(i$oyw&1~SRNwIuoahAZ?2
zdTHI$Zv-~<W47XWP)El2dwUCEKlP`{$QGO(gWWFqefa*kPRO8Es#urlr=z5$Y<|*q
z=JB@INzW)L^VUw^BJNwQ(^})j%Vf^WHw6<DCXXADBeHSxeQtN=_L*y*#-4UxUZ5n<
zt)%96m79-=Sr|*VeG@z;JmCGoSSu+kx1FRmJ7eL(g~er2EZWb8rod%9zbAEzD#Drn
z>4BojcXmS1Q+gi56}??(*SYda^z`bi3F<|1zu8T)OfO;1ry=hu)sphGGn%M)@7|q$
zyPeSbzMUOIe_Pv@TF?~sC{L{u+unYJ&{Ia7f=RYxa-N8C{hdxf-_zw9naC{2{U|JA
zyC!7uOqu*f_BU7aA56kL#t{TJ@dd+USZb0AeO~)aN!-BKe&1%u+IM-fdH&7%<cru$
zkbBZOo9K|B$mG-ORzF$$(mqegDdYN|S^zs_k7SZ&KDUVPa)m%bWa8y)<U_n;b!mi9
ztN{hVCBh=XQ@$=|BA6V>H+&KF3?{~Et_2Z!CHCCbgGD{|F2!5|0$P>UugUVqvy*_l
zy;~*g#m<(Bs<XtF`G}N6;X=-F!h@D?NFXM{eJSCQ)V@5BnEVD8R}vrQnC2%SozhtM
zYbN$nrbi~ko}}dda)}VnQt7SN&h~P@Rv2!=i5)l?vb^U|6%m_Qz~UCpeT0@=z@%2`
z?j_amB+HsyzJA)-7V+si?(tR@(TmlnzFG9ZT$}I#NZ2Aldw6_IMxUPwrbt(KtGd>T
zG%N3@hl`76KO+ZvS;Y||E6V))tgN2pI5|X6o<H9@Z+BOKOSWx?wr7c2)S|5(TKG*W
z&c{5k9#C<5H)DF*loNr_X7BLyc@BE(aoFZzEK#eIy5-U6te1A7c(m+wYl=CPiCkxE
zdoNFjNDA~Wp8x&UcbMI?RD{yC&i5OyhO>S$A+nw~julOt>z^=U(Owz4Kk)Mv!S+V)
zhW+e(_l=Ug`hVF()^3s1GG%(rDpZ$SkS~iSq+E!~k8lYcIf3mLyV1+~=FLBI-Jum~
zvfqX+e^J?(H5VTAJ*f6q$J)jTN=ZdL7<<pX|GQrc)tpZNZE>GnuJ0Np9q*IV{Iej=
zLIk=IYhK~B>XSw69;NotH2cOGH(3Ypt@u`FvDLGS)0~77_B03RFXH040%ps_&lh<;
zwi`nNuH<Ja+EAv0#`URK?qU+BC9ZiGw)tJu<SrLqMfLMP-P)6{r2Le;Bv1EcYkH#-
z*L|9+Vp(-mI>~(cbE6SW`@!NoA-3ve%!=+|K7kIayxW>5wyW-0kZ}P~;(;kNZ_TLJ
zxGTiImRfOQD906KIU>`{#ko25rAdwYPxCVb>MzEp5;@i70e$;sHs(`-DiK8{h1PFR
z#5uMkQ0+$13qAW=Q>&9{m^@3IeQeK>W<Ull3=H^%97~?9(J`>ykBzqb1?hytz}Pqa
zy;fGwnT0y_TpiR73xxNV{Ajo4-abRKpEjOFAQo=JT+J5)scUt`cvlwMvuiPlrK>;P
zBeu-L#NW32>eX2GIYD9fjhB7sPoZekXQz>JXs1?IDaYFgJ=dsj(_!wHCYCSB_q1@u
z5D@2Us9l8)9k^X_;mMl%``-D8#)%?9sW&kSv-lx(ECnIffs1KVlZ>Stwjn(|y^*Ug
zneGW`<1}!vSEbxHn~cnj+cWtr0a1zUI;?8kN1sbeq&f6xuIlF8_f0RKAVTfRWnQ5c
z&OQJ6VPq^Zd9%Zj)8E&k&17hHou&FX+Dk$%D7x!P^x4Loc<8e0+zGkBW^(MPdnT(l
zBj7}U9QDej!~Se1aKQ9BtZQaZYR|fyc&&6clyBj)MxzHe$h@old@6kG?Cm2%Lk&>x
z2sS!98|JJiVH$%p99e8V9zJ}i_NZ9f$P<sYlb;1?{?|{v`I|&%J3B(75)1m};{g7t
zzDo-hDtO|uN9tVv2s`B<=(2GIsf^cong2oqkionO7AMW^TA;|5+9O{-7G-42q72ur
z_3=4V#kw{%j}W?!Z-WnCIlxD)-ki;@A?Dq8)ttCj)zh>7@bK`(#(GW{NvMHASo-E@
ztxU5(sJ&xw_S=hl8}_+R-dypksa-D_F)Fd8k+D%%{e3XBOvd;sQkymU;3G73r_&P^
z4ZQwYreHky`|q8d=zAaOuM6qHg*QjKU)JhMz;Esz9#fNzBa7Ad*MwG2Hh}9d+G)E^
z8@Ohc%I0{|vDVRw_~VOb)(8*J=iZ8IzoVzuiS<wjVCLJXkfbDxlNTW!W%Du})eI}$
z*lm$94)qkVRrWPtnhE<(F#E8H-^w?jf4=C(`GfH`(qr|pHU2fbS@-zvqV?LJ%K;)|
z<{KYd^h3`V8dLZltbf`#L*K%F?^yRt;^$cyOs6pa*XHId^)39Hm1i&>hin&eKY6nz
z7*Y5}Ese05F!b}vJQmJ{O`$VX9f{l7|69b93AC^Ei+Mb7R5DjD7a6On`La=4E%Gpo
zJ9V{B?Ozg01d0NmO-VA5%E^y@exA1f9uM<DYK2}zsh-{nHtP<^US|-WD(Vj(1p@Ps
zq*jb{9(!zu6<DLBZtv*Mzf7isxir8_gY4avY@<ioU{4wde(w|(eE6f5vL%?DU)=+d
zKInrc{S4{Uy=wNq%2*l7M%Ul4MjuygG3p$&oWWtB%@5txRf?;W|DhUH)NoT+yokZW
zbF8U9ZoW);(z>{Bx9qxgvdcwuDDh6JG^kIlTb_jUTAq8BiXSKE*77m1(PrtDG9E=(
z`lbvVJEb1G_p+n^r4d93U4#d1TEIVVe$1t4-ciB4IT~NJZ+rqx+E4UjCJR@(bX6F~
zGFGg#2J~hA3?73IkIPeZ!}l9;tyJ26li6OBl~$$aYfk~^$I^-kVSsTd`7E<_twv1e
z)P(zZ$}<QXiv<64+(!ug1@{X&v*bv$GNkj`&9Or8)B^zRbBjQiZJKLM3QdQhWQ~^H
zo1ZTp0LCXTFZARm@s7Ov)3ejaurmt-(IH7mP5$~BU${)@kq=W+PR06)-)}BXdZd(?
zx4XFbeT8$(q%*V3qFgGUiKtqQC$*c#e%LNV#4BXc=jcBAr^!se%qN_2nu>%z>5~{b
z4S}d7DU|!8s}P?gp8S`a%`f%8{H(;1|E0L23GgBRGZP3Q5`4CQe|2Ue_}6qI)|5zU
z|9$so3e$fGWB>mPyXRl|nd5o5E}fB;^^E$g+9z)L>6w|<@o_y75fSy~S8ssWOjT2p
z2}yAA;`{b*<DY7uf*zu886O}2MPC%wHm$6yn@Ha>FreXBgZ=fWD6DrLrwh-VJqtU(
z&sIOBwpOT08(LL^j>4_+=%_X@+O!NrR}75(n0Wbp;^kxId^Z~)msUX2A+n7@ad{UN
zu`dso$<{b77Yh>PzBP-jeexQVq+D-Ay4IEtVLy}W{k74bSh-$@XntRVH9y+5SiCUj
zcJE)#?TDP+4WD*nG%Cc;xIOkMHJ%s`=Z{pATZlEgPT`$+GFELSWUR1TxuD~csqJlq
z`be=+;7Zjj4YFr#qHznOqWCoaqL>vs@W*GQr+;s5mZsvhEE2rpt)<_e_tbT@ZigP(
zDSA?So`~`Ve0=PmMx}p+>i@YUHFC@MF8qa1qM~PHT*Tz40LKH17LH%**CRw;R6;^N
z@Y}0}VTueSc&vso@TU3euU{XLLZ>oz!ZMrju?2pF`76Wb;Q6`kG+@`6Gkx>s4LpT6
zU&{Y%l%{J$|M$W_ZGl1{iduo&ZR;DAO~eCZpq@9!R-=d*&M1xC1UhutsQcXzjC5EU
z<lr&KJg#%w=`S15MSuVP7(I!XjR(Du>7?FOCV%7`cQ5%CO#Q6gAAupuG&2Wo@%`-)
zn=+G58Y2@EW=>9J9K%@h!`CFFq(-Pm2wfi^a_<37c@>4Mjaoy5Gi@*T*4}KAMMu0q
zQ-Pv_g1WnV?XStnGX2kI_V#}(6Re$|b*ocKS3|4!#h(+ins~{-IcOMi_wL=C!7U(g
z!;$nw>v`t5t=4G(1AH;l-w2y8HP;Zjz+bh}OT;&-k0L~reLvjW;Bi{f26EGd2|wEY
zNDbW_Z*K%CczfSN>9k5sL;J4cVg3>o74_Ka(=ZzIgr}B|=YQgBkS!a{%JYHHJ5pqC
z>N>E)-DP5WtSGep&v3aFv!0$_f`GjgaK$OZpvSyUJalZENX<tpKla(;VQlk&Drs@Z
zQi$}-ZsB!C1=E$$s<3om2-L8(wIv`Vj4+`1y|JMpC-?4nGfUKVlNEvN$yGzsBM0=n
z>VS@}%m9J;z#tp-IP8vMVjgg_0c)-z49N3npOVY?9q_m_B8N-O!Xr8GoG)p?(tF$g
zND|fS87kuz_;p^)MXnd)gMXRcU-2xS?pN6GQ&W$_;7Ok>e4r6_eGC;u#pd_P5v!-_
zb&aZ*yX<)r0kQ|y*us_)QKs|!d8(bA9SIFh!G{kN=t)HO-*E3|71m+tM+ZC6f_MJ@
z`(YTS-s6C>4gZtd`j9w6*Jh~LXn%jPFfu+q5B5$rtb6rLtOfj&+YMg>u-QCJbaWDc
ztHRR11yNaaCJM)jdexWPj2Ha;sgGK~ASHYW?{1h<BgUrN*cu|}v962xwZ@3rAOBps
z9`XKtGDw<=qUcfSb3?=F$;q_%_?sR_TLZ)O9yMC!mh-x~c!=C^1G@lcgzkbLt!I}-
zVV(NSGLRm2m2`@<xw?A8A)}w(G_G@hOFnvJ3iN`pyjEH{lh*Zn#zC~A^~Q~O*=q2_
z+~yc2WKT1#SDx$E0!~rLKR_)+7<VS-uQr_24{iZfWA)MY2#;ykEuhHLjJV>>H*rxH
zq00<{K`gZ94Fryv@bs{6Va82iH^6OTJ@@As>yGyq_eaK$jZvxy9(w3|90p{H1!ldO
zX1i+>)yw50B-gH`9v-@ToE+`Mig;9$-PcT3ND#~c0VHEQuOGR!I3OEV?JzH1b=f&*
zYU;%iOf^G6X9rNBDC1TW$jUq}hvBX9f(@)y@2-A2OK|0iIxv`Fij8EkUMELd71jwo
zJw5#0*M2et1Q**@X)koA`K{C+SnpWj;o()UR4)RnbUv}ju2_jlXa9rMT0=Ee1X*B6
zNVLQDTTjH82`n&fWVh+)wpy7Bvn2z`es685DJTT1=WAtn9d9JzZejf6LkdiQCN;Sj
zbu+RbkxbIv-Mw<gd1nPnmW8LJA}%3;jrO}jt+nvs_~<}3><Wz#lV9T#x+(;Q6}GjE
zynH0qdiI=*H+gGYTkqz4F9T8~OF=D;*J^5Zma&nY<mSyhDx2CTPFPel5Mgt>ZCeKR
zo>@B&kFDgJ8qS2XaiA&W1$GQ0h^Tn)qvT$`EIP33zBGhzjy7tGLWcPg-uSW~>rF1`
zyp~L#11ms4Ow5LoOA_&*PY_<ckR@_P2@B{Unebp|HJV;T8Q=M;mzNhPX4No<{mpsh
z=BcSE5-Q?PHT($*ELx?H!?=MEG_V&@B;L*FOxVa|FV}4au8b5X-kLRTQ~{->+*YHN
zPr@*n@>KAD*+|JnGcr>Ch6>j$bEI-x6wCBt+dLt{ro`jG5p+^63@IQWpjhw<;xb7>
z5}klR)eq5&g+K`!MsoeSiuF}OgxV|6q8hE=zjrMSm7sop!UMA4n!-+)Fp$5^S<#cH
zVGd2C8rjJ3JL*m6^k2b7y+=>~$f7^5d$h_vGA@pJeyLlKS)%~+BZiA1XYkS>!Z`$t
zo^}}b^RE{J!x2R?j~@LK%V)!ibY1$`Txv5eB9NefG#OWPKUl7a<*|sOPh>`52peB;
zP#|>YV3jEOthK8QK_R7)>pH&*(Lbv@2dXtz$i?pCG(`uhnc8X1muOA=25<*HaLl8>
zD<Y!%>eVYU#xweny;+KjpD)m=N=Zq@818Q`qg%om?a#4!n@fWRwE((OGcZiUm9}Zy
ztK)?LHF&H?HJ3*!G)|6>P<zwS8_x;Xo&&$9R;}~L`jcZfY7uu$0|SGF;WD%@5mn|X
zL?HIY2Uw6D!)21Im?Yx5*)5$PeD>3GuwrP9!v5As=E#VMbkHc0Qc}4RepjktulD7t
zF{A0utX=zGm{Rlg|N8*cn?TTE?mZX;csVW&8lY|#R-hKucRJ}97@7bVK>RZ(8HWO<
z!6ZcMI#1v?H8tfJH2Q*#QB_rq0EtcWqj7z_zP^qVc)8wELGvIkc@=o}D}HHT>|Yog
zSIf45pr{!Lc1C~rFgM2M@9#fL4L*4z6>SJye@&rY9mnidrebWIDfNybB||nks*;6;
zMcxdGgi4Hq#$0Bpkv+988{Dp&bFsls8A(XLDrEKkqA!zuV(`r225yW<+%zWt15P$-
zbxfzmX|?>11Co66&mVTiy~cG`_V@PY`!w|W0I2n)Bd>$KZap<!bAWq>KN|W0YDx#g
z9jJ231G96Aei&-kzfrBUUAlB>p({n=;|Nis(&NW(#l^)jsgjqm4!!a`)K_5yh+7K9
z5Ows-Dk@UCKui-sk6P@{e^mAChJb*;{_bQ5s*R-r$Hfa$2_PnDrQyWWYPd9K)T+W7
zyC5>5E+Z2_Zw5*XM7o%@W{DI3!UeQRXQC_YhscBkkFon8+FHdg{BhPAT~IA^_3BmJ
z3ZT{VxedOY%hxJvo$U9>UpR^j3rl58a#x0Tl=bzKfj>4qTGthsrI=KoECc`w<c?Cn
zF8%iH+c=F*L@lTaE4Q-cTUfIqIJE2-bW@<G4%?k3ZB#MRSD^D?2L&+NewKOvlqWYX
zrFaMDPh~Ij<9sC&!;+(&)rRBaV>RB@RLMYYyQ#-umPl$uI5?ibD&xQ|_lxg;H8nRg
zv$Gd{-JVT+35R>lW`4v}&nCJB1Pv7#yymqWOwLO3)B?g`7WKap(aFhTM3mf2TjhWW
z%~!|ja8Mi8J;kPePbwuPg|w=TC->e?p;;_$L!Kw5(YE{n=%v)K`AXg4@3!Guml6O^
z&x{kn)?w}1Sm3{GKQ5BP?tF8ZJQu8;R*hpma9O~M;PU0bAm&!(uhVud({`jaEAEUL
z=p_%_VhZ6&gSY7Y+>$p=&}rp95+{aNV6Nr`U`_J>mhg>WqV=+w`0|F5U0=-q_Ug`<
z#}4+|^hy=r%rsV=Y9=0@GI5tc%fX_DNBdjoE$QFtTgOLhjl{AfX|t^@El<F3%0v`P
z0*Wd!Xq1VlnH{c1p}}&b0@boziK?p6DozA{-o?3@VwZ6K!hs!V@-2`g3-~$9@|SEl
ztn{VpcOHjPa+@X-iykUT2j2{BoOloZl(luiwQJWf4PIWz9?;m_UdM-av+<lp0q#KA
zV9|d+LBvA~=Zgbx5X_ew@)D38)vQ#be+b})aB?^+D(Jpj1TiwK^jl+NWAjE|qU!wx
zG(!T*?PxYV&+eU?Ox5ONkA<8p5=zS4-CgIMV-){QnD;n9PYlTBVCr5R5?$B^@$4x!
zY6HWE9Iq11g6(;<*{isKt*MA&(<^{25;=GrCd(RX7P}899XoJzxY5ZeUotolaaSdG
z`83Z;(+GJhrlv+>QA?S7>!bklgQ@sB4uL=QM?^%#Mh6$rCaZ+qAR{wQ6h!6zum#O<
zlmBJ%-4;tw>;5CamK`t+{HxMA8zdwoAE+NVI)pp`3f4E{^W)R=edy@BEwy_yakA9*
zVCHsqh5$~!sqX|$K58$t`|~w<od;@3uLf(KnUN7G*a;g0I|+ODSh<;Gx#IVBVDCKl
z<gZ_&lQJXC;A`5A*Mk^yu0BJJ;xyVPm6zurj=<CxXH%c^&p+pgZt@omZsp|U#B!Mg
zgI9wF59l5%s{(L;XLNw;!EMrU6WN0<tk;^UuIL%c_&d>M<_6MaAc$r4kr@CIjTlBR
zX2IvB-*H%VMriuukfrO|=L-x-5bOM|n|=A!+p92f4lNAP5w6c(yhsO($$-Rd2mg_h
zk`k=+bTEyu7EHl?u1+w_Z9kNrMpofM(%y`xs?MnTR{(RV<q)S4tT`wOSMXMVu2u<j
zQB8B6Zk^W0j~_FllHmbXwsco(Rz=_su(7+2rdL<l0K<U(fEU<@stRG-Uyov{0Ejz{
zjY^hV4k^!6P6C7Q(a!h)2t20N#UB6s`RbCvWk;-SGj&);!GM;MlG4gfsW@tRd^~!;
zVE|~8cXm$d=%j-y*+Qrvl#jZAY=O(Fy7D0)_s5@I>R;Fg>PT#v9I!25kX)Pb`u#<{
zlY70~ID?CuCD1^L^k!>6KMQ&uR(|>NWppIrSt1a=3ET|6e4IE<nhVO&tZHP~k{2k~
zm0=-<D{S;~3~_dD=W0Ir7YHv{SLm(n#wzI?@NsjOj#>dhIz3V$tK_TIELk%Ub5C~|
zQuxBB3WKQlq8=37zGqJhGm23BauqDn&s4}C4-RV99jw&2t~Zgx4r&#FE%1CY7cIm^
zn5Jwfa=6wQc+bg#nR&<Nsa$W~$5o>yT$D-7N&**|x@UB_yS~yUdc14+aRiVtHX4M_
zEhf|DvmPvlk!L5siO4umUuK{x!_sHk<Dx6`O|7tC#XLYX1i>fa<l>rk5S+HF8odFa
z|K-DjPxt!DyqM{3-~QsU=Pn9vUX{H`*%WTjwfMNWS9ZjgFXB7X@$vD`D<q|*YuLPi
zBVBD<7zQzEXEP3V67a4TT2djNKLy8q-2<2pHJjjc@6H{mzyJQbc*wFT_%JUx<7}=n
zEM}`|PrCoXZ!q7j)HSd`c8c4M1*v7eI6FcD5DcDiIwM-wPHzPO#>bH}XU^a-BC_XT
zwV|*Nt)bDf160GAj$qXntRO5A%Ws={Fy0_K)L*rTA<J*nif!~GUfdecD}2$jpr&(B
z!HF}?`Oo-}J<_xu>3W`f85mp8bvt9Y2vLIDnR1A8sOdb0yW`S40BjgG=ei{d>$YRn
z3w84`7}aUucOOvakynX{iNSfmU}N$?A5>RRiMTuaSLwP;lkQAcR`ScoawEaWf#2+;
zx9T8>+TSz-i;nXq>XM(A!H@l^aK%w+Z@~nGQ>6#G29z>A^6Ys+Cfn3PsUXTKwsDu)
z1g$#PGMuX2S)OVM$6;T%sr?g17G~Lzaw{;y@M;tO_*7DD;5h#s1S5t#s$jnZcuqW+
zowBPcg-{467NxdmHa4fkwY8T!(>W?x_k@M(T+iTa2+cd3sg5YJos<&({hcl#Fi<7@
zjywqk#hh;aC*zX4+}x_z8mEC@O@Y{vQLtjwU`1trkY2x@2Cmwtqt%KrS6O(jCt_mW
zgpKbZssKKjrsjugr_};*W3|EP+~pfjM~15^DJ75CH0Xc?tO!7CEW169Pc{q;&Ss(4
z5&CO6#P&R<fDW9Il$<;(Ir-L(N8sDHU0?mL!_{z(gxZT8QWc02`Zd}<G+GDuaFX?P
z=Z;#s;I@sCj(t1N!gxaieGZ@l!*B0wPvYUim}-<ozUD`92zcOJP5KsN=DlKWetu2x
z%HPwo=v1ptOiXYRHr5d*y8mu5ufy$2Z^lzF8@Y`$q~G4%fM5(ruKC{YX+Q($uvI>@
z9{Rei{%1yGx`n#J4D|HbPfLt5#*QJ1RT<dWbMD+ZQ&^S#N!k<5`8;xTHlfJ)Kvk~m
ziF%X)1;8_;|1?Jj4mZKwF&;0RO-)U6COd@1Yae%uycA4OyG{XPSkwO(@G7dSM>(#H
z=E%qKp86mFuyH;Jmq`aG9W?mA3xgliz}Zdz{Q3eKvPH8A_Mg1*ptP?yfRuT2)$8OL
zSerg@b3y;&<SEDFV0&StqHup}F$=2wq8~&XQ0#9co#cUTG>3=X{}srFi@2dyMxUlh
zhx7~YjHZE!h^+Deq*!i^&E4CW#ZAgEz%rPQz7i8fB3cpc>v!aoOii<as@<vkpyQre
zl^0qZq#+6z)Jd7u=de!WeJ~!}`s42N;{yk!uxlx-&nV;9z@Q)&R3b!#OvmdzaG{DQ
zmS>p&?9^L&PGP%_!3Rk{d6*N-^JnFjwn<-?0h$K{2bc3-y=MVv*nFfM*u}e3C9461
zvnT)(KseJxwITy15C}YX<fDhGrtp?n5cB=}nBDbB?8(YdNfycg-a`;e;dG1ly$bg|
zyCYbyF1CilRPe!KGVg+Uh>XmDm|qTPPu#=On++BjfSX>y&jFwO8S}>xQ)6Xjmc>Qm
z5}kkf+ciTx#uU6Ao(NoKBirp10VAT`zI%rO{D~<wQ^92%V7~>@;Zg%i)4o@zbHOGo
z1hUn4feTn>F~Ew8=Y@ubF7#%Fz;b24vbI4Gd?6b?<}J-7*UO!M1*aBqUzN`HPsyTi
fzyE(PGjSyPUYSK7rRW`p8y`|q;ZcFOzR&*wVbAso

diff --git a/dev/articles/lazy_tensor.html b/dev/articles/lazy_tensor.html
index 13b35658..c59a49d7 100644
--- a/dev/articles/lazy_tensor.html
+++ b/dev/articles/lazy_tensor.html
@@ -387,7 +387,7 @@ <h2 id="digging-into-internals">Digging Into Internals<a class="anchor" aria-lab
 <span><span class="co">#&gt; &lt;DataDescriptor: 1 ops&gt;</span></span>
 <span><span class="co">#&gt; * dataset_shapes: [x: (NA,1)]</span></span>
 <span><span class="co">#&gt; * input_map: (x) -&gt; Graph</span></span>
-<span><span class="co">#&gt; * pointer: nop.3354b5.x.output</span></span>
+<span><span class="co">#&gt; * pointer: nop.563486.x.output</span></span>
 <span><span class="co">#&gt; * shape: [(NA,1)]</span></span></code></pre></div>
 <p>The printed output of the data descriptor informs us about:</p>
 <ul>
@@ -434,7 +434,7 @@ <h2 id="digging-into-internals">Digging Into Internals<a class="anchor" aria-lab
 <span><span class="co">#&gt; &lt;DataDescriptor: 1 ops&gt;</span></span>
 <span><span class="co">#&gt; * dataset_shapes: [x: (NA,1)]</span></span>
 <span><span class="co">#&gt; * input_map: (x) -&gt; Graph</span></span>
-<span><span class="co">#&gt; * pointer: nop.3354b5.x.output</span></span>
+<span><span class="co">#&gt; * pointer: nop.563486.x.output</span></span>
 <span><span class="co">#&gt; * shape: [(NA,1)]</span></span></code></pre></div>
 <p>What happens during <code>materialize(lt[1])</code> is the
 following:</p>
@@ -484,8 +484,8 @@ <h2 id="digging-into-internals">Digging Into Internals<a class="anchor" aria-lab
 <span><span class="co">#&gt; Graph with 2 PipeOps:</span></span>
 <span><span class="co">#&gt;            ID         State sccssors    prdcssors</span></span>
 <span><span class="co">#&gt;        &lt;char&gt;        &lt;char&gt;   &lt;char&gt;       &lt;char&gt;</span></span>
-<span><span class="co">#&gt;  nop.3354b5.x        &lt;list&gt;   poly.x             </span></span>
-<span><span class="co">#&gt;        poly.x &lt;&lt;UNTRAINED&gt;&gt;          nop.3354b5.x</span></span></code></pre></div>
+<span><span class="co">#&gt;  nop.563486.x        &lt;list&gt;   poly.x             </span></span>
+<span><span class="co">#&gt;        poly.x &lt;&lt;UNTRAINED&gt;&gt;          nop.563486.x</span></span></code></pre></div>
 <p>We see that the <code>$graph</code> has a new pipeop with id
 <code>"poly.x"</code> and the output <code>pointer</code> points to
 <code>poly.x</code>. Also we see that the shape of the tensor is now
diff --git a/dev/pkgdown.yml b/dev/pkgdown.yml
index a9e61e82..a84a9cac 100644
--- a/dev/pkgdown.yml
+++ b/dev/pkgdown.yml
@@ -7,7 +7,7 @@ articles:
   articles/internals_pipeop_torch: internals_pipeop_torch.html
   articles/lazy_tensor: lazy_tensor.html
   articles/pipeop_torch: pipeop_torch.html
-last_built: 2024-10-18T15:11Z
+last_built: 2024-10-29T11:35Z
 urls:
   reference: https://mlr3torch.mlr-org.com/reference
   article: https://mlr3torch.mlr-org.com/articles
diff --git a/dev/reference/DataDescriptor.html b/dev/reference/DataDescriptor.html
index 6b9685f1..7ddde268 100644
--- a/dev/reference/DataDescriptor.html
+++ b/dev/reference/DataDescriptor.html
@@ -263,14 +263,14 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> &lt;DataDescriptor: 1 ops&gt;</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * dataset_shapes: [x: (NA,3,3)]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * input_map: (x) -&gt; Graph</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.922020.x.output</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.6c4b31.x.output</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * shape: [(NA,3,3)]</span>
 <span class="r-in"><span><span class="co"># is the same as using the converter:</span></span></span>
 <span class="r-in"><span><span class="fu"><a href="as_data_descriptor.html">as_data_descriptor</a></span><span class="op">(</span><span class="va">ds</span>, <span class="fu"><a href="https://rdrr.io/r/base/list.html" class="external-link">list</a></span><span class="op">(</span>x <span class="op">=</span> <span class="fu"><a href="https://rdrr.io/r/base/c.html" class="external-link">c</a></span><span class="op">(</span><span class="cn">NA</span>, <span class="fl">3</span>, <span class="fl">3</span><span class="op">)</span><span class="op">)</span><span class="op">)</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> &lt;DataDescriptor: 1 ops&gt;</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * dataset_shapes: [x: (NA,3,3)]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * input_map: (x) -&gt; Graph</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.922020.x.output</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.6c4b31.x.output</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * shape: [(NA,3,3)]</span>
 </code></pre></div>
     </div>
diff --git a/dev/reference/TorchCallback.html b/dev/reference/TorchCallback.html
index 73412310..18536efe 100644
--- a/dev/reference/TorchCallback.html
+++ b/dev/reference/TorchCallback.html
@@ -210,7 +210,7 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-in"><span><span class="va">torch_callback</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> &lt;TorchCallback:checkpoint&gt; Checkpoint</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * Generator: CallbackSetCheckpoint</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> * Parameters: path=/tmp/RtmpSGhCDU/file1c9a4be2a6de, freq=1</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> * Parameters: path=/tmp/Rtmpmt7gXp/file657f35f35d4b, freq=1</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * Packages: mlr3torch,torch</span>
 <span class="r-in"><span><span class="va">torch_callback</span><span class="op">$</span><span class="va">label</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [1] "Checkpoint"</span>
diff --git a/dev/reference/as_data_descriptor.html b/dev/reference/as_data_descriptor.html
index fca6c1a2..3a347f5f 100644
--- a/dev/reference/as_data_descriptor.html
+++ b/dev/reference/as_data_descriptor.html
@@ -93,7 +93,7 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> &lt;DataDescriptor: 1 ops&gt;</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * dataset_shapes: [x: (NA,4)]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * input_map: (x) -&gt; Graph</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.550a0d.x.output</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.fe081b.x.output</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * shape: [(NA,4)]</span>
 <span class="r-in"><span></span></span>
 <span class="r-in"><span><span class="co"># if the dataset has a .getbatch method, the shapes are inferred</span></span></span>
@@ -106,7 +106,7 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> &lt;DataDescriptor: 1 ops&gt;</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * dataset_shapes: [x: (NA,4)]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * input_map: (x) -&gt; Graph</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.da7663.x.output</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> * pointer: nop.d2a8ff.x.output</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> * shape: [(NA,4)]</span>
 </code></pre></div>
     </div>
diff --git a/dev/reference/materialize.html b/dev/reference/materialize.html
index ac83bf97..1231531f 100644
--- a/dev/reference/materialize.html
+++ b/dev/reference/materialize.html
@@ -132,86 +132,86 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
     <div class="sourceCode"><pre class="sourceCode r"><code><span class="r-in"><span><span class="va">lt1</span> <span class="op">=</span> <span class="fu"><a href="as_lazy_tensor.html">as_lazy_tensor</a></span><span class="op">(</span><span class="fu"><a href="https://rdrr.io/pkg/torch/man/torch_randn.html" class="external-link">torch_randn</a></span><span class="op">(</span><span class="fl">10</span>, <span class="fl">3</span><span class="op">)</span><span class="op">)</span></span></span>
 <span class="r-in"><span><span class="fu">materialize</span><span class="op">(</span><span class="va">lt1</span>, rbind <span class="op">=</span> <span class="cn">TRUE</span><span class="op">)</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -2.2687  0.4539  0.0872</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4005  1.0713  0.1125</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.0352  0.6216  0.1302</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9565  1.0647 -1.1269</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968  0.8098 -0.9682</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7730 -1.3947  0.5044</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9440 -0.9359 -0.1327</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5167  0.8462  0.2562</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2682 -0.7513  1.3673</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9112 -1.9628 -0.3506</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3420 -0.7548 -0.2027</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8642 -0.6203 -1.1163</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1052  0.5600 -0.3974</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3943 -0.0705 -1.7041</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.5337  0.1588 -0.8624</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4815 -0.1104 -0.3846</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  2.3301  1.1064  0.2867</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1371 -1.4983  0.6197</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968  1.8850  0.5676</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4147  1.3346 -1.3396</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{10,3} ]</span>
 <span class="r-in"><span><span class="fu">materialize</span><span class="op">(</span><span class="va">lt1</span>, rbind <span class="op">=</span> <span class="cn">FALSE</span><span class="op">)</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[1]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -2.2687</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4539</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0872</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3420</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.7548</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2027</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[2]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4005</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.0713</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1125</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8642</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.6203</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1163</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[3]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.0352</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6216</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1302</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1052</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5600</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3974</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[4]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9565</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.0647</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1269</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3943</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0705</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7041</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[5]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8098</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.9682</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.5337</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1588</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.8624</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[6]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7730</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3947</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5044</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4815</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1104</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3846</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[7]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9440</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.9359</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1327</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  2.3301</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.1064</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2867</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[8]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5167</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8462</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2562</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1371</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.4983</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6197</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[9]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2682</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.7513</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3673</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.8850</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5676</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [[10]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9112</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.9628</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3506</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4147</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3346</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3396</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-in"><span><span class="va">lt2</span> <span class="op">=</span> <span class="fu"><a href="as_lazy_tensor.html">as_lazy_tensor</a></span><span class="op">(</span><span class="fu"><a href="https://rdrr.io/pkg/torch/man/torch_randn.html" class="external-link">torch_randn</a></span><span class="op">(</span><span class="fl">10</span>, <span class="fl">4</span><span class="op">)</span><span class="op">)</span></span></span>
@@ -219,184 +219,184 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-in"><span><span class="fu">materialize</span><span class="op">(</span><span class="va">d</span>, rbind <span class="op">=</span> <span class="cn">TRUE</span><span class="op">)</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -2.2687  0.4539  0.0872</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4005  1.0713  0.1125</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.0352  0.6216  0.1302</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9565  1.0647 -1.1269</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968  0.8098 -0.9682</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7730 -1.3947  0.5044</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9440 -0.9359 -0.1327</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5167  0.8462  0.2562</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2682 -0.7513  1.3673</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9112 -1.9628 -0.3506</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3420 -0.7548 -0.2027</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8642 -0.6203 -1.1163</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1052  0.5600 -0.3974</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3943 -0.0705 -1.7041</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.5337  0.1588 -0.8624</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4815 -0.1104 -0.3846</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  2.3301  1.1064  0.2867</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1371 -1.4983  0.6197</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968  1.8850  0.5676</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4147  1.3346 -1.3396</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{10,3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1387 -1.7121  0.3313  0.8510</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.2791  1.2661  0.4940  0.2577</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6875  0.0049 -1.4214  1.0401</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2687 -0.6857  0.2180 -0.0905</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3564 -1.3467 -0.8512  0.3529</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6444  0.2139  0.6715  3.3102</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7180 -1.6664  0.4273 -0.3261</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4558 -0.1893  0.6411  1.7529</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.2947  0.1367 -0.1888 -0.0066</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4252 -0.6875  0.3217 -0.6429</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1006  1.3168  0.1388  0.5232</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2336 -0.0717 -1.0036  0.6683</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.7921  1.4544  0.1958 -0.2346</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6385 -0.0318  0.6482 -0.1380</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3797  0.7916 -0.2373 -1.0011</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3849 -0.2163 -0.0237 -0.0863</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4032  0.1983 -1.6133 -2.2197</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2452  0.7073  0.3545  0.0833</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.8829  0.2502 -0.3007 -0.1676</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.9140 -0.7256 -0.7844 -1.4341</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{10,4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-in"><span><span class="fu">materialize</span><span class="op">(</span><span class="va">d</span>, rbind <span class="op">=</span> <span class="cn">FALSE</span><span class="op">)</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[1]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -2.2687</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4539</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0872</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3420</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.7548</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2027</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[2]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4005</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.0713</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1125</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8642</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.6203</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1163</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[3]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.0352</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6216</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1302</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1052</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5600</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3974</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[4]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9565</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.0647</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1269</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3943</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0705</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7041</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[5]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8098</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.9682</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.5337</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1588</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.8624</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[6]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7730</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3947</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5044</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4815</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1104</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3846</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[7]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9440</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.9359</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1327</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  2.3301</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.1064</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2867</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[8]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5167</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8462</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2562</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1371</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.4983</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6197</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[9]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2682</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.7513</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3673</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3968</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.8850</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5676</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt1[[10]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9112</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.9628</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3506</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4147</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3346</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3396</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[1]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1387</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7121</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3313</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8510</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1006</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3168</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1388</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5232</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[2]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.2791</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.2661</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4940</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2577</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2336</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0717</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.0036</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6683</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[3]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6875</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0049</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.4214</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.0401</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.7921</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4544</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1958</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2346</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[4]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2687</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.6857</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2180</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0905</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6385</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0318</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6482</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1380</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[5]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3564</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.3467</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.8512</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3529</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.3797</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.7916</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2373</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.0011</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[6]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6444</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2139</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6715</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  3.3102</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3849</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2163</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0237</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0863</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[7]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.7180</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.6664</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4273</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3261</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4032</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1983</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.6133</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -2.2197</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[8]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4558</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1893</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6411</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.7529</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2452</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.7073</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3545</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0833</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[9]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.2947</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1367</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1888</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0066</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.8829</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2502</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3007</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1676</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $lt2[[10]]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4252</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.6875</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3217</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.6429</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.9140</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.7256</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.7844</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -1.4341</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
diff --git a/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html b/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html
index 829e757a..3480a19b 100644
--- a/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html
+++ b/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html
@@ -274,35 +274,35 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-in"><span><span class="va">x_batch2</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> (1,1,.,.) = </span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.1699 -0.0348 -1.3127  0.0449 -1.3391 -0.2706</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.4008 -0.2498  0.2398  0.7099 -0.2593 -0.6988</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.7197  0.2285  0.2184  0.5368 -0.0414  0.5573</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.4601 -0.7059 -1.3604  0.3196 -0.3387 -0.7384</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.2113  1.3375  0.5421  0.5011  0.3896 -0.3945</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.3065 -0.1602 -0.6599 -0.0274 -0.3330  0.1745</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.1683  0.9521 -0.3007 -0.6531 -0.5466 -0.2548</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.8280  0.7439 -1.1441 -0.2175  0.6100 -1.8303</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.6169  1.3783  0.2681 -0.2965 -0.6230  0.1022</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.4736  0.9308  0.0779  0.6800 -0.6026  0.1194</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.7146  0.7433 -0.7192  0.7677  0.5538  0.4407</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.3748  0.1611 -0.8856 -0.3067 -0.1230  0.5929</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> (2,1,.,.) = </span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.4998  0.6063  1.7977 -0.0074  0.8868  1.1315</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -1.2362  0.4990 -0.0351 -0.4337 -0.5377  0.1622</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.3271 -0.2007  0.5288  0.5646  0.2186 -0.6196</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.8539 -0.6276 -0.7953 -0.2205 -0.3521 -0.1657</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.1857 -0.1610  0.2109 -0.0741 -0.0214  0.5110</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.4607  0.5773  0.6792  0.5576  0.0791  0.5881</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.1821 -0.2982  1.1122  0.6025 -0.0759 -0.8357</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.1558 -0.9088 -0.7307  0.6930 -0.1729  0.8293</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.4416 -0.2946  0.4691  0.5564 -0.7042 -1.1576</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.5102  0.1238 -0.2319  0.7806 -1.2194 -1.4314</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   1.5757  0.0112 -1.0029  0.4919  0.4147 -0.0619</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   1.1526  0.4710 -0.9762 -1.2955 -0.1328  0.0883</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> (1,2,.,.) = </span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.0530  1.2853 -1.4618 -0.5761 -0.6764  0.5012</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.0895  0.7634  0.7542  1.1485  1.9902  0.2917</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.0751  1.5195 -0.0507  0.0306  0.3952 -0.3084</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.5689  0.5605 -0.1022  0.2474  0.3521  0.8797</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.4937 -0.4953  1.4223 -0.3710 -1.0329 -0.1044</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   1.1720  0.5355 -0.6309  0.3580 -0.7996  0.0643</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.2915 -0.8742  0.2805 -0.3131 -0.0847  0.3623</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.8970  1.4281  1.1198 -0.4452 -0.3298 -0.7621</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.6495  0.3189  0.3577  0.3509  0.4461  0.3858</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.2846 -1.6956 -0.5109  1.1558  0.6108  0.1983</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.3220  0.4607  0.1320  0.0521 -1.6308 -0.7962</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.9437  1.8788 -0.5748  0.0037 -0.1264 -0.8996</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> (2,2,.,.) = </span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.3858  2.2418 -0.5567 -0.3174 -0.0823  0.4582</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.5028  0.0923 -0.2350 -0.1985  0.2016 -0.7900</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  -0.5887 -0.6755  0.2675 -0.1618 -0.0426  1.5266</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.1032  0.4991 -0.1085 -0.1012  0.0068 -0.4961</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   0.4799 -0.3508 -0.5198  0.0623  0.5824  0.3236</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.2720  0.9749 -1.7862  0.7320 -0.4558 -0.2841</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.6081  0.0487  0.1334 -0.1368 -0.8086 -0.4014</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.7191  0.0231 -0.3324 -0.8849 -0.8758 -0.2848</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  -1.2964  0.1189  0.2938  0.0605 -0.3840 -0.5428</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   0.3190  0.1164 -0.5326  1.0797  0.4521 -0.6165</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> ... [the output was truncated (use n=-1 to disable)]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{2,3,6,6} ]</span>
 </code></pre></div>
diff --git a/dev/reference/mlr_pipeops_torch_model_classif.html b/dev/reference/mlr_pipeops_torch_model_classif.html
index 82f103ed..f324daf3 100644
--- a/dev/reference/mlr_pipeops_torch_model_classif.html
+++ b/dev/reference/mlr_pipeops_torch_model_classif.html
@@ -274,17 +274,17 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`1`$exp_avg</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0651 -0.0071 -0.2694 -0.1905</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3519 -0.1111 -0.4643 -0.2070</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4170  0.1182  0.7338  0.3975</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0813 -0.0115 -0.3071 -0.2139</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6080  0.2067  0.9018  0.4727</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.5266 -0.1952 -0.5947 -0.2588</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3,4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`1`$exp_avg_sq</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> 0.01 *</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1201  0.0036  1.3689  0.6378</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   1.7104  0.1656  3.3447  0.7363</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>   1.6698  0.1561  4.0499  1.1255</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1392  0.0041  1.6147  0.7540</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   3.0754  0.3983  5.4061  1.4248</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>   3.0210  0.4019  4.2777  0.8787</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3,4} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
@@ -294,17 +294,17 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`2`$exp_avg</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0548</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0768</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1315</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0619</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1498</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0879</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`2`$exp_avg_sq</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> 0.001 *</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5453</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9515</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  1.2573</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.6434</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  1.4496</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.9899</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{3} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
@@ -336,7 +336,7 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> Empty data.table (0 rows and 3 cols): stage,class,msg</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $train_time</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> [1] 0.055</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> [1] 0.053</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $param_vals</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $param_vals$epochs</span>
diff --git a/dev/reference/mlr_pipeops_torch_model_regr.html b/dev/reference/mlr_pipeops_torch_model_regr.html
index 4b472b9b..fc4ca703 100644
--- a/dev/reference/mlr_pipeops_torch_model_regr.html
+++ b/dev/reference/mlr_pipeops_torch_model_regr.html
@@ -274,14 +274,14 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`1`$exp_avg</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -1.1899 -2.5584 -16.8081 -583.6016 -12.5174 -11.6554 -219.9824 -66.3019 -2.5536 -9.4319</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -8.7515 -70.6545 -155.1045 -6165.2896 -82.2042 -85.4106 -3880.7810 -405.2719 -7.2561 -78.5860</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{1,10} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`1`$exp_avg_sq</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> Columns 1 to 6 1.3291e-01  4.0389e+00  2.2521e+01  2.8091e+04  1.1109e+01  1.0300e+01</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> Columns 1 to 6 4.7060e+00  2.7571e+02  1.3329e+03  2.1144e+06  3.7363e+02  4.0293e+02</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> Columns 7 to 10 8.3546e+03  3.0016e+02  4.2510e-01  7.0911e+00</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> Columns 7 to 10 8.3214e+05  9.1789e+03  3.2087e+00  3.4837e+02</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{1,10} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
@@ -291,12 +291,12 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`2`$exp_avg</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -3.4187</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -23.3342</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{1} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $optimizer$state$`2`$exp_avg_sq</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.8167</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  30.1836</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{1} ]</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
@@ -333,7 +333,7 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-out co"><span class="r-pr">#&gt;</span> Empty data.table (0 rows and 3 cols): stage,class,msg</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $train_time</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> [1] 0.04</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> [1] 0.039</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> </span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $param_vals</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> $param_vals$epochs</span>
diff --git a/dev/reference/nn_graph.html b/dev/reference/nn_graph.html
index 67d7caef..e7c11748 100644
--- a/dev/reference/nn_graph.html
+++ b/dev/reference/nn_graph.html
@@ -119,22 +119,22 @@ <h2 id="ref-examples">Examples<a class="anchor" aria-label="anchor" href="#ref-e
 <span class="r-in"><span></span></span>
 <span class="r-in"><span><span class="fu">network</span><span class="op">(</span>module_1.input <span class="op">=</span> <span class="va">x</span><span class="op">)</span></span></span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> torch_tensor</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4208</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3031</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4180</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3396</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0821</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2099</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3070</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3691</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1470</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.4891</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5106</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.3846</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.2878</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.4239</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0854</span>
-<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0169</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2515</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0529</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0081</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2387</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0763</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0844</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.2905</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1252</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0067</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.1168</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0734</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.3246</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.1031</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.5290</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span> -0.0672</span>
+<span class="r-out co"><span class="r-pr">#&gt;</span>  0.0823</span>
 <span class="r-out co"><span class="r-pr">#&gt;</span> [ CPUFloatType{16,1} ][ grad_fn = &lt;AddmmBackward0&gt; ]</span>
 </code></pre></div>
     </div>
diff --git a/dev/search.json b/dev/search.json
index e4e218d5..c9871a54 100644
--- a/dev/search.json
+++ b/dev/search.json
@@ -1 +1 @@
-[{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":null,"dir":"","previous_headings":"","what":"GNU Lesser General Public License","title":"GNU Lesser General Public License","text":"Version 3, 29 June 2007Copyright © 2007 Free Software Foundation, Inc. <http://fsf.org/> Everyone permitted copy distribute verbatim copies license document, changing allowed. version GNU Lesser General Public License incorporates terms conditions version 3 GNU General Public License, supplemented additional permissions listed .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_0-additional-definitions","dir":"","previous_headings":"","what":"0. Additional Definitions","title":"GNU Lesser General Public License","text":"used herein, “License” refers version 3 GNU Lesser General Public License, “GNU GPL” refers version 3 GNU General Public License. “Library” refers covered work governed License, Application Combined Work defined . “Application” work makes use interface provided Library, otherwise based Library. Defining subclass class defined Library deemed mode using interface provided Library. “Combined Work” work produced combining linking Application Library. particular version Library Combined Work made also called “Linked Version”. “Minimal Corresponding Source” Combined Work means Corresponding Source Combined Work, excluding source code portions Combined Work , considered isolation, based Application, Linked Version. “Corresponding Application Code” Combined Work means object code /source code Application, including data utility programs needed reproducing Combined Work Application, excluding System Libraries Combined Work.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_1-exception-to-section-3-of-the-gnu-gpl","dir":"","previous_headings":"","what":"1. Exception to Section 3 of the GNU GPL","title":"GNU Lesser General Public License","text":"may convey covered work sections 3 4 License without bound section 3 GNU GPL.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_2-conveying-modified-versions","dir":"","previous_headings":"","what":"2. Conveying Modified Versions","title":"GNU Lesser General Public License","text":"modify copy Library, , modifications, facility refers function data supplied Application uses facility (argument passed facility invoked), may convey copy modified version: ) License, provided make good faith effort ensure , event Application supply function data, facility still operates, performs whatever part purpose remains meaningful, b) GNU GPL, none additional permissions License applicable copy.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_3-object-code-incorporating-material-from-library-header-files","dir":"","previous_headings":"","what":"3. Object Code Incorporating Material from Library Header Files","title":"GNU Lesser General Public License","text":"object code form Application may incorporate material header file part Library. may convey object code terms choice, provided , incorporated material limited numerical parameters, data structure layouts accessors, small macros, inline functions templates (ten fewer lines length), following: ) Give prominent notice copy object code Library used Library use covered License. b) Accompany object code copy GNU GPL license document.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_4-combined-works","dir":"","previous_headings":"","what":"4. Combined Works","title":"GNU Lesser General Public License","text":"may convey Combined Work terms choice , taken together, effectively restrict modification portions Library contained Combined Work reverse engineering debugging modifications, also following: ) Give prominent notice copy Combined Work Library used Library use covered License. b) Accompany Combined Work copy GNU GPL license document. c) Combined Work displays copyright notices execution, include copyright notice Library among notices, well reference directing user copies GNU GPL license document. d) one following: 0) Convey Minimal Corresponding Source terms License, Corresponding Application Code form suitable , terms permit, user recombine relink Application modified version Linked Version produce modified Combined Work, manner specified section 6 GNU GPL conveying Corresponding Source. 1) Use suitable shared library mechanism linking Library. suitable mechanism one () uses run time copy Library already present user’s computer system, (b) operate properly modified version Library interface-compatible Linked Version. e) Provide Installation Information, otherwise required provide information section 6 GNU GPL, extent information necessary install execute modified version Combined Work produced recombining relinking Application modified version Linked Version. (use option 4d0, Installation Information must accompany Minimal Corresponding Source Corresponding Application Code. use option 4d1, must provide Installation Information manner specified section 6 GNU GPL conveying Corresponding Source.)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_5-combined-libraries","dir":"","previous_headings":"","what":"5. Combined Libraries","title":"GNU Lesser General Public License","text":"may place library facilities work based Library side side single library together library facilities Applications covered License, convey combined library terms choice, following: ) Accompany combined library copy work based Library, uncombined library facilities, conveyed terms License. b) Give prominent notice combined library part work based Library, explaining find accompanying uncombined form work.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_6-revised-versions-of-the-gnu-lesser-general-public-license","dir":"","previous_headings":"","what":"6. Revised Versions of the GNU Lesser General Public License","title":"GNU Lesser General Public License","text":"Free Software Foundation may publish revised /new versions GNU Lesser General Public License time time. new versions similar spirit present version, may differ detail address new problems concerns. version given distinguishing version number. Library received specifies certain numbered version GNU Lesser General Public License “later version” applies , option following terms conditions either published version later version published Free Software Foundation. Library received specify version number GNU Lesser General Public License, may choose version GNU Lesser General Public License ever published Free Software Foundation. Library received specifies proxy can decide whether future versions GNU Lesser General Public License shall apply, proxy’s public statement acceptance version permanent authorization choose version Library.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/articles/callbacks.html","id":"building-blocks","dir":"Articles","previous_headings":"","what":"Building Blocks","title":"Custom Callbacks","text":"heart callback mechanism three R6 classes: CallbackSet contains methods executed different stages training loop TorchCallback wraps CallbackSet annotates meta information, including ParamSet ContextTorch defines information training process CallbackSet access using predefined callbacks, one usually interacts TorchCallback class, constructed using t_clbk(<id>): wrapped CallbackSet accessible via field $generator, can create new instance calling $new() Within training loop torch model, different stages exist callback can execute code. , list stages available, also described documentation CallbackSet. history callback, see runs code beginning, validation starts, end epoch.","code":"tc_hist = t_clbk(\"history\") tc_hist ## <TorchCallback:history> History ## * Generator: CallbackSetHistory ## * Parameters: list() ## * Packages: mlr3torch,torch cbs_hist = tc_hist$generator$new() mlr_reflections$torch$callback_stages ##  [1] \"on_begin\"             \"on_epoch_begin\"       \"on_batch_begin\"       ##  [4] \"on_after_backward\"    \"on_batch_end\"         \"on_before_valid\"      ##  [7] \"on_batch_valid_begin\" \"on_batch_valid_end\"   \"on_valid_end\"         ## [10] \"on_epoch_end\"         \"on_end\"               \"on_exit\" cbs_hist ## <CallbackSetHistory> ## * Stages: on_begin, on_before_valid, on_epoch_end"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/callbacks.html","id":"writing-a-custom-logger","dir":"Articles","previous_headings":"","what":"Writing a Custom Logger","title":"Custom Callbacks","text":"order define custom callback, can use helper torch_callback() helper function. example, create custom logger keeps track exponential moving average train loss prints end every epoch. callback takes one argument alpha smoothing parameter store moving average self$moving_loss. value alpha can later configured Learner. , implement main logic callback using two stages: on_batch_end(), called optimizer updates network parameters using mini-batch. , access last loss via ContextTorch’s (accessible via self$ctx) $last_loss field update value self$moving_loss. on_before_valid(), run validation loop. point simply print exponential moving average training loss. Finally, order make final value moving average accessible Learner training finishes, implement state_dict() method return value load_state_dict(state_dict) method, takes previously retrieved state dict sets callback. Note access final $moving_loss via learner, otherwise implement methods. created TorchCallback object associated CallbackSet R6-class us:","code":"custom_logger = torch_callback(\"custom_logger\",   initialize = function(alpha = 0.1) {     self$alpha = alpha     self$moving_loss = NULL   },   on_batch_end = function() {     if (is.null(self$moving_training_loss)) {       self$moving_loss = self$ctx$last_loss     } else {       self$moving_loss = self$alpha * self$last_loss + (1 - self$alpha) * self$moving_loss     }   },   on_before_valid = function() {     cat(sprintf(\"Epoch %s: %.2f\\n\", self$ctx$epoch, self$moving_loss))   },   state_dict = function() {     self$moving_loss   },   load_state_dict = function(state_dict) {     self$moving_loss = state_dict   } ) custom_logger ## <TorchCallback:custom_logger> Custom_logger ## * Generator: CallbackSetCustom_logger ## * Parameters: list() ## * Packages: mlr3torch,torch custom_logger$generator ## <CallbackSetCustom_logger> object generator ##   Inherits from: <inherit> ##   Public: ##     initialize: function (alpha = 0.1)  ##     state_dict: function ()  ##     load_state_dict: function (state_dict)  ##     on_before_valid: function ()  ##     on_batch_end: function ()  ##   Parent env: <environment: 0x55c7a37847b0> ##   Locked objects: FALSE ##   Locked class: FALSE ##   Portable: TRUE"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/callbacks.html","id":"using-the-custom-logger","dir":"Articles","previous_headings":"","what":"Using the Custom Logger","title":"Custom Callbacks","text":"order showcase effects, create train simple multi layer perceptron pass callback. set smoothing parameter alpha 0.05. information returnede state_dict() now accessible via Learner’s $model-slot:","code":"task = tsk(\"iris\") mlp = lrn(\"classif.mlp\",   callbacks = custom_logger, cb.custom_logger.alpha = 0.05,   epochs = 5, batch_size = 64, neurons = 20)  mlp$train(task) ## Epoch 1: 1.22 ## Epoch 2: 1.27 ## Epoch 3: 1.48 ## Epoch 4: 1.07 ## Epoch 5: 1.09 mlp$model$callbacks$custom_logger ## [1] 1.090038"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"quickstart","dir":"Articles","previous_headings":"","what":"Quickstart","title":"Get Started","text":"vignette show get started mlr3torch training simple neural network tabular regression problem. assume familiar mlr3 framework, see e.g. mlr3 book. first example, train simple multi-layer perceptron (MLP) well-known “mtcars” task, goal predict miles per galleon (‘mpg’) cars. architecture comes predfined learner mlr3torch, can also easily create new network architectures, see Neural Networks Graphs vignette detailed introduoduion. first set seed reproducibility, load library construct task. Learners mlr3torch work similary mlr3 learners. , construct simple multi layer perceptron regression. usual calling lrn() configuring parameters: use two hidden layers 50 neurons, training, set batch size 32, number training epochs 30 device \"cpu\". complete description available parameters see ?mlr3torch::LearnerTorchMLP. can use learner training prediction just like regression learner. , split observations training test set, train learner training set create predictions test set. Finally, compute mean squared error predictions.","code":"set.seed(314) library(mlr3torch) task = tsk(\"mtcars\") task$head() #>      mpg    am  carb   cyl  disp  drat  gear    hp  qsec    vs    wt #>    <num> <num> <num> <num> <num> <num> <num> <num> <num> <num> <num> #> 1:  21.0     1     4     6   160  3.90     4   110 16.46     0 2.620 #> 2:  21.0     1     4     6   160  3.90     4   110 17.02     0 2.875 #> 3:  22.8     1     1     4   108  3.85     4    93 18.61     1 2.320 #> 4:  21.4     0     1     6   258  3.08     3   110 19.44     1 3.215 #> 5:  18.7     0     2     8   360  3.15     3   175 17.02     0 3.440 #> 6:  18.1     0     1     6   225  2.76     3   105 20.22     1 3.460 mlp = lrn(\"regr.mlp\",   # architecture parameters   neurons = c(50, 50),   # training arguments   batch_size = 32, epochs = 30, device = \"cpu\" ) mlp #> <LearnerTorchMLP[regr]:regr.mlp>: My Little Powny #> * Model: - #> * Parameters: epochs=30, device=cpu, num_threads=1, #>   num_interop_threads=1, seed=random, eval_freq=1, #>   measures_train=<list>, measures_valid=<list>, patience=0, #>   min_delta=0, batch_size=32, neurons=50,50, p=0.5, #>   activation=<nn_relu>, activation_args=<list> #> * Validate: NULL #> * Packages: mlr3, mlr3torch, torch #> * Predict Types:  [response] #> * Feature Types: integer, numeric, lazy_tensor #> * Properties: internal_tuning, marshal, validation #> * Optimizer: adam #> * Loss: mse #> * Callbacks: - # Split the obersevations into training and test set splits = partition(task) # Train the learner on the train set mlp$train(task, row_ids = splits$train) # Predict the test set prediction = mlp$predict(task, row_ids = splits$test) # Compute the mse prediction$score(msr(\"regr.mse\")) #> regr.mse  #>  283.838"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"configuring-a-learner","dir":"Articles","previous_headings":"","what":"Configuring a Learner","title":"Get Started","text":"Although torch learners quite like mlr3 learners, differences. One LearnerTorch classes construction arguments, .e. torch learners modular learners. learners free implement construction arguments, common torch learners, namely loss, optimizer callbacks. object can parameters included LearnerTorch’s parameter set. previous example, specify explicitly used default values, Adam optimizer, MSE loss callbacks. now show configure three aspects learner mlr3torch::TorchOptimizer, mlr3torch::TorchLoss, mlr3torch::TorchCallback classes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"loss","dir":"Articles","previous_headings":"Configuring a Learner","what":"Loss","title":"Get Started","text":"loss function, also known objective function cost function, measures discrepancy predicted output true output. quantifies well model performing training. R package torch, underpins mlr3torch framework, already provides number predefined loss functions Mean Squared Error (nn_mse_loss), Mean Absolute Error (nn_l1_loss), cross entropy loss (nn_cross_entropy_loss). mlr3torch, represent loss functions using mlr3torch::TorchLoss class. provides thin wrapper around torch loss functions annotates meta information, importantly paradox::ParamSet allows configure loss function. object can constructed using t_loss(<key>). , construct L1 loss function, also known Mean Absolute Error (MAE). printed output informs us wrapped loss function (nn_l1_loss), configured parameters, packages depends task types can used. ParamSet contains one parameter, namely reduction, specifies loss reduced batch. wrapped loss module generator accessible slot $generator. can pass TorchLoss argument loss initialization learner. parameters loss added learner’s ParamSet, prefixed \"loss.\". predefined loss functions stored mlr3torch_losses dictionary, can retrieved using t_loss(<key>).","code":"l1 = t_loss(\"l1\") l1 #> <TorchLoss:l1> Absolute Error #> * Generator: nn_l1_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # the paradox::ParamSet of the loss l1$param_set #> <ParamSet(1)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <num> <num>   <num>  <list> <list> #> 1: reduction ParamFct    NA    NA       2    mean [NULL] l1$generator #> <nn_l1_loss> object generator #>   Inherits from: <inherit> #>   Public: #>     .classes: nn_l1_loss nn_loss nn_module #>     initialize: function (reduction = \"mean\")  #>     forward: function (input, target)  #>     clone: function (deep = FALSE, ..., replace_values = TRUE)  #>   Private: #>     .__clone_r6__: function (deep = FALSE)  #>   Parent env: <environment: 0x556f0dd5ea18> #>   Locked objects: FALSE #>   Locked class: FALSE #>   Portable: TRUE mlp_l1 = lrn(\"regr.mlp\", loss = l1) mlp_l1$param_set$values$loss.reduction #> NULL mlr3torch_losses #> <DictionaryMlr3torchLosses> with 3 stored values #> Keys: cross_entropy, l1, mse"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"optimizer","dir":"Articles","previous_headings":"Configuring a Learner","what":"Optimizer","title":"Get Started","text":"optimizer determines model’s weights updated based calculated loss. adjusts parameters model minimize loss function, optimizing model’s performance. Optimizers work analogous loss functions, .e. mlr3torch provides thin wrapper – TorchOptimizer class – around optimizers Adam (optim_adam) SGD (optim_sgd). TorchLoss objects can constructed using t_opt(<key>). optimizers, associated ParamSet interesting see : wrapped torch optimizer can accessed slot generator. Parameters TorchOptimizer (also TorchLoss TorchCallback) can set usual mlr3 way, .e. either construction, afterwards using $set_values() method parameter set. see optimizer’s parameters added learner’s ParamSet (prefixed \"opt.\") values set values specified. exposing optimizer’s parameters, can conveniently tuned using mlr3tuning. available optimizers stored mlr3torch_optimizers dictionary.","code":"sgd = t_opt(\"sgd\") sgd #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: list() #> * Packages: torch,mlr3torch  sgd$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamDbl     0   Inf     Inf <NoDefault[0]> [NULL] #> 2:     momentum ParamDbl     0     1     Inf              0 [NULL] #> 3:    dampening ParamDbl     0     1     Inf              0 [NULL] #> 4: weight_decay ParamDbl     0     1     Inf              0 [NULL] #> 5:     nesterov ParamLgl    NA    NA       2          FALSE [NULL] sgd$param_set$set_values(   lr = 0.5, # increase learning rate   nesterov = FALSE # no nesterov momentum ) mlp_sgd = lrn(\"regr.mlp\", optimizer = sgd) as.data.table(mlp_sgd$param_set)[   startsWith(id, \"opt.\")][[1L]] #> [1] \"opt.lr\"           \"opt.momentum\"     \"opt.dampening\"    \"opt.weight_decay\" #> [5] \"opt.nesterov\" mlp_sgd$param_set$values[c(\"opt.lr\", \"opt.nesterov\")] #> $opt.lr #> [1] 0.5 #>  #> $opt.nesterov #> [1] FALSE mlr3torch_optimizers #> <DictionaryMlr3torchOptimizers> with 7 stored values #> Keys: adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"callbacks","dir":"Articles","previous_headings":"Configuring a Learner","what":"Callbacks","title":"Get Started","text":"third important configuration option callbacks allow customize training process. allows saving model checkpoints, logging metrics, implementing custom functionality specific training scenarios. tutorial implement custom callback, see Custom Callbacks vignette. , show use predefined callbacks. , retrieve \"history\" callback using t_clbk(), parameters merely saves training validation history learner can accessed afterwards. wanted learn callback , can access help page wrapped object using $help() method. Note also possible loss optimizer. predefined callbacks stored mlr3torch_callbacks dictionary.","code":"history = t_clbk(\"history\") history #> <TorchCallback:history> History #> * Generator: CallbackSetHistory #> * Parameters: list() #> * Packages: mlr3torch,torch history$help() mlr3torch_callbacks #> <DictionaryMlr3torchCallbacks> with 4 stored values #> Keys: checkpoint, history, progress, tb"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"putting-it-together","dir":"Articles","previous_headings":"Configuring a Learner","what":"Putting it Together","title":"Get Started","text":"now define customized MLP learner using loss, optimizer callback just covered. keep track performance, use 30% training data validation evaluate using MAE Measure. Note mearures_valid measures_train parameters LearnerTorch take common mlr3::Measures, whereas loss function must TorchLoss. now train learner “mtcars” task use train-test split . make predictions unseen test data compare scores. directly optimized L1 (aka MAE) loss tweaked learning rate, configured mlp_custom learner lower MAE default mlp learner. configured learner use history callback, can find validation history $model slot: plot shows epochs 6 30.  important information stored Learner’s model $network, underlying nn_module. full description model, see ?LearnerTorch.","code":"mlp_custom = lrn(\"regr.mlp\",   # construction arguments   optimizer = sgd, loss = l1, callbacks = history,   # scores to keep track of   measures_valid = msr(\"regr.mae\"),   # other parameters are left as-is:   # architecture   neurons = c(50, 50),   # training arguments   batch_size = 32, epochs = 30, device = \"cpu\",   # validation proportion   validate = 0.3 )  mlp_custom #> <LearnerTorchMLP[regr]:regr.mlp>: My Little Powny #> * Model: - #> * Parameters: epochs=30, device=cpu, num_threads=1, #>   num_interop_threads=1, seed=random, eval_freq=1, #>   measures_train=<list>, measures_valid=<MeasureRegrSimple>, #>   patience=0, min_delta=0, batch_size=32, neurons=50,50, p=0.5, #>   activation=<nn_relu>, activation_args=<list>, opt.lr=0.5, #>   opt.nesterov=FALSE #> * Validate: 0.3 #> * Packages: mlr3, mlr3torch, torch #> * Predict Types:  [response] #> * Feature Types: integer, numeric, lazy_tensor #> * Properties: internal_tuning, marshal, validation #> * Optimizer: sgd #> * Loss: l1 #> * Callbacks: history mlp_custom$train(task, row_ids = splits$train) prediction_custom = mlp_custom$predict(task, row_ids = splits$test) prediction_custom$score(msr(\"regr.mae\")) #> regr.mae  #> 7.122375 prediction$score(msr(\"regr.mae\")) #> regr.mae  #> 15.27983 head(mlp_custom$model$callbacks$history$valid) #>    epoch     regr.mae #>    <num>        <num> #> 1:     1 1.777395e+04 #> 2:     2 3.955504e+08 #> 3:     3 1.143863e+04 #> 4:     4 1.792927e+01 #> 5:     5 1.759594e+01 #> 6:     6 1.726260e+01"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"a-torch-primer","dir":"Articles","previous_headings":"","what":"A torch Primer","title":"Internals","text":"start sampling input tensor: 2 batches 3 features: nn_module constructed nn_module_generator. nn_linear one simpler generators: Applying module gives 2-batch 4 units: neural network one (4-unit) hidden layer two outputs needs following ingredients can pipe tensor layers follows. now continue showing neural network can represented mlr3torch.","code":"input = torch_randn(2, 3) input #> torch_tensor #>  1.0799 -2.3132 -0.0767 #> -1.0029 -0.4515  0.3045 #> [ CPUFloatType{2,3} ] module_1 = nn_linear(in_features = 3, out_features = 4, bias = TRUE) output = module_1(input) output #> torch_tensor #>  0.9228 -1.1292 -0.5401  0.8716 #>  0.1488 -0.7452 -0.0094  0.2273 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ] activation = nn_sigmoid() module_2 = nn_linear(4, 3, bias = TRUE) softmax = nn_softmax(2) output = module_1(input) output = activation(output) output = module_2(output) output = softmax(output) output #> torch_tensor #>  0.2143  0.2882  0.4975 #>  0.2435  0.2812  0.4753 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"neural-networks-as-graphs","dir":"Articles","previous_headings":"","what":"Neural Networks as Graphs","title":"Internals","text":"mlr3torch, nn_modules wrapped PipeOpModule. advantage network structure can represented mlr3pipelines::Graph object made explicit (can plotted, can extended manipulated), compared e.g. writing function pipes input series modules. PipeOpModule can used wrap module directly, usually constructed PipeOpTorch (see later). typically single input single output, although multiple inputs possible (module called multiple arguments), multiple outputs possible module-function returns list. input output channels must explicitly declared construction. now continue recreate network using PipeOpModules. can wrap linear module_1 layers like : Note po(\"module_1\") equivalent po(\"module\", id = \"module_1\"). mechanism convenient avoid ID clashes graphs contain PipeOp multiple times. can use generated PipeOp familiar way: Note use $train(), since torch modules anything maps state (filled empty list). single hidden layer neural network can constructed Graph, can training . can now use graph’s $train() method pipe tensor whole Graph. object allows easily perform forward pass, inherit nn_module, useful various reasons. Instead class inherits nn_module Graph (work R6, since multiple inheritance available), class inherits nn_module contains Graph member slot composition. class nn_graph. constructed Graph, well information shape(s) torch_tensor(s) expects inputs. Shape info communicated integer-valued numeric vector; dimensions arbitrary, e.g. batch-size, given NA. network expects input shape c(NA, 3), since first layer created nn_linear(in_features = 3, ...). Graph multiple outputs, also possible select subset outputs use, change output order, giving output_map argument. module gives us convenience torch nn_module objects, e.g.: can used transform tensors just torch::nn_module:","code":"library(mlr3torch) po_module_1 = po(\"module_1\", module = module_1) output = po_module_1$train(list(input))[[1]] output #> torch_tensor #>  0.9228 -1.1292 -0.5401  0.8716 #>  0.1488 -0.7452 -0.0094  0.2273 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ] po_activation = po(\"module\", id = \"activation\", activation) po_module_2 = po(\"module_2\", module = module_2) po_softmax = po(\"module\", id = \"softmax\", module = softmax)  module_graph = po_module_1 %>>%   po_activation %>>%   po_module_2 %>>%   po_softmax  module_graph$plot(html = TRUE) output = module_graph$train(input)[[1]] output #> torch_tensor #>  0.2143  0.2882  0.4975 #>  0.2435  0.2812  0.4753 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ] # the name of the single input is: module_graph$input #>              name        train predict    op.id channel.name #>            <char>       <char>  <char>   <char>       <char> #> 1: module_1.input torch_tensor    NULL module_1        input  graph_module = nn_graph(   module_graph,   shapes_in = list(module_1.input = c(NA, 3)) ) graph_module$children #> $module_list #> An `nn_module` containing 31 parameters. #>  #> ── Modules ───────────────────────────────────────────────────────────────────── #> • 0: <nn_linear> #16 parameters #> • 1: <nn_sigmoid> #0 parameters #> • 2: <nn_linear> #15 parameters #> • 3: <nn_softmax> #0 parameters graph_module(input) #> torch_tensor #>  0.2143  0.2882  0.4975 #>  0.2435  0.2812  0.4753 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ]"},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"modeldescriptor","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"ModelDescriptor","title":"Internals","text":"PipeOpModule represents nn_module fixed specific tensor shape hyperparameters. constructing neural network using operators, one take care output shape operations match input shapes following operations. complete Graph matching PipeOpModules can constructed using operators mostly inherit PipeOpTorch, making use ModelDescriptor class. ModelDescriptor class contains Graph (mostly) PipeOpModules information. PipeOpTorch transforms ModelDescriptor adds PipeOpModules Graph. ModelDescriptors always build Graph specific Task. easiest way initialize proper ModelDescriptor use appropriate PipeOpTorchIngress given datatype. use PipeOpTorchIngressNumeric, used numeric data. ModelDescriptor S3 object contains Graph, information generate data ($ingress $task), tags build model unrelated architecture ($optimizer, $loss $callbacks) well information necessary extend graph along given output ($pointer $pointer_shape). $pointer identifies output $graph PipeOpTorch extend. Piping ModelDescriptor PipeOpTorchLinear, example, adds PipeOpModule wrapping torch::nn_linear. $pointer now updated identify output PipeOpModule, $pointer_shape shows shape changed 4 units (3 input ). model_descriptor_to_module() function converts nn_graph, functional torch::nn_module.","code":"task = tsk(\"iris\")$select(colnames(iris)[1:3])  po_torch_in = po(\"torch_ingress_num\") md = po_torch_in$train(list(task))[[1]]  md #> <ModelDescriptor: 1 ops> #> * Ingress:  torch_ingress_num.input: [(NA,3)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  N/A #> * Loss:  N/A #> * pointer:  torch_ingress_num.output [(NA,3)] unclass(md) #> $graph #> Graph with 1 PipeOps: #>                 ID         State sccssors prdcssors #>             <char>        <char>   <char>    <char> #>  torch_ingress_num <<UNTRAINED>>                    #>  #> $ingress #> $ingress$torch_ingress_num.input #> Ingress: Task[Petal.Length,Sepal.Length,Sepal.Width] --> Tensor(NA, 3) #>  #>  #> $task #> <TaskClassif:iris> (150 x 4): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (3): #>   - dbl (3): Petal.Length, Sepal.Length, Sepal.Width #>  #> $optimizer #> NULL #>  #> $loss #> NULL #>  #> $callbacks #> named list() #>  #> $pointer #> [1] \"torch_ingress_num\" \"output\"            #>  #> $pointer_shape #> [1] NA  3 po_torch_linear = po(\"nn_linear\", out_features = 4) md = po_torch_linear$train(list(md))[[1]]  md$graph #> Graph with 2 PipeOps: #>                 ID         State  sccssors         prdcssors #>             <char>        <char>    <char>            <char> #>  torch_ingress_num <<UNTRAINED>> nn_linear                   #>          nn_linear <<UNTRAINED>>           torch_ingress_num md$pointer #> [1] \"nn_linear\" \"output\" md$pointer_shape #> [1] NA  4 small_module = model_descriptor_to_module(md, list(md$pointer))  small_module(input) #> torch_tensor #> -0.3948  0.0355 -1.1148  0.9649 #> -0.3343 -1.2209  0.1999  0.1987 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"using-modeldescriptor-to-get-data","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"Using ModelDescriptor to get Data","title":"Internals","text":"ModelDescriptor represent Graph nn_module created, also way Task processed get input batches. torch::dataset can created calling task_dataset(); task feature_ingress_tokens arguments can retrieved ModelDescriptor. target_batchgetter needs created extra (necessary), since depends ultimate machine learning model, looked far. Use $.getbatch() method get batch can given nn_module. Note $x $y slot, latter used, account possible target batches. $x slot also list, since able handle NNs multiple inputs (see ).","code":"td = task_dataset(   task = md$task,   feature_ingress_tokens = md$ingress,   device = \"cpu\" )  td #> <dataset> #>   Public: #>     .getbatch: function (index)  #>     .getitem: function (index)  #>     .length: function ()  #>     all_features: Petal.Length Sepal.Length Sepal.Width Species #>     cache_lazy_tensors: FALSE #>     clone: function (deep = FALSE)  #>     device: cpu #>     feature_ingress_tokens: list #>     initialize: function (task, feature_ingress_tokens, target_batchgetter = NULL,  #>     load_state_dict: function (x, ..., .refer_to_state_dict = FALSE)  #>     state_dict: function ()  #>     target_batchgetter: NULL #>     task: TaskClassif, TaskSupervised, Task, R6 batch = td$.getbatch(1:3) batch #> $x #> $x$torch_ingress_num.input #> torch_tensor #>  1.4000  5.1000  3.5000 #>  1.4000  4.9000  3.0000 #>  1.3000  4.7000  3.2000 #> [ CPUFloatType{3,3} ] #>  #>  #> $.index #> torch_tensor #>  1 #>  2 #>  3 #> [ CPULongType{3} ]  small_module(batch$x[[1]]) #> torch_tensor #>  4.6313 -0.4342  2.2361 -0.9563 #>  4.2838 -0.3334  2.0962 -0.8742 #>  4.2675 -0.4344  2.0631 -0.8572 #> [ CPUFloatType{3,4} ][ grad_fn = <AddmmBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"building-sequential-nns","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"Building sequential NNs","title":"Internals","text":"sequential NN can easily implemented follows: Note second nn_linear need informed output dimension first nn_linear, since ModelDescriptor passed along Graph edges knows info ($pointer_shape slot).","code":"graph_generator = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 4, id = \"linear1\") %>>%     po(\"nn_sigmoid\") %>>%   po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%   po(\"nn_softmax\", dim = 2) md_sequential = graph_generator$train(task)[[1]]  graph_module = model_descriptor_to_module(md_sequential, list(md_sequential$pointer))  graph_module(input) #> torch_tensor #>  0.4210  0.2343  0.3447 #>  0.4048  0.2254  0.3698 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"building-more-interesting-nns","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"Building more interesting NNs","title":"Internals","text":"One main features mlr3pipelines ability easily represent computational Graphs. ModelDescriptor / PipeOpTorch setup built make full use functionality. possible multiple inputs NN using multiple PipeOpTorchIngress inputs, possible parallel alternative path branching, possible multiple outputs. Consider following (bit nonsensical) network operates differently \"Petal\" \"Sepal\" features tsk(\"iris\") manually split task , shown wholly integrated mlr3pipelines pipeline can automatically. can use create neural network iris tasks created . set $keep_results debug flag can inspection happening: make multiple observations : can observe ModelDescriptor grows passed along edges graph_iris. Note $graph slot ModelDescriptor often updated -reference, time inspect intermediate results, may contain complete graph. However, see $ingress, $pointer $pointer_shape ModelDescriptors take sepal.-path differ ones take petal.-path: po(\"nn_merge_cat\") unites two ModelDescriptors contains common ingress. pointer_shape now reflects output “cat”-operation: 2nd dimension added : Multiple ModelDescriptors created, since graph_iris multiple outpus. makes possible create neural network multiple outputs. need unite outputs graph_iris using model_descriptor_union() can pass model_descriptor_to_module(). need collect output_pointers separately. parameter list_output must set TRUE since module multiple outputs. PipeOpBranch disappears resulting Graph PipeOpModule iris_module. PipeOpTorchs graph_iris add anything ModelDescriptors. branch interpeted graph_iris runs, nn_relu path followed. iris_module therefore contains Graph “relu” activation: ModelDescriptor’s $task slot contains Task features used create input data NN inputs. can given task_dataset(), along $ingress, create torch dataset creates batches. , output graph_iris can used: resulting module multiple inputs multiple outputs. call first two rows iris, set debug $keep_results flag can inspect happening nn_module’s $graph: first linear layer takes “Sepal” input (\"linear1\") creates 2x4 tensor (batch size 2, 4 units), \"linear3\" layer 2x5 output: observe po(\"nn_merge_cat\") concatenates , expected:","code":"iris_petal = tsk(\"iris\")$select(c(\"Petal.Length\", \"Petal.Width\")) iris_sepal = tsk(\"iris\")$select(c(\"Sepal.Length\", \"Sepal.Width\")) graph_sepal = po(\"torch_ingress_num\", id = \"sepal.in\") %>>%   po(\"nn_linear\", out_features = 4, id = \"linear1\")  graph_petal = po(\"torch_ingress_num\", id = \"petal.in\") %>>%   po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%   po(\"nn_tanh\") %>>%   po(\"nn_linear\", out_features = 5, id = \"linear3\")  graph_common = ppl(\"branch\", graphs = list(     sigmoid = po(\"nn_sigmoid\"),     relu = po(\"nn_relu\")   )) %>>%   gunion(list(     po(\"nn_linear\", out_features = 1, id = \"lin_out\"),     po(\"nn_linear\", out_features = 3, id = \"cat_out\") %>>%       po(\"nn_softmax\", dim = 2)   ))   graph_iris = gunion(list(graph_sepal, graph_petal)) %>>%   po(\"nn_merge_cat\") %>>%   graph_common  graph_iris$plot(html = TRUE) graph_iris$param_set$values$branch.selection = \"relu\"  graph_iris$keep_results = TRUE  iris_mds = graph_iris$train(   input = list(sepal.in.input = iris_sepal, petal.in.input = iris_petal),   single_input = FALSE )  iris_mds #> $lin_out.output #> <ModelDescriptor: 11 ops> #> * Ingress:  sepal.in.input: [(NA,2)], petal.in.input: [(NA,2)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  N/A #> * Loss:  N/A #> * pointer:  lin_out.output [(NA,1)] #>  #> $nn_softmax.output #> <ModelDescriptor: 11 ops> #> * Ingress:  sepal.in.input: [(NA,2)], petal.in.input: [(NA,2)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  N/A #> * Loss:  N/A #> * pointer:  nn_softmax.output [(NA,3)] # sepal.in path graph_iris$pipeops$linear1$.result[[1]]$ingress #> $sepal.in.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) graph_iris$pipeops$linear1$.result[[1]]$pointer #> [1] \"linear1\" \"output\" graph_iris$pipeops$linear1$.result[[1]]$pointer_shape #> [1] NA  4  # petal.in path graph_iris$pipeops$linear3$.result[[1]]$ingress #> $petal.in.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) graph_iris$pipeops$linear3$.result[[1]]$pointer #> [1] \"linear3\" \"output\" graph_iris$pipeops$linear3$.result[[1]]$pointer_shape #> [1] NA  5 graph_iris$pipeops$nn_merge_cat$.result[[1]]$ingress #> $sepal.in.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) #>  #> $petal.in.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) graph_iris$pipeops$nn_merge_cat$.result[[1]]$pointer_shape #> [1] NA  9 iris_mds_union = model_descriptor_union(iris_mds[[1]], iris_mds[[2]]) output_pointers = list(iris_mds[[1]]$pointer, iris_mds[[2]]$pointer) output_pointers #> [[1]] #> [1] \"lin_out\" \"output\"  #>  #> [[2]] #> [1] \"nn_softmax\" \"output\" iris_module = model_descriptor_to_module(iris_mds_union, output_pointers, list_output = TRUE) iris_module$graph$plot(html = TRUE) iris_mds_union$task  # contains all features #> <TaskClassif:iris> (150 x 5): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (4): #>   - dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Width  iris_td = task_dataset(   task = iris_mds_union$task,   feature_ingress_tokens = iris_mds_union$ingress,   device = \"cpu\" )  batch = iris_td$.getbatch(1:2) batch #> $x #> $x$sepal.in.input #> torch_tensor #>  5.1000  3.5000 #>  4.9000  3.0000 #> [ CPUFloatType{2,2} ] #>  #> $x$petal.in.input #> torch_tensor #>  1.4000  0.2000 #>  1.4000  0.2000 #> [ CPUFloatType{2,2} ] #>  #>  #> $.index #> torch_tensor #>  1 #>  2 #> [ CPULongType{2} ] iris_module$graph$keep_results = TRUE  iris_module(   sepal.in.input = batch$x$sepal.in.input,   petal.in.input = batch$x$petal.in.input ) #> $lin_out.output #> torch_tensor #> -1.1582 #> -1.0255 #> [ CPUFloatType{2,1} ][ grad_fn = <AddmmBackward0> ] #>  #> $nn_softmax.output #> torch_tensor #>  0.9275  0.0577  0.0148 #>  0.9175  0.0639  0.0187 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ] iris_module$graph$pipeops$linear1$.result #> $output #> torch_tensor #>  1.6322  4.9484  4.6457  0.6479 #>  1.5350  4.6447  4.3141  0.4427 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ] iris_module$graph$pipeops$linear3$.result #> $output #> torch_tensor #>  0.1811 -0.2108  0.3277  0.3537  0.0080 #>  0.1811 -0.2108  0.3277  0.3537  0.0080 #> [ CPUFloatType{2,5} ][ grad_fn = <AddmmBackward0> ] iris_module$graph$pipeops$nn_merge_cat$.result #> $output #> torch_tensor #>  1.6322  4.9484  4.6457  0.6479  0.1811 -0.2108  0.3277  0.3537  0.0080 #>  1.5350  4.6447  4.3141  0.4427  0.1811 -0.2108  0.3277  0.3537  0.0080 #> [ CPUFloatType{2,9} ][ grad_fn = <CatBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"building-torch-learners","dir":"Articles","previous_headings":"","what":"Building Torch Learners","title":"Internals","text":"now seen NN Graphs PipeOpModule created turned nn_modules. Using PipeOpTorch even creates ModelDescriptor objects contain additional info batch tensors extracted Tasks. complete Learner, still necessary define loss-function used optimization, optimizer, optionally callbacks. already covered class representations – TorchLoss, TorchOptimizer, TorchCallbacks, Get Started vignette. use adam optimizer, cross-entropy loss function, history callback.","code":"adam = t_opt(\"adam\", lr = 0.02) adam #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: lr=0.02 #> * Packages: torch,mlr3torch  xe = t_loss(\"cross_entropy\") xe #> <TorchLoss:cross_entropy> Cross Entropy #> * Generator: nn_cross_entropy_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: classif  history = t_clbk(\"history\") history #> <TorchCallback:history> History #> * Generator: CallbackSetHistory #> * Parameters: list() #> * Packages: mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"learnertorchmodel","dir":"Articles","previous_headings":"Building Torch Learners","what":"LearnerTorchModel","title":"Internals","text":"LearnerTorchModel represents supervised model (regression classification) using torch NNs. needs nn_module, well list TorchIngressToken define batches created Task. TorchIngressToken hard-code column-names Task used data-input, Learner created like therefore works specific Task created. (Generally full mlr3pipelines-UI used problem, see .) following uses sequential NN : training model, set hyperparameters. following calls $predict_newdata function plot response surface along Sepal.Width = mean(Sepal.Width) plane, along ground-truth values:","code":"lr_sequential = lrn(\"classif.torch_model\",   task_type = \"classif\",   network = model_descriptor_to_module(md_sequential, list(md_sequential$pointer)),   ingress_tokens = md_sequential$ingress,   optimizer = adam,   callbacks = history,   loss = xe )  lr_sequential #> <LearnerTorchModel[classif]:classif.model>: Torch Model #> * Model: - #> * Parameters: device=auto, num_threads=1, num_interop_threads=1, #>   seed=random, eval_freq=1, measures_train=<list>, #>   measures_valid=<list>, patience=0, min_delta=0, opt.lr=0.02 #> * Validate: NULL #> * Packages: mlr3, mlr3torch, torch #> * Predict Types:  [response], prob #> * Feature Types: logical, integer, numeric, character, factor, ordered, #>   POSIXct, lazy_tensor #> * Properties: featureless, hotstart_backward, hotstart_forward, #>   importance, internal_tuning, loglik, marshal, missings, multiclass, #>   oob_error, selected_features, twoclass, validation, weights #> * Optimizer: adam #> * Loss: cross_entropy #> * Callbacks: history lr_sequential$param_set$set_values(   batch_size = 50,   epochs = 100,   measures_train = msrs(c(\"classif.logloss\", \"classif.ce\")) ) # This is required to evaluate the logloss during training lr_sequential$predict_type = \"prob\"  lr_sequential$train(md_sequential$task) library(data.table) library(ggplot2)  newdata = cbind(data.table(Sepal.Width = mean(iris$Sepal.Width)), CJ(   Sepal.Length = seq(min(iris$Sepal.Length), max(iris$Sepal.Length), length.out = 30),   Petal.Length = seq(min(iris$Petal.Length), max(iris$Petal.Length), length.out = 30) ))  predictions = lr_sequential$predict_newdata(newdata)  plot_predictions = function(predictions) {   ggplot(cbind(newdata, Species = predictions$response),       aes(x = Sepal.Length, y = Petal.Length, fill = Species)) +     geom_tile(alpha = .3) +     geom_tile(alpha = .3) +     geom_point(data = iris,       aes(x = Sepal.Length, y = Petal.Length, fill = Species),       color = \"black\", pch = 21, size = 3) +     theme_bw() } plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"torch-learner-pipelines","dir":"Articles","previous_headings":"","what":"Torch Learner Pipelines","title":"Internals","text":"model shown constructed using ModelDescriptor generated Graph PipeOpTorch operators. ModelDescriptor furthermore contains Task pertains. makes possible use create NN model gets trained right away, using PipeOpTorchModelClassif. missing prerequisite now add desired TorchOptimizer TorchLoss information ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"adding-optimizer-loss-and-callback-meta-info-to-modeldescriptor","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"Adding Optimizer, Loss and Callback Meta-Info to ModelDescriptor","title":"Internals","text":"Remember ModelDescriptor $optimizer, $loss $callbacks slots necessary build complete Learner NN. can set corresponding PipeOpTorch operators. po(\"torch_optimizer\") used set $optimizer slot ModelDescriptor; takes desired TorchOptimizer object construction exports ParamSet. works analogously loss-function. also callbacks:","code":"po_adam = po(\"torch_optimizer\", optimizer = adam)  # hyperparameters are made available and can be changed: po_adam$param_set$values #> $lr #> [1] 0.02  md_sequential = po_adam$train(list(md_sequential))[[1]]  md_sequential$optimizer #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: lr=0.02 #> * Packages: torch,mlr3torch po_xe = po(\"torch_loss\", loss = xe) md_sequential = po_xe$train(list(md_sequential))[[1]]  md_sequential$loss #> <TorchLoss:cross_entropy> Cross Entropy #> * Generator: nn_cross_entropy_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: classif po_history = po(\"torch_callbacks\", callbacks = t_clbk(\"history\")) md_sequential = po_history$train(list(md_sequential))[[1]]  md_sequential$callbacks #> $history #> <TorchCallback:history> History #> * Generator: CallbackSetHistory #> * Parameters: list() #> * Packages: mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"combined-instantiation-and-training-of-learnertorchmodel","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"Combined Instantiation and Training of LearnerTorchModel","title":"Internals","text":"ModelDescriptor can now given po(\"torch_model_classif\"). po(\"torch_model_classif\") behaves similarly PipeOpLearner: returns NULL training, prediction $predict(). po(\"torch_model_classif\") behaves similarly PipeOpLearner: returns NULL training, prediction $predict().","code":"po_model = po(\"torch_model_classif\", batch_size = 50, epochs = 50)  po_model$train(list(md_sequential)) #> $output #> NULL newtask = TaskClassif$new(\"newdata\", cbind(newdata, Species = factor(NA, levels = levels(iris$Species))), target = \"Species\")  predictions = po_model$predict(list(newtask))[[1]]  plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"the-whole-pipeline","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"The whole Pipeline","title":"Internals","text":"Remember md_sequential created using Graph initial Task piped . combine Graph PipeOpTorchModelClassif, get Graph behaves like Graph ends PipeOpLearner, can therefore wrapped GraphLearner. following uses one hidden layer : Compare resulting Graph Graph trained model: Predictions, (can use predict_newdata ):","code":"graph_sequential_full = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 4, id = \"linear1\") %>>%     po(\"nn_sigmoid\") %>>%     po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%     po(\"nn_softmax\", dim = 2, id = \"softmax\") %>>%     po(\"nn_linear\", out_features = 3, id = \"linear3\") %>>%     po(\"nn_softmax\", dim = 2, id = \"softmax2\") %>>%     po(\"torch_optimizer\", optimizer = adam) %>>%     po(\"torch_loss\", loss = xe) %>>%     po(\"torch_callbacks\", callbacks = history) %>>%     po(\"torch_model_classif\", batch_size = 50, epochs = 100) lr_sequential_full = as_learner(graph_sequential_full)  lr_sequential_full$train(task) graph_sequential_full$plot(html = TRUE) model = lr_sequential_full$graph_model$state$torch_model_classif$model model$network$graph$plot(html = TRUE) predictions = lr_sequential_full$predict_newdata(newdata)  plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"mixed-pipelines","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"Mixed Pipelines","title":"Internals","text":"just limited PipeOpTorch kinds Graphs, also limited single PipeOpTorchIngress. following pipeline, example, removes Petal.Length columns Task fits model:  using Petal.Length Sepal.Length separately first?  examples hopefully demonstrated possibilities come representation neural network layers PipeOps. Even though vignette quite technical, hope given -depth understanding underlying mechanisms.","code":"gr = po(\"select\", selector = selector_name(\"Petal.Length\")) %>>%   po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 5, id = \"linear1\") %>>%   po(\"nn_relu\") %>>%   po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%   po(\"nn_softmax\", dim = 2) %>>%   po(\"torch_optimizer\", optimizer = adam) %>>%   po(\"torch_loss\", loss = xe) %>>%   po(\"torch_model_classif\", batch_size = 50, epochs = 50)  gr$plot(html = TRUE) lr = as_learner(gr) lr$train(task) predictions = lr$predict_newdata(newdata)  plot_predictions(predictions) gr = gunion(list(       po(\"select\", selector = selector_name(\"Petal.Length\"), id = \"sel1\") %>>%         po(\"torch_ingress_num\", id = \"ingress.petal\") %>>%         po(\"nn_linear\", out_features = 3, id = \"linear1\"),       po(\"select\", selector = selector_name(\"Sepal.Length\"), id = \"sel2\") %>>%         po(\"torch_ingress_num\", id = \"ingress.sepal\") %>>%         po(\"nn_linear\", out_features = 3, id = \"linear2\")     )) %>>%     po(\"nn_merge_cat\")  %>>%     po(\"nn_relu\", id = \"act1\") %>>%     po(\"nn_linear\", out_features = 3, id = \"linear3\") %>>%     po(\"nn_softmax\", dim = 2, id = \"act3\") %>>%     po(\"torch_optimizer\", optimizer = adam, lr = 0.1) %>>%     po(\"torch_loss\", loss = xe) %>>%     po(\"torch_model_classif\", batch_size = 50, epochs = 50)  gr$plot(html = TRUE) lr = as_learner(gr) lr$train(task) predictions = lr$predict_newdata(newdata)  plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/lazy_tensor.html","id":"creating-a-lazy-tensor","dir":"Articles","previous_headings":"","what":"Creating a Lazy Tensor","title":"Lazy Tensor","text":"Every lazy_tensor built top torch::dataset, assume familiar . information create torch::datasets, recommend reading torch package documentation. additional restriction impose dataset must .getitem .getbatch method returns list named tensors. example, create lazy_tensor length 1000, whose elements drawn uniform distribution \\([0, 1]\\). data stored -memory example, necessary $.getitem() method can e.g. load images disk. order create lazy_tensor mydata, annotate returned shapes dataset passing named list dataset_shapes. first dimension must NA, batch dimension. can also set shape NULL indicate unknown, .e. varies elements. Note case, implemented .getbatch method, even omitted specifying dataset_shapes auto-inferred. can convert vector torch_tensor just like : added preprocessing, calling $.getbatch() method mydata selecting element x. continue creating example task lt, relationship x y variable polynomial. Note target variable, classification regression, lazy_tensor, must factor numeric respectively. , plot data:  next section, create custom PipeOp fit polynomial regression model.","code":"mydata = dataset(   initialize = function() {     self$x = runif(1000, -1, 1)   },   .getbatch = function(i) list(x = torch_tensor(self$x[i])$unsqueeze(2)),   .length = function() 1000 )() lt = as_lazy_tensor(mydata, dataset_shapes = list(x = c(NA, 1))) lt[1:5] #> <ltnsr[5]> #> [1] <tnsr[1]> <tnsr[1]> <tnsr[1]> <tnsr[1]> <tnsr[1]> materialize(lt[1:5], rbind = TRUE) #> torch_tensor #> -0.9852 #> -0.0672 #> -0.0044 #> -0.4205 #>  0.4658 #> [ CPUFloatType{5,1} ] torch_equal(   materialize(lt[1], rbind = TRUE),   mydata$.getbatch(1)$x ) #> [1] TRUE library(data.table) x = mydata$x y = 0.2 + 0.1 * x - 0.1 * x^2 - 0.3 * x^3 + 0.5 * x^4 + 0.5 * x^7 + 0.6 * x^11 + rnorm(length(mydata)) * 0.1 dt = data.table(y = y, x = lt) task_poly = as_task_regr(dt, target = \"y\", id = \"poly\") task_poly #> <TaskRegr:poly> (1000 x 2) #> * Target: y #> * Properties: - #> * Features (1): #>   - lt (1): x library(ggplot2) ggplot(data = data.frame(x = x, y = y)) +   geom_point(aes(x = x, y = y), alpha = 0.5)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/lazy_tensor.html","id":"custom-preprocessing","dir":"Articles","previous_headings":"","what":"Custom Preprocessing","title":"Lazy Tensor","text":"order create custom preprocessing operator lazy tensor, create new PipeOp class. make convenient possible, mlr3torch offters pipeop_preproc_torch() function recommend using purpose. important arguments : id - Used default identifier PipeOp fn - preprocessing function. default, first argument assumed torch_tensor remaining arguments part PipeOp’s parameter set. shapes_out - function returns shapes output tensors given input shapes. can also set NULL unknown shape \"infer\" auto-inference, see ?pipeop_preproc_torch information. , create PipeOp, transforms vector x matrix \\((x^{d_1} ..., x^{d_n})\\), \\(d\\) degrees parameter PipeOp. can now create new instance PipeOp calling $new(), set parameter degrees degrees used simulating data . , set parameter stages, always available, \"\", means preprocessing applied training prediction. data augmentation can set \"train\". create polynomial regression learner, combine polynomial preprocessor lrn(\"regr.mlp\") hidden layer (.e. linear model) train learner task. , visualize predictions see model captured non-linear relationship reasonably:  next section, briefly cover implementation details lazy_tensor, necessary work data-type, feel free skip part.","code":"PipeOpPreprocTorchPoly = pipeop_preproc_torch(\"poly\",   fn = function(x, degrees) {     torch_cat(lapply(degrees, function(d) torch_pow(x, d)), dim = 2L)   },   shapes_out = \"infer\" ) po_poly = PipeOpPreprocTorchPoly$new()  po_poly$param_set$set_values(   degrees = c(0, 1, 2, 3, 4, 7, 11),   stages = \"both\" ) lrn_poly = as_learner(   po_poly %>>% lrn(\"regr.mlp\", batch_size = 256, epochs = 100,   neurons = integer(0)) )  lrn_poly$train(task_poly) pred = lrn_poly$predict(task_poly) dt = melt(data.table(   truth = pred$truth,   response = pred$response,   x = x),   id.vars = \"x\", measure.vars = c(\"truth\", \"response\") ) dt$variable = factor(dt$variable, levels = c(\"truth\", \"response\"))  ggplot(data = dt) +   geom_point(aes(x = x, y = value, color = variable))"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/lazy_tensor.html","id":"digging-into-internals","dir":"Articles","previous_headings":"","what":"Digging Into Internals","title":"Lazy Tensor","text":"Internally, lazy_tensor vector uses DataDescriptor class represent (possibly) preprocessed data. similar ModelDescriptor class used build neural nerworks using PipeOpTorch objects. DataDescriptor stores torch::dataset, mlr3pipelines::Graph metadata. Per default, preprocessing graph contains single PipOpNop nothing. printed output data descriptor informs us : number PipeOps contained preprocessing graph output shapes dataset input map, .e. data passed preprocessing graph, important multiple inputs pointer, points specific channel output PipeOp. output channel tensor represented DataDescriptor. Note id input po(\"nop\") randomly generated, needed prevent id clashes one input preprocessing graph. shape, shape tensor position pointer lazy tensor can constructed integer vector DataDescriptor. integer vector specifies element DataDescriptor lazy_tensor contains. , first two elements lazy_tensor vector represent element DataDescriptor, third element represents different element. Note indices refer DataDescriptor. Internally, lazy tensor represented list lists, element containing id DataDescriptor Currently, can single DataDescriptor lazy_tensor vector. happens materialize(lt[1]) following: Preprocessing lazy_tensor vector adds new PipeOps preprocessing graph updates metainformation like pointer output shape. show , create simple example task, using lt vector feature. see $graph new pipeop id \"poly.x\" output pointer points poly.x. Also see shape tensor now c(NA, 7) c(NA, 1) , can verify calling materialize() :","code":"desc = DataDescriptor$new(   dataset = mydata,   dataset_shapes = list(x = c(NA, 1)) ) desc #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,1)] #> * input_map: (x) -> Graph #> * pointer: nop.3354b5.x.output #> * shape: [(NA,1)] lt = lazy_tensor(desc, ids = c(1, 1, 2)) materialize(lt, rbind = TRUE) #> torch_tensor #> -0.9852 #> -0.9852 #> -0.0672 #> [ CPUFloatType{3,1} ] unclass(lt[[1]]) #> [[1]] #> [1] 1 #>  #> [[2]] #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,1)] #> * input_map: (x) -> Graph #> * pointer: nop.3354b5.x.output #> * shape: [(NA,1)] # get index and data descriptor desc = lt[[1]][[2]] id = lt[[1]][[1]]  # retrieve the batch <id> from the datast dataset_output = desc$dataset$.getbatch(id)  # batch is reorganized according to the input map graph_input = dataset_output[desc$input_map] names(graph_input) = names(desc$graph$input$name)  # the reorganized batch is fed into the preprocessing graph graph_output = desc$graph$train(graph_input, single_input = FALSE)  # the output pointed to by the pointer is returned tensor = graph_output[[paste0(desc$pointer, collapse = \".\")]] tensor #> torch_tensor #> -0.9852 #> [ CPUFloatType{1,1} ] taskin = as_task_regr(data.table(x = lt, y = 1:3), target = \"y\")  taskout = po_poly$train(list(taskin))[[1L]]  lt_out = taskout$data(cols = \"x\")$x  descout = lt_out[[1]][[2]]  descout #> <DataDescriptor: 2 ops> #> * dataset_shapes: [x: (NA,1)] #> * input_map: (x) -> Graph #> * pointer: poly.x.output #> * shape: [(NA,7)]  descout$graph #> Graph with 2 PipeOps: #>            ID         State sccssors    prdcssors #>        <char>        <char>   <char>       <char> #>  nop.3354b5.x        <list>   poly.x              #>        poly.x <<UNTRAINED>>          nop.3354b5.x materialize(lt_out[1:2], rbind = TRUE) #> torch_tensor #>  1.0000 -0.9852  0.9706 -0.9563  0.9421 -0.9009 -0.8487 #>  1.0000 -0.9852  0.9706 -0.9563  0.9421 -0.9009 -0.8487 #> [ CPUFloatType{2,7} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/authors.html","id":null,"dir":"","previous_headings":"","what":"Authors","title":"Authors and Citation","text":"Sebastian Fischer. Maintainer, author. Bernd Bischl. Contributor. Lukas Burk. Contributor. Martin Binder. Author. Florian Pfisterer. Contributor. Carson Zhang. Contributor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/authors.html","id":"citation","dir":"","previous_headings":"","what":"Citation","title":"Authors and Citation","text":"Fischer S, Binder M (2024). mlr3torch: Deep Learning 'mlr3'. R package version 0.1.2-9000, https://github.com/mlr-org/mlr3torch/, https://mlr3torch.mlr-org.com/.","code":"@Manual{,   title = {mlr3torch: Deep Learning with 'mlr3'},   author = {Sebastian Fischer and Martin Binder},   year = {2024},   note = {R package version 0.1.2-9000, https://github.com/mlr-org/mlr3torch/},   url = {https://mlr3torch.mlr-org.com/}, }"},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"mlr3torch-","dir":"","previous_headings":"","what":"Deep Learning with mlr3","title":"Deep Learning with mlr3","text":"Package website: release | dev Deep Learning torch mlr3.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"installation","dir":"","previous_headings":"","what":"Installation","title":"Deep Learning with mlr3","text":"Afterwards, also need run command : information installing torch can found .","code":"# Install from CRAN install.packages(\"mlr3torch\") # Install the development version from GitHub: pak::pak(\"mlr-org/mlr3torch\") torch::install_torch()"},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"what-is-mlr3torch","dir":"","previous_headings":"","what":"What is mlr3torch?","title":"Deep Learning with mlr3","text":"mlr3torch deep learning framework mlr3 ecosystem built top torch. allows easily build, train evaluate deep learning models lines codes, without needing worry low-level details. --shelf learners readily available, custom architectures can defined connecting PipeOpTorch operators mlr3pipelines::Graph. Using predefined learners simple multi layer perceptron (MLP) works just like mlr3 Learner. , train learner sonar example task: Next, construct architecture using PipeOpTorch objects. first pipeop – PipeOpTorchIngress – defines entrypoint network. subsequent pipeops define neural network layers. turn learner, configure loss, optimizer, callbacks well training arguments. work generic tensors, lazy_tensor type can used. wraps torch::dataset, allows preprocess data (lazily) using PipeOp objects. , flatten MNIST task, can train multi-layer perceptron . Note transform data -memory, applied data actually loaded. actually access tensors, can call materialize(). show slice resulting tensor readability: , define complex architecture one single input lazy_tensor. , first define single residual block: Next, create neural network takes input lazy_tensor (po(\"torch_ingress_ltnsr\")). first applies linear layer repeats layer using special PipeOpTorchBlock, followed network’s head. , configure loss, optimizer training parameters. Note po(\"nn_linear_0\") equivalent po(\"nn_linear\", id = \"nn_linear_0\") need avoid ID clashes linear layer po(\"nn_block\"). Next, prepend preprocessing step flattens images can directly apply learner unflattened MNIST task. order keep track performance training, use 20% data evaluate using classification accuracy. left train learner:","code":"library(mlr3torch) learner_mlp = lrn(\"classif.mlp\",   # defining network parameters   activation     = nn_relu,   neurons        = c(20, 20),   # training parameters   batch_size     = 16,   epochs         = 50,   device         = \"cpu\",   # Proportion of data to use for validation   validate = 0.3,   # Defining the optimizer, loss, and callbacks   optimizer      = t_opt(\"adam\", lr = 0.1),   loss           = t_loss(\"cross_entropy\"),   callbacks      = t_clbk(\"history\"), # this saves the history in the learner   # Measures to track   measures_valid = msrs(c(\"classif.logloss\", \"classif.ce\")),   measures_train = msrs(c(\"classif.acc\")),   # predict type (required by logloss)   predict_type = \"prob\" ) learner_mlp$train(tsk(\"sonar\")) architecture = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 20) %>>%   po(\"nn_relu\") %>>%   po(\"nn_head\") graph_mlp = architecture %>>%   po(\"torch_loss\", loss = t_loss(\"cross_entropy\")) %>>%   po(\"torch_optimizer\", optimizer = t_opt(\"adam\", lr = 0.1)) %>>%   po(\"torch_callbacks\", callbacks = t_clbk(\"history\")) %>>%   po(\"torch_model_classif\",     batch_size = 16, epochs = 50, device = \"cpu\")  graph_lrn = as_learner(graph_mlp) # load the predefined mnist task mnist = tsk(\"mnist\") mnist$head(3L) #>     label           image #>    <fctr>   <lazy_tensor> #> 1:      5 <tnsr[1x28x28]> #> 2:      0 <tnsr[1x28x28]> #> 3:      4 <tnsr[1x28x28]>  # Flatten the images flattener = po(\"trafo_reshape\", shape = c(-1, 28 * 28)) mnist_flat = flattener$train(list(mnist))[[1L]]  mnist_flat$head(3L) #>     label         image #>    <fctr> <lazy_tensor> #> 1:      5   <tnsr[784]> #> 2:      0   <tnsr[784]> #> 3:      4   <tnsr[784]> materialize(   mnist_flat$data(1:2, cols = \"image\")[[1L]],   rbind = TRUE )[1:2, 1:4] #> torch_tensor #>  0  0  0  0 #>  0  0  0  0 #> [ CPUFloatType{2,4} ] layer = list(   po(\"nop\"),   po(\"nn_linear\", out_features = 50L) %>>%     po(\"nn_dropout\") %>>% po(\"nn_relu\") ) %>>% po(\"nn_merge_sum\") deep_network = po(\"torch_ingress_ltnsr\") %>>%   po(\"nn_linear_0\", out_features = 50L) %>>%   po(\"nn_block\", layer, n_blocks = 5L) %>>%   po(\"nn_head\") %>>%   po(\"torch_loss\", loss = t_loss(\"cross_entropy\")) %>>%   po(\"torch_optimizer\", optimizer = t_opt(\"adam\")) %>>%   po(\"torch_model_classif\",     epochs = 100L, batch_size = 32   ) deep_learner = as_learner(   flattener %>>% deep_network ) deep_learner$id = \"deep_network\" set_validate(deep_learner, 0.2) deep_learner$param_set$set_values(   torch_model_classif.measures_valid = msr(\"classif.acc\") ) deep_learner$train(mnist)"},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"feature-overview","dir":"","previous_headings":"","what":"Feature Overview","title":"Deep Learning with mlr3","text":"--shelf architectures readily available mlr3::Learners. Currently, supervised regression classification supported. Custom learners can defined using Graph language mlr3pipelines. package supports tabular data, well generic tensors via lazy_tensor type. Multi-modal data can handled conveniently, lazy_tensor objects can stored alongside tabular data. possible customize training process via (predefined custom) callbacks. package fully integrated mlr3 ecosystem. Neural network architectures, well hyperparameters can easily tuned via mlr3tuning friends.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"documentation","dir":"","previous_headings":"","what":"Documentation","title":"Deep Learning with mlr3","text":"Start reading one vignettes package website!","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"contributing","dir":"","previous_headings":"","what":"Contributing:","title":"Deep Learning with mlr3","text":"run tests one needs set environment variable TEST_TORCH = 1, e.g. adding .Renviron.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"acknowledgements","dir":"","previous_headings":"","what":"Acknowledgements","title":"Deep Learning with mlr3","text":"Without great R package torch none possible. names callback stages taken luz, another high-level deep learning framework R torch. Building neural networks using PipeOpTorch operators inspired keras.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"bugs-questions-feedback","dir":"","previous_headings":"","what":"Bugs, Questions, Feedback","title":"Deep Learning with mlr3","text":"mlr3torch free open source software project encourages participation feedback. issues, questions, suggestions feedback, please hesitate open “issue” GitHub page! case problems / bugs, often helpful provide “minimum working example” showcases behaviour (don’t worry bug obvious). Please understand resources project limited: response may sometimes delayed days, feature suggestions may rejected deemed tangential vision behind project.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Data Descriptor — DataDescriptor","title":"Data Descriptor — DataDescriptor","text":"data descriptor rather internal data structure used lazy_tensor data type. essence annotated torch::dataset preprocessing graph (consisting mosty PipeOpModule operators). additional meta data (e.g. pointer, shapes) allows preprocess lazy_tensors mlr3pipelines::Graph just like (non-lazy) data types. preprocessing applied materialize() called lazy_tensor. create data descriptor, can also use as_data_descriptor() function.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Data Descriptor — DataDescriptor","text":"natural define S3 class, opted R6 class avoid usual trouble serializing S3 objects. row contained DataDescriptor S3 class, copy object serializing.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Data Descriptor — DataDescriptor","text":"dataset (torch::dataset) dataset. graph (Graph) preprocessing graph. dataset_shapes (named list() (integer() NULL)) shapes output. input_map (character()) input map dataset preprocessing graph. pointer (character(2)) output pointer. pointer_shape (integer() | NULL) shape output indicated pointer. dataset_hash (character(1)) Hash wrapped dataset. hash (character(1)) Hash data descriptor. graph_input (character()) input channels preprocessing graph (cached save time). pointer_shape_predict (integer() NULL) Internal use .","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Data Descriptor — DataDescriptor","text":"DataDescriptor$new() DataDescriptor$print() DataDescriptor$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Data Descriptor — DataDescriptor","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Data Descriptor — DataDescriptor","text":"","code":"DataDescriptor$new(   dataset,   dataset_shapes = NULL,   graph = NULL,   input_map = NULL,   pointer = NULL,   pointer_shape = NULL,   pointer_shape_predict = NULL,   clone_graph = TRUE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Data Descriptor — DataDescriptor","text":"dataset (torch::dataset) torch dataset. return named list() torch_tensor objects. dataset_shapes (named list() (integer() NULL)) shapes output. Names elements list returned dataset. shape NULL (unknown, e.g. images different sizes) first dimension must NA indicate batch dimension. graph (Graph) preprocessing graph. left NULL, preprocessing applied data input_map, pointer, pointer_shape, pointer_shape_predict inferred case dataset returns one element. input_map (character()) Character vector must length input graph. Specifies data dataset fed preprocessing graph. pointer (character(2) | NULL) Points output channel within graph: Element 1 PipeOp's id element 2 PipeOp's output channel. pointer_shape (integer() | NULL) Shape output indicated pointer. pointer_shape_predict (integer() NULL) Internal use . Used Graph anticipate possible mismatches train predict shapes. clone_graph (logical(1)) Whether clone preprocessing graph.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Data Descriptor — DataDescriptor","text":"Prints object","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Data Descriptor — DataDescriptor","text":"","code":"DataDescriptor$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Data Descriptor — DataDescriptor","text":"... () Unused","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Data Descriptor — DataDescriptor","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Data Descriptor — DataDescriptor","text":"","code":"DataDescriptor$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Data Descriptor — DataDescriptor","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Data Descriptor — DataDescriptor","text":"","code":"# Create a dataset ds = dataset(   initialize = function() self$x = torch_randn(10, 3, 3),   .getitem = function(i) list(x = self$x[i, ]),   .length = function() nrow(self$x) )() dd = DataDescriptor$new(ds, list(x = c(NA, 3, 3))) dd #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,3,3)] #> * input_map: (x) -> Graph #> * pointer: nop.922020.x.output #> * shape: [(NA,3,3)] # is the same as using the converter: as_data_descriptor(ds, list(x = c(NA, 3, 3))) #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,3,3)] #> * input_map: (x) -> Graph #> * pointer: nop.922020.x.output #> * shape: [(NA,3,3)]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Represent a Model with Meta-Info — ModelDescriptor","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"Represents model; possibly complete model, possibly one process built . model takes input tensors shapes shapes_in pipes graph. Input shapes get mapped input channels graph. Output shapes named output channels graph; also possible represent -ops tensors, case names input output identical. ModelDescriptor objects typically represent partial models built , case pointer slot indicates specific point graph produces tensor shape pointer_shape, graph extended. allowed graph structure modified -reference different parts code. However, modifications may never add edges elements Graph destination. particular, element graph$input may removed reference, e.g. adding edge Graph input channel PipeOp previously without parent destination. cases better create specific ModelDescriptor training Graph consisting (mostly) operators PipeOpTorchIngress, PipeOpTorch, PipeOpTorchLoss, PipeOpTorchOptimizer, PipeOpTorchCallbacks. ModelDescriptor can converted nn_graph via model_descriptor_to_module.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"","code":"ModelDescriptor(   graph,   ingress,   task,   optimizer = NULL,   loss = NULL,   callbacks = NULL,   pointer = NULL,   pointer_shape = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"graph (Graph)Graph PipeOpModule PipeOpNOP operators. ingress (uniquely named list TorchIngressToken) List inputs go graph. Names must subset graph$input$name. task (Task) (Training)-Task model built. May necessary aspects loss use etc. optimizer (TorchOptimizer | NULL) Additional info: optimizer use. loss (TorchLoss | NULL) Additional info: loss use. callbacks (list CallbackSet NULL) Additional info: callbacks use. pointer (character(2) | NULL) Indicating element model . Points output channel within graph: Element 1 PipeOp's id element 2 PipeOp's output channel. pointer_shape (integer | NULL) Shape output indicated pointer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"(ModelDescriptor)","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCenterCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentCenterCrop — mlr_pipeops_augment_center_crop","title":"PipeOpPreprocTorchAugmentCenterCrop — mlr_pipeops_augment_center_crop","text":"Calls torchvision::transform_center_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCenterCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentCenterCrop — mlr_pipeops_augment_center_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentColorJitter.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentColorJitter — mlr_pipeops_augment_color_jitter","title":"PipeOpPreprocTorchAugmentColorJitter — mlr_pipeops_augment_color_jitter","text":"Calls torchvision::transform_color_jitter, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentColorJitter.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentColorJitter — mlr_pipeops_augment_color_jitter","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentCrop — mlr_pipeops_augment_crop","title":"PipeOpPreprocTorchAugmentCrop — mlr_pipeops_augment_crop","text":"Calls torchvision::transform_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentCrop — mlr_pipeops_augment_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentHflip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentHflip — mlr_pipeops_augment_hflip","title":"PipeOpPreprocTorchAugmentHflip — mlr_pipeops_augment_hflip","text":"Calls torchvision::transform_hflip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentHflip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentHflip — mlr_pipeops_augment_hflip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomAffine.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomAffine — mlr_pipeops_augment_random_affine","title":"PipeOpPreprocTorchAugmentRandomAffine — mlr_pipeops_augment_random_affine","text":"Calls torchvision::transform_random_affine, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomAffine.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomAffine — mlr_pipeops_augment_random_affine","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomChoice.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomChoice — mlr_pipeops_augment_random_choice","title":"PipeOpPreprocTorchAugmentRandomChoice — mlr_pipeops_augment_random_choice","text":"Calls torchvision::transform_random_choice, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomChoice.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomChoice — mlr_pipeops_augment_random_choice","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomCrop — mlr_pipeops_augment_random_crop","title":"PipeOpPreprocTorchAugmentRandomCrop — mlr_pipeops_augment_random_crop","text":"Calls torchvision::transform_random_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomCrop — mlr_pipeops_augment_random_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomHorizontalFlip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomHorizontalFlip — mlr_pipeops_augment_random_horizontal_flip","title":"PipeOpPreprocTorchAugmentRandomHorizontalFlip — mlr_pipeops_augment_random_horizontal_flip","text":"Calls torchvision::transform_random_horizontal_flip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomHorizontalFlip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomHorizontalFlip — mlr_pipeops_augment_random_horizontal_flip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomOrder.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomOrder — mlr_pipeops_augment_random_order","title":"PipeOpPreprocTorchAugmentRandomOrder — mlr_pipeops_augment_random_order","text":"Calls torchvision::transform_random_order, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomOrder.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomOrder — mlr_pipeops_augment_random_order","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomResizedCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomResizedCrop — mlr_pipeops_augment_random_resized_crop","title":"PipeOpPreprocTorchAugmentRandomResizedCrop — mlr_pipeops_augment_random_resized_crop","text":"Calls torchvision::transform_random_resized_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomResizedCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomResizedCrop — mlr_pipeops_augment_random_resized_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomVerticalFlip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomVerticalFlip — mlr_pipeops_augment_random_vertical_flip","title":"PipeOpPreprocTorchAugmentRandomVerticalFlip — mlr_pipeops_augment_random_vertical_flip","text":"Calls torchvision::transform_random_vertical_flip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomVerticalFlip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomVerticalFlip — mlr_pipeops_augment_random_vertical_flip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentResizedCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentResizedCrop — mlr_pipeops_augment_resized_crop","title":"PipeOpPreprocTorchAugmentResizedCrop — mlr_pipeops_augment_resized_crop","text":"Calls torchvision::transform_resized_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentResizedCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentResizedCrop — mlr_pipeops_augment_resized_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRotate.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRotate — mlr_pipeops_augment_rotate","title":"PipeOpPreprocTorchAugmentRotate — mlr_pipeops_augment_rotate","text":"Calls torchvision::transform_rotate, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRotate.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRotate — mlr_pipeops_augment_rotate","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentVflip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentVflip — mlr_pipeops_augment_vflip","title":"PipeOpPreprocTorchAugmentVflip — mlr_pipeops_augment_vflip","text":"Calls torchvision::transform_vflip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentVflip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentVflip — mlr_pipeops_augment_vflip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustBrightness.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustBrightness — mlr_pipeops_trafo_adjust_brightness","title":"PipeOpPreprocTorchTrafoAdjustBrightness — mlr_pipeops_trafo_adjust_brightness","text":"Calls torchvision::transform_adjust_brightness, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustBrightness.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustBrightness — mlr_pipeops_trafo_adjust_brightness","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustGamma.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustGamma — mlr_pipeops_trafo_adjust_gamma","title":"PipeOpPreprocTorchTrafoAdjustGamma — mlr_pipeops_trafo_adjust_gamma","text":"Calls torchvision::transform_adjust_gamma, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustGamma.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustGamma — mlr_pipeops_trafo_adjust_gamma","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustHue.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustHue — mlr_pipeops_trafo_adjust_hue","title":"PipeOpPreprocTorchTrafoAdjustHue — mlr_pipeops_trafo_adjust_hue","text":"Calls torchvision::transform_adjust_hue, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustHue.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustHue — mlr_pipeops_trafo_adjust_hue","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustSaturation.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustSaturation — mlr_pipeops_trafo_adjust_saturation","title":"PipeOpPreprocTorchTrafoAdjustSaturation — mlr_pipeops_trafo_adjust_saturation","text":"Calls torchvision::transform_adjust_saturation, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustSaturation.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustSaturation — mlr_pipeops_trafo_adjust_saturation","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoGrayscale.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoGrayscale — mlr_pipeops_trafo_grayscale","title":"PipeOpPreprocTorchTrafoGrayscale — mlr_pipeops_trafo_grayscale","text":"Calls torchvision::transform_grayscale, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoGrayscale.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoGrayscale — mlr_pipeops_trafo_grayscale","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoNop — mlr_pipeops_trafo_nop","title":"PipeOpPreprocTorchTrafoNop — mlr_pipeops_trafo_nop","text":"nothing.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoNop — mlr_pipeops_trafo_nop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNormalize.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoNormalize — mlr_pipeops_trafo_normalize","title":"PipeOpPreprocTorchTrafoNormalize — mlr_pipeops_trafo_normalize","text":"Calls torchvision::transform_normalize, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNormalize.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoNormalize — mlr_pipeops_trafo_normalize","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoPad.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoPad — mlr_pipeops_trafo_pad","title":"PipeOpPreprocTorchTrafoPad — mlr_pipeops_trafo_pad","text":"Calls torchvision::transform_pad, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoPad.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoPad — mlr_pipeops_trafo_pad","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoReshape.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","title":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","text":"Reshapes tensor according parameter shape, calling torch_reshape(). preprocessing function applied batch-wise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoReshape.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoReshape.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","text":"shape :: integer() desired output shape. first dimension batch dimension usually -1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoResize.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoResize — mlr_pipeops_trafo_resize","title":"PipeOpPreprocTorchTrafoResize — mlr_pipeops_trafo_resize","text":"Calls torchvision::transform_resize, see information parameters. preprocessing applied whole batch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoResize.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoResize — mlr_pipeops_trafo_resize","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoRgbToGrayscale.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoRgbToGrayscale — mlr_pipeops_trafo_rgb_to_grayscale","title":"PipeOpPreprocTorchTrafoRgbToGrayscale — mlr_pipeops_trafo_rgb_to_grayscale","text":"Calls torchvision::transform_rgb_to_grayscale, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoRgbToGrayscale.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoRgbToGrayscale — mlr_pipeops_trafo_rgb_to_grayscale","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Callback — TorchCallback","title":"Torch Callback — TorchCallback","text":"wraps CallbackSet annotates metadata, importantly ParamSet. callback created given parameter values calling $generate() method. class usually used configure callback torch learner, e.g. constructing learner ModelDescriptor. list available callbacks, see mlr3torch_callbacks. conveniently retrieve TorchCallback, use t_clbk().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Callback — TorchCallback","text":"Defined constructor argument param_set. parameter set provided construction, parameter set constructed creating parameter argument wrapped loss function, parametes type ParamUty.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Torch Callback — TorchCallback","text":"mlr3torch::TorchDescriptor -> TorchCallback","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Callback — TorchCallback","text":"mlr3torch::TorchDescriptor$generate() mlr3torch::TorchDescriptor$help() mlr3torch::TorchDescriptor$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Callback — TorchCallback","text":"TorchCallback$new() TorchCallback$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Callback — TorchCallback","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Callback — TorchCallback","text":"","code":"TorchCallback$new(   callback_generator,   param_set = NULL,   id = NULL,   label = NULL,   packages = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Callback — TorchCallback","text":"callback_generator (R6ClassGenerator) class generator callback wrapped. param_set (ParamSet NULL) parameter set. NULL (default) inferred callback_generator. id (character(1)) id new object. label (character(1)) Label new instance. packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Callback — TorchCallback","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Callback — TorchCallback","text":"","code":"TorchCallback$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Callback — TorchCallback","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Callback — TorchCallback","text":"","code":"# Create a new torch callback from an existing callback set torch_callback = TorchCallback$new(CallbackSetCheckpoint) # The parameters are inferred torch_callback$param_set #> <ParamSet(3)> #>           id    class lower upper nlevels        default  value #>       <char>   <char> <num> <num>   <num>         <list> <list> #> 1:      path ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:      freq ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 3: freq_type ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]  # Retrieve a torch callback from the dictionary torch_callback = t_clbk(\"checkpoint\",   path = tempfile(), freq = 1 ) torch_callback #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: path=/tmp/RtmpSGhCDU/file1c9a4be2a6de, freq=1 #> * Packages: mlr3torch,torch torch_callback$label #> [1] \"Checkpoint\" torch_callback$id #> [1] \"checkpoint\"  # open the help page of the wrapped callback set # torch_callback$help()  # Create the callback set callback = torch_callback$generate() callback #> <CallbackSetCheckpoint> #> * Stages: on_batch_end, on_epoch_end, on_exit # is the same as CallbackSetCheckpoint$new(   path = tempfile(), freq = 1 ) #> <CallbackSetCheckpoint> #> * Stages: on_batch_end, on_epoch_end, on_exit  # Use in a learner learner = lrn(\"regr.mlp\", callbacks = t_clbk(\"checkpoint\")) # the parameters of the callback are added to the learner's parameter set learner$param_set #> <ParamSetCollection(36)> #>                          id    class lower upper nlevels        default #>                      <char>   <char> <num> <num>   <num>         <list> #>  1:                  epochs ParamInt 0e+00   Inf     Inf <NoDefault[0]> #>  2:                  device ParamFct    NA    NA      12 <NoDefault[0]> #>  3:             num_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  4:     num_interop_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  5:                    seed ParamInt  -Inf   Inf     Inf <NoDefault[0]> #>  6:               eval_freq ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  7:          measures_train ParamUty    NA    NA     Inf <NoDefault[0]> #>  8:          measures_valid ParamUty    NA    NA     Inf <NoDefault[0]> #>  9:                patience ParamInt 0e+00   Inf     Inf <NoDefault[0]> #> 10:               min_delta ParamDbl 0e+00   Inf     Inf <NoDefault[0]> #> 11:              batch_size ParamInt 1e+00   Inf     Inf <NoDefault[0]> #> 12:                 shuffle ParamLgl    NA    NA       2          FALSE #> 13:                 sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 14:           batch_sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 15:             num_workers ParamInt 0e+00   Inf     Inf              0 #> 16:              collate_fn ParamUty    NA    NA     Inf         [NULL] #> 17:              pin_memory ParamLgl    NA    NA       2          FALSE #> 18:               drop_last ParamLgl    NA    NA       2          FALSE #> 19:                 timeout ParamDbl  -Inf   Inf     Inf             -1 #> 20:          worker_init_fn ParamUty    NA    NA     Inf <NoDefault[0]> #> 21:          worker_globals ParamUty    NA    NA     Inf <NoDefault[0]> #> 22:         worker_packages ParamUty    NA    NA     Inf <NoDefault[0]> #> 23:                 neurons ParamUty    NA    NA     Inf <NoDefault[0]> #> 24:                       p ParamDbl 0e+00 1e+00     Inf <NoDefault[0]> #> 25:              activation ParamUty    NA    NA     Inf <NoDefault[0]> #> 26:         activation_args ParamUty    NA    NA     Inf <NoDefault[0]> #> 27:                   shape ParamUty    NA    NA     Inf <NoDefault[0]> #> 28:                  opt.lr ParamDbl 0e+00   Inf     Inf          0.001 #> 29:               opt.betas ParamUty    NA    NA     Inf    0.900,0.999 #> 30:                 opt.eps ParamDbl 1e-16 1e-04     Inf          1e-08 #> 31:        opt.weight_decay ParamDbl 0e+00 1e+00     Inf              0 #> 32:             opt.amsgrad ParamLgl    NA    NA       2          FALSE #> 33:          loss.reduction ParamFct    NA    NA       2           mean #> 34:      cb.checkpoint.path ParamUty    NA    NA     Inf <NoDefault[0]> #> 35:      cb.checkpoint.freq ParamInt 1e+00   Inf     Inf <NoDefault[0]> #> 36: cb.checkpoint.freq_type ParamFct    NA    NA       2          epoch #>                          id    class lower upper nlevels        default #>            value #>           <list> #>  1:       [NULL] #>  2:         auto #>  3:            1 #>  4:            1 #>  5:       random #>  6:            1 #>  7:    <list[0]> #>  8:    <list[0]> #>  9:            0 #> 10:            0 #> 11:       [NULL] #> 12:       [NULL] #> 13:       [NULL] #> 14:       [NULL] #> 15:       [NULL] #> 16:       [NULL] #> 17:       [NULL] #> 18:       [NULL] #> 19:       [NULL] #> 20:       [NULL] #> 21:       [NULL] #> 22:       [NULL] #> 23:              #> 24:          0.5 #> 25: <nn_relu[1]> #> 26:    <list[0]> #> 27:       [NULL] #> 28:       [NULL] #> 29:       [NULL] #> 30:       [NULL] #> 31:       [NULL] #> 32:       [NULL] #> 33:       [NULL] #> 34:       [NULL] #> 35:       [NULL] #> 36:       [NULL] #>            value"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Torch Descriptors — TorchDescriptor","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Abstract Base Class TorchLoss, TorchOptimizer, TorchCallback inherit. class wraps generator (R6Class Generator torch version generator) annotates metadata ParamSet, label, ID, packages, manual page. parameters construction arguments wrapped generator parameter $values passed generator calling public method $generate().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Defined constructor argument param_set. parameters tagged \"train\", done automatically initialize.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"label (character(1)) Label object. Can used tables, plot text output instead ID. param_set (ParamSet) Set hyperparameters. packages (character(1)) Set required packages. packages loaded, attached. id (character(1)) Identifier object. Used tables, plot text output. generator wrapped generator described. man (character(1)) String format [pkg]::[topic] pointing manual page object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"phash (character(1)) Hash (unique identifier) partial object, excluding components varied systematically (e.g. parameter values).","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"TorchDescriptor$new() TorchDescriptor$print() TorchDescriptor$generate() TorchDescriptor$help() TorchDescriptor$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$new(   generator,   id = NULL,   param_set = NULL,   packages = NULL,   label = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"generator wrapped generator described. id (character(1)) id new object. param_set (ParamSet) parameter set. packages (character()) R packages object depends . label (character(1)) Label new instance. man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Prints object","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"... ","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-generate-","dir":"Reference","previous_headings":"","what":"Method generate()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Calls generator given parameter values.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$generate()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-help-","dir":"Reference","previous_headings":"","what":"Method help()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Displays help file wrapped object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$help()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Ingress Token — TorchIngressToken","title":"Torch Ingress Token — TorchIngressToken","text":"function creates S3 class class \"TorchIngressToken\", internal data structure. contains (meta-)information batch generated Task fed entry point neural network. stored ingress field ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Ingress Token — TorchIngressToken","text":"","code":"TorchIngressToken(features, batchgetter, shape)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Ingress Token — TorchIngressToken","text":"features (character) Features batchgetter operate. batchgetter (function) Function two arguments: data device. function given output Task$data(rows = batch_indices, cols = features) produce tensor shape shape_out. shape (integer) Shape batchgetter produce. Batch-dimension included NA.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Torch Ingress Token — TorchIngressToken","text":"TorchIngressToken object.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Ingress Token — TorchIngressToken","text":"","code":"# Define a task for which we want to define an ingress token task = tsk(\"iris\")  # We create an ingress token for two feature Sepal.Length and Petal.Length: # We have to specify the features, the batchgetter and the shape features = c(\"Sepal.Length\", \"Petal.Length\") # As a batchgetter we use batchgetter_num  batch_dt = task$data(rows = 1:10, cols =features) batch_dt #>     Sepal.Length Petal.Length #>            <num>        <num> #>  1:          5.1          1.4 #>  2:          4.9          1.4 #>  3:          4.7          1.3 #>  4:          4.6          1.5 #>  5:          5.0          1.4 #>  6:          5.4          1.7 #>  7:          4.6          1.4 #>  8:          5.0          1.5 #>  9:          4.4          1.4 #> 10:          4.9          1.5 batch_tensor = batchgetter_num(batch_dt, \"cpu\") batch_tensor #> torch_tensor #>  5.1000  1.4000 #>  4.9000  1.4000 #>  4.7000  1.3000 #>  4.6000  1.5000 #>  5.0000  1.4000 #>  5.4000  1.7000 #>  4.6000  1.4000 #>  5.0000  1.5000 #>  4.4000  1.4000 #>  4.9000  1.5000 #> [ CPUFloatType{10,2} ]  # The shape is unknown in the first dimension (batch dimension)  ingress_token = TorchIngressToken(   features = features,   batchgetter = batchgetter_num,   shape = c(NA, 2) ) ingress_token #> Ingress: Task[Sepal.Length,Petal.Length] --> Tensor(NA, 2)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Loss — TorchLoss","title":"Torch Loss — TorchLoss","text":"wraps torch::nn_loss annotates metadata, importantly ParamSet. loss function created given parameter values calling $generate() method. class usually used configure loss function torch learner, e.g. construcing learner ModelDescriptor. list available losses, see mlr3torch_losses. Items dictionary can retrieved using t_loss().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Loss — TorchLoss","text":"Defined constructor argument param_set. parameter set provided construction, parameter set constructed creating parameter argument wrapped loss function, parametes type ParamUty.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Torch Loss — TorchLoss","text":"mlr3torch::TorchDescriptor -> TorchLoss","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Torch Loss — TorchLoss","text":"task_types (character()) task types loss supports.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Loss — TorchLoss","text":"mlr3torch::TorchDescriptor$generate() mlr3torch::TorchDescriptor$help()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Loss — TorchLoss","text":"TorchLoss$new() TorchLoss$print() TorchLoss$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Loss — TorchLoss","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Loss — TorchLoss","text":"","code":"TorchLoss$new(   torch_loss,   task_types = NULL,   param_set = NULL,   id = NULL,   label = NULL,   packages = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Loss — TorchLoss","text":"torch_loss (nn_loss) loss module. task_types (character()) task types supported loss. param_set (ParamSet NULL) parameter set. NULL (default) inferred torch_loss. id (character(1)) id new object. label (character(1)) Label new instance. packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Torch Loss — TorchLoss","text":"Prints object","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Loss — TorchLoss","text":"","code":"TorchLoss$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Loss — TorchLoss","text":"... ","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Loss — TorchLoss","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Loss — TorchLoss","text":"","code":"TorchLoss$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Loss — TorchLoss","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Loss — TorchLoss","text":"","code":"# Create a new torch loss torch_loss = TorchLoss$new(torch_loss = nn_mse_loss, task_types = \"regr\") torch_loss #> <TorchLoss:nn_mse_loss> nn_mse_loss #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # the parameters are inferred torch_loss$param_set #> <ParamSet(1)> #>           id    class lower upper nlevels        default  value #>       <char>   <char> <num> <num>   <num>         <list> <list> #> 1: reduction ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]  # Retrieve a loss from the dictionary: torch_loss = t_loss(\"mse\", reduction = \"mean\") # is the same as torch_loss #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: reduction=mean #> * Packages: torch,mlr3torch #> * Task Types: regr torch_loss$param_set #> <ParamSet(1)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <num> <num>   <num>  <list> <list> #> 1: reduction ParamFct    NA    NA       2    mean   mean torch_loss$label #> [1] \"Mean Squared Error\" torch_loss$task_types #> [1] \"regr\" torch_loss$id #> [1] \"mse\"  # Create the loss function loss_fn = torch_loss$generate() loss_fn #> An `nn_module` containing 0 parameters. # Is the same as nn_mse_loss(reduction = \"mean\") #> An `nn_module` containing 0 parameters.  # open the help page of the wrapped loss function # torch_loss$help()  # Use in a learner learner = lrn(\"regr.mlp\", loss = t_loss(\"mse\")) # The parameters of the loss are added to the learner's parameter set learner$param_set #> <ParamSetCollection(33)> #>                      id    class lower upper nlevels        default #>                  <char>   <char> <num> <num>   <num>         <list> #>  1:              epochs ParamInt 0e+00   Inf     Inf <NoDefault[0]> #>  2:              device ParamFct    NA    NA      12 <NoDefault[0]> #>  3:         num_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  4: num_interop_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  5:                seed ParamInt  -Inf   Inf     Inf <NoDefault[0]> #>  6:           eval_freq ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  7:      measures_train ParamUty    NA    NA     Inf <NoDefault[0]> #>  8:      measures_valid ParamUty    NA    NA     Inf <NoDefault[0]> #>  9:            patience ParamInt 0e+00   Inf     Inf <NoDefault[0]> #> 10:           min_delta ParamDbl 0e+00   Inf     Inf <NoDefault[0]> #> 11:          batch_size ParamInt 1e+00   Inf     Inf <NoDefault[0]> #> 12:             shuffle ParamLgl    NA    NA       2          FALSE #> 13:             sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 14:       batch_sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 15:         num_workers ParamInt 0e+00   Inf     Inf              0 #> 16:          collate_fn ParamUty    NA    NA     Inf         [NULL] #> 17:          pin_memory ParamLgl    NA    NA       2          FALSE #> 18:           drop_last ParamLgl    NA    NA       2          FALSE #> 19:             timeout ParamDbl  -Inf   Inf     Inf             -1 #> 20:      worker_init_fn ParamUty    NA    NA     Inf <NoDefault[0]> #> 21:      worker_globals ParamUty    NA    NA     Inf <NoDefault[0]> #> 22:     worker_packages ParamUty    NA    NA     Inf <NoDefault[0]> #> 23:             neurons ParamUty    NA    NA     Inf <NoDefault[0]> #> 24:                   p ParamDbl 0e+00 1e+00     Inf <NoDefault[0]> #> 25:          activation ParamUty    NA    NA     Inf <NoDefault[0]> #> 26:     activation_args ParamUty    NA    NA     Inf <NoDefault[0]> #> 27:               shape ParamUty    NA    NA     Inf <NoDefault[0]> #> 28:              opt.lr ParamDbl 0e+00   Inf     Inf          0.001 #> 29:           opt.betas ParamUty    NA    NA     Inf    0.900,0.999 #> 30:             opt.eps ParamDbl 1e-16 1e-04     Inf          1e-08 #> 31:    opt.weight_decay ParamDbl 0e+00 1e+00     Inf              0 #> 32:         opt.amsgrad ParamLgl    NA    NA       2          FALSE #> 33:      loss.reduction ParamFct    NA    NA       2           mean #>                      id    class lower upper nlevels        default #>            value #>           <list> #>  1:       [NULL] #>  2:         auto #>  3:            1 #>  4:            1 #>  5:       random #>  6:            1 #>  7:    <list[0]> #>  8:    <list[0]> #>  9:            0 #> 10:            0 #> 11:       [NULL] #> 12:       [NULL] #> 13:       [NULL] #> 14:       [NULL] #> 15:       [NULL] #> 16:       [NULL] #> 17:       [NULL] #> 18:       [NULL] #> 19:       [NULL] #> 20:       [NULL] #> 21:       [NULL] #> 22:       [NULL] #> 23:              #> 24:          0.5 #> 25: <nn_relu[1]> #> 26:    <list[0]> #> 27:       [NULL] #> 28:       [NULL] #> 29:       [NULL] #> 30:       [NULL] #> 31:       [NULL] #> 32:       [NULL] #> 33:       [NULL] #>            value"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Optimizer — TorchOptimizer","title":"Torch Optimizer — TorchOptimizer","text":"wraps torch::torch_optimizer_generatora annotates metadata, importantly ParamSet. optimizer created given parameter values calling $generate() method. class usually used configure optimizer torch learner, e.g. construcing learner ModelDescriptor. list available optimizers, see mlr3torch_optimizers. Items dictionary can retrieved using t_opt().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Optimizer — TorchOptimizer","text":"Defined constructor argument param_set. parameter set provided construction, parameter set constructed creating parameter argument wrapped loss function, parametes type ParamUty.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Torch Optimizer — TorchOptimizer","text":"mlr3torch::TorchDescriptor -> TorchOptimizer","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Optimizer — TorchOptimizer","text":"mlr3torch::TorchDescriptor$help() mlr3torch::TorchDescriptor$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Optimizer — TorchOptimizer","text":"TorchOptimizer$new() TorchOptimizer$generate() TorchOptimizer$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Optimizer — TorchOptimizer","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Optimizer — TorchOptimizer","text":"","code":"TorchOptimizer$new(   torch_optimizer,   param_set = NULL,   id = NULL,   label = NULL,   packages = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Optimizer — TorchOptimizer","text":"torch_optimizer (torch_optimizer_generator) torch optimizer. param_set (ParamSet NULL) parameter set. NULL (default) inferred torch_optimizer. id (character(1)) id new object. label (character(1)) Label new instance. packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"method-generate-","dir":"Reference","previous_headings":"","what":"Method generate()","title":"Torch Optimizer — TorchOptimizer","text":"Instantiates optimizer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Optimizer — TorchOptimizer","text":"","code":"TorchOptimizer$generate(params)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Optimizer — TorchOptimizer","text":"params (named list() torch_tensors) parameters network.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Torch Optimizer — TorchOptimizer","text":"torch_optimizer","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Optimizer — TorchOptimizer","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Optimizer — TorchOptimizer","text":"","code":"TorchOptimizer$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Optimizer — TorchOptimizer","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Optimizer — TorchOptimizer","text":"","code":"# Create a new torch loss torch_opt = TorchOptimizer$new(optim_adam, label = \"adam\") torch_opt #> <TorchOptimizer:optim_adam> adam #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch # If the param set is not specified, parameters are inferred but are of class ParamUty torch_opt$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:        betas ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 3:          eps ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 4: weight_decay ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 5:      amsgrad ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]  # open the help page of the wrapped optimizer # torch_opt$help()  # Retrieve an optimizer from the dictionary torch_opt = t_opt(\"sgd\", lr = 0.1) torch_opt #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: lr=0.1 #> * Packages: torch,mlr3torch torch_opt$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamDbl     0   Inf     Inf <NoDefault[0]>    0.1 #> 2:     momentum ParamDbl     0     1     Inf              0 [NULL] #> 3:    dampening ParamDbl     0     1     Inf              0 [NULL] #> 4: weight_decay ParamDbl     0     1     Inf              0 [NULL] #> 5:     nesterov ParamLgl    NA    NA       2          FALSE [NULL] torch_opt$label #> [1] \"Stochastic Gradient Descent\" torch_opt$id #> [1] \"sgd\"  # Create the optimizer for a network net = nn_linear(10, 1) opt = torch_opt$generate(net$parameters)  # is the same as optim_sgd(net$parameters, lr = 0.1) #> <optim_sgd> #>   Inherits from: <torch_optimizer> #>   Public: #>     add_param_group: function (param_group)  #>     clone: function (deep = FALSE)  #>     defaults: list #>     initialize: function (params, lr = optim_required(), momentum = 0, dampening = 0,  #>     load_state_dict: function (state_dict, ..., .refer_to_state_dict = FALSE)  #>     param_groups: list #>     state: State, R6 #>     state_dict: function ()  #>     step: function (closure = NULL)  #>     zero_grad: function ()  #>   Private: #>     step_helper: function (closure, loop_fun)   # Use in a learner learner = lrn(\"regr.mlp\", optimizer = t_opt(\"sgd\")) # The parameters of the optimizer are added to the learner's parameter set learner$param_set #> <ParamSetCollection(33)> #>                      id    class lower upper nlevels        default #>                  <char>   <char> <num> <num>   <num>         <list> #>  1:              epochs ParamInt     0   Inf     Inf <NoDefault[0]> #>  2:              device ParamFct    NA    NA      12 <NoDefault[0]> #>  3:         num_threads ParamInt     1   Inf     Inf <NoDefault[0]> #>  4: num_interop_threads ParamInt     1   Inf     Inf <NoDefault[0]> #>  5:                seed ParamInt  -Inf   Inf     Inf <NoDefault[0]> #>  6:           eval_freq ParamInt     1   Inf     Inf <NoDefault[0]> #>  7:      measures_train ParamUty    NA    NA     Inf <NoDefault[0]> #>  8:      measures_valid ParamUty    NA    NA     Inf <NoDefault[0]> #>  9:            patience ParamInt     0   Inf     Inf <NoDefault[0]> #> 10:           min_delta ParamDbl     0   Inf     Inf <NoDefault[0]> #> 11:          batch_size ParamInt     1   Inf     Inf <NoDefault[0]> #> 12:             shuffle ParamLgl    NA    NA       2          FALSE #> 13:             sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 14:       batch_sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 15:         num_workers ParamInt     0   Inf     Inf              0 #> 16:          collate_fn ParamUty    NA    NA     Inf         [NULL] #> 17:          pin_memory ParamLgl    NA    NA       2          FALSE #> 18:           drop_last ParamLgl    NA    NA       2          FALSE #> 19:             timeout ParamDbl  -Inf   Inf     Inf             -1 #> 20:      worker_init_fn ParamUty    NA    NA     Inf <NoDefault[0]> #> 21:      worker_globals ParamUty    NA    NA     Inf <NoDefault[0]> #> 22:     worker_packages ParamUty    NA    NA     Inf <NoDefault[0]> #> 23:             neurons ParamUty    NA    NA     Inf <NoDefault[0]> #> 24:                   p ParamDbl     0     1     Inf <NoDefault[0]> #> 25:          activation ParamUty    NA    NA     Inf <NoDefault[0]> #> 26:     activation_args ParamUty    NA    NA     Inf <NoDefault[0]> #> 27:               shape ParamUty    NA    NA     Inf <NoDefault[0]> #> 28:              opt.lr ParamDbl     0   Inf     Inf <NoDefault[0]> #> 29:        opt.momentum ParamDbl     0     1     Inf              0 #> 30:       opt.dampening ParamDbl     0     1     Inf              0 #> 31:    opt.weight_decay ParamDbl     0     1     Inf              0 #> 32:        opt.nesterov ParamLgl    NA    NA       2          FALSE #> 33:      loss.reduction ParamFct    NA    NA       2           mean #>                      id    class lower upper nlevels        default #>            value #>           <list> #>  1:       [NULL] #>  2:         auto #>  3:            1 #>  4:            1 #>  5:       random #>  6:            1 #>  7:    <list[0]> #>  8:    <list[0]> #>  9:            0 #> 10:            0 #> 11:       [NULL] #> 12:       [NULL] #> 13:       [NULL] #> 14:       [NULL] #> 15:       [NULL] #> 16:       [NULL] #> 17:       [NULL] #> 18:       [NULL] #> 19:       [NULL] #> 20:       [NULL] #> 21:       [NULL] #> 22:       [NULL] #> 23:              #> 24:          0.5 #> 25: <nn_relu[1]> #> 26:    <list[0]> #> 27:       [NULL] #> 28:       [NULL] #> 29:       [NULL] #> 30:       [NULL] #> 31:       [NULL] #> 32:       [NULL] #> 33:       [NULL] #>            value"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to Data Descriptor — as_data_descriptor","title":"Convert to Data Descriptor — as_data_descriptor","text":"Converts input DataDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to Data Descriptor — as_data_descriptor","text":"","code":"as_data_descriptor(x, dataset_shapes, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to Data Descriptor — as_data_descriptor","text":"x () Object convert. dataset_shapes (named list() (integer() NULL)) shapes output. Names elements list returned dataset. shape NULL (unknown, e.g. images different sizes) first dimension must NA indicate batch dimension. ... () arguments passed DataDescriptor constructor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Convert to Data Descriptor — as_data_descriptor","text":"","code":"ds = dataset(\"example\",   initialize = function() self$iris = iris[, -5],   .getitem = function(i) list(x = torch_tensor(as.numeric(self$iris[i, ]))),   .length = function() nrow(self$iris) )() as_data_descriptor(ds, list(x = c(NA, 4L))) #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,4)] #> * input_map: (x) -> Graph #> * pointer: nop.550a0d.x.output #> * shape: [(NA,4)]  # if the dataset has a .getbatch method, the shapes are inferred ds2 = dataset(\"example\",   initialize = function() self$iris = iris[, -5],   .getbatch = function(i) list(x = torch_tensor(as.matrix(self$iris[i, ]))),   .length = function() nrow(self$iris) )() as_data_descriptor(ds2) #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,4)] #> * input_map: (x) -> Graph #> * pointer: nop.da7663.x.output #> * shape: [(NA,4)]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to Lazy Tensor — as_lazy_tensor","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"Convert object lazy_tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"","code":"as_lazy_tensor(x, ...)  # S3 method for class 'dataset' as_lazy_tensor(x, dataset_shapes = NULL, ids = NULL, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"x () Object convert lazy_tensor ... () Additional arguments passed method. dataset_shapes (named list() (integer() NULL)) shapes output. Names elements list returned dataset. shape NULL (unknown, e.g. images different sizes) first dimension must NA indicate batch dimension. ids (integer()) ids include lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"","code":"iris_ds = dataset(\"iris\",   initialize = function() {     self$iris = iris[, -5]   },   .getbatch = function(i) {     list(x = torch_tensor(as.matrix(self$iris[i, ])))   },   .length = function() nrow(self$iris) )() # no need to specify the dataset shapes as they can be inferred from the .getbatch method # only first 5 observations as_lazy_tensor(iris_ds, ids = 1:5) #> <ltnsr[5]> #> [1] <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> # all observations head(as_lazy_tensor(iris_ds)) #> <ltnsr[6]> #> [1] <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]>  iris_ds2 = dataset(\"iris\",   initialize = function() self$iris = iris[, -5],   .getitem = function(i) list(x = torch_tensor(as.numeric(self$iris[i, ]))),   .length = function() nrow(self$iris) )() # if .getitem is implemented we cannot infer the shapes as they might vary, # so we have to annotate them explicitly as_lazy_tensor(iris_ds2, dataset_shapes = list(x = c(NA, 4L)))[1:5] #> <ltnsr[5]> #> [1] <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]>  # Convert a matrix lt = as_lazy_tensor(matrix(rnorm(100), nrow = 20)) materialize(lt[1:5], rbind = TRUE) #> torch_tensor #> -1.4000  0.4682  0.0700  1.0743  1.9243 #>  0.2553  0.3630 -0.6391 -0.6651  1.2984 #> -2.4373 -1.3045 -0.0500  1.1140  0.7488 #> -0.0056  0.7378 -0.2515 -0.2459  0.5562 #>  0.6216  1.8885  0.4448 -1.1776 -0.5483 #> [ CPUFloatType{5,5} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to a TorchCallback — as_torch_callback","title":"Convert to a TorchCallback — as_torch_callback","text":"Converts object TorchCallback.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to a TorchCallback — as_torch_callback","text":"","code":"as_torch_callback(x, clone = FALSE, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to a TorchCallback — as_torch_callback","text":"x () Object converted. clone (logical(1)) Whether make deep clone. ... () Additional arguments","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to a TorchCallback — as_torch_callback","text":"TorchCallback.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to a list of Torch Callbacks — as_torch_callbacks","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"Converts object list TorchCallback.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"","code":"as_torch_callbacks(x, clone, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"x () Object convert. clone (logical(1)) Whether create deep clone. ... () Additional arguments.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"list() TorchCallbacks","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to TorchLoss — as_torch_loss","title":"Convert to TorchLoss — as_torch_loss","text":"Converts object TorchLoss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to TorchLoss — as_torch_loss","text":"","code":"as_torch_loss(x, clone = FALSE, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to TorchLoss — as_torch_loss","text":"x () Object convert TorchLoss. clone (logical(1)) Whether make deep clone. ... () Additional arguments. Currently used pass additional constructor arguments TorchLoss objects type nn_loss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to TorchLoss — as_torch_loss","text":"TorchLoss.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to TorchOptimizer — as_torch_optimizer","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"Converts object TorchOptimizer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"","code":"as_torch_optimizer(x, clone = FALSE, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"x () Object convert TorchOptimizer. clone (logical(1)) Whether make deep clone. Default FALSE. ... () Additional arguments. Currently used pass additional constructor arguments TorchOptimizer objects type torch_optimizer_generator.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"TorchOptimizer","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/assert_lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Assert Lazy Tensor — assert_lazy_tensor","title":"Assert Lazy Tensor — assert_lazy_tensor","text":"Asserts whether something lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/assert_lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Assert Lazy Tensor — assert_lazy_tensor","text":"","code":"assert_lazy_tensor(x)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/assert_lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Assert Lazy Tensor — assert_lazy_tensor","text":"x () Object check.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/auto_device.html","id":null,"dir":"Reference","previous_headings":"","what":"Auto Device — auto_device","title":"Auto Device — auto_device","text":"First tries cuda, cpu.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/auto_device.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Auto Device — auto_device","text":"","code":"auto_device(device = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/auto_device.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Auto Device — auto_device","text":"device (character(1)) device. NULL, returned .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_categ.html","id":null,"dir":"Reference","previous_headings":"","what":"Batchgetter for Categorical data — batchgetter_categ","title":"Batchgetter for Categorical data — batchgetter_categ","text":"Converts data frame categorical data long tensor converting data integers. input checks performed.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_categ.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Batchgetter for Categorical data — batchgetter_categ","text":"","code":"batchgetter_categ(data, device, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_categ.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Batchgetter for Categorical data — batchgetter_categ","text":"data (data.table)data.table converted tensor. device (character(1)) device. ... () Unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_num.html","id":null,"dir":"Reference","previous_headings":"","what":"Batchgetter for Numeric Data — batchgetter_num","title":"Batchgetter for Numeric Data — batchgetter_num","text":"Converts data frame numeric data float tensor calling .matrix(). input checks performed","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_num.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Batchgetter for Numeric Data — batchgetter_num","text":"","code":"batchgetter_num(data, device, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_num.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Batchgetter for Numeric Data — batchgetter_num","text":"data (data.table())data.table converted tensor. device (character(1)) device tensor created. ... () Unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Set of Callbacks for Torch — callback_set","title":"Create a Set of Callbacks for Torch — callback_set","text":"Creates R6ClassGenerator inheriting CallbackSet. Additionally performs checks stages accidentally misspelled. create TorchCallback use torch_callback(). order resulting class cloneable, private method $deep_clone() must provided.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Set of Callbacks for Torch — callback_set","text":"","code":"callback_set(   classname,   on_begin = NULL,   on_end = NULL,   on_exit = NULL,   on_epoch_begin = NULL,   on_before_valid = NULL,   on_epoch_end = NULL,   on_batch_begin = NULL,   on_batch_end = NULL,   on_after_backward = NULL,   on_batch_valid_begin = NULL,   on_batch_valid_end = NULL,   on_valid_end = NULL,   state_dict = NULL,   load_state_dict = NULL,   initialize = NULL,   public = NULL,   private = NULL,   active = NULL,   parent_env = parent.frame(),   inherit = CallbackSet,   lock_objects = FALSE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Set of Callbacks for Torch — callback_set","text":"classname (character(1)) class name. on_begin, on_end, on_epoch_begin, on_before_valid, on_epoch_end, on_batch_begin, on_batch_end, on_after_backward, on_batch_valid_begin, on_batch_valid_end, on_valid_end, on_exit (function) Function execute given stage, see section Stages. state_dict (function()) function retrieves state dict callback. available learner training. load_state_dict (function(state_dict)) Function loads callback state. initialize (function()) initialization method callback. public, private, active (list()) Additional public, private, active fields add callback. parent_env (environment()) parent environment R6Class. inherit (R6ClassGenerator) class inherit. class must either CallbackSet (default) inherit . lock_objects (logical(1)) Whether lock objects resulting R6Class. FALSE (default), values can freely assigned self without declaring class definition.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Set of Callbacks for Torch — callback_set","text":"CallbackSet","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/equals-.lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Compare lazy tensors — ==.lazy_tensor","title":"Compare lazy tensors — ==.lazy_tensor","text":"Compares lazy tensors using indices data descriptor's hash. means two lazy_tensors: equal: mateterialize tensors. unequal: might materialize tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/equals-.lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Compare lazy tensors — ==.lazy_tensor","text":"","code":"# S3 method for class 'lazy_tensor' x == y"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/equals-.lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Compare lazy tensors — ==.lazy_tensor","text":"x, y (lazy_tensor) Values compare.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/is_lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Check for lazy tensor — is_lazy_tensor","title":"Check for lazy tensor — is_lazy_tensor","text":"Checks whether object lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/is_lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Check for lazy tensor — is_lazy_tensor","text":"","code":"is_lazy_tensor(x)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/is_lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Check for lazy tensor — is_lazy_tensor","text":"x () Object check.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a lazy tensor — lazy_tensor","title":"Create a lazy tensor — lazy_tensor","text":"Create lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a lazy tensor — lazy_tensor","text":"","code":"lazy_tensor(data_descriptor = NULL, ids = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a lazy tensor — lazy_tensor","text":"data_descriptor (DataDescriptor NULL) data descriptor NULL lazy tensor length 0. ids (integer()) elements data_descriptor included lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a lazy tensor — lazy_tensor","text":"","code":"ds = dataset(\"example\",   initialize = function() self$iris = iris[, -5],   .getitem = function(i) list(x = torch_tensor(as.numeric(self$iris[i, ]))),   .length = function() nrow(self$iris) )() dd = as_data_descriptor(ds, list(x = c(NA, 4L))) lt = as_lazy_tensor(dd)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":null,"dir":"Reference","previous_headings":"","what":"Materialize Lazy Tensor Columns — materialize","title":"Materialize Lazy Tensor Columns — materialize","text":"materialize lazy_tensor() data.frame() / list() containing – among things – lazy_tensor() columns. .e. data described underlying DataDescriptors loaded indices lazy_tensor(), preprocessed put unto specified device. elements lazy tensor must shape, list tensors returned default. elements shape, tensors can also rbinded single tensor (parameter rbind).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Materialize Lazy Tensor Columns — materialize","text":"","code":"materialize(x, device = \"cpu\", rbind = FALSE, ...)  # S3 method for class 'list' materialize(x, device = \"cpu\", rbind = FALSE, cache = \"auto\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Materialize Lazy Tensor Columns — materialize","text":"x () object materialize. Either lazy_tensor list() / data.frame() containing lazy_tensor columns. device (character(1)) torch device. rbind (logical(1)) Whether rbind lazy tensor columns (TRUE) return list tensors (FALSE). second case, batch dimension. ... () Additional arguments. cache (character(1) environment() NULL) Optional cache (intermediate) materialization results. Per default, caching enabled dataset data descriptor (different output pointer) used one lazy tensor column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Materialize Lazy Tensor Columns — materialize","text":"(list() lazy_tensors lazy_tensor)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Materialize Lazy Tensor Columns — materialize","text":"Materializing lazy tensor consists : Loading data internal dataset DataDescriptor. Processing batches preprocessing Graphs. Returning result PipeOp pointed DataDescriptor (pointer). multiple lazy_tensor columns can benefit caching : ) Output(s) dataset might input multiple graphs. b) Different lazy tensors might outputs graph. reason possible provide cache environment. hash key ) hash indices dataset. hash key b) hash indices, dataset preprocessing graph.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Materialize Lazy Tensor Columns — materialize","text":"","code":"lt1 = as_lazy_tensor(torch_randn(10, 3)) materialize(lt1, rbind = TRUE) #> torch_tensor #> -2.2687  0.4539  0.0872 #> -0.4005  1.0713  0.1125 #> -1.0352  0.6216  0.1302 #>  0.9565  1.0647 -1.1269 #>  0.3968  0.8098 -0.9682 #> -1.7730 -1.3947  0.5044 #>  0.9440 -0.9359 -0.1327 #>  0.5167  0.8462  0.2562 #> -0.2682 -0.7513  1.3673 #>  0.9112 -1.9628 -0.3506 #> [ CPUFloatType{10,3} ] materialize(lt1, rbind = FALSE) #> [[1]] #> torch_tensor #> -2.2687 #>  0.4539 #>  0.0872 #> [ CPUFloatType{3} ] #>  #> [[2]] #> torch_tensor #> -0.4005 #>  1.0713 #>  0.1125 #> [ CPUFloatType{3} ] #>  #> [[3]] #> torch_tensor #> -1.0352 #>  0.6216 #>  0.1302 #> [ CPUFloatType{3} ] #>  #> [[4]] #> torch_tensor #>  0.9565 #>  1.0647 #> -1.1269 #> [ CPUFloatType{3} ] #>  #> [[5]] #> torch_tensor #>  0.3968 #>  0.8098 #> -0.9682 #> [ CPUFloatType{3} ] #>  #> [[6]] #> torch_tensor #> -1.7730 #> -1.3947 #>  0.5044 #> [ CPUFloatType{3} ] #>  #> [[7]] #> torch_tensor #>  0.9440 #> -0.9359 #> -0.1327 #> [ CPUFloatType{3} ] #>  #> [[8]] #> torch_tensor #>  0.5167 #>  0.8462 #>  0.2562 #> [ CPUFloatType{3} ] #>  #> [[9]] #> torch_tensor #> -0.2682 #> -0.7513 #>  1.3673 #> [ CPUFloatType{3} ] #>  #> [[10]] #> torch_tensor #>  0.9112 #> -1.9628 #> -0.3506 #> [ CPUFloatType{3} ] #>  lt2 = as_lazy_tensor(torch_randn(10, 4)) d = data.table::data.table(lt1 = lt1, lt2 = lt2) materialize(d, rbind = TRUE) #> $lt1 #> torch_tensor #> -2.2687  0.4539  0.0872 #> -0.4005  1.0713  0.1125 #> -1.0352  0.6216  0.1302 #>  0.9565  1.0647 -1.1269 #>  0.3968  0.8098 -0.9682 #> -1.7730 -1.3947  0.5044 #>  0.9440 -0.9359 -0.1327 #>  0.5167  0.8462  0.2562 #> -0.2682 -0.7513  1.3673 #>  0.9112 -1.9628 -0.3506 #> [ CPUFloatType{10,3} ] #>  #> $lt2 #> torch_tensor #> -1.1387 -1.7121  0.3313  0.8510 #>  1.2791  1.2661  0.4940  0.2577 #>  0.6875  0.0049 -1.4214  1.0401 #> -0.2687 -0.6857  0.2180 -0.0905 #>  0.3564 -1.3467 -0.8512  0.3529 #>  0.6444  0.2139  0.6715  3.3102 #> -1.7180 -1.6664  0.4273 -0.3261 #> -0.4558 -0.1893  0.6411  1.7529 #> -1.2947  0.1367 -0.1888 -0.0066 #> -0.4252 -0.6875  0.3217 -0.6429 #> [ CPUFloatType{10,4} ] #>  materialize(d, rbind = FALSE) #> $lt1 #> $lt1[[1]] #> torch_tensor #> -2.2687 #>  0.4539 #>  0.0872 #> [ CPUFloatType{3} ] #>  #> $lt1[[2]] #> torch_tensor #> -0.4005 #>  1.0713 #>  0.1125 #> [ CPUFloatType{3} ] #>  #> $lt1[[3]] #> torch_tensor #> -1.0352 #>  0.6216 #>  0.1302 #> [ CPUFloatType{3} ] #>  #> $lt1[[4]] #> torch_tensor #>  0.9565 #>  1.0647 #> -1.1269 #> [ CPUFloatType{3} ] #>  #> $lt1[[5]] #> torch_tensor #>  0.3968 #>  0.8098 #> -0.9682 #> [ CPUFloatType{3} ] #>  #> $lt1[[6]] #> torch_tensor #> -1.7730 #> -1.3947 #>  0.5044 #> [ CPUFloatType{3} ] #>  #> $lt1[[7]] #> torch_tensor #>  0.9440 #> -0.9359 #> -0.1327 #> [ CPUFloatType{3} ] #>  #> $lt1[[8]] #> torch_tensor #>  0.5167 #>  0.8462 #>  0.2562 #> [ CPUFloatType{3} ] #>  #> $lt1[[9]] #> torch_tensor #> -0.2682 #> -0.7513 #>  1.3673 #> [ CPUFloatType{3} ] #>  #> $lt1[[10]] #> torch_tensor #>  0.9112 #> -1.9628 #> -0.3506 #> [ CPUFloatType{3} ] #>  #>  #> $lt2 #> $lt2[[1]] #> torch_tensor #> -1.1387 #> -1.7121 #>  0.3313 #>  0.8510 #> [ CPUFloatType{4} ] #>  #> $lt2[[2]] #> torch_tensor #>  1.2791 #>  1.2661 #>  0.4940 #>  0.2577 #> [ CPUFloatType{4} ] #>  #> $lt2[[3]] #> torch_tensor #>  0.6875 #>  0.0049 #> -1.4214 #>  1.0401 #> [ CPUFloatType{4} ] #>  #> $lt2[[4]] #> torch_tensor #> -0.2687 #> -0.6857 #>  0.2180 #> -0.0905 #> [ CPUFloatType{4} ] #>  #> $lt2[[5]] #> torch_tensor #>  0.3564 #> -1.3467 #> -0.8512 #>  0.3529 #> [ CPUFloatType{4} ] #>  #> $lt2[[6]] #> torch_tensor #>  0.6444 #>  0.2139 #>  0.6715 #>  3.3102 #> [ CPUFloatType{4} ] #>  #> $lt2[[7]] #> torch_tensor #> -1.7180 #> -1.6664 #>  0.4273 #> -0.3261 #> [ CPUFloatType{4} ] #>  #> $lt2[[8]] #> torch_tensor #> -0.4558 #> -0.1893 #>  0.6411 #>  1.7529 #> [ CPUFloatType{4} ] #>  #> $lt2[[9]] #> torch_tensor #> -1.2947 #>  0.1367 #> -0.1888 #> -0.0066 #> [ CPUFloatType{4} ] #>  #> $lt2[[10]] #> torch_tensor #> -0.4252 #> -0.6875 #>  0.3217 #> -0.6429 #> [ CPUFloatType{4} ] #>  #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":null,"dir":"Reference","previous_headings":"","what":"Materialize a Lazy Tensor — materialize_internal","title":"Materialize a Lazy Tensor — materialize_internal","text":"Convert lazy_tensor torch_tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Materialize a Lazy Tensor — materialize_internal","text":"","code":"materialize_internal(x, device = \"cpu\", cache = NULL, rbind)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Materialize a Lazy Tensor — materialize_internal","text":"x (lazy_tensor()) lazy tensor materialize. device (character(1L)) device put materialized tensor (running preprocessing graph). cache (NULL environment()) Whether cache (intermediate) results materialization. can make data loading faster multiple lazy_tensors reference dataset graph. rbind (logical(1)) Whtether rbind resulting tensors (TRUE) return list tensors (FALSE).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Materialize a Lazy Tensor — materialize_internal","text":"lazy_tensor()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Materialize a Lazy Tensor — materialize_internal","text":"Materializing lazy tensor consists : Loading data internal dataset DataDescriptor. Processing batches preprocessing Graphs. Returning result PipeOp pointed DataDescriptor (pointer). materializing multiple lazy_tensor columns, caching can useful : ) Output(s) dataset might input multiple graphs. (task_dataset shoudl rarely case try merge ). b) Different lazy tensors might outputs graph. reason possible provide cache environment. hash key ) hash indices dataset. hash key b) hash indices dataset preprocessing graph.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch-package.html","id":null,"dir":"Reference","previous_headings":"","what":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","title":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","text":"Deep Learning library extends mlr3 framework building upon 'torch' package. allows conveniently build, train, evaluate deep learning models without worry low level details. Custom architectures can created using graph language defined 'mlr3pipelines'.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch-package.html","id":"options","dir":"Reference","previous_headings":"","what":"Options","title":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","text":"mlr3torch.cache: Whether cache downloaded data (TRUE) (FALSE, default). can also set specific folder file system used cache directory.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch-package.html","id":"author","dir":"Reference","previous_headings":"","what":"Author","title":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","text":"Maintainer: Sebastian Fischer sebf.fischer@gmail.com (ORCID) Authors: Martin Binder mlr.developer@mb706.com contributors: Bernd Bischl bernd_bischl@gmx.net (ORCID) [contributor] Lukas Burk github@quantenbrot.de (ORCID) [contributor] Florian Pfisterer pfistererf@googlemail.com (ORCID) [contributor] Carson Zhang carsonzhang4@gmail.com [contributor]","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":null,"dir":"Reference","previous_headings":"","what":"Dictionary of Torch Callbacks — mlr3torch_callbacks","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"mlr3misc::Dictionary torch callbacks. Use t_clbk() conveniently retrieve callbacks. Can converted data.table using .data.table.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"","code":"mlr3torch_callbacks"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"object class DictionaryMlr3torchCallbacks (inherits Dictionary, R6) length 13.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"","code":"mlr3torch_callbacks$get(\"checkpoint\") #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch # is the same as t_clbk(\"checkpoint\") #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch # convert to a data.table as.data.table(mlr3torch_callbacks) #> Key: <key> #>           key       label                 packages #>        <char>      <char>                   <char> #> 1: checkpoint  Checkpoint          mlr3torch,torch #> 2:    history     History          mlr3torch,torch #> 3:   progress    Progress progress,mlr3torch,torch #> 4:         tb TensorBoard tfevents,mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":null,"dir":"Reference","previous_headings":"","what":"Loss Functions — mlr3torch_losses","title":"Loss Functions — mlr3torch_losses","text":"Dictionary torch loss descriptors. See t_loss() conveniently retrieving loss function. Can converted data.table using .data.table.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Functions — mlr3torch_losses","text":"","code":"mlr3torch_losses"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Loss Functions — mlr3torch_losses","text":"object class DictionaryMlr3torchLosses (inherits Dictionary, R6) length 13.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"available-loss-functions","dir":"Reference","previous_headings":"","what":"Available Loss Functions","title":"Loss Functions — mlr3torch_losses","text":"cross_entropy, l1, mse","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Loss Functions — mlr3torch_losses","text":"","code":"mlr3torch_losses$get(\"mse\") #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # is equivalent to t_loss(\"mse\") #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # convert to a data.table as.data.table(mlr3torch_losses) #> Key: <key> #>              key              label task_types        packages #>           <char>             <char>     <list>          <char> #> 1: cross_entropy      Cross Entropy    classif torch,mlr3torch #> 2:            l1     Absolute Error       regr torch,mlr3torch #> 3:           mse Mean Squared Error       regr torch,mlr3torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":null,"dir":"Reference","previous_headings":"","what":"Optimizers — mlr3torch_optimizers","title":"Optimizers — mlr3torch_optimizers","text":"Dictionary torch optimizers. Use t_opt conveniently retrieving optimizers. Can converted data.table using .data.table.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizers — mlr3torch_optimizers","text":"","code":"mlr3torch_optimizers"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Optimizers — mlr3torch_optimizers","text":"object class DictionaryMlr3torchOptimizers (inherits Dictionary, R6) length 13.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"available-optimizers","dir":"Reference","previous_headings":"","what":"Available Optimizers","title":"Optimizers — mlr3torch_optimizers","text":"adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Optimizers — mlr3torch_optimizers","text":"","code":"mlr3torch_optimizers$get(\"adam\") #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch # is equivalent to t_opt(\"adam\") #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch # convert to a data.table as.data.table(mlr3torch_optimizers) #> Key: <key> #>         key                         label        packages #>      <char>                        <char>          <char> #> 1: adadelta Adaptive Learning Rate Method torch,mlr3torch #> 2:  adagrad   Adaptive Gradient algorithm torch,mlr3torch #> 3:     adam    Adaptive Moment Estimation torch,mlr3torch #> 4:     asgd  SGD with Adaptive Batch Size torch,mlr3torch #> 5:  rmsprop  Root Mean Square Propagation torch,mlr3torch #> 6:    rprop     Resilient Backpropagation torch,mlr3torch #> 7:      sgd   Stochastic Gradient Descent torch,mlr3torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":null,"dir":"Reference","previous_headings":"","what":"Lazy Data Backend — mlr_backends_lazy","title":"Lazy Data Backend — mlr_backends_lazy","text":"lazy data backend wraps constructor lazily creates another backend, e.g. downloading (caching) data internet. backend used, metadata backend known advance accessible downloading actual data. backend first constructed, verified provided metadata correct, otherwise informative error message thrown. construction lazily constructed backend, calls like $data(), $missings(), $distinct(), $hash() redirected . Information available backend constructed : nrow - number rows (set length rownames). ncol - number columns (provided via id column col_info). colnames - column names. rownames - row names. col_info - column information, can obtained via mlr3::col_info(). Beware accessing backend's hash also contructs backend. Note cases data contains lazy_tensor columns, necessary naming class nothing lazy_tensor data type. Important constructor generates factor() variables important ordering levels data corresponds ordering levels col_info argument.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Lazy Data Backend — mlr_backends_lazy","text":"mlr3::DataBackend -> DataBackendLazy","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Lazy Data Backend — mlr_backends_lazy","text":"backend (DataBackend) wrapped backend lazily constructed first accessed. nrow (integer(1)) Number rows (observations). ncol (integer(1)) Number columns (variables), including primary key column. rownames (integer()) Returns vector distinct row identifiers, .e. contents primary key column. colnames (character()) Returns vector column names, including primary key column. is_constructed (logical(1)) Whether backend already constructed.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Lazy Data Backend — mlr_backends_lazy","text":"mlr3::DataBackend$format()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Lazy Data Backend — mlr_backends_lazy","text":"DataBackendLazy$new() DataBackendLazy$data() DataBackendLazy$head() DataBackendLazy$distinct() DataBackendLazy$missings() DataBackendLazy$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$new(constructor, rownames, col_info, primary_key)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"constructor (function) function argument backend (lazy backend), whose return value must actual backend. function called first time field $backend accessed. rownames (integer()) row names. Must permutation rownames lazily constructed backend. col_info (data.table::data.table()) data.table columns id, type levels containing column id, type levels. Note levels must provided correct order. primary_key (character(1)) Name primary key column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-data-","dir":"Reference","previous_headings":"","what":"Method data()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Returns slice data specified format. rows must addressed vector primary key values, columns must referred via column names. Queries rows matching row id queries columns matching column name silently ignored. Rows guaranteed returned order rows, columns may returned arbitrary order. Duplicated row ids result duplicated rows, duplicated column names lead exception. Accessing data triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$data(rows, cols)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"rows (integer()) Row indices. cols (character()) Column names.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-head-","dir":"Reference","previous_headings":"","what":"Method head()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Retrieve first n rows. triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$head(n = 6L)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"n (integer(1)) Number rows.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Lazy Data Backend — mlr_backends_lazy","text":"data.table::data.table() first n rows.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-distinct-","dir":"Reference","previous_headings":"","what":"Method distinct()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Returns named list vectors distinct values column specified. na_rm TRUE, missing values removed returned vectors distinct values. Non-existing rows columns silently ignored. triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$distinct(rows, cols, na_rm = TRUE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-3","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"rows (integer()) Row indices. cols (character()) Column names. na_rm (logical(1)) Whether remove NAs .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"returns-1","dir":"Reference","previous_headings":"","what":"Returns","title":"Lazy Data Backend — mlr_backends_lazy","text":"Named list() distinct values.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-missings-","dir":"Reference","previous_headings":"","what":"Method missings()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Returns number missing values per column specified slice data. Non-existing rows columns silently ignored. triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$missings(rows, cols)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-4","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"rows (integer()) Row indices. cols (character()) Column names.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"returns-2","dir":"Reference","previous_headings":"","what":"Returns","title":"Lazy Data Backend — mlr_backends_lazy","text":"Total missing values per column (named numeric()).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Printer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$print()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"# We first define a backend constructor constructor = function(backend) {   cat(\"Data is constructed!\\n\")   DataBackendDataTable$new(     data.table(x = rnorm(10), y = rnorm(10), row_id = 1:10),     primary_key = \"row_id\"   ) }  # to wrap this backend constructor in a lazy backend, we need to provide the correct metadata for it column_info = data.table(   id = c(\"x\", \"y\", \"row_id\"),   type = c(\"numeric\", \"numeric\", \"integer\"),   levels = list(NULL, NULL, NULL) ) backend_lazy = DataBackendLazy$new(   constructor = constructor,   rownames = 1:10,   col_info = column_info,   primary_key = \"row_id\" )  # Note that the constructor is not called for the calls below # as they can be read from the metadata backend_lazy$nrow #> [1] 10 backend_lazy$rownames #>  [1]  1  2  3  4  5  6  7  8  9 10 backend_lazy$ncol #> [1] 3 backend_lazy$colnames #> [1] \"x\"      \"y\"      \"row_id\" col_info(backend_lazy) #>        id    type levels #>    <char>  <char> <list> #> 1:      x numeric [NULL] #> 2:      y numeric [NULL] #> 3: row_id integer [NULL]  # Only now the backend is constructed backend_lazy$data(1, \"x\") #> Data is constructed! #>             x #>         <num> #> 1: -0.3872136 # Is the same as: backend_lazy$backend$data(1, \"x\") #>             x #>         <num> #> 1: -0.3872136"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":null,"dir":"Reference","previous_headings":"","what":"Checkpoint Callback — mlr_callback_set.checkpoint","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves optimizer network states training. final network optimizer always stored.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saving learner callback trained model impossible, model slot set last callback step executed.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"mlr3torch::CallbackSet -> CallbackSetCheckpoint","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"mlr3torch::CallbackSet$load_state_dict() mlr3torch::CallbackSet$print() mlr3torch::CallbackSet$state_dict()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"CallbackSetCheckpoint$new() CallbackSetCheckpoint$on_epoch_end() CallbackSetCheckpoint$on_batch_end() CallbackSetCheckpoint$on_exit() CallbackSetCheckpoint$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$new(path, freq, freq_type = \"epoch\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"path (character(1)) path folder models saved. freq (integer(1)) frequency often model saved. Frequency either per step epoch, can configured freq_type parameter. freq_type (character(1)) Can either \"epoch\" (default) \"step\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves network optimizer state dict. nothing freq_type freq met.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-on-batch-end-","dir":"Reference","previous_headings":"","what":"Method on_batch_end()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves selected objects defined save. nothing freq_type freq met.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$on_batch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-on-exit-","dir":"Reference","previous_headings":"","what":"Method on_exit()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$on_exit()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":null,"dir":"Reference","previous_headings":"","what":"History Callback — mlr_callback_set.history","title":"History Callback — mlr_callback_set.history","text":"Saves training validation history training. history saved data.table $train $valid slots. first column always epoch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"History Callback — mlr_callback_set.history","text":"mlr3torch::CallbackSet -> CallbackSetHistory","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"History Callback — mlr_callback_set.history","text":"mlr3torch::CallbackSet$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"History Callback — mlr_callback_set.history","text":"CallbackSetHistory$on_begin() CallbackSetHistory$state_dict() CallbackSetHistory$load_state_dict() CallbackSetHistory$on_before_valid() CallbackSetHistory$on_epoch_end() CallbackSetHistory$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-on-begin-","dir":"Reference","previous_headings":"","what":"Method on_begin()","title":"History Callback — mlr_callback_set.history","text":"Initializes lists train validation metrics stored.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$on_begin()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-state-dict-","dir":"Reference","previous_headings":"","what":"Method state_dict()","title":"History Callback — mlr_callback_set.history","text":"Converts lists data.tables.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$state_dict()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-load-state-dict-","dir":"Reference","previous_headings":"","what":"Method load_state_dict()","title":"History Callback — mlr_callback_set.history","text":"Sets field $train $valid contained state dict.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$load_state_dict(state_dict)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"History Callback — mlr_callback_set.history","text":"state_dict (callback_state_history) state dict retrieved via $state_dict().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-on-before-valid-","dir":"Reference","previous_headings":"","what":"Method on_before_valid()","title":"History Callback — mlr_callback_set.history","text":"Add latest training scores history.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$on_before_valid()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"History Callback — mlr_callback_set.history","text":"Add latest validation scores history.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"History Callback — mlr_callback_set.history","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"History Callback — mlr_callback_set.history","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Callbacks — mlr_callback_set","title":"Base Class for Callbacks — mlr_callback_set","text":"Base class callbacks inherit (see section Inheriting). callback set collection functions executed different stages training loop. can used gain control training process neural network without write everything scratch. used torch learner, CallbackSet wrapped TorchCallback. latters parameter set represents arguments CallbackSet's $initialize() method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Callbacks — mlr_callback_set","text":"available stage (see section Stages) public method $on_<stage>() can defined. evaluation context (ContextTorch) can accessed via self$ctx, contains current state training loop. context assigned beginning training loop removed afterwards. Different stages callback can communicate assigning values $self. State: able store information $model slot LearnerTorch, callbacks support state API. can overload $state_dict() public method define stored learner$model$callbacks$<id> training finishes. also requires implement $load_state_dict(state_dict) method defines load previously saved callback state different callback. Note $state_dict() include parameter values used initialize callback. creating custom callbacks, function torch_callback() recommended, creates CallbackSet wraps TorchCallback. create CallbackSet convenience function callback_set() can used. functions perform checks stages accidentally misspelled.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"stages","dir":"Reference","previous_headings":"","what":"Stages","title":"Base Class for Callbacks — mlr_callback_set","text":"begin :: Run training loop begins. epoch_begin :: Run beginning epoch. batch_begin :: Run forward call. after_backward :: Run backward call. batch_end :: Run optimizer step. batch_valid_begin :: Run forward call validation loop. batch_valid_end :: Run forward call validation loop. valid_end :: Run end validation. epoch_end :: Run end epoch. end :: Run last epoch. exit :: Run last, using .exit().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"terminate-training","dir":"Reference","previous_headings":"","what":"Terminate Training","title":"Base Class for Callbacks — mlr_callback_set","text":"training stopped, possible set field $terminate ContextTorch. end every epoch field checked TRUE, training stops. can example used implement custom early stopping.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Base Class for Callbacks — mlr_callback_set","text":"ctx (ContextTorch NULL) evaluation context callback. field always NULL except $train() call torch learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Callbacks — mlr_callback_set","text":"stages (character()) active stages callback set.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Callbacks — mlr_callback_set","text":"CallbackSet$print() CallbackSet$state_dict() CallbackSet$load_state_dict() CallbackSet$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Base Class for Callbacks — mlr_callback_set","text":"Prints object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Callbacks — mlr_callback_set","text":"... () Currently unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-state-dict-","dir":"Reference","previous_headings":"","what":"Method state_dict()","title":"Base Class for Callbacks — mlr_callback_set","text":"Returns information kept LearnerTorch's state training. information loadable callback using $load_state_dict() able continue training. returns NULL default.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$state_dict()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-load-state-dict-","dir":"Reference","previous_headings":"","what":"Method load_state_dict()","title":"Base Class for Callbacks — mlr_callback_set","text":"Loads state dict callback continue training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$load_state_dict(state_dict)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Callbacks — mlr_callback_set","text":"state_dict () state dict retrieved via $state_dict().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Callbacks — mlr_callback_set","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Callbacks — mlr_callback_set","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":null,"dir":"Reference","previous_headings":"","what":"Progress Callback — mlr_callback_set.progress","title":"Progress Callback — mlr_callback_set.progress","text":"Prints progress bar metrics training validation.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Progress Callback — mlr_callback_set.progress","text":"mlr3torch::CallbackSet -> CallbackSetProgress","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Progress Callback — mlr_callback_set.progress","text":"mlr3torch::CallbackSet$load_state_dict() mlr3torch::CallbackSet$print() mlr3torch::CallbackSet$state_dict()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Progress Callback — mlr_callback_set.progress","text":"CallbackSetProgress$on_epoch_begin() CallbackSetProgress$on_batch_end() CallbackSetProgress$on_before_valid() CallbackSetProgress$on_batch_valid_end() CallbackSetProgress$on_epoch_end() CallbackSetProgress$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-epoch-begin-","dir":"Reference","previous_headings":"","what":"Method on_epoch_begin()","title":"Progress Callback — mlr_callback_set.progress","text":"Initializes progress bar training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_epoch_begin()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-batch-end-","dir":"Reference","previous_headings":"","what":"Method on_batch_end()","title":"Progress Callback — mlr_callback_set.progress","text":"Increments training progress bar.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_batch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-before-valid-","dir":"Reference","previous_headings":"","what":"Method on_before_valid()","title":"Progress Callback — mlr_callback_set.progress","text":"Creates progress bar validation.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_before_valid()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-batch-valid-end-","dir":"Reference","previous_headings":"","what":"Method on_batch_valid_end()","title":"Progress Callback — mlr_callback_set.progress","text":"Increments validation progress bar.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_batch_valid_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"Progress Callback — mlr_callback_set.progress","text":"Prints summary training validation process.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Progress Callback — mlr_callback_set.progress","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Progress Callback — mlr_callback_set.progress","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":null,"dir":"Reference","previous_headings":"","what":"TensorBoard Logging Callback — mlr_callback_set.tb","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Logs training loss, training measures, validation measures events. view , use TensorBoard tensorflow::tensorboard() (requires tensorflow) CLI.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Logs events every epoch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"mlr3torch::CallbackSet -> CallbackSetTB","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"mlr3torch::CallbackSet$load_state_dict() mlr3torch::CallbackSet$print() mlr3torch::CallbackSet$state_dict()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"CallbackSetTB$new() CallbackSetTB$on_epoch_end() CallbackSetTB$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"","code":"CallbackSetTB$new(path, log_train_loss)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"path (character(1)) path folder events logged. Point TensorBoard folder view . log_train_loss (logical(1)) Whether log training loss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Logs training loss, training measures, validation measures TensorBoard events.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"","code":"CallbackSetTB$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"","code":"CallbackSetTB$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Context for Torch Learner — mlr_context_torch","title":"Context for Torch Learner — mlr_context_torch","text":"Context training torch learner. - mostly read-- information callbacks access argument ctx. information callbacks, see CallbackSet.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Context for Torch Learner — mlr_context_torch","text":"learner (Learner) torch learner. task_train (Task) training task. task_valid (Task NULL) validation task. loader_train (torch::dataloader) data loader training. loader_valid (torch::dataloader) data loader validation. measures_train (list() Measures) Measures used training. measures_valid (list() Measures) Measures used validation. network (torch::nn_module) torch network. optimizer (torch::optimizer) optimizer. loss_fn (torch::nn_module) loss function. total_epochs (integer(1)) total number epochs learner trained . last_scores_train (named list() NULL) scores last training batch. Names ids training measures. LearnerTorch sets eval_freq different 1, NULL epochs evaluate model. last_scores_valid (list()) scores last validation batch. Names ids validation measures. LearnerTorch sets eval_freq different 1, NULL epochs evaluate model. epoch (integer(1)) current epoch. step (integer(1)) current iteration. prediction_encoder (function()) learner's prediction encoder. batch (named list() torch_tensors) current batch. terminate (logical(1)) field set TRUE end epoch, training stops.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Context for Torch Learner — mlr_context_torch","text":"ContextTorch$new() ContextTorch$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Context for Torch Learner — mlr_context_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Context for Torch Learner — mlr_context_torch","text":"","code":"ContextTorch$new(   learner,   task_train,   task_valid = NULL,   loader_train,   loader_valid = NULL,   measures_train = NULL,   measures_valid = NULL,   network,   optimizer,   loss_fn,   total_epochs,   prediction_encoder,   eval_freq = 1L )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Context for Torch Learner — mlr_context_torch","text":"learner (Learner) torch learner. task_train (Task) training task. task_valid (Task NULL) validation task. loader_train (torch::dataloader) data loader training. loader_valid (torch::dataloader NULL) data loader validation. measures_train (list() Measures NULL) Measures used training. Default NULL. measures_valid (list() Measures NULL) Measures used validation. network (torch::nn_module) torch network. optimizer (torch::optimizer) optimizer. loss_fn (torch::nn_module) loss function. total_epochs (integer(1)) total number epochs learner trained . prediction_encoder (function()) learner's prediction encoder. eval_freq (integer(1)) evaluation frequency.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Context for Torch Learner — mlr_context_torch","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Context for Torch Learner — mlr_context_torch","text":"","code":"ContextTorch$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Context for Torch Learner — mlr_context_torch","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":null,"dir":"Reference","previous_headings":"","what":"My Little Pony — mlr_learners.mlp","title":"My Little Pony — mlr_learners.mlp","text":"Fully connected feed forward network dropout activation function. features can either single lazy_tensor one numeric columns ().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"dictionary","dir":"Reference","previous_headings":"","what":"Dictionary","title":"My Little Pony — mlr_learners.mlp","text":"Learner can instantiated using sugar function lrn():","code":"lrn(\"classif.mlp\", ...) lrn(\"regr.mlp\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"My Little Pony — mlr_learners.mlp","text":"Supported task types: 'classif', 'regr' Predict Types: classif: 'response', 'prob' regr: 'response' Feature Types: “integer”, “numeric”, “lazy_tensor” Required Packages: mlr3, mlr3torch, torch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"My Little Pony — mlr_learners.mlp","text":"Parameters LearnerTorch, well : activation :: [nn_module] activation function. initialized nn_relu. activation_args :: named list() named list initialization arguments activation function. intialized empty list. neurons :: integer() number neurons per hidden layer. default hidden layer. Setting c(10, 20) first hidden layer 10 neurons second 20. p :: numeric(1) dropout probability. initialized 0.5. shape :: integer() NULL input shape length 2, e.g. c(NA, 5). needs present lazy tensor input unknown shape (NULL). Otherwise input shape inferred number numeric features.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"My Little Pony — mlr_learners.mlp","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchMLP","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"My Little Pony — mlr_learners.mlp","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"My Little Pony — mlr_learners.mlp","text":"LearnerTorchMLP$new() LearnerTorchMLP$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"My Little Pony — mlr_learners.mlp","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"My Little Pony — mlr_learners.mlp","text":"","code":"LearnerTorchMLP$new(   task_type,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"My Little Pony — mlr_learners.mlp","text":"task_type (character(1)) task type, either \"classif\" \"regr\". optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"My Little Pony — mlr_learners.mlp","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"My Little Pony — mlr_learners.mlp","text":"","code":"LearnerTorchMLP$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"My Little Pony — mlr_learners.mlp","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"My Little Pony — mlr_learners.mlp","text":"","code":"# Define the Learner and set parameter values learner = lrn(\"classif.mlp\") learner$param_set$set_values(   epochs = 1, batch_size = 16, device = \"cpu\",   neurons = 10 )  # Define a Task task = tsk(\"iris\")  # Create train and test set ids = partition(task)  # Train the learner on the training ids learner$train(task, row_ids = ids$train)  # Make predictions for the test rows predictions = learner$predict(task, row_ids = ids$test)  # Score the predictions predictions$score() #> classif.ce  #>       0.68"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":null,"dir":"Reference","previous_headings":"","what":"Tabular ResNet — mlr_learners.tab_resnet","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Tabular resnet.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"dictionary","dir":"Reference","previous_headings":"","what":"Dictionary","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Learner can instantiated using sugar function lrn():","code":"lrn(\"classif.tab_resnet\", ...) lrn(\"regr.tab_resnet\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Supported task types: 'classif', 'regr' Predict Types: classif: 'response', 'prob' regr: 'response' Feature Types: “integer”, “numeric” Required Packages: mlr3, mlr3torch, torch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Parameters LearnerTorch, well : n_blocks :: integer(1) number blocks. d_block :: integer(1) input output dimension block. d_hidden :: integer(1) latent dimension block. d_hidden_multiplier :: integer(1) Alternative way specify latent dimension d_block * d_hidden_multiplier. dropout1 :: numeric(1) First dropout ratio. dropout2 :: numeric(1) Second dropout ratio.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Gorishniy Y, Rubachev , Khrulkov V, Babenko (2021). “Revisiting Deep Learning  Tabular Data.” arXiv, 2106.11959.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchTabResNet","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"LearnerTorchTabResNet$new() LearnerTorchTabResNet$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"","code":"LearnerTorchTabResNet$new(   task_type,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"task_type (character(1)) task type, either \"classif\" \"regr\". optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"","code":"LearnerTorchTabResNet$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"","code":"# Define the Learner and set parameter values learner = lrn(\"classif.tab_resnet\") learner$param_set$set_values(   epochs = 1, batch_size = 16, device = \"cpu\",   n_blocks = 2, d_block = 10, d_hidden = 20, dropout1 = 0.3, dropout2 = 0.3 )  # Define a Task task = tsk(\"iris\")  # Create train and test set ids = partition(task)  # Train the learner on the training ids learner$train(task, row_ids = ids$train)  # Make predictions for the test rows predictions = learner$predict(task, row_ids = ids$test)  # Score the predictions predictions$score() #> classif.ce  #>        0.9"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":null,"dir":"Reference","previous_headings":"","what":"Featureless Torch Learner — mlr_learners.torch_featureless","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Featureless torch learner. Output constant weight learned training. classification, (asymptoptically) result majority class prediction using standard cross-entropy loss. regression, result median L1 loss mean L2 loss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"dictionary","dir":"Reference","previous_headings":"","what":"Dictionary","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Learner can instantiated using sugar function lrn():","code":"lrn(\"classif.torch_featureless\", ...) lrn(\"regr.torch_featureless\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Supported task types: 'classif', 'regr' Predict Types: classif: 'response', 'prob' regr: 'response' Feature Types: “logical”, “integer”, “numeric”, “character”, “factor”, “ordered”, “POSIXct”, “lazy_tensor” Required Packages: mlr3, mlr3torch, torch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"LearnerTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchFeatureless","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"LearnerTorchFeatureless$new() LearnerTorchFeatureless$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"","code":"LearnerTorchFeatureless$new(   task_type,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"task_type (character(1)) task type, either \"classif\" \"regr\". optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"","code":"LearnerTorchFeatureless$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"","code":"# Define the Learner and set parameter values learner = lrn(\"classif.torch_featureless\") learner$param_set$set_values(   epochs = 1, batch_size = 16, device = \"cpu\"    )  # Define a Task task = tsk(\"iris\")  # Create train and test set ids = partition(task)  # Train the learner on the training ids learner$train(task, row_ids = ids$train)  # Make predictions for the test rows predictions = learner$predict(task, row_ids = ids$test)  # Score the predictions predictions$score() #> classif.ce  #>       0.64"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":null,"dir":"Reference","previous_headings":"","what":"AlexNet Image Classifier — mlr_learners.torchvision","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Classic image classification networks torchvision.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Parameters LearnerTorchImage pretrained :: logical(1) Whether use pretrained model. final linear layer replaced new nn_linear number classes inferred Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Supported task types: \"classif\" Predict Types: \"response\" \"prob\" Feature Types: \"lazy_tensor\" Required packages: \"mlr3torch\", \"torch\", \"torchvision\"","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> mlr3torch::LearnerTorchImage -> LearnerTorchVision","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"LearnerTorchVision$new() LearnerTorchVision$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"","code":"LearnerTorchVision$new(   name,   module_generator,   label,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"name (character(1)) name network. module_generator (function(pretrained, num_classes)) Function generates network. label (character(1)) label network. #' @references Krizhevsky, Alex, Sutskever, Ilya, Hinton, E. G (2017). “Imagenet classification deep convolutional neural networks.” Communications ACM, 60(6), 84–90. Sandler, Mark, Howard, Andrew, Zhu, Menglong, Zhmoginov, Andrey, Chen, Liang-Chieh (2018). “Mobilenetv2: Inverted residuals linear bottlenecks.” Proceedings IEEE conference computer vision pattern recognition, 4510–4520. , Kaiming, Zhang, Xiangyu, Ren, Shaoqing, Sun, Jian (2016). “Deep residual learning image recognition.” Proceedings IEEE conference computer vision pattern recognition, 770–778. Simonyan, Karen, Zisserman, Andrew (2014). “deep convolutional networks large-scale image recognition.” arXiv preprint arXiv:1409.1556. optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"","code":"LearnerTorchVision$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Torch Learners — mlr_learners_torch","title":"Base Class for Torch Learners — mlr_learners_torch","text":"base class provides basic functionality training prediction neural network. torch learners inherit class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"validation","dir":"Reference","previous_headings":"","what":"Validation","title":"Base Class for Torch Learners — mlr_learners_torch","text":"specify validation data, can set $validate field Learner, can set : NULL: validation ratio: proportion 1 - ratio task used training ratio used validation. \"test\" means \"test\" task resampling used possible calling $train() manually. \"predefined\": use predefined $internal_valid_task mlr3::Task. validation data can also used early stopping, see description Learner's parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"saving-a-learner","dir":"Reference","previous_headings":"","what":"Saving a Learner","title":"Base Class for Torch Learners — mlr_learners_torch","text":"order save LearnerTorch later usage, necessary call $marshal() method Learner writing disk, object otherwise saved correctly. loading marshaled LearnerTorch R , need call $unmarshal() transform useable state.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"early-stopping-and-tuning","dir":"Reference","previous_headings":"","what":"Early Stopping and Tuning","title":"Base Class for Torch Learners — mlr_learners_torch","text":"order prevent overfitting, LearnerTorch class allows use early stopping via patience min_delta parameters, see Learner's parameters. tuning LearnerTorch also possible combine explicit tuning via mlr3tuning LearnerTorch's internal tuning epochs via early stopping. , just need include epochs = to_tune(upper = <upper>, internal = TRUE) search space, <upper> maximally allowed number epochs, configure early stopping.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"model","dir":"Reference","previous_headings":"","what":"Model","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Model list class \"learner_torch_model\" following elements: network :: trained network. optimizer :: $state_dict() optimizer used train network. loss_fn :: $state_dict() loss used train network. callbacks :: callbacks used train network. seed :: seed / used training prediction. epochs :: many epochs model trained (early stopping). task_col_info :: data.table() containing information train-task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Torch Learners — mlr_learners_torch","text":"General: parameters optimizer, loss callbacks, prefixed \"opt.\", \"loss.\" \"cb.<callback id>.\" respectively, well : epochs :: integer(1) number epochs. device :: character(1) device. One \"auto\", \"cpu\", \"cuda\" values defined mlr_reflections$torch$devices. value initialized \"auto\", select \"cuda\" possible, try \"mps\" otherwise fall back \"cpu\". num_threads :: integer(1) number threads intraop pararallelization (device \"cpu\"). value initialized 1. num_interop_threads :: integer(1) number threads intraop interop pararallelization (device \"cpu\"). value initialized 1. Note can set session changing value within R session raise warning. seed :: integer(1) \"random\" NULL torch seed used training prediction. value initialized \"random\", means random seed sampled beginning training phase. seed (either set randomly sampled) available via $model$seed training used prediction. Note setting seed training phase mean default (.e. seed \"random\"), clones learner use different seed. set NULL, seeding done. Evaluation: measures_train :: Measure list() Measures. Measures evaluated training. measures_valid :: Measure list() Measures. Measures evaluated validation. eval_freq :: integer(1) often train / validation predictions evaluated using measures_train / measures_valid. initialized 1. Note final model always evaluated. Early Stopping: patience :: integer(1) activates early stopping using validation scores. performance model improve patience evaluation steps, training ended. Note final model stored learner, best model. initialized 0, means early stopping. first entry measures_valid used metric. also requires specify $validate field Learner, well measures_valid. min_delta :: double(1) minimum improvement threshold (>) early stopping. initialized 0. Dataloader: batch_size :: integer(1) batch size (required). shuffle :: logical(1) Whether shuffle instances dataset. Default FALSE. impact validation. sampler :: torch::sampler Object defines dataloader draw samples. batch_sampler :: torch::sampler Object defines dataloader draws batches. num_workers :: integer(1) number workers data loading (batches loaded parallel). default 0, means data loaded main process. collate_fn :: function merge list samples form batch. pin_memory :: logical(1) Whether dataloader copies tensors CUDA pinned memory returning . drop_last :: logical(1) Whether drop last training batch epoch training. Default FALSE. timeout :: numeric(1) timeout value collecting batch workers. Negative values mean timeout default -1. worker_init_fn :: function(id) function receives worker id ([1, num_workers]) exectued seeding worker data loading. worker_globals :: list() | character() loading data parallel, allows export globals workers. character vector, objects global environment names copied workers. worker_packages :: character() packages load workers. Also see torch::dataloder information.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Torch Learners — mlr_learners_torch","text":"seperate classes classification regression inherit . Instead, task_type must specified  construction argument. Currently, classification regression supported. inheriting class, one overload two private methods: .network(task, param_vals) (Task, list()) -> nn_module Construct torch::nn_module object given task parameter values, .e. neural network trained learner. classification, output network expected scores application final softmax layer. .dataset(task, param_vals) (Task, list()) -> torch::dataset Create dataset task. Must respect parameter value device. Moreover, one needs pay attention respect row ids provided task. also possible overwrite private .dataloader() method instead .dataset() method. Per default, dataloader constructed using output .dataset() method. However, respect dataloader parameters ParamSet. .dataloader(task, param_vals) (Task, list()) -> torch::dataloader Create dataloader task. Needs respect least batch_size shuffle (otherwise predictions can permuted). change predict types, private .encode_prediction() method can overwritten: .encode_prediction(predict_tensor, task, param_vals) (torch_tensor, Task, list()) -> list() Take raw predictions self$network (predict_tensor) encode format can converted valid mlr3 predictions using mlr3::as_prediction_data(). method must take self$predict_type account. possible add parameters specifying param_set construction argument, currently possible remove existing parameters, .e. listed section Parameters. None parameters provided param_set can id starts \"loss.\", \"opt.\", \"cb.\"`, preserved dynamically constructed parameters optimizer, loss function, callbacks. perform additional input checks task, private .verify_train_task(task, param_vals) .verify_predict_task(task, param_vals) can overwritten. learners construction arguments change hash learner, required implement private $.additional_phash_input().","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Base Class for Torch Learners — mlr_learners_torch","text":"mlr3::Learner -> LearnerTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Torch Learners — mlr_learners_torch","text":"validate construct internal validation data. parameter can either NULL, ratio $(0, 1)$, \"test\", \"predefined\". loss (TorchLoss) torch loss. optimizer (TorchOptimizer) torch optimizer. callbacks (list() TorchCallbacks) List torch callbacks. ids set names. internal_valid_scores Retrieves internal validation scores named list(). Specify $validate field measures_valid parameter configure . Returns NULL learner trained yet. internal_tuned_values early stopping activate, returns named list early-stopped epochs, otherwise empty list returned. Returns NULL learner trained yet. marshaled (logical(1)) Whether learner marshaled. network (nn_module()) Shortcut learner$model$network. param_set (ParamSet) parameter set hash (character(1)) Hash (unique identifier) object. phash (character(1)) Hash (unique identifier) partial object, excluding components varied systematically tuning (parameter values).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Base Class for Torch Learners — mlr_learners_torch","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Torch Learners — mlr_learners_torch","text":"LearnerTorch$new() LearnerTorch$format() LearnerTorch$print() LearnerTorch$marshal() LearnerTorch$unmarshal() LearnerTorch$dataset() LearnerTorch$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$new(   id,   task_type,   param_set,   properties,   man,   label,   feature_types,   optimizer = NULL,   loss = NULL,   packages = character(),   predict_types = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"id (character(1)) id new object. task_type (character(1)) task type. param_set (ParamSet alist()) Either parameter set, alist() containing different values self, e.g. alist(private$.param_set1, private$.param_set2), ParamSet collection created. properties (character()) properties object. See mlr_reflections$learner_properties available values. man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help(). label (character(1)) Label new instance. feature_types (character()) feature types. See mlr_reflections$task_feature_types available values, Additionally, \"lazy_tensor\" supported. optimizer (NULL TorchOptimizer) optimizer use training. Defaults adam. loss (NULL TorchLoss) loss use training. Defaults MSE regression cross entropy classification. packages (character()) R packages object depends . predict_types (character()) predict types. See mlr_reflections$learner_predict_types available values. regression, default \"response\". classification, defaults \"response\" \"prob\". deviate defaults, necessary overwrite private $.encode_prediction() method, see section Inheriting. callbacks (list() TorchCallbacks) callbacks use training. Defaults empty list(), .e. callbacks.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-format-","dir":"Reference","previous_headings":"","what":"Method format()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Helper print outputs.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$format(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... (ignored).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Prints object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... () Currently unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-marshal-","dir":"Reference","previous_headings":"","what":"Method marshal()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Marshal learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$marshal(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-3","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... () Additional parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Learners — mlr_learners_torch","text":"self","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-unmarshal-","dir":"Reference","previous_headings":"","what":"Method unmarshal()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Unmarshal learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$unmarshal(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-4","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... () Additional parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"returns-1","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Learners — mlr_learners_torch","text":"self","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-dataset-","dir":"Reference","previous_headings":"","what":"Method dataset()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Create dataset task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$dataset(task)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-5","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"task Task task","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"returns-2","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Learners — mlr_learners_torch","text":"dataset","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-6","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-6","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":null,"dir":"Reference","previous_headings":"","what":"Image Learner — mlr_learners_torch_image","title":"Image Learner — mlr_learners_torch_image","text":"Base Class Image Learners. features assumed single lazy_tensor column RGB format.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Image Learner — mlr_learners_torch_image","text":"Parameters include inherited LearnerTorch param_set construction argument.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Image Learner — mlr_learners_torch_image","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchImage","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Image Learner — mlr_learners_torch_image","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Image Learner — mlr_learners_torch_image","text":"LearnerTorchImage$new() LearnerTorchImage$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Image Learner — mlr_learners_torch_image","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Image Learner — mlr_learners_torch_image","text":"","code":"LearnerTorchImage$new(   id,   task_type,   param_set = ps(),   label,   optimizer = NULL,   loss = NULL,   callbacks = list(),   packages = c(\"torchvision\", \"magick\"),   man,   properties = NULL,   predict_types = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Image Learner — mlr_learners_torch_image","text":"id (character(1)) id new object. task_type (character(1)) task type. param_set (ParamSet) parameter set. label (character(1)) Label new instance. optimizer (TorchOptimizer) torch optimizer. loss (TorchLoss) loss use training. callbacks (list() TorchCallbacks) callbacks used training. Must unique ids. executed order provided packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help(). properties (character()) properties object. See mlr_reflections$learner_properties available values. predict_types (character()) predict types. See mlr_reflections$learner_predict_types available values.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Image Learner — mlr_learners_torch_image","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Image Learner — mlr_learners_torch_image","text":"","code":"LearnerTorchImage$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Image Learner — mlr_learners_torch_image","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":null,"dir":"Reference","previous_headings":"","what":"Learner Torch Model — mlr_learners_torch_model","title":"Learner Torch Model — mlr_learners_torch_model","text":"Create torch learner instantiated nn_module(). classification, output network must scores (softmax).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Learner Torch Model — mlr_learners_torch_model","text":"See LearnerTorch","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Learner Torch Model — mlr_learners_torch_model","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchModel","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Learner Torch Model — mlr_learners_torch_model","text":"network_stored (nn_module NULL) network trained. calling $train(), NULL. ingress_tokens (named list() TorchIngressToken NULL) ingress tokens. Must non-NULL calling $train().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Learner Torch Model — mlr_learners_torch_model","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Learner Torch Model — mlr_learners_torch_model","text":"LearnerTorchModel$new() LearnerTorchModel$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Learner Torch Model — mlr_learners_torch_model","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Learner Torch Model — mlr_learners_torch_model","text":"","code":"LearnerTorchModel$new(   network = NULL,   ingress_tokens = NULL,   task_type,   properties = NULL,   optimizer = NULL,   loss = NULL,   callbacks = list(),   packages = character(0),   feature_types = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Learner Torch Model — mlr_learners_torch_model","text":"network (nn_module) instantiated nn_module. cloned construction. classification, outputs must scores (softmax). ingress_tokens (list TorchIngressToken()) list ingress tokens defines dataloader defined. task_type (character(1)) task type. properties (NULL character()) properties learner. Defaults available properties given task type. optimizer (TorchOptimizer) torch optimizer. loss (TorchLoss) loss use training. callbacks (list() TorchCallbacks) callbacks used training. Must unique ids. executed order provided packages (character()) R packages object depends . feature_types (NULL character()) feature types. Defaults available feature types.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Learner Torch Model — mlr_learners_torch_model","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Learner Torch Model — mlr_learners_torch_model","text":"","code":"LearnerTorchModel$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Learner Torch Model — mlr_learners_torch_model","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Learner Torch Model — mlr_learners_torch_model","text":"","code":"# We show the learner using a classification task  # The iris task has 4 features and 3 classes network = nn_linear(4, 3) task = tsk(\"iris\")  # This defines the dataloader. # It loads all 4 features, which are also numeric. # The shape is (NA, 4) because the batch dimension is generally NA ingress_tokens = list(   input = TorchIngressToken(task$feature_names, batchgetter_num, c(NA, 4)) )  # Creating the learner and setting required parameters learner = lrn(\"classif.torch_model\",   network = network,   ingress_tokens = ingress_tokens,   batch_size = 16,   epochs = 1,   device = \"cpu\" )  # A simple train-predict ids = partition(task) learner$train(task, ids$train) learner$predict(task, ids$test) #> <PredictionClassif> for 50 observations: #>  row_ids     truth response #>        3    setosa   setosa #>        9    setosa   setosa #>       10    setosa   setosa #>      ---       ---      --- #>      147 virginica   setosa #>      149 virginica   setosa #>      150 virginica   setosa"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":null,"dir":"Reference","previous_headings":"","what":"Class for Torch Module Wrappers — mlr_pipeops_module","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"PipeOpModule wraps nn_module function called train phase mlr3pipelines::PipeOp. , allows assemble PipeOpModules computational mlr3pipelines::Graph represents either neural network preprocessing graph lazy_tensor. cases easier create network creating graph generates graph. cases easier create network creating structurally related graph consisting nodes class PipeOpTorchIngress PipeOpTorch. graph generate graph consisting PipeOpModules part ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"number names input output channels can set construction. input output \"torch_tensor\" training, NULL prediction prediction phase currently serves meaningful purpose.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"training, wrapped nn_module / function called provided inputs order channels defined. Arguments matched name.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"mlr3pipelines::PipeOp -> PipeOpModule","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"module (nn_module) torch module called training phase.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"PipeOpModule$new() PipeOpModule$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"","code":"PipeOpModule$new(   id = \"module\",   module = nn_identity(),   inname = \"input\",   outname = \"output\",   param_vals = list(),   packages = character(0) )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"id (character(1)) id new object. module (nn_module function()) torch module function wrapped. inname (character()) names input channels. outname (character()) names output channels. parameter length 1, parameter module must return tensor. Otherwise must return list() tensors corresponding length. param_vals (named list()) Parameter values set construction. packages (character()) R packages object depends .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"","code":"PipeOpModule$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"","code":"## creating an PipeOpModule manually  # one input and output channel po_module = po(\"module\",   id = \"linear\",   module = torch::nn_linear(10, 20),   inname = \"input\",   outname = \"output\" ) x = torch::torch_randn(16, 10) # This calls the forward function of the wrapped module. y = po_module$train(list(input = x)) str(y) #> List of 1 #>  $ output:Float [1:16, 1:20]  # multiple input and output channels nn_custom = torch::nn_module(\"nn_custom\",   initialize = function(in_features, out_features) {     self$lin1 = torch::nn_linear(in_features, out_features)     self$lin2 = torch::nn_linear(in_features, out_features)   },   forward = function(x, z) {     list(out1 = self$lin1(x), out2 = torch::nnf_relu(self$lin2(z)))   } )  module = nn_custom(3, 2) po_module = po(\"module\",   id = \"custom\",   module = module,   inname = c(\"x\", \"z\"),   outname = c(\"out1\", \"out2\") ) x = torch::torch_randn(1, 3) z = torch::torch_randn(1, 3) out = po_module$train(list(x = x, z = z)) str(out) #> List of 2 #>  $ out1:Float [1:1, 1:2] #>  $ out2:Float [1:1, 1:2]  # How such a PipeOpModule is usually generated graph = po(\"torch_ingress_num\") %>>% po(\"nn_linear\", out_features = 10L) result = graph$train(tsk(\"iris\")) # The PipeOpTorchLinear generates a PipeOpModule and adds it to a new (module) graph result[[1]]$graph #> Graph with 2 PipeOps: #>                 ID         State  sccssors         prdcssors #>             <char>        <char>    <char>            <char> #>  torch_ingress_num <<UNTRAINED>> nn_linear                   #>          nn_linear <<UNTRAINED>>           torch_ingress_num linear_module = result[[1L]]$graph$pipeops$nn_linear linear_module #> PipeOp: <nn_linear> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [torch_tensor,NULL] #> Output channels <name [train type, predict type]>: #>   output [torch_tensor,NULL] formalArgs(linear_module$module) #> [1] \"input\" linear_module$input$name #> [1] \"input\"  # Constructing a PipeOpModule using a simple function po_add1 = po(\"module\",   id = \"add_one\",   module = function(x) x + 1 ) input = list(torch_tensor(1)) po_add1$train(input)$output #> torch_tensor #>  2 #> [ CPUFloatType{1} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Applies 1D adaptive average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"output_size :: integer() target output size. single number.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Calls nn_adaptive_avg_pool1d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAdaptiveAvgPool -> PipeOpTorchAdaptiveAvgPool1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"PipeOpTorchAdaptiveAvgPool1D$new() PipeOpTorchAdaptiveAvgPool1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"","code":"PipeOpTorchAdaptiveAvgPool1D$new(   id = \"nn_adaptive_avg_pool1d\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"","code":"PipeOpTorchAdaptiveAvgPool1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_adaptive_avg_pool1d\") pipeop #> PipeOp: <nn_adaptive_avg_pool1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: output_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Applies 2D adaptive average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"output_size :: integer() target output size. Can single number vector.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Calls nn_adaptive_avg_pool2d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAdaptiveAvgPool -> PipeOpTorchAdaptiveAvgPool2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"PipeOpTorchAdaptiveAvgPool2D$new() PipeOpTorchAdaptiveAvgPool2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"","code":"PipeOpTorchAdaptiveAvgPool2D$new(   id = \"nn_adaptive_avg_pool2d\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"","code":"PipeOpTorchAdaptiveAvgPool2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_adaptive_avg_pool2d\") pipeop #> PipeOp: <nn_adaptive_avg_pool2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: output_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Applies 3D adaptive average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"output_size :: integer() target output size. Can single number vector.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Calls nn_adaptive_avg_pool3d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAdaptiveAvgPool -> PipeOpTorchAdaptiveAvgPool3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"PipeOpTorchAdaptiveAvgPool3D$new() PipeOpTorchAdaptiveAvgPool3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"","code":"PipeOpTorchAdaptiveAvgPool3D$new(   id = \"nn_adaptive_avg_pool3d\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"","code":"PipeOpTorchAdaptiveAvgPool3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_adaptive_avg_pool3d\") pipeop #> PipeOp: <nn_adaptive_avg_pool3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: output_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Applies 1D average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"kernel_size :: (integer()) size window. Can single number vector. stride :: integer() stride window. Can single number vector. Default: kernel_size. padding :: integer() Implicit zero paddings sides input. Can single number vector. Default: 0. ceil_mode :: integer() TRUE, use ceil instead floor compute output shape. Default: FALSE. count_include_pad :: logical(1) TRUE, include zero-padding averaging calculation. Default: TRUE. divisor_override :: logical(1) specified, used divisor, otherwise size pooling region used. Default: NULL. available dimension greater 1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Calls nn_avg_pool1d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAvgPool -> PipeOpTorchAvgPool1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"PipeOpTorchAvgPool1D$new() PipeOpTorchAvgPool1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"","code":"PipeOpTorchAvgPool1D$new(id = \"nn_avg_pool1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"","code":"PipeOpTorchAvgPool1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_avg_pool1d\") pipeop #> PipeOp: <nn_avg_pool1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>                   id    class lower upper nlevels        default  value #>               <char>   <char> <num> <num>   <num>         <list> <list> #> 1:       kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:            stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 3:           padding ParamUty    NA    NA     Inf              0 [NULL] #> 4:         ceil_mode ParamLgl    NA    NA       2          FALSE [NULL] #> 5: count_include_pad ParamLgl    NA    NA       2           TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Applies 2D average-pooling operation \\(kH * kW\\) regions step size \\(sH * sW\\) steps. number output features equal number input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Calls nn_avg_pool2d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"kernel_size :: (integer()) size window. Can single number vector. stride :: integer() stride window. Can single number vector. Default: kernel_size. padding :: integer() Implicit zero paddings sides input. Can single number vector. Default: 0. ceil_mode :: integer() TRUE, use ceil instead floor compute output shape. Default: FALSE. count_include_pad :: logical(1) TRUE, include zero-padding averaging calculation. Default: TRUE. divisor_override :: logical(1) specified, used divisor, otherwise size pooling region used. Default: NULL. available dimension greater 1.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAvgPool -> PipeOpTorchAvgPool2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"PipeOpTorchAvgPool2D$new() PipeOpTorchAvgPool2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"","code":"PipeOpTorchAvgPool2D$new(id = \"nn_avg_pool2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"","code":"PipeOpTorchAvgPool2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_avg_pool2d\") pipeop #> PipeOp: <nn_avg_pool2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(6)> #>                   id    class lower upper nlevels        default  value #>               <char>   <char> <num> <num>   <num>         <list> <list> #> 1:       kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:            stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 3:           padding ParamUty    NA    NA     Inf              0 [NULL] #> 4:         ceil_mode ParamLgl    NA    NA       2          FALSE [NULL] #> 5: count_include_pad ParamLgl    NA    NA       2           TRUE [NULL] #> 6:  divisor_override ParamDbl     0   Inf     Inf         [NULL] [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Applies 3D average-pooling operation \\(kT * kH * kW\\) regions step size \\(sT * sH * sW\\) steps. number output features equal \\(\\lfloor \\frac{ \\mbox{input planes} }{sT} \\rfloor\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Calls nn_avg_pool3d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"kernel_size :: (integer()) size window. Can single number vector. stride :: integer() stride window. Can single number vector. Default: kernel_size. padding :: integer() Implicit zero paddings sides input. Can single number vector. Default: 0. ceil_mode :: integer() TRUE, use ceil instead floor compute output shape. Default: FALSE. count_include_pad :: logical(1) TRUE, include zero-padding averaging calculation. Default: TRUE. divisor_override :: logical(1) specified, used divisor, otherwise size pooling region used. Default: NULL. available dimension greater 1.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAvgPool -> PipeOpTorchAvgPool3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"PipeOpTorchAvgPool3D$new() PipeOpTorchAvgPool3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"","code":"PipeOpTorchAvgPool3D$new(id = \"nn_avg_pool3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"","code":"PipeOpTorchAvgPool3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_avg_pool3d\") pipeop #> PipeOp: <nn_avg_pool3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(6)> #>                   id    class lower upper nlevels        default  value #>               <char>   <char> <num> <num>   <num>         <list> <list> #> 1:       kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:            stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 3:           padding ParamUty    NA    NA     Inf              0 [NULL] #> 4:         ceil_mode ParamLgl    NA    NA       2          FALSE [NULL] #> 5: count_include_pad ParamLgl    NA    NA       2           TRUE [NULL] #> 6:  divisor_override ParamDbl     0   Inf     Inf         [NULL] [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Applies Batch Normalization channel across batch data.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"eps :: numeric(1) value added denominator numerical stability. Default: 1e-5. momentum :: numeric(1) value used running_mean running_var computation. Can set NULL cumulative moving average (.e. simple average). Default: 0.1 affine :: logical(1) boolean value set TRUE, module learnable affine parameters. Default: TRUE track_running_stats :: logical(1) boolean value set TRUE, module tracks running mean variance, set FALSE, module track statistics always uses batch statistics training eval modes. Default: TRUE","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Calls torch::nn_batch_norm1d(). parameter num_features inferred second dimension input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchBatchNorm -> PipeOpTorchBatchNorm1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"PipeOpTorchBatchNorm1D$new() PipeOpTorchBatchNorm1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"","code":"PipeOpTorchBatchNorm1D$new(id = \"nn_batch_norm1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"","code":"PipeOpTorchBatchNorm1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_batch_norm1d\") pipeop #> PipeOp: <nn_batch_norm1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(4)> #>                     id    class lower upper nlevels default  value #>                 <char>   <char> <num> <num>   <num>  <list> <list> #> 1:                 eps ParamDbl     0   Inf     Inf   1e-05 [NULL] #> 2:            momentum ParamDbl     0   Inf     Inf     0.1 [NULL] #> 3:              affine ParamLgl    NA    NA       2    TRUE [NULL] #> 4: track_running_stats ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Applies Batch Normalization channel across batch data.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Calls torch::nn_batch_norm2d(). parameter num_features inferred second dimension input shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"eps :: numeric(1) value added denominator numerical stability. Default: 1e-5. momentum :: numeric(1) value used running_mean running_var computation. Can set NULL cumulative moving average (.e. simple average). Default: 0.1 affine :: logical(1) boolean value set TRUE, module learnable affine parameters. Default: TRUE track_running_stats :: logical(1) boolean value set TRUE, module tracks running mean variance, set FALSE, module track statistics always uses batch statistics training eval modes. Default: TRUE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchBatchNorm -> PipeOpTorchBatchNorm2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"PipeOpTorchBatchNorm2D$new() PipeOpTorchBatchNorm2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"","code":"PipeOpTorchBatchNorm2D$new(id = \"nn_batch_norm2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"","code":"PipeOpTorchBatchNorm2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_batch_norm2d\") pipeop #> PipeOp: <nn_batch_norm2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(4)> #>                     id    class lower upper nlevels default  value #>                 <char>   <char> <num> <num>   <num>  <list> <list> #> 1:                 eps ParamDbl     0   Inf     Inf   1e-05 [NULL] #> 2:            momentum ParamDbl     0   Inf     Inf     0.1 [NULL] #> 3:              affine ParamLgl    NA    NA       2    TRUE [NULL] #> 4: track_running_stats ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Applies Batch Normalization channel across batch data.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Calls torch::nn_batch_norm3d(). parameter num_features inferred second dimension input shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"eps :: numeric(1) value added denominator numerical stability. Default: 1e-5. momentum :: numeric(1) value used running_mean running_var computation. Can set NULL cumulative moving average (.e. simple average). Default: 0.1 affine :: logical(1) boolean value set TRUE, module learnable affine parameters. Default: TRUE track_running_stats :: logical(1) boolean value set TRUE, module tracks running mean variance, set FALSE, module track statistics always uses batch statistics training eval modes. Default: TRUE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchBatchNorm -> PipeOpTorchBatchNorm3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"PipeOpTorchBatchNorm3D$new() PipeOpTorchBatchNorm3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"","code":"PipeOpTorchBatchNorm3D$new(id = \"nn_batch_norm3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"","code":"PipeOpTorchBatchNorm3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_batch_norm3d\") pipeop #> PipeOp: <nn_batch_norm3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(4)> #>                     id    class lower upper nlevels default  value #>                 <char>   <char> <num> <num>   <num>  <list> <list> #> 1:                 eps ParamDbl     0   Inf     Inf   1e-05 [NULL] #> 2:            momentum ParamDbl     0   Inf     Inf     0.1 [NULL] #> 3:              affine ParamLgl    NA    NA       2    TRUE [NULL] #> 4: track_running_stats ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":null,"dir":"Reference","previous_headings":"","what":"Block Repetition — mlr_pipeops_nn_block","title":"Block Repetition — mlr_pipeops_nn_block","text":"Repeat block n_blocks times.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Block Repetition — mlr_pipeops_nn_block","text":"parameters available block , well n_blocks :: integer(1) often repeat block.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Block Repetition — mlr_pipeops_nn_block","text":"PipeOp sets input output channels block (Graph) received construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Block Repetition — mlr_pipeops_nn_block","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Block Repetition — mlr_pipeops_nn_block","text":"Part documentation copied adapted documentation torch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Block Repetition — mlr_pipeops_nn_block","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchBlock","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Block Repetition — mlr_pipeops_nn_block","text":"block (Graph) neural network segment repeated PipeOp.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Block Repetition — mlr_pipeops_nn_block","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Block Repetition — mlr_pipeops_nn_block","text":"PipeOpTorchBlock$new() PipeOpTorchBlock$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Block Repetition — mlr_pipeops_nn_block","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Block Repetition — mlr_pipeops_nn_block","text":"","code":"PipeOpTorchBlock$new(block, id = \"nn_block\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Block Repetition — mlr_pipeops_nn_block","text":"block (Graph) graph consisting primarily PipeOpTorch objects repeated. id (character(1)) id new object. param_vals (named list()) Parameter values set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Block Repetition — mlr_pipeops_nn_block","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Block Repetition — mlr_pipeops_nn_block","text":"","code":"PipeOpTorchBlock$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Block Repetition — mlr_pipeops_nn_block","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Block Repetition — mlr_pipeops_nn_block","text":"","code":"block = po(\"nn_linear\") %>>% po(\"nn_relu\") po_block = po(\"nn_block\", block, nn_linear.out_features = 10L, n_blocks = 3) network = po(\"torch_ingress_num\") %>>% po_block %>>% po(\"nn_head\") %>>% po(\"torch_loss\", t_loss(\"cross_entropy\")) %>>% po(\"torch_optimizer\", t_opt(\"adam\")) %>>% po(\"torch_model_classif\",   batch_size = 50,   epochs = 3)  task = tsk(\"iris\") network$train(task) #> $torch_model_classif.output #> NULL #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":null,"dir":"Reference","previous_headings":"","what":"CELU Activation Function — mlr_pipeops_nn_celu","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Applies element-wise, \\(CELU(x) = max(0,x) + min(0, \\alpha * (exp(x \\alpha) - 1))\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"alpha :: numeric(1) alpha value ELU formulation. Default: 1.0 inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Calls torch::nn_celu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchCELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"PipeOpTorchCELU$new() PipeOpTorchCELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"","code":"PipeOpTorchCELU$new(id = \"nn_celu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"","code":"PipeOpTorchCELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_celu\") pipeop #> PipeOp: <nn_celu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1:   alpha ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 2: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Convolution — mlr_pipeops_nn_conv1d","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Applies 1D convolution input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"out_channels :: integer(1) Number channels produced convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. default 1. padding :: integer() ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘TRUE’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Calls torch::nn_conv1d() trained. paramter in_channels inferred second dimension input tensor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConv -> PipeOpTorchConv1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"PipeOpTorchConv1D$new() PipeOpTorchConv1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"","code":"PipeOpTorchConv1D$new(id = \"nn_conv1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"","code":"PipeOpTorchConv1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv1d\", kernel_size = 10, out_channels = 1) pipeop #> PipeOp: <nn_conv1d> (not trained) #> values: <out_channels=1, kernel_size=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(8)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2:  kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>     10 #> 3:       stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:      padding ParamUty    NA    NA     Inf              0 [NULL] #> 5:     dilation ParamUty    NA    NA     Inf              1 [NULL] #> 6:       groups ParamInt     1   Inf     Inf              1 [NULL] #> 7:         bias ParamLgl    NA    NA       2           TRUE [NULL] #> 8: padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Convolution — mlr_pipeops_nn_conv2d","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Applies 2D convolution input image composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Calls torch::nn_conv2d() trained. paramter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"out_channels :: integer(1) Number channels produced convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. default 1. padding :: integer() ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘TRUE’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConv -> PipeOpTorchConv2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"PipeOpTorchConv2D$new() PipeOpTorchConv2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"","code":"PipeOpTorchConv2D$new(id = \"nn_conv2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"","code":"PipeOpTorchConv2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv2d\", kernel_size = 10, out_channels = 1) pipeop #> PipeOp: <nn_conv2d> (not trained) #> values: <out_channels=1, kernel_size=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(8)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2:  kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>     10 #> 3:       stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:      padding ParamUty    NA    NA     Inf              0 [NULL] #> 5:     dilation ParamUty    NA    NA     Inf              1 [NULL] #> 6:       groups ParamInt     1   Inf     Inf              1 [NULL] #> 7:         bias ParamLgl    NA    NA       2           TRUE [NULL] #> 8: padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Convolution — mlr_pipeops_nn_conv3d","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Applies 3D convolution input image composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Calls torch::nn_conv3d() trained. paramter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"out_channels :: integer(1) Number channels produced convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. default 1. padding :: integer() ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘TRUE’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConv -> PipeOpTorchConv3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"PipeOpTorchConv3D$new() PipeOpTorchConv3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"","code":"PipeOpTorchConv3D$new(id = \"nn_conv3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"","code":"PipeOpTorchConv3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv3d\", kernel_size = 10, out_channels = 1) pipeop #> PipeOp: <nn_conv3d> (not trained) #> values: <out_channels=1, kernel_size=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(8)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2:  kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>     10 #> 3:       stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:      padding ParamUty    NA    NA     Inf              0 [NULL] #> 5:     dilation ParamUty    NA    NA     Inf              1 [NULL] #> 6:       groups ParamInt     1   Inf     Inf              1 [NULL] #> 7:         bias ParamLgl    NA    NA       2           TRUE [NULL] #> 8: padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":null,"dir":"Reference","previous_headings":"","what":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Transpose 1D Convolution Transpose 1D Convolution","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"out_channels :: integer(1) Number output channels produce convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. Default: 1. padding ::  integer()` ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. output_padding ::integer() Additional size added one side output shape. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘True’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Calls nn_conv_transpose1d. parameter in_channels inferred second dimension input tensor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConvTranspose -> PipeOpTorchConvTranspose1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"PipeOpTorchConvTranspose1D$new() PipeOpTorchConvTranspose1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"","code":"PipeOpTorchConvTranspose1D$new(id = \"nn_conv_transpose1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"","code":"PipeOpTorchConvTranspose1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv_transpose1d\", kernel_size = 3, out_channels = 2) pipeop #> PipeOp: <nn_conv_transpose1d> (not trained) #> values: <out_channels=2, kernel_size=3> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(9)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:   out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      2 #> 2:    kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>      3 #> 3:         stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:        padding ParamUty    NA    NA     Inf              0 [NULL] #> 5: output_padding ParamUty    NA    NA     Inf              0 [NULL] #> 6:       dilation ParamUty    NA    NA     Inf              1 [NULL] #> 7:         groups ParamInt     1   Inf     Inf              1 [NULL] #> 8:           bias ParamLgl    NA    NA       2           TRUE [NULL] #> 9:   padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":null,"dir":"Reference","previous_headings":"","what":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Applies 2D transposed convolution operator input image composed several input planes, sometimes also called \"deconvolution\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Calls nn_conv_transpose2d. parameter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"out_channels :: integer(1) Number output channels produce convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. Default: 1. padding ::  integer()` ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. output_padding ::integer() Additional size added one side output shape. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘True’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConvTranspose -> PipeOpTorchConvTranspose2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"PipeOpTorchConvTranspose2D$new() PipeOpTorchConvTranspose2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"","code":"PipeOpTorchConvTranspose2D$new(id = \"nn_conv_transpose2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"","code":"PipeOpTorchConvTranspose2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv_transpose2d\", kernel_size = 3, out_channels = 2) pipeop #> PipeOp: <nn_conv_transpose2d> (not trained) #> values: <out_channels=2, kernel_size=3> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(9)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:   out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      2 #> 2:    kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>      3 #> 3:         stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:        padding ParamUty    NA    NA     Inf              0 [NULL] #> 5: output_padding ParamUty    NA    NA     Inf              0 [NULL] #> 6:       dilation ParamUty    NA    NA     Inf              1 [NULL] #> 7:         groups ParamInt     1   Inf     Inf              1 [NULL] #> 8:           bias ParamLgl    NA    NA       2           TRUE [NULL] #> 9:   padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":null,"dir":"Reference","previous_headings":"","what":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Applies 3D transposed convolution operator input image composed several input planes, sometimes also called \"deconvolution\"","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Calls nn_conv_transpose3d. parameter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"out_channels :: integer(1) Number output channels produce convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. Default: 1. padding ::  integer()` ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. output_padding ::integer() Additional size added one side output shape. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘True’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConvTranspose -> PipeOpTorchConvTranspose3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"PipeOpTorchConvTranspose3D$new() PipeOpTorchConvTranspose3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"","code":"PipeOpTorchConvTranspose3D$new(id = \"nn_conv_transpose3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"","code":"PipeOpTorchConvTranspose3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv_transpose3d\", kernel_size = 3, out_channels = 2) pipeop #> PipeOp: <nn_conv_transpose3d> (not trained) #> values: <out_channels=2, kernel_size=3> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(9)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:   out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      2 #> 2:    kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>      3 #> 3:         stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:        padding ParamUty    NA    NA     Inf              0 [NULL] #> 5: output_padding ParamUty    NA    NA     Inf              0 [NULL] #> 6:       dilation ParamUty    NA    NA     Inf              1 [NULL] #> 7:         groups ParamInt     1   Inf     Inf              1 [NULL] #> 8:           bias ParamLgl    NA    NA       2           TRUE [NULL] #> 9:   padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":null,"dir":"Reference","previous_headings":"","what":"Dropout — mlr_pipeops_nn_dropout","title":"Dropout — mlr_pipeops_nn_dropout","text":"training, randomly zeroes elements input tensor probability p using samples Bernoulli distribution.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Dropout — mlr_pipeops_nn_dropout","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Dropout — mlr_pipeops_nn_dropout","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Dropout — mlr_pipeops_nn_dropout","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Dropout — mlr_pipeops_nn_dropout","text":"p :: numeric(1) Probability element zeroed. Default: 0.5 inplace inplace :: logical(1) set TRUE, operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Dropout — mlr_pipeops_nn_dropout","text":"Calls torch::nn_dropout() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Dropout — mlr_pipeops_nn_dropout","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchDropout","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Dropout — mlr_pipeops_nn_dropout","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Dropout — mlr_pipeops_nn_dropout","text":"PipeOpTorchDropout$new() PipeOpTorchDropout$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Dropout — mlr_pipeops_nn_dropout","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Dropout — mlr_pipeops_nn_dropout","text":"","code":"PipeOpTorchDropout$new(id = \"nn_dropout\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Dropout — mlr_pipeops_nn_dropout","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Dropout — mlr_pipeops_nn_dropout","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Dropout — mlr_pipeops_nn_dropout","text":"","code":"PipeOpTorchDropout$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Dropout — mlr_pipeops_nn_dropout","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Dropout — mlr_pipeops_nn_dropout","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_dropout\") pipeop #> PipeOp: <nn_dropout> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1:       p ParamDbl     0     1     Inf     0.5 [NULL] #> 2: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":null,"dir":"Reference","previous_headings":"","what":"ELU Activation Function — mlr_pipeops_nn_elu","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Applies element-wise, $$ELU(x) = max(0,x) + min(0, \\alpha * (exp(x) - 1))$$.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"alpha :: numeric(1) alpha value ELU formulation. Default: 1.0 inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Calls torch::nn_elu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"PipeOpTorchELU$new() PipeOpTorchELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"","code":"PipeOpTorchELU$new(id = \"nn_elu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"","code":"PipeOpTorchELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_elu\") pipeop #> PipeOp: <nn_elu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1:   alpha ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 2: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":null,"dir":"Reference","previous_headings":"","what":"Flattens a Tensor — mlr_pipeops_nn_flatten","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"use nn_sequential.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"start_dim :: integer(1) wich dimension start flattening. Default 2. end_dim :: integer(1) wich dimension stop flattening. Default -1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"Calls torch::nn_flatten() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchFlatten","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"PipeOpTorchFlatten$new() PipeOpTorchFlatten$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"","code":"PipeOpTorchFlatten$new(id = \"nn_flatten\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"","code":"PipeOpTorchFlatten$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_flatten\") pipeop #> PipeOp: <nn_flatten> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <int> <num>   <num>  <list> <list> #> 1: start_dim ParamInt     1   Inf     Inf       2 [NULL] #> 2:   end_dim ParamInt     1   Inf     Inf      -1 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":null,"dir":"Reference","previous_headings":"","what":"GELU Activation Function — mlr_pipeops_nn_gelu","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Gelu","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"approximate :: character(1) Whether use approximation algorithm. Default \"none\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Calls torch::nn_gelu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchGELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"PipeOpTorchGELU$new() PipeOpTorchGELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"","code":"PipeOpTorchGELU$new(id = \"nn_gelu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"","code":"PipeOpTorchGELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_gelu\") pipeop #> PipeOp: <nn_gelu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels default  value #>         <char>   <char> <num> <num>   <num>  <list> <list> #> 1: approximate ParamFct    NA    NA       2    none [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":null,"dir":"Reference","previous_headings":"","what":"GLU Activation Function — mlr_pipeops_nn_glu","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"gated linear unit. Computes:","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"dim :: integer(1) Dimension split input. Default: -1","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"Calls torch::nn_glu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchGLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"PipeOpTorchGLU$new() PipeOpTorchGLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"","code":"PipeOpTorchGLU$new(id = \"nn_glu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"","code":"PipeOpTorchGLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_glu\") pipeop #> PipeOp: <nn_glu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <int> <num>   <num>  <list> <list> #> 1:    dim ParamInt     1   Inf     Inf      -1 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":null,"dir":"Reference","previous_headings":"","what":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Applies hard shrinkage function element-wise","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"lambd :: numeric(1) lambda value Hardshrink formulation formulation. Default 0.5.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Calls torch::nn_hardshrink() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHardShrink","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"PipeOpTorchHardShrink$new() PipeOpTorchHardShrink$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"","code":"PipeOpTorchHardShrink$new(id = \"nn_hardshrink\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"","code":"PipeOpTorchHardShrink$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_hardshrink\") pipeop #> PipeOp: <nn_hardshrink> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:  lambd ParamDbl  -Inf   Inf     Inf     0.5 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":null,"dir":"Reference","previous_headings":"","what":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Applies element-wise function \\(\\mbox{Hardsigmoid}(x) = \\frac{ReLU6(x + 3)}{6}\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Calls torch::nn_hardsigmoid() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHardSigmoid","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"PipeOpTorchHardSigmoid$new() PipeOpTorchHardSigmoid$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"","code":"PipeOpTorchHardSigmoid$new(id = \"nn_hardsigmoid\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"","code":"PipeOpTorchHardSigmoid$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_hardsigmoid\") pipeop #> PipeOp: <nn_hardsigmoid> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":null,"dir":"Reference","previous_headings":"","what":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Applies HardTanh function element-wise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"min_val :: numeric(1) Minimum value linear region range. Default: -1. max_val :: numeric(1) Maximum value linear region range. Default: 1. inplace :: logical(1) Can optionally operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Calls torch::nn_hardtanh() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHardTanh","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"PipeOpTorchHardTanh$new() PipeOpTorchHardTanh$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"","code":"PipeOpTorchHardTanh$new(id = \"nn_hardtanh\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"","code":"PipeOpTorchHardTanh$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_hardtanh\") pipeop #> PipeOp: <nn_hardtanh> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: min_val ParamDbl  -Inf   Inf     Inf      -1 [NULL] #> 2: max_val ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 3: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":null,"dir":"Reference","previous_headings":"","what":"Output Head — mlr_pipeops_nn_head","title":"Output Head — mlr_pipeops_nn_head","text":"Output head classification regresssion. NOTE method $shapes_out() access task, returns c(NA, NA). PipeOp trained however, model descriptor correct output shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Output Head — mlr_pipeops_nn_head","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Output Head — mlr_pipeops_nn_head","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Output Head — mlr_pipeops_nn_head","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Output Head — mlr_pipeops_nn_head","text":"bias :: logical(1) Whether use bias. Default TRUE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Output Head — mlr_pipeops_nn_head","text":"Calls torch::nn_linear() input output features inferred input shape / task.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Output Head — mlr_pipeops_nn_head","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHead","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Output Head — mlr_pipeops_nn_head","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Output Head — mlr_pipeops_nn_head","text":"PipeOpTorchHead$new() PipeOpTorchHead$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Output Head — mlr_pipeops_nn_head","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Output Head — mlr_pipeops_nn_head","text":"","code":"PipeOpTorchHead$new(id = \"nn_head\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Output Head — mlr_pipeops_nn_head","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Output Head — mlr_pipeops_nn_head","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Output Head — mlr_pipeops_nn_head","text":"","code":"PipeOpTorchHead$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Output Head — mlr_pipeops_nn_head","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Output Head — mlr_pipeops_nn_head","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_head\") pipeop #> PipeOp: <nn_head> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:   bias ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":null,"dir":"Reference","previous_headings":"","what":"Layer Normalization — mlr_pipeops_nn_layer_norm","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Applies Layer Normalization last certain number dimensions.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"dims :: integer(1) number dimensions normalized (starting last dimension). elementwise_affine :: logical(1) Whether learn affine-linear parameters initialized 1 weights 0 biases. default TRUE. eps :: numeric(1) value added denominator numerical stability.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Calls torch::nn_layer_norm() trained. parameter normalized_shape inferred dimensions last dims dimensions input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLayerNorm","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"PipeOpTorchLayerNorm$new() PipeOpTorchLayerNorm$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"","code":"PipeOpTorchLayerNorm$new(id = \"nn_layer_norm\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"","code":"PipeOpTorchLayerNorm$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_layer_norm\", dims = 1) pipeop #> PipeOp: <nn_layer_norm> (not trained) #> values: <dims=1> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>                    id    class lower upper nlevels        default  value #>                <char>   <char> <num> <num>   <num>         <list> <list> #> 1:               dims ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2: elementwise_affine ParamLgl    NA    NA       2           TRUE [NULL] #> 3:                eps ParamDbl     0   Inf     Inf          1e-05 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":null,"dir":"Reference","previous_headings":"","what":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Applies element-wise, \\(LeakyReLU(x) = max(0, x) + negative_slope * min(0, x)\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"negative_slope :: numeric(1) Controls angle negative slope. Default: 1e-2. inplace :: logical(1) Can optionally operation -place. Default: ‘FALSE’.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Calls torch::nn_hardswish() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLeakyReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"PipeOpTorchLeakyReLU$new() PipeOpTorchLeakyReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"","code":"PipeOpTorchLeakyReLU$new(id = \"nn_leaky_relu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"","code":"PipeOpTorchLeakyReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_leaky_relu\") pipeop #> PipeOp: <nn_leaky_relu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>                id    class lower upper nlevels default  value #>            <char>   <char> <num> <num>   <num>  <list> <list> #> 1: negative_slope ParamDbl  -Inf   Inf     Inf    0.01 [NULL] #> 2:        inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":null,"dir":"Reference","previous_headings":"","what":"Linear Layer — mlr_pipeops_nn_linear","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Applies linear transformation incoming data: \\(y = xA^T + b\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Linear Layer — mlr_pipeops_nn_linear","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Linear Layer — mlr_pipeops_nn_linear","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Linear Layer — mlr_pipeops_nn_linear","text":"out_features :: integer(1) output features linear layer. bias :: logical(1) Whether use bias. Default TRUE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Calls torch::nn_linear() trained parameter in_features inferred second last dimension input tensor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Linear Layer — mlr_pipeops_nn_linear","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLinear","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Linear Layer — mlr_pipeops_nn_linear","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Linear Layer — mlr_pipeops_nn_linear","text":"PipeOpTorchLinear$new() PipeOpTorchLinear$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Linear Layer — mlr_pipeops_nn_linear","text":"","code":"PipeOpTorchLinear$new(id = \"nn_linear\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Linear Layer — mlr_pipeops_nn_linear","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Linear Layer — mlr_pipeops_nn_linear","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Linear Layer — mlr_pipeops_nn_linear","text":"","code":"PipeOpTorchLinear$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Linear Layer — mlr_pipeops_nn_linear","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Linear Layer — mlr_pipeops_nn_linear","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_linear\", out_features = 10) pipeop #> PipeOp: <nn_linear> (not trained) #> values: <out_features=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_features ParamInt     1   Inf     Inf <NoDefault[0]>     10 #> 2:         bias ParamLgl    NA    NA       2           TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":null,"dir":"Reference","previous_headings":"","what":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Applies element-wise \\(LogSigmoid(x_i) = log(\\frac{1}{1 + exp(-x_i)})\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Calls torch::nn_log_sigmoid() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLogSigmoid","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"PipeOpTorchLogSigmoid$new() PipeOpTorchLogSigmoid$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"","code":"PipeOpTorchLogSigmoid$new(id = \"nn_log_sigmoid\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"","code":"PipeOpTorchLogSigmoid$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_log_sigmoid\") pipeop #> PipeOp: <nn_log_sigmoid> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Applies 1D max pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"return_indices FALSE construction, one input channel 'input' one output channel 'output'. return_indices TRUE, two output channels 'output' 'indices'. explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"kernel_size :: integer() size window. Can single number vector. stride :: (integer(1)) stride window. Can single number vector. Default: kernel_size padding :: integer() Implicit zero paddings sides input. Can single number tuple (padW,). Default: 0 dilation :: integer() Controls spacing kernel points; also known à trous algorithm. Default: 1 ceil_mode :: logical(1) True, use ceil instead floor compute output shape. Default: FALSE","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Calls torch::nn_max_pool1d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMaxPool -> PipeOpTorchMaxPool1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"PipeOpTorchMaxPool1D$new() PipeOpTorchMaxPool1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"","code":"PipeOpTorchMaxPool1D$new(   id = \"nn_max_pool1d\",   return_indices = FALSE,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"id (character(1)) Identifier resulting  object. return_indices (logical(1)) Whether return indices. TRUE, two output channels \"output\" \"indices\". param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"","code":"PipeOpTorchMaxPool1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_max_pool1d\") pipeop #> PipeOp: <nn_max_pool1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:     padding ParamUty    NA    NA     Inf              0 [NULL] #> 3:      stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 4:    dilation ParamInt  -Inf   Inf     Inf              1 [NULL] #> 5:   ceil_mode ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Applies 2D max pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Calls torch::nn_max_pool2d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"return_indices FALSE construction, one input channel 'input' one output channel 'output'. return_indices TRUE, two output channels 'output' 'indices'. explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"kernel_size :: integer() size window. Can single number vector. stride :: (integer(1)) stride window. Can single number vector. Default: kernel_size padding :: integer() Implicit zero paddings sides input. Can single number tuple (padW,). Default: 0 dilation :: integer() Controls spacing kernel points; also known à trous algorithm. Default: 1 ceil_mode :: logical(1) True, use ceil instead floor compute output shape. Default: FALSE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMaxPool -> PipeOpTorchMaxPool2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"PipeOpTorchMaxPool2D$new() PipeOpTorchMaxPool2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"","code":"PipeOpTorchMaxPool2D$new(   id = \"nn_max_pool2d\",   return_indices = FALSE,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"id (character(1)) Identifier resulting  object. return_indices (logical(1)) Whether return indices. TRUE, two output channels \"output\" \"indices\". param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"","code":"PipeOpTorchMaxPool2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_max_pool2d\") pipeop #> PipeOp: <nn_max_pool2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:     padding ParamUty    NA    NA     Inf              0 [NULL] #> 3:      stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 4:    dilation ParamInt  -Inf   Inf     Inf              1 [NULL] #> 5:   ceil_mode ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Applies 3D max pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Calls torch::nn_max_pool3d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"return_indices FALSE construction, one input channel 'input' one output channel 'output'. return_indices TRUE, two output channels 'output' 'indices'. explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"kernel_size :: integer() size window. Can single number vector. stride :: (integer(1)) stride window. Can single number vector. Default: kernel_size padding :: integer() Implicit zero paddings sides input. Can single number tuple (padW,). Default: 0 dilation :: integer() Controls spacing kernel points; also known à trous algorithm. Default: 1 ceil_mode :: logical(1) True, use ceil instead floor compute output shape. Default: FALSE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMaxPool -> PipeOpTorchMaxPool3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"PipeOpTorchMaxPool3D$new() PipeOpTorchMaxPool3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"","code":"PipeOpTorchMaxPool3D$new(   id = \"nn_max_pool3d\",   return_indices = FALSE,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"id (character(1)) Identifier resulting  object. return_indices (logical(1)) Whether return indices. TRUE, two output channels \"output\" \"indices\". param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"","code":"PipeOpTorchMaxPool3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_max_pool3d\") pipeop #> PipeOp: <nn_max_pool3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:     padding ParamUty    NA    NA     Inf              0 [NULL] #> 3:      stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 4:    dilation ParamInt  -Inf   Inf     Inf              1 [NULL] #> 5:   ceil_mode ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge Operation — mlr_pipeops_nn_merge","title":"Merge Operation — mlr_pipeops_nn_merge","text":"Base class merge operations addition (PipeOpTorchMergeSum), multiplication (PipeOpTorchMergeProd concatenation (PipeOpTorchMergeCat).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge Operation — mlr_pipeops_nn_merge","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge Operation — mlr_pipeops_nn_merge","text":"PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge Operation — mlr_pipeops_nn_merge","text":"See respective child class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge Operation — mlr_pipeops_nn_merge","text":"Per default, private$.shapes_out() method outputs broadcasted tensors. two things aware: NAs assumed batch (almost always batch size first dimension). Tensors expected number dimensions, .e. missing dimensions filled 1s. reason first dimension batch dimension. private method can overwritten PipeOpTorchs inheriting class.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge Operation — mlr_pipeops_nn_merge","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchMerge","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge Operation — mlr_pipeops_nn_merge","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge Operation — mlr_pipeops_nn_merge","text":"PipeOpTorchMerge$new() PipeOpTorchMerge$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge Operation — mlr_pipeops_nn_merge","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge Operation — mlr_pipeops_nn_merge","text":"","code":"PipeOpTorchMerge$new(   id,   module_generator,   param_set = ps(),   innum = 0,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge Operation — mlr_pipeops_nn_merge","text":"id (character(1)) Identifier resulting  object. module_generator (nn_module_generator) torch module generator. param_set (ParamSet) parameter set. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge Operation — mlr_pipeops_nn_merge","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge Operation — mlr_pipeops_nn_merge","text":"","code":"PipeOpTorchMerge$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge Operation — mlr_pipeops_nn_merge","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Concatenates multiple tensors given dimension. broadcasting rules applied , must reshape tensors shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch. PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"dim :: integer(1) dimension along concatenate tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Calls nn_merge_cat() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMerge -> PipeOpTorchMergeCat","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"PipeOpTorchMergeCat$new() PipeOpTorchMergeCat$speak() PipeOpTorchMergeCat$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"PipeOpTorchMergeCat$new(id = \"nn_merge_cat\", innum = 0, param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"id (character(1)) Identifier resulting  object. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"method-speak-","dir":"Reference","previous_headings":"","what":"Method speak()","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"cat say?","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"PipeOpTorchMergeCat$speak()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"PipeOpTorchMergeCat$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_merge_cat\") pipeop #> PipeOp: <nn_merge_cat> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   ... [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:    dim ParamInt  -Inf   Inf     Inf      -1 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge by Product — mlr_pipeops_nn_merge_prod","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Calculates product input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch. PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Calls nn_merge_prod() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMerge -> PipeOpTorchMergeProd","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"PipeOpTorchMergeProd$new() PipeOpTorchMergeProd$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"","code":"PipeOpTorchMergeProd$new(id = \"nn_merge_prod\", innum = 0, param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"id (character(1)) Identifier resulting  object. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"","code":"PipeOpTorchMergeProd$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_merge_prod\") pipeop #> PipeOp: <nn_merge_prod> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   ... [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge by Summation — mlr_pipeops_nn_merge_sum","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Calculates sum input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch. PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Calls nn_merge_sum() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMerge -> PipeOpTorchMergeSum","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"PipeOpTorchMergeSum$new() PipeOpTorchMergeSum$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"","code":"PipeOpTorchMergeSum$new(id = \"nn_merge_sum\", innum = 0, param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"id (character(1)) Identifier resulting  object. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"","code":"PipeOpTorchMergeSum$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_merge_sum\") pipeop #> PipeOp: <nn_merge_sum> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   ... [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":null,"dir":"Reference","previous_headings":"","what":"PReLU Activation Function — mlr_pipeops_nn_prelu","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Applies element-wise function \\(PReLU(x) = max(0,x) + weight * min(0,x)\\) weight learnable parameter.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"num_parameters :: integer(1): Number learn. Although takes int input, two values legitimate: 1, number channels input. Default: 1. init :: numeric(1) T initial value . Default: 0.25.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Calls torch::nn_prelu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchPReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"PipeOpTorchPReLU$new() PipeOpTorchPReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"","code":"PipeOpTorchPReLU$new(id = \"nn_prelu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"","code":"PipeOpTorchPReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_prelu\") pipeop #> PipeOp: <nn_prelu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>                id    class lower upper nlevels default  value #>            <char>   <char> <num> <num>   <num>  <list> <list> #> 1: num_parameters ParamInt     1   Inf     Inf       1 [NULL] #> 2:           init ParamDbl  -Inf   Inf     Inf    0.25 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":null,"dir":"Reference","previous_headings":"","what":"ReLU Activation Function — mlr_pipeops_nn_relu","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Applies rectified linear unit function element-wise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Calls torch::nn_relu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"PipeOpTorchReLU$new() PipeOpTorchReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"","code":"PipeOpTorchReLU$new(id = \"nn_relu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"","code":"PipeOpTorchReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_relu\") pipeop #> PipeOp: <nn_relu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":null,"dir":"Reference","previous_headings":"","what":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Applies element-wise function \\(ReLU6(x) = min(max(0,x), 6)\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Calls torch::nn_relu6() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchReLU6","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"PipeOpTorchReLU6$new() PipeOpTorchReLU6$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"","code":"PipeOpTorchReLU6$new(id = \"nn_relu6\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"","code":"PipeOpTorchReLU6$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_relu6\") pipeop #> PipeOp: <nn_relu6> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":null,"dir":"Reference","previous_headings":"","what":"Reshape a Tensor — mlr_pipeops_nn_reshape","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Reshape tensor given shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"shape :: integer(1) desired output shape. Unknown dimension (one ) can either specified -1 NA.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Calls nn_reshape() trained. internally calls torch::torch_reshape() given shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchReshape","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"PipeOpTorchReshape$new() PipeOpTorchReshape$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"","code":"PipeOpTorchReshape$new(id = \"nn_reshape\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"","code":"PipeOpTorchReshape$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_reshape\") pipeop #> PipeOp: <nn_reshape> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <num> <num>   <num>         <list> <list> #> 1:  shape ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":null,"dir":"Reference","previous_headings":"","what":"RReLU Activation Function — mlr_pipeops_nn_rrelu","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Randomized leaky ReLU.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"lower:: numeric(1) Lower bound uniform distribution. Default: 1/8. upper:: numeric(1) Upper bound uniform distribution. Default: 1/3. inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Calls torch::nn_rrelu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchRReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"PipeOpTorchRReLU$new() PipeOpTorchRReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"","code":"PipeOpTorchRReLU$new(id = \"nn_rrelu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"","code":"PipeOpTorchRReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_rrelu\") pipeop #> PipeOp: <nn_rrelu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>         id    class lower upper nlevels   default  value #>     <char>   <char> <num> <num>   <num>    <list> <list> #> 1:   lower ParamDbl  -Inf   Inf     Inf     0.125 [NULL] #> 2:   upper ParamDbl  -Inf   Inf     Inf 0.3333333 [NULL] #> 3: inplace ParamLgl    NA    NA       2     FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":null,"dir":"Reference","previous_headings":"","what":"SELU Activation Function — mlr_pipeops_nn_selu","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Applies element-wise, $$SELU(x) = scale * (max(0,x) + min(0, \\alpha * (exp(x) - 1)))$$, \\(\\alpha=1.6732632423543772848170429916717\\) \\(scale=1.0507009873554804934193349852946\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Calls torch::nn_selu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"PipeOpTorchSELU$new() PipeOpTorchSELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"","code":"PipeOpTorchSELU$new(id = \"nn_selu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"","code":"PipeOpTorchSELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_selu\") pipeop #> PipeOp: <nn_selu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":null,"dir":"Reference","previous_headings":"","what":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Applies element-wise \\(Sigmoid(x_i) = \\frac{1}{1 + exp(-x_i)}\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Calls torch::nn_sigmoid() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSigmoid","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"PipeOpTorchSigmoid$new() PipeOpTorchSigmoid$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"","code":"PipeOpTorchSigmoid$new(id = \"nn_sigmoid\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"","code":"PipeOpTorchSigmoid$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_sigmoid\") pipeop #> PipeOp: <nn_sigmoid> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":null,"dir":"Reference","previous_headings":"","what":"Softmax — mlr_pipeops_nn_softmax","title":"Softmax — mlr_pipeops_nn_softmax","text":"Applies softmax function.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Softmax — mlr_pipeops_nn_softmax","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Softmax — mlr_pipeops_nn_softmax","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Softmax — mlr_pipeops_nn_softmax","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Softmax — mlr_pipeops_nn_softmax","text":"dim :: integer(1) dimension along Softmax computed (every slice along dim sum 1).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Softmax — mlr_pipeops_nn_softmax","text":"Calls torch::nn_softmax() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Softmax — mlr_pipeops_nn_softmax","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftmax","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Softmax — mlr_pipeops_nn_softmax","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Softmax — mlr_pipeops_nn_softmax","text":"PipeOpTorchSoftmax$new() PipeOpTorchSoftmax$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Softmax — mlr_pipeops_nn_softmax","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Softmax — mlr_pipeops_nn_softmax","text":"","code":"PipeOpTorchSoftmax$new(id = \"nn_softmax\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Softmax — mlr_pipeops_nn_softmax","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Softmax — mlr_pipeops_nn_softmax","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Softmax — mlr_pipeops_nn_softmax","text":"","code":"PipeOpTorchSoftmax$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Softmax — mlr_pipeops_nn_softmax","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Softmax — mlr_pipeops_nn_softmax","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softmax\") pipeop #> PipeOp: <nn_softmax> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <int> <num>   <num>         <list> <list> #> 1:    dim ParamInt     1   Inf     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":null,"dir":"Reference","previous_headings":"","what":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Applies element-wise, function \\(Softplus(x) = 1/\\beta * log(1 + exp(\\beta * x))\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"beta :: numeric(1) beta value Softplus formulation. Default: 1 threshold :: numeric(1) Values revert linear function. Default: 20","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Calls torch::nn_softplus() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftPlus","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"PipeOpTorchSoftPlus$new() PipeOpTorchSoftPlus$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"","code":"PipeOpTorchSoftPlus$new(id = \"nn_softplus\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"","code":"PipeOpTorchSoftPlus$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softplus\") pipeop #> PipeOp: <nn_softplus> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <num> <num>   <num>  <list> <list> #> 1:      beta ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 2: threshold ParamDbl  -Inf   Inf     Inf      20 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":null,"dir":"Reference","previous_headings":"","what":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Applies soft shrinkage function elementwise","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"lamd :: numeric(1) lambda (must less zero) value Softshrink formulation. Default: 0.5","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Calls torch::nn_softshrink() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftShrink","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"PipeOpTorchSoftShrink$new() PipeOpTorchSoftShrink$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"","code":"PipeOpTorchSoftShrink$new(id = \"nn_softshrink\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"","code":"PipeOpTorchSoftShrink$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softshrink\") pipeop #> PipeOp: <nn_softshrink> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:  lambd ParamDbl  -Inf     1     Inf     0.5 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":null,"dir":"Reference","previous_headings":"","what":"SoftSign Activation Function — mlr_pipeops_nn_softsign","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Applies element-wise, function \\(SoftSign(x) = x/(1 + |x|\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Calls torch::nn_softsign() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftSign","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"PipeOpTorchSoftSign$new() PipeOpTorchSoftSign$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"","code":"PipeOpTorchSoftSign$new(id = \"nn_softsign\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"","code":"PipeOpTorchSoftSign$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softsign\") pipeop #> PipeOp: <nn_softsign> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Squeezes tensor calling torch::torch_squeeze() given dimension dim.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"dim :: integer(1) dimension squeeze. NULL, dimensions size 1 squeezed. Negative values interpreted downwards last dimension.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Calls nn_squeeze() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSqueeze","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"PipeOpTorchSqueeze$new() PipeOpTorchSqueeze$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"","code":"PipeOpTorchSqueeze$new(id = \"nn_squeeze\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"","code":"PipeOpTorchSqueeze$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_squeeze\") pipeop #> PipeOp: <nn_squeeze> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <num> <num>   <num>         <list> <list> #> 1:    dim ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":null,"dir":"Reference","previous_headings":"","what":"Tanh Activation Function — mlr_pipeops_nn_tanh","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Applies element-wise function:","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Calls torch::nn_tanh() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchTanh","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"PipeOpTorchTanh$new() PipeOpTorchTanh$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"","code":"PipeOpTorchTanh$new(id = \"nn_tanh\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"","code":"PipeOpTorchTanh$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_tanh\") pipeop #> PipeOp: <nn_tanh> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":null,"dir":"Reference","previous_headings":"","what":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Applies element-wise, \\(Tanhshrink(x) = x - Tanh(x)\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Calls torch::nn_tanhshrink() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchTanhShrink","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"PipeOpTorchTanhShrink$new() PipeOpTorchTanhShrink$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"","code":"PipeOpTorchTanhShrink$new(id = \"nn_tanhshrink\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"","code":"PipeOpTorchTanhShrink$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_tanhshrink\") pipeop #> PipeOp: <nn_tanhshrink> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":null,"dir":"Reference","previous_headings":"","what":"Treshold Activation Function — mlr_pipeops_nn_threshold","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Thresholds element input Tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"threshold :: numeric(1) value threshold . value :: numeric(1) value replace . inplace :: logical(1) Can optionally operation -place. Default: ‘FALSE’.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Calls torch::nn_threshold() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchThreshold","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"PipeOpTorchThreshold$new() PipeOpTorchThreshold$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"","code":"PipeOpTorchThreshold$new(id = \"nn_threshold\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"","code":"PipeOpTorchThreshold$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_threshold\", threshold = 1, value = 2) pipeop #> PipeOp: <nn_threshold> (not trained) #> values: <threshold=1, value=2> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>           id    class lower upper nlevels        default  value #>       <char>   <char> <num> <num>   <num>         <list> <list> #> 1: threshold ParamDbl  -Inf   Inf     Inf <NoDefault[0]>      1 #> 2:     value ParamDbl  -Inf   Inf     Inf <NoDefault[0]>      2 #> 3:   inplace ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Unqueeze Tensor Unqueeze Tensor","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"dim :: integer(1) dimension unsqueeze. Negative values interpreted downwards last dimension.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Calls nn_unsqueeze() trained. internally calls torch::torch_unsqueeze().","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchUnsqueeze","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"PipeOpTorchUnsqueeze$new() PipeOpTorchUnsqueeze$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"","code":"PipeOpTorchUnsqueeze$new(id = \"nn_unsqueeze\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"","code":"PipeOpTorchUnsqueeze$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_unsqueeze\") pipeop #> PipeOp: <nn_unsqueeze> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <num> <num>   <num>         <list> <list> #> 1:    dim ParamInt  -Inf   Inf     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"PipeOp can used preprocess (one ) lazy_tensor columns contained mlr3::Task. preprocessing function specified construction argument fn additional arguments function can defined PipeOp's parameter set. preprocessing done per column, .e. number lazy tensor output columns equal number lazy tensor input columns. create custom preprocessing PipeOps can use pipeop_preproc_torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"addition specifying construction arguments, can overwrite private .shapes_out() method. overwrite , output shapes assumed unknown (NULL). .shapes_out(shapes_in, param_vals, task) (list(), list(), TaskorNULL) -> list()\\cr private method calculates output shapes lazy tensor columns created applying preprocessing function provided parameter values (param_vals). taskis rarely needed, checked notNULL`. private method responsibility calculate output shapes one input column, .e. input shapes_in can assumed exactly one shape vector must calculate output shapes return list() length 1. can also assumed shape NULL (.e. unknown). Also, first dimension can NA, .e. unknown (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"See PipeOpTaskPreproc.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"addition state elements PipeOpTaskPreprocSimple, state also contains $param_vals set training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"addition parameters inherited PipeOpTaskPreproc well specified construction argument param_set following parameters: stages :: character(1) stages apply preprocessing. Can one \"train\", \"predict\" \"\". initial value parameter set \"train\" PipeOp's id starts \"augment_\" \"\" otherwise. Note preprocessing applied $predict() uses parameters set $train() set performing prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"$train() / $predict(), PipeOpModule one input one output channel created. pipeop applies function fn input tensor additionally passing parameter values (minus stages affect_columns) fn. preprocessing graph lazy tensor columns shallowly cloned PipeOpModule added. done avoid modifying user input means identical PipeOpModules can part different preprocessing graphs. possible, created PipeOpModule stateless. later point graph, preprocessing graphs merged possible avoid unnecessary computation. best illustrated example: One lazy tensor column's preprocessing graph -> B. , two branches created B -> C B -> D, creating two preprocessing graphs -> B -> C -> B -> D. loading data, want run preprocessing , .e. want run -> B part twice. reason, task_dataset() try merge graphs cache results graphs. However, graphs using dataset can currently merged. Also, shapes created $train() $predict() might differ. avoid creation graphs predict shapes incompatible train shapes, hypothetical predict shapes already calculated $train() (parameters set train also used predict) PipeOpTorchModel check train predict shapes compatibility starting training. Otherwise, mechanism similar ModelDescriptor construct.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpTaskPreproc -> PipeOpTaskPreprocTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"fn preprocessing function. rowwise Whether preprocessing applied rowwise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"PipeOpTaskPreprocTorch$new() PipeOpTaskPreprocTorch$shapes_out() PipeOpTaskPreprocTorch$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"PipeOpTaskPreprocTorch$new(   fn,   id = \"preproc_torch\",   param_vals = list(),   param_set = ps(),   packages = character(0),   rowwise = FALSE,   stages_init = NULL,   tags = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"fn (function character(2)) preprocessing function. Must modify input -place. character(2), first element namespace second element name. preprocessing function applied tensor, tensor passed position first argument. param_set inferred (left NULL) assumed first argument torch_tensor. id (character(1)) id new object. param_vals (named list()) Parameter values set construction. param_set (ParamSet) case function fn takes additional parameter besides torch_tensor can specfied parameters. None parameters can \"predict\" tag. tags include \"train\". packages (character()) packages preprocessing function depends . rowwise (logical(1)) Whether preprocessing function applied rowwise (concatenated row) directly whole tensor. first case batch dimension. stages_init (character(1)) Initial value stages parameter. tags (character()) Tags pipeop.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"method-shapes-out-","dir":"Reference","previous_headings":"","what":"Method shapes_out()","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"Calculates output shapes result applying preprocessing one lazy tensor columns provided shape. Names ignored order matters. uses parameter values currently set.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"PipeOpTaskPreprocTorch$shapes_out(shapes_in, stage = NULL, task = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"shapes_in (list() (integer() NULL)) input input shapes lazy tensors. NULL indicates shape unknown. First dimension must NA (NULL). stage (character(1)) stage: either \"train\" \"predict\". task (Task NULL) task, rarely needed.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"list() (integer() NULL)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"PipeOpTaskPreprocTorch$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"# Creating a simple task d = data.table(   x1 = as_lazy_tensor(rnorm(10)),   x2 = as_lazy_tensor(rnorm(10)),   x3 = as_lazy_tensor(as.double(1:10)),   y = rnorm(10) )  taskin = as_task_regr(d, target = \"y\")  # Creating a simple preprocessing pipeop po_simple = po(\"preproc_torch\",   # get rid of environment baggage   fn = mlr3misc::crate(function(x, a) x + a),   param_set = paradox::ps(a = paradox::p_int(tags = c(\"train\", \"required\"))) )  po_simple$param_set$set_values(   a = 100,   affect_columns = selector_name(c(\"x1\", \"x2\")),   stages = \"both\" # use during train and predict )  taskout_train = po_simple$train(list(taskin))[[1L]] materialize(taskout_train$data(cols = c(\"x1\", \"x2\")), rbind = TRUE) #> $x1 #> torch_tensor #>  100.6242 #>   99.3046 #>   99.2164 #>   99.0469 #>  101.7928 #>  100.3490 #>  100.2591 #>   99.1940 #>  100.1057 #>   99.6664 #> [ CPUFloatType{10,1} ] #>  #> $x2 #> torch_tensor #>  101.6418 #>   99.3561 #>  100.5870 #>   99.8496 #>   98.2892 #>  101.4310 #>   97.3548 #>   98.9675 #>   99.2925 #>   99.2994 #> [ CPUFloatType{10,1} ] #>   taskout_predict_noaug = po_simple$predict(list(taskin))[[1L]] materialize(taskout_predict_noaug$data(cols = c(\"x1\", \"x2\")), rbind = TRUE) #> $x1 #> torch_tensor #>  100.6242 #>   99.3046 #>   99.2164 #>   99.0469 #>  101.7928 #>  100.3490 #>  100.2591 #>   99.1940 #>  100.1057 #>   99.6664 #> [ CPUFloatType{10,1} ] #>  #> $x2 #> torch_tensor #>  101.6418 #>   99.3561 #>  100.5870 #>   99.8496 #>   98.2892 #>  101.4310 #>   97.3548 #>   98.9675 #>   99.2925 #>   99.2994 #> [ CPUFloatType{10,1} ] #>   po_simple$param_set$set_values(   stages = \"train\" )  # transformation is not applied taskout_predict_aug = po_simple$predict(list(taskin))[[1L]] materialize(taskout_predict_aug$data(cols = c(\"x1\", \"x2\")), rbind = TRUE) #> $x1 #> torch_tensor #>  100.6242 #>   99.3046 #>   99.2164 #>   99.0469 #>  101.7928 #>  100.3490 #>  100.2591 #>   99.1940 #>  100.1057 #>   99.6664 #> [ CPUFloatType{10,1} ] #>  #> $x2 #> torch_tensor #>  101.6418 #>   99.3561 #>  100.5870 #>   99.8496 #>   98.2892 #>  101.4310 #>   97.3548 #>   98.9675 #>   99.2925 #>   99.2994 #> [ CPUFloatType{10,1} ] #>   # Creating a more complex preprocessing PipeOp PipeOpPreprocTorchPoly = R6::R6Class(\"PipeOpPreprocTorchPoly\",  inherit = PipeOpTaskPreprocTorch,  public = list(    initialize = function(id = \"preproc_poly\", param_vals = list()) {      param_set = paradox::ps(        n_degree = paradox::p_int(lower = 1L, tags = c(\"train\", \"required\"))      )      param_set$set_values(        n_degree = 1L      )      fn = mlr3misc::crate(function(x, n_degree) {        torch::torch_cat(          lapply(seq_len(n_degree), function(d) torch::torch_pow(x, d)),          dim = 2L        )      })       super$initialize(        fn = fn,        id = id,        packages = character(0),        param_vals = param_vals,        param_set = param_set,        stages_init = \"both\"      )    }  ),  private = list(    .shapes_out = function(shapes_in, param_vals, task) {      # shapes_in is a list of length 1 containing the shapes      checkmate::assert_true(length(shapes_in[[1L]]) == 2L)      if (shapes_in[[1L]][2L] != 1L) {        stop(\"Input shape must be (NA, 1)\")      }      list(c(NA, param_vals$n_degree))    }  ) )  po_poly = PipeOpPreprocTorchPoly$new(   param_vals = list(n_degree = 3L, affect_columns = selector_name(\"x3\")) )  po_poly$shapes_out(list(c(NA, 1L)), stage = \"train\") #> [[1]] #> [1] NA  3 #>   taskout = po_poly$train(list(taskin))[[1L]] materialize(taskout$data(cols = \"x3\"), rbind = TRUE) #> $x3 #> torch_tensor #>     1     1     1 #>     2     4     8 #>     3     9    27 #>     4    16    64 #>     5    25   125 #>     6    36   216 #>     7    49   343 #>     8    64   512 #>     9    81   729 #>    10   100  1000 #> [ CPUFloatType{10,3} ] #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"PipeOpTorch base class PipeOps represent neural network layers Graph. training, generates PipeOpModule wraps nn_module attaches architecture, also represented Graph consisting mostly PipeOpModules PipeOpNOPs. former Graph operates ModelDescriptors, latter operates tensors. relationship PipeOpTorch PipeOpModule similar relationshop nn_module_generator (like nn_linear) nn_module (like output nn_linear(...)). crucial difference PipeOpTorch infers auxiliary parameters (like in_features nn_linear) automatically intermediate tensor shapes communicated ModelDescriptor. prediction, PipeOpTorch takes Task channel outputs new Task resulting feature union channel. one input output channel, task simply piped .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"inheriting class, one overload either private$.shapes_out() private$.shape_dependent_params() methods, overload private$.make_module(). .make_module(shapes_in, param_vals, task) (list(), list()) -> nn_module private method called generated nn_module passed argument module PipeOpModule. must overwritten, module_generator provided. left , calls provided module_generator arguments obtained private method .shape_dependent_params(). .shapes_out(shapes_in, param_vals, task) (list(), list(), Task NULL) -> named list() private method gets list numeric vectors (shapes_in), parameter values (param_vals), well (optional) Task. shapes_in can assumed order input names PipeOp. output shapes must order output names PipeOp. case output shapes depends task (case PipeOpTorchHead), function return valid output shapes (possibly containing NAs) task argument provided . .shape_dependent_params(shapes_in, param_vals, task) (list(), list()) -> named list() private method inputs .shapes_out. .make_module() overwritten, constructs arguments passed module_generator. Usually means must infer auxiliary parameters can inferred input shapes add user-supplied parameter values (param_vals).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"training, inputs outputs class ModelDescriptor. prediction, input output channels class Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"ParamSet specified child class inheriting PipeOpTorch. Usually parameters arguments wrapped nn_module minus auxiliary parameter can automatically inferred shapes input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"training, PipeOpTorch creates PipeOpModule given parameter specification input shapes incoming ModelDescriptors using private method .make_module(). input shapes provided slot pointer_shape incoming ModelDescriptors. channel names PipeOpModule identical channel names generating PipeOpTorch. model descriptor union incoming ModelDescriptors created. Note modifies graph first ModelDescriptor place efficiency. PipeOpModule added graph slot union edges connect sending PipeOpModules input channel PipeOpModule addeded graph. possible every incoming ModelDescriptor contains information id channel name sending PipeOp slot pointer. new graph model_descriptor_union represents current state neural network architecture. structurally similar subgraph consists pipeops class PipeOpTorch PipeOpTorchIngress ancestors PipeOpTorch. output, shallow copy ModelDescriptor created pointer pointer_shape updated accordingly. shallow copy means ModelDescriptors point Graph allows graph modified -reference different parts code.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"mlr3pipelines::PipeOp -> PipeOpTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"module_generator (nn_module_generator NULL) module generator wrapped PipeOpTorch. NULL, private method private$.make_module(shapes_in, param_vals) must overwritte, see section 'Inheriting'. change construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"PipeOpTorch$new() PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"","code":"PipeOpTorch$new(   id,   module_generator,   param_set = ps(),   param_vals = list(),   inname = \"input\",   outname = \"output\",   packages = \"torch\",   tags = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"id (character(1)) Identifier resulting  object. module_generator (nn_module_generator) torch module generator. param_set (ParamSet) parameter set. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction. inname (character()) names PipeOp's input channels. input channels generated PipeOpModule. Unless wrapped module_generator's forward method (present) argument ..., inname must identical argument names order avoid ambiguity. forward method argument ..., order input channels determines tensors passed wrapped nn_module. left NULL (default), argument module_generator must given argument names modue_generator's forward function set inname. outname (character())  names output channels channels. ouput channels generated PipeOpModule therefore also names list returned $train(). case one output channel, nn_module constructed PipeOp training must return named list(), names list names output channels. default \"output\". packages (character()) R packages object depends . tags (character()) tags PipeOp. tags \"torch\" always added.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"method-shapes-out-","dir":"Reference","previous_headings":"","what":"Method shapes_out()","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"Calculates output shapes given input shapes, parameters task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"","code":"PipeOpTorch$shapes_out(shapes_in, task = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"shapes_in (list() integer()) input input shapes, must order input channel names PipeOp. task (Task NULL) task, rarely used (default NULL). exception PipeOpTorchHead.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"named list() containing output shapes. names names output channels PipeOp.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"","code":"## Creating a neural network # In torch  task = tsk(\"iris\")  network_generator = torch::nn_module(   initialize = function(task, d_hidden) {     d_in = length(task$feature_names)     self$linear = torch::nn_linear(d_in, d_hidden)     self$output = if (task$task_type == \"regr\") {       torch::nn_linear(d_hidden, 1)     } else if (task$task_type == \"classif\") {       torch::nn_linear(d_hidden, length(task$class_names))     }   },   forward = function(x) {     x = self$linear(x)     x = torch::nnf_relu(x)     self$output(x)   } )  network = network_generator(task, d_hidden = 50) x = torch::torch_tensor(as.matrix(task$data(1, task$feature_names))) y = torch::with_no_grad(network(x))   # In mlr3torch network_generator = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 50) %>>%   po(\"nn_head\") md = network_generator$train(task)[[1L]] network = model_descriptor_to_module(md) y = torch::with_no_grad(network(torch_ingress_num.input = x))    ## Implementing a custom PipeOpTorch  # defining a custom module nn_custom = nn_module(\"nn_custom\",   initialize = function(d_in1, d_in2, d_out1, d_out2, bias = TRUE) {     self$linear1 = nn_linear(d_in1, d_out1, bias)     self$linear2 = nn_linear(d_in2, d_out2, bias)   },   forward = function(input1, input2) {     output1 = self$linear1(input1)     output2 = self$linear1(input2)      list(output1 = output1, output2 = output2)   } )  # wrapping the module into a custom PipeOpTorch  library(paradox)  PipeOpTorchCustom = R6::R6Class(\"PipeOpTorchCustom\",   inherit = PipeOpTorch,   public = list(     initialize = function(id = \"nn_custom\", param_vals = list()) {       param_set = ps(         d_out1 = p_int(lower = 1, tags = c(\"required\", \"train\")),         d_out2 = p_int(lower = 1, tags = c(\"required\", \"train\")),         bias = p_lgl(default = TRUE, tags = \"train\")       )       super$initialize(         id = id,         param_vals = param_vals,         param_set = param_set,         inname = c(\"input1\", \"input2\"),         outname = c(\"output1\", \"output2\"),         module_generator = nn_custom       )     }   ),   private = list(     .shape_dependent_params = function(shapes_in, param_vals, task) {       c(param_vals,         list(d_in1 = tail(shapes_in[[\"input1\"]], 1)), d_in2 = tail(shapes_in[[\"input2\"]], 1)       )     },     .shapes_out = function(shapes_in, param_vals, task) {       list(         input1 = c(head(shapes_in[[\"input1\"]], -1), param_vals$d_out1),         input2 = c(head(shapes_in[[\"input2\"]], -1), param_vals$d_out2)       )     }   ) )  ## Training  # generate input task = tsk(\"iris\") task1 = task$clone()$select(paste0(\"Sepal.\", c(\"Length\", \"Width\"))) task2 = task$clone()$select(paste0(\"Petal.\", c(\"Length\", \"Width\"))) graph = gunion(list(po(\"torch_ingress_num_1\"), po(\"torch_ingress_num_2\"))) mds_in = graph$train(list(task1, task2), single_input = FALSE)  mds_in[[1L]][c(\"graph\", \"task\", \"ingress\", \"pointer\", \"pointer_shape\")] #> $graph #> Graph with 1 PipeOps: #>                   ID         State sccssors prdcssors #>               <char>        <char>   <char>    <char> #>  torch_ingress_num_1 <<UNTRAINED>>                    #>  #> $task #> <TaskClassif:iris> (150 x 3): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (2): #>   - dbl (2): Sepal.Length, Sepal.Width #>  #> $ingress #> $ingress$torch_ingress_num_1.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) #>  #>  #> $pointer #> [1] \"torch_ingress_num_1\" \"output\"              #>  #> $pointer_shape #> [1] NA  2 #>  mds_in[[2L]][c(\"graph\", \"task\", \"ingress\", \"pointer\", \"pointer_shape\")] #> $graph #> Graph with 1 PipeOps: #>                   ID         State sccssors prdcssors #>               <char>        <char>   <char>    <char> #>  torch_ingress_num_2 <<UNTRAINED>>                    #>  #> $task #> <TaskClassif:iris> (150 x 3): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (2): #>   - dbl (2): Petal.Length, Petal.Width #>  #> $ingress #> $ingress$torch_ingress_num_2.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) #>  #>  #> $pointer #> [1] \"torch_ingress_num_2\" \"output\"              #>  #> $pointer_shape #> [1] NA  2 #>   # creating the PipeOpTorch and training it po_torch = PipeOpTorchCustom$new() po_torch$param_set$values = list(d_out1 = 10, d_out2 = 20) train_input = list(input1 = mds_in[[1L]], input2 = mds_in[[2L]]) mds_out = do.call(po_torch$train, args = list(input = train_input)) po_torch$state #> $output1 #> [1] NA 10 #>  #> $output2 #> [1] NA 20 #>   # the new model descriptors  # the resulting graphs are identical identical(mds_out[[1L]]$graph, mds_out[[2L]]$graph) #> [1] TRUE # not that as a side-effect, also one of the input graphs is modified in-place for efficiency mds_in[[1L]]$graph$edges #>                 src_id src_channel    dst_id dst_channel #>                 <char>      <char>    <char>      <char> #> 1: torch_ingress_num_1      output nn_custom      input1 #> 2: torch_ingress_num_2      output nn_custom      input2  # The new task has both Sepal and Petal features identical(mds_out[[1L]]$task, mds_out[[2L]]$task) #> [1] TRUE mds_out[[2L]]$task #> <TaskClassif:iris> (150 x 5): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (4): #>   - dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Width  # The new ingress slot contains all ingressors identical(mds_out[[1L]]$ingress, mds_out[[2L]]$ingress) #> [1] TRUE mds_out[[1L]]$ingress #> $torch_ingress_num_1.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) #>  #> $torch_ingress_num_2.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) #>   # The pointer and pointer_shape slots are different mds_out[[1L]]$pointer #> [1] \"nn_custom\" \"output1\"   mds_out[[2L]]$pointer #> [1] \"nn_custom\" \"output2\"    mds_out[[1L]]$pointer_shape #> [1] NA 10 mds_out[[2L]]$pointer_shape #> [1] NA 20  ## Prediction predict_input = list(input1 = task1, input2 = task2) tasks_out = do.call(po_torch$predict, args = list(input = predict_input)) identical(tasks_out[[1L]], tasks_out[[2L]]) #> [1] TRUE"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":null,"dir":"Reference","previous_headings":"","what":"Callback Configuration — mlr_pipeops_torch_callbacks","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"Configures callbacks deep learning model.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"one input channel \"input\" one output channel \"output\". training, channels class ModelDescriptor. prediction, channels class Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"parameters defined dynamically callbacks, id respective callbacks respective set id.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"training callbacks cloned added ModelDescriptor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"mlr3pipelines::PipeOp -> PipeOpTorchCallbacks","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"PipeOpTorchCallbacks$new() PipeOpTorchCallbacks$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"","code":"PipeOpTorchCallbacks$new(   callbacks = list(),   id = \"torch_callbacks\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"callbacks (list TorchCallbacks)  callbacks (something convertible via as_torch_callbacks()). Must unique ids. callbacks cloned construction. id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"","code":"PipeOpTorchCallbacks$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"","code":"po_cb = po(\"torch_callbacks\", \"checkpoint\") po_cb$param_set #> <ParamSetCollection(3)> #>                      id    class lower upper nlevels        default  value #>                  <char>   <char> <num> <num>   <num>         <list> <list> #> 1:      checkpoint.path ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:      checkpoint.freq ParamInt     1   Inf     Inf <NoDefault[0]> [NULL] #> 3: checkpoint.freq_type ParamFct    NA    NA       2          epoch [NULL] mdin = po(\"torch_ingress_num\")$train(list(tsk(\"iris\"))) mdin[[1L]]$callbacks #> named list() mdout = po_cb$train(mdin)[[1L]] mdout$callbacks #> $checkpoint #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch #>  # Can be called again po_cb1 = po(\"torch_callbacks\", t_clbk(\"progress\")) mdout1 = po_cb1$train(list(mdout))[[1L]] mdout1$callbacks #> $progress #> <TorchCallback:progress> Progress #> * Generator: CallbackSetProgress #> * Parameters: list() #> * Packages: progress,mlr3torch,torch #>  #> $checkpoint #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":null,"dir":"Reference","previous_headings":"","what":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Use entry-point mlr3torch-networks. Unless advanced user, need use directly PipeOpTorchIngressNumeric, PipeOpTorchIngressCategorical PipeOpTorchIngressLazyTensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"state set input shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Defined construction argument param_set.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Creates object class TorchIngressToken given task. purpuse store information construct torch dataloader task entry point network.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"mlr3pipelines::PipeOp -> PipeOpTorchIngress","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"feature_types (character(1)) features types can consumed PipeOpTorchIngress.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"PipeOpTorchIngress$new() PipeOpTorchIngress$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"","code":"PipeOpTorchIngress$new(   id,   param_set = ps(),   param_vals = list(),   packages = character(0),   feature_types )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"id (character(1)) Identifier resulting  object. param_set (ParamSet) parameter set. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction. packages (character()) R packages object depends . feature_types (character()) feature types. See mlr_reflections$task_feature_types available values, Additionally, \"lazy_tensor\" supported.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"","code":"PipeOpTorchIngress$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"Ingress PipeOp represents categorical (factor(), ordered() logical()) entry point torch network.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"select :: logical(1) Whether PipeOp selected supported feature types. Otherwise err receiving tasks unsupported feature types.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"Uses batchgetter_categ().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"state set input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorchIngress -> PipeOpTorchIngressCategorical","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"PipeOpTorchIngressCategorical$new() PipeOpTorchIngressCategorical$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"","code":"PipeOpTorchIngressCategorical$new(   id = \"torch_ingress_categ\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"","code":"PipeOpTorchIngressCategorical$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"","code":"graph = po(\"select\", selector = selector_type(\"factor\")) %>>%   po(\"torch_ingress_categ\") task = tsk(\"german_credit\") # The output is a model descriptor md = graph$train(task)[[1L]] ingress = md$ingress[[1L]] ingress$batchgetter(task$data(1, ingress$features), \"cpu\") #> torch_tensor #>  5  5  1  2  3  1  3  1  3  1  4  5  1  2 #> [ CPULongType{1,14} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":null,"dir":"Reference","previous_headings":"","what":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"Ingress single lazy_tensor column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"shape :: integer() shape tensor, first dimension (batch) must NA. specified, lazy tensor input column needs known shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"returned batchgetter materializes lazy tensor column tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"state set input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorchIngress -> PipeOpTorchIngressLazyTensor","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"PipeOpTorchIngressLazyTensor$new() PipeOpTorchIngressLazyTensor$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"","code":"PipeOpTorchIngressLazyTensor$new(   id = \"torch_ingress_ltnsr\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"","code":"PipeOpTorchIngressLazyTensor$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"","code":"po_ingress = po(\"torch_ingress_ltnsr\") task = tsk(\"lazy_iris\")  md = po_ingress$train(list(task))[[1L]]  ingress = md$ingress x_batch = ingress[[1L]]$batchgetter(data = task$data(1, \"x\"), device = \"cpu\", cache = NULL) x_batch #> torch_tensor #>  5.1000  3.5000  1.4000  0.2000 #> [ CPUFloatType{1,4} ]  # Now we try a lazy tensor with unknown shape, i.e. the shapes between the rows can differ  ds = dataset(   initialize = function() self$x = list(torch_randn(3, 10, 10), torch_randn(3, 8, 8)),   .getitem = function(i) list(x = self$x[[i]]),   .length = function() 2)()  task_unknown = as_task_regr(data.table(   x = as_lazy_tensor(ds, dataset_shapes = list(x = NULL)),   y = rnorm(2) ), target = \"y\", id = \"example2\")  # this task (as it is) can NOT be processed by PipeOpTorchIngressLazyTensor # It therefore needs to be preprocessed po_resize = po(\"trafo_resize\", size = c(6, 6)) task_unknown_resize = po_resize$train(list(task_unknown))[[1L]]  # printing the transformed column still shows unknown shapes, # because the preprocessing pipeop cannot infer them, # however we know that the shape is now (3, 10, 10) for all rows task_unknown_resize$data(1:2, \"x\") #>                x #>    <lazy_tensor> #> 1:      <tnsr[]> #> 2:      <tnsr[]> po_ingress$param_set$set_values(shape = c(NA, 3, 6, 6))  md2 = po_ingress$train(list(task_unknown_resize))[[1L]]  ingress2 = md2$ingress x_batch2 = ingress2[[1L]]$batchgetter(   data = task_unknown_resize$data(1:2, \"x\"),   device = \"cpu\",   cache = NULL )  x_batch2 #> torch_tensor #> (1,1,.,.) =  #>   0.1699 -0.0348 -1.3127  0.0449 -1.3391 -0.2706 #>  -0.4008 -0.2498  0.2398  0.7099 -0.2593 -0.6988 #>  -0.7197  0.2285  0.2184  0.5368 -0.0414  0.5573 #>  -0.4601 -0.7059 -1.3604  0.3196 -0.3387 -0.7384 #>   0.2113  1.3375  0.5421  0.5011  0.3896 -0.3945 #>   0.3065 -0.1602 -0.6599 -0.0274 -0.3330  0.1745 #>  #> (2,1,.,.) =  #>   0.4998  0.6063  1.7977 -0.0074  0.8868  1.1315 #>  -1.2362  0.4990 -0.0351 -0.4337 -0.5377  0.1622 #>  -0.3271 -0.2007  0.5288  0.5646  0.2186 -0.6196 #>  -0.8539 -0.6276 -0.7953 -0.2205 -0.3521 -0.1657 #>   0.1857 -0.1610  0.2109 -0.0741 -0.0214  0.5110 #>  -0.4607  0.5773  0.6792  0.5576  0.0791  0.5881 #>  #> (1,2,.,.) =  #>   0.0530  1.2853 -1.4618 -0.5761 -0.6764  0.5012 #>   0.0895  0.7634  0.7542  1.1485  1.9902  0.2917 #>   0.0751  1.5195 -0.0507  0.0306  0.3952 -0.3084 #>   0.5689  0.5605 -0.1022  0.2474  0.3521  0.8797 #>   0.4937 -0.4953  1.4223 -0.3710 -1.0329 -0.1044 #>   1.1720  0.5355 -0.6309  0.3580 -0.7996  0.0643 #>  #> (2,2,.,.) =  #>  -0.3858  2.2418 -0.5567 -0.3174 -0.0823  0.4582 #>   0.5028  0.0923 -0.2350 -0.1985  0.2016 -0.7900 #>  -0.5887 -0.6755  0.2675 -0.1618 -0.0426  1.5266 #>   0.1032  0.4991 -0.1085 -0.1012  0.0068 -0.4961 #>   0.4799 -0.3508 -0.5198  0.0623  0.5824  0.3236 #> ... [the output was truncated (use n=-1 to disable)] #> [ CPUFloatType{2,3,6,6} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"Ingress PipeOp represents numeric (integer() numeric()) entry point torch network.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"Uses batchgetter_num().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"state set input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorchIngress -> PipeOpTorchIngressNumeric","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"PipeOpTorchIngressNumeric$new() PipeOpTorchIngressNumeric$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"","code":"PipeOpTorchIngressNumeric$new(id = \"torch_ingress_num\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"","code":"PipeOpTorchIngressNumeric$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"","code":"graph = po(\"select\", selector = selector_type(c(\"numeric\", \"integer\"))) %>>%   po(\"torch_ingress_num\") task = tsk(\"german_credit\") # The output is a model descriptor md = graph$train(task)[[1L]] ingress = md$ingress[[1L]] ingress$batchgetter(task$data(1:5, ingress$features), \"cpu\") #> torch_tensor #>    67  1169     6 #>    22  5951    48 #>    49  2096    12 #>    45  7882    42 #>    53  4870    24 #> [ CPUFloatType{5,3} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":null,"dir":"Reference","previous_headings":"","what":"Loss Configuration — mlr_pipeops_torch_loss","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"Configures loss deep learning model.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"parameters defined dynamically loss set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"training loss cloned added ModelDescriptor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"mlr3pipelines::PipeOp -> PipeOpTorchLoss","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"PipeOpTorchLoss$new() PipeOpTorchLoss$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"","code":"PipeOpTorchLoss$new(loss, id = \"torch_loss\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"loss (TorchLoss character(1) nn_loss) loss (something convertible via as_torch_loss()). id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"","code":"PipeOpTorchLoss$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"","code":"po_loss = po(\"torch_loss\", loss = t_loss(\"cross_entropy\")) po_loss$param_set #> <ParamSet(3)> #>              id    class lower upper nlevels default  value #>          <char>   <char> <num> <num>   <num>  <list> <list> #> 1:       weight ParamUty    NA    NA     Inf  [NULL] [NULL] #> 2: ignore_index ParamInt  -Inf   Inf     Inf    -100 [NULL] #> 3:    reduction ParamFct    NA    NA       2    mean [NULL] mdin = po(\"torch_ingress_num\")$train(list(tsk(\"iris\"))) mdin[[1L]]$loss #> NULL mdout = po_loss$train(mdin)[[1L]] mdout$loss #> <TorchLoss:cross_entropy> Cross Entropy #> * Generator: nn_cross_entropy_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: classif"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOp Torch Model — mlr_pipeops_torch_model","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"Builds Torch Learner ModelDescriptor trains given parameter specification. task type must specified construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"one input channel \"input\" takes ModelDescriptor traing Task specified task_type prediction. output NULL training Prediction given task_type prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"trained LearnerTorchModel.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"General: parameters optimizer, loss callbacks, prefixed \"opt.\", \"loss.\" \"cb.<callback id>.\" respectively, well : epochs :: integer(1) number epochs. device :: character(1) device. One \"auto\", \"cpu\", \"cuda\" values defined mlr_reflections$torch$devices. value initialized \"auto\", select \"cuda\" possible, try \"mps\" otherwise fall back \"cpu\". num_threads :: integer(1) number threads intraop pararallelization (device \"cpu\"). value initialized 1. num_interop_threads :: integer(1) number threads intraop interop pararallelization (device \"cpu\"). value initialized 1. Note can set session changing value within R session raise warning. seed :: integer(1) \"random\" NULL torch seed used training prediction. value initialized \"random\", means random seed sampled beginning training phase. seed (either set randomly sampled) available via $model$seed training used prediction. Note setting seed training phase mean default (.e. seed \"random\"), clones learner use different seed. set NULL, seeding done. Evaluation: measures_train :: Measure list() Measures. Measures evaluated training. measures_valid :: Measure list() Measures. Measures evaluated validation. eval_freq :: integer(1) often train / validation predictions evaluated using measures_train / measures_valid. initialized 1. Note final model always evaluated. Early Stopping: patience :: integer(1) activates early stopping using validation scores. performance model improve patience evaluation steps, training ended. Note final model stored learner, best model. initialized 0, means early stopping. first entry measures_valid used metric. also requires specify $validate field Learner, well measures_valid. min_delta :: double(1) minimum improvement threshold (>) early stopping. initialized 0. Dataloader: batch_size :: integer(1) batch size (required). shuffle :: logical(1) Whether shuffle instances dataset. Default FALSE. impact validation. sampler :: torch::sampler Object defines dataloader draw samples. batch_sampler :: torch::sampler Object defines dataloader draws batches. num_workers :: integer(1) number workers data loading (batches loaded parallel). default 0, means data loaded main process. collate_fn :: function merge list samples form batch. pin_memory :: logical(1) Whether dataloader copies tensors CUDA pinned memory returning . drop_last :: logical(1) Whether drop last training batch epoch training. Default FALSE. timeout :: numeric(1) timeout value collecting batch workers. Negative values mean timeout default -1. worker_init_fn :: function(id) function receives worker id ([1, num_workers]) exectued seeding worker data loading. worker_globals :: list() | character() loading data parallel, allows export globals workers. character vector, objects global environment names copied workers. worker_packages :: character() packages load workers. Also see torch::dataloder information.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"LearnerTorchModel created calling model_descriptor_to_learner() provided ModelDescriptor received input channel. parameters set according parameters specified PipeOpTorchModel '$train() method called [Task][mlr3::Task] stored [ModelDescriptor`].","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpLearner -> PipeOpTorchModel","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"PipeOpTorchModel$new() PipeOpTorchModel$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"","code":"PipeOpTorchModel$new(task_type, id = \"torch_model\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"task_type (character(1)) task type model. id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"","code":"PipeOpTorchModel$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"Builds torch classifier trains .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"See LearnerTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"one input channel \"input\" takes ModelDescriptor traing Task specified task_type prediction. output NULL training Prediction given task_type prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"trained LearnerTorchModel.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"LearnerTorchModel created calling model_descriptor_to_learner() provided ModelDescriptor received input channel. parameters set according parameters specified PipeOpTorchModel '$train() method called [Task][mlr3::Task] stored [ModelDescriptor`].","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpLearner -> mlr3torch::PipeOpTorchModel -> PipeOpTorchModelClassif","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"PipeOpTorchModelClassif$new() PipeOpTorchModelClassif$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"","code":"PipeOpTorchModelClassif$new(id = \"torch_model_classif\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"","code":"PipeOpTorchModelClassif$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"","code":"# simple logistic regression  # configure the model descriptor md = as_graph(po(\"torch_ingress_num\") %>>%   po(\"nn_head\") %>>%   po(\"torch_loss\", \"cross_entropy\") %>>%   po(\"torch_optimizer\", \"adam\"))$train(tsk(\"iris\"))[[1L]]  print(md) #> <ModelDescriptor: 2 ops> #> * Ingress:  torch_ingress_num.input: [(NA,4)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  Adaptive Moment Estimation #> * Loss:  Cross Entropy #> * pointer:  nn_head.output [(NA,3)]  # build the learner from the model descriptor and train it po_model = po(\"torch_model_classif\", batch_size = 50, epochs = 1) po_model$train(list(md)) #> $output #> NULL #>  po_model$state #> $model #> $network #> An `nn_module` containing 15 parameters. #>  #> ── Modules ───────────────────────────────────────────────────────────────────── #> • module_list: <nn_module_list> #15 parameters #>  #> $internal_valid_scores #> NULL #>  #> $loss_fn #> list() #>  #> $optimizer #> $optimizer$param_groups #> $optimizer$param_groups[[1]] #> $optimizer$param_groups[[1]]$params #> [1] 1 2 #>  #> $optimizer$param_groups[[1]]$lr #> [1] 0.001 #>  #> $optimizer$param_groups[[1]]$betas #> [1] 0.900 0.999 #>  #> $optimizer$param_groups[[1]]$eps #> [1] 1e-08 #>  #> $optimizer$param_groups[[1]]$weight_decay #> [1] 0 #>  #> $optimizer$param_groups[[1]]$amsgrad #> [1] FALSE #>  #>  #>  #> $optimizer$state #> $optimizer$state$`1` #> $optimizer$state$`1`$step #> [1] 3 #>  #> $optimizer$state$`1`$exp_avg #> torch_tensor #> -0.0651 -0.0071 -0.2694 -0.1905 #> -0.3519 -0.1111 -0.4643 -0.2070 #>  0.4170  0.1182  0.7338  0.3975 #> [ CPUFloatType{3,4} ] #>  #> $optimizer$state$`1`$exp_avg_sq #> torch_tensor #> 0.01 * #>  0.1201  0.0036  1.3689  0.6378 #>   1.7104  0.1656  3.3447  0.7363 #>   1.6698  0.1561  4.0499  1.1255 #> [ CPUFloatType{3,4} ] #>  #>  #> $optimizer$state$`2` #> $optimizer$state$`2`$step #> [1] 3 #>  #> $optimizer$state$`2`$exp_avg #> torch_tensor #> -0.0548 #> -0.0768 #>  0.1315 #> [ CPUFloatType{3} ] #>  #> $optimizer$state$`2`$exp_avg_sq #> torch_tensor #> 0.001 * #>  0.5453 #>  0.9515 #>  1.2573 #> [ CPUFloatType{3} ] #>  #>  #>  #>  #> $epochs #> [1] 1 #>  #> $callbacks #> named list() #>  #> $seed #> [1] 1285070263 #>  #> $task_col_info #> Key: <id> #>              id    type                      levels #>          <char>  <char>                      <list> #> 1: Petal.Length numeric                      [NULL] #> 2:  Petal.Width numeric                      [NULL] #> 3: Sepal.Length numeric                      [NULL] #> 4:  Sepal.Width numeric                      [NULL] #> 5:      Species  factor setosa,versicolor,virginica #>  #> attr(,\"class\") #> [1] \"learner_torch_model\" \"list\"                #>  #> $log #> Empty data.table (0 rows and 3 cols): stage,class,msg #>  #> $train_time #> [1] 0.055 #>  #> $param_vals #> $param_vals$epochs #> [1] 1 #>  #> $param_vals$device #> [1] \"auto\" #>  #> $param_vals$num_threads #> [1] 1 #>  #> $param_vals$num_interop_threads #> [1] 1 #>  #> $param_vals$seed #> [1] \"random\" #>  #> $param_vals$eval_freq #> [1] 1 #>  #> $param_vals$measures_train #> list() #>  #> $param_vals$measures_valid #> list() #>  #> $param_vals$patience #> [1] 0 #>  #> $param_vals$min_delta #> [1] 0 #>  #> $param_vals$batch_size #> [1] 50 #>  #>  #> $task_hash #> [1] \"34a6beb27c3181c3\" #>  #> $feature_names #> [1] \"Petal.Length\" \"Petal.Width\"  \"Sepal.Length\" \"Sepal.Width\"  #>  #> $validate #> NULL #>  #> $mlr3_version #> [1] ‘0.21.1’ #>  #> $internal_tuned_values #> named list() #>  #> $data_prototype #> Empty data.table (0 rows and 5 cols): Species,Petal.Length,Petal.Width,Sepal.Length,Sepal.Width #>  #> $task_prototype #> Empty data.table (0 rows and 5 cols): Species,Petal.Length,Petal.Width,Sepal.Length,Sepal.Width #>  #> $train_task #> <TaskClassif:iris> (150 x 5): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (4): #>   - dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Width #>  #> attr(,\"class\") #> [1] \"learner_state\" \"list\""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Regression Model — mlr_pipeops_torch_model_regr","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"Builds torch regression model trains .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"See LearnerTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"one input channel \"input\" takes ModelDescriptor traing Task specified task_type prediction. output NULL training Prediction given task_type prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"trained LearnerTorchModel.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"LearnerTorchModel created calling model_descriptor_to_learner() provided ModelDescriptor received input channel. parameters set according parameters specified PipeOpTorchModel '$train() method called [Task][mlr3::Task] stored [ModelDescriptor`].","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpLearner -> mlr3torch::PipeOpTorchModel -> PipeOpTorchModelRegr","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"PipeOpTorchModelRegr$new() PipeOpTorchModelRegr$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"","code":"PipeOpTorchModelRegr$new(id = \"torch_model_regr\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"","code":"PipeOpTorchModelRegr$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"","code":"# simple linear regression  # build the model descriptor md = as_graph(po(\"torch_ingress_num\") %>>%   po(\"nn_head\") %>>%   po(\"torch_loss\", \"mse\") %>>%   po(\"torch_optimizer\", \"adam\"))$train(tsk(\"mtcars\"))[[1L]]  print(md) #> <ModelDescriptor: 2 ops> #> * Ingress:  torch_ingress_num.input: [(NA,10)] #> * Task:  mtcars [regr] #> * Callbacks:  N/A #> * Optimizer:  Adaptive Moment Estimation #> * Loss:  Mean Squared Error #> * pointer:  nn_head.output [(NA,1)]  # build the learner from the model descriptor and train it po_model = po(\"torch_model_regr\", batch_size = 20, epochs = 1) po_model$train(list(md)) #> $output #> NULL #>  po_model$state #> $model #> $network #> An `nn_module` containing 11 parameters. #>  #> ── Modules ───────────────────────────────────────────────────────────────────── #> • module_list: <nn_module_list> #11 parameters #>  #> $internal_valid_scores #> NULL #>  #> $loss_fn #> list() #>  #> $optimizer #> $optimizer$param_groups #> $optimizer$param_groups[[1]] #> $optimizer$param_groups[[1]]$params #> [1] 1 2 #>  #> $optimizer$param_groups[[1]]$lr #> [1] 0.001 #>  #> $optimizer$param_groups[[1]]$betas #> [1] 0.900 0.999 #>  #> $optimizer$param_groups[[1]]$eps #> [1] 1e-08 #>  #> $optimizer$param_groups[[1]]$weight_decay #> [1] 0 #>  #> $optimizer$param_groups[[1]]$amsgrad #> [1] FALSE #>  #>  #>  #> $optimizer$state #> $optimizer$state$`1` #> $optimizer$state$`1`$step #> [1] 2 #>  #> $optimizer$state$`1`$exp_avg #> torch_tensor #> -1.1899 -2.5584 -16.8081 -583.6016 -12.5174 -11.6554 -219.9824 -66.3019 -2.5536 -9.4319 #> [ CPUFloatType{1,10} ] #>  #> $optimizer$state$`1`$exp_avg_sq #> torch_tensor #> Columns 1 to 6 1.3291e-01  4.0389e+00  2.2521e+01  2.8091e+04  1.1109e+01  1.0300e+01 #>  #> Columns 7 to 10 8.3546e+03  3.0016e+02  4.2510e-01  7.0911e+00 #> [ CPUFloatType{1,10} ] #>  #>  #> $optimizer$state$`2` #> $optimizer$state$`2`$step #> [1] 2 #>  #> $optimizer$state$`2`$exp_avg #> torch_tensor #> -3.4187 #> [ CPUFloatType{1} ] #>  #> $optimizer$state$`2`$exp_avg_sq #> torch_tensor #>  0.8167 #> [ CPUFloatType{1} ] #>  #>  #>  #>  #> $epochs #> [1] 1 #>  #> $callbacks #> named list() #>  #> $seed #> [1] 1984753776 #>  #> $task_col_info #>         id    type levels #>     <char>  <char> <list> #>  1:     am numeric [NULL] #>  2:   carb numeric [NULL] #>  3:    cyl numeric [NULL] #>  4:   disp numeric [NULL] #>  5:   drat numeric [NULL] #>  6:   gear numeric [NULL] #>  7:     hp numeric [NULL] #>  8:   qsec numeric [NULL] #>  9:     vs numeric [NULL] #> 10:     wt numeric [NULL] #> 11:    mpg numeric [NULL] #>  #> attr(,\"class\") #> [1] \"learner_torch_model\" \"list\"                #>  #> $log #> Empty data.table (0 rows and 3 cols): stage,class,msg #>  #> $train_time #> [1] 0.04 #>  #> $param_vals #> $param_vals$epochs #> [1] 1 #>  #> $param_vals$device #> [1] \"auto\" #>  #> $param_vals$num_threads #> [1] 1 #>  #> $param_vals$num_interop_threads #> [1] 1 #>  #> $param_vals$seed #> [1] \"random\" #>  #> $param_vals$eval_freq #> [1] 1 #>  #> $param_vals$measures_train #> list() #>  #> $param_vals$measures_valid #> list() #>  #> $param_vals$patience #> [1] 0 #>  #> $param_vals$min_delta #> [1] 0 #>  #> $param_vals$batch_size #> [1] 20 #>  #>  #> $task_hash #> [1] \"6fbb93eed42adb31\" #>  #> $feature_names #>  [1] \"am\"   \"carb\" \"cyl\"  \"disp\" \"drat\" \"gear\" \"hp\"   \"qsec\" \"vs\"   \"wt\"   #>  #> $validate #> NULL #>  #> $mlr3_version #> [1] ‘0.21.1’ #>  #> $internal_tuned_values #> named list() #>  #> $data_prototype #> Empty data.table (0 rows and 11 cols): mpg,am,carb,cyl,disp,drat... #>  #> $task_prototype #> Empty data.table (0 rows and 11 cols): mpg,am,carb,cyl,disp,drat... #>  #> $train_task #> <TaskRegr:mtcars> (32 x 11): Motor Trends #> * Target: mpg #> * Properties: - #> * Features (10): #>   - dbl (10): am, carb, cyl, disp, drat, gear, hp, qsec, vs, wt #>  #> attr(,\"class\") #> [1] \"learner_state\" \"list\""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":null,"dir":"Reference","previous_headings":"","what":"Optimizer Configuration — mlr_pipeops_torch_optimizer","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"Configures optimizer deep learning model.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"one input channel \"input\" one output channel \"output\". training, channels class ModelDescriptor. prediction, channels class Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"parameters defined dynamically optimizer set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"training, optimizer cloned added ModelDescriptor. Note parameter set stored TorchOptimizer reference-identical parameter set pipeop .","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"mlr3pipelines::PipeOp -> PipeOpTorchOptimizer","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"PipeOpTorchOptimizer$new() PipeOpTorchOptimizer$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"","code":"PipeOpTorchOptimizer$new(   optimizer = t_opt(\"adam\"),   id = \"torch_optimizer\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"optimizer (TorchOptimizer character(1) torch_optimizer_generator) optimizer (something convertible via as_torch_optimizer()). id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"","code":"PipeOpTorchOptimizer$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"","code":"po_opt = po(\"torch_optimizer\", \"sgd\", lr = 0.01) po_opt$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamDbl     0   Inf     Inf <NoDefault[0]>   0.01 #> 2:     momentum ParamDbl     0     1     Inf              0 [NULL] #> 3:    dampening ParamDbl     0     1     Inf              0 [NULL] #> 4: weight_decay ParamDbl     0     1     Inf              0 [NULL] #> 5:     nesterov ParamLgl    NA    NA       2          FALSE [NULL] mdin = po(\"torch_ingress_num\")$train(list(tsk(\"iris\"))) mdin[[1L]]$optimizer #> NULL mdout = po_opt$train(mdin) mdout[[1L]]$optimizer #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: lr=0.01 #> * Packages: torch,mlr3torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":null,"dir":"Reference","previous_headings":"","what":"Iris Classification Task — mlr_tasks_lazy_iris","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"classification task popular datasets::iris data set. Just like iris task, features represented one lazy tensor column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"R6::R6Class inheriting mlr3::TaskClassif.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"https://en.wikipedia.org/wiki/Iris_flower_data_set","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"construction","dir":"Reference","previous_headings":"","what":"Construction","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"","code":"tsk(\"lazy_iris\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"Task type: “classif” Properties: “multiclass” Missings: Target: “Species” Features: “x” Data Dimension: 150x3","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"Anderson E (1936). “Species Problem Iris.” Annals Missouri Botanical Garden, 23(3), 457. doi:10.2307/2394164 .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"","code":"task = tsk(\"lazy_iris\") task #> <TaskClassif:lazy_iris> (150 x 2): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (1): #>   - lt (1): x df = task$data() materialize(df$x[1:6], rbind = TRUE) #> torch_tensor #>  5.1000  3.5000  1.4000  0.2000 #>  4.9000  3.0000  1.4000  0.2000 #>  4.7000  3.2000  1.3000  0.2000 #>  4.6000  3.1000  1.5000  0.2000 #>  5.0000  3.6000  1.4000  0.2000 #>  5.4000  3.9000  1.7000  0.4000 #> [ CPUFloatType{6,4} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":null,"dir":"Reference","previous_headings":"","what":"MNIST Image classification — mlr_tasks_mnist","title":"MNIST Image classification — mlr_tasks_mnist","text":"Classic MNIST image classification. underlying DataBackend contains columns \"label\", \"image\", \"row_id\", \"split\", last column indicates whether row belongs train test set. first 60000 rows belong training set, last 10000 rows test set.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"MNIST Image classification — mlr_tasks_mnist","text":"https://torchvision.mlverse.org/reference/mnist_dataset.html","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"construction","dir":"Reference","previous_headings":"","what":"Construction","title":"MNIST Image classification — mlr_tasks_mnist","text":"","code":"tsk(\"mnist\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"download","dir":"Reference","previous_headings":"","what":"Download","title":"MNIST Image classification — mlr_tasks_mnist","text":"task's backend DataBackendLazy download data requested. meta-data already available . can cache datasets setting mlr3torch.cache option TRUE specific path used cache directory.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"MNIST Image classification — mlr_tasks_mnist","text":"Task type: “classif” Properties: “multiclass” Missings: Target: “label” Features: “image” Data Dimension: 70000x4","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"MNIST Image classification — mlr_tasks_mnist","text":"Lecun, Y., Bottou, L., Bengio, Y., Haffner, P. (1998). “Gradient-based learning applied document recognition.” Proceedings IEEE, 86(11), 2278-2324. doi:10.1109/5.726791 .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"MNIST Image classification — mlr_tasks_mnist","text":"","code":"task = tsk(\"mnist\") task #> <TaskClassif:mnist> (70000 x 2): MNIST Digit Classification #> * Target: label #> * Properties: multiclass #> * Features (1): #>   - lt (1): image"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":null,"dir":"Reference","previous_headings":"","what":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"Subset famous ImageNet dataset. data obtained torchvision::tiny_imagenet_dataset(). underlying DataBackend contains columns \"class\", \"image\", \"..row_id\", \"split\", last column indicates whether row belongs train, validation test set defined provided torchvision. labels test rows, default, observations inactive, means task uses 110000 120000 observations defined underlying data backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"construction","dir":"Reference","previous_headings":"","what":"Construction","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"","code":"tsk(\"tiny_imagenet\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"download","dir":"Reference","previous_headings":"","what":"Download","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"task's backend DataBackendLazy download data requested. meta-data already available . can cache datasets setting mlr3torch.cache option TRUE specific path used cache directory.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"Task type: “classif” Properties: “multiclass” Missings: Target: “class” Features: “image” Data Dimension: 120000x4","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"Deng, Jia, Dong, Wei, Socher, Richard, Li, Li-Jia, Li, Kai, Fei-Fei, Li (2009). “Imagenet: large-scale hierarchical image database.” 2009 IEEE conference computer vision pattern recognition, 248–255. IEEE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"","code":"task = tsk(\"tiny_imagenet\") task #> <TaskClassif:tiny_imagenet> (110000 x 2): ImageNet Subset #> * Target: class #> * Properties: multiclass #> * Features (1): #>   - lt (1): image"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"First nn_graph created using model_descriptor_to_module learner created module remaining information model descriptor, must include optimizer loss function optionally callbacks.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"","code":"model_descriptor_to_learner(model_descriptor)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"model_descriptor (ModelDescriptor) model descriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"Learner","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"Creates nn_graph ModelDescriptor. Mostly internal use, since ModelDescriptor circumstances harder use just creating nn_graph directly.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"","code":"model_descriptor_to_module(   model_descriptor,   output_pointers = NULL,   list_output = FALSE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"model_descriptor (ModelDescriptor) Model Descriptor. pointer ignored, instead output_pointer values used. $graph member modified -reference. output_pointers (list character) Collection pointers indicate part model_descriptor$graph used output. Entries format ModelDescriptor$pointer. list_output (logical(1)) Whether output list tensors. FALSE, length(output_pointers) must 1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"nn_graph","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":null,"dir":"Reference","previous_headings":"","what":"Union of ModelDescriptors — model_descriptor_union","title":"Union of ModelDescriptors — model_descriptor_union","text":"mostly internal function used PipeOpTorchs multiple input channels. creates union multiple ModelDescriptors: graphs combinded (identical begin ). first entry's graph modified reference. PipeOps ID must identical. new input edges may added PipeOps. Drops pointer / pointer_shape entries. new task feature union two incoming tasks. optimizer loss ModelDescriptors must identical. Ingress tokens callbacks merged, objects \"id\" must identical.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Union of ModelDescriptors — model_descriptor_union","text":"","code":"model_descriptor_union(md1, md2)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Union of ModelDescriptors — model_descriptor_union","text":"md1 (ModelDescriptor) first ModelDescriptor. md2 (ModelDescriptor) second ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Union of ModelDescriptors — model_descriptor_union","text":"ModelDescriptor","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Union of ModelDescriptors — model_descriptor_union","text":"requirement new input edgedes may added PipeOps  theoretically necessary, since assume ModelDescriptor built beginning end (.e. PipeOps never get new ancestors) can make assumption simplify things. Otherwise need treat \"...\"-inputs special.)","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Neural Network Layer — nn","title":"Create a Neural Network Layer — nn","text":"Retrieve neural network layer mlr_pipeops dictionary.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Neural Network Layer — nn","text":"","code":"nn(.key, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Neural Network Layer — nn","text":".key (character(1)) ... () Additional parameters, constructor arguments fields.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a Neural Network Layer — nn","text":"","code":"po1 = po(\"nn_linear\", id = \"linear\") # is the same as: po2 = nn(\"linear\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":null,"dir":"Reference","previous_headings":"","what":"Graph Network — nn_graph","title":"Graph Network — nn_graph","text":"Represents neural network using Graph usually costains mostly PipeOpModules.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Graph Network — nn_graph","text":"","code":"nn_graph(graph, shapes_in, output_map = graph$output$name, list_output = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Graph Network — nn_graph","text":"graph (Graph) Graph wrap. cloned. shapes_in (named integer) Shape info tensors go graph. Names must graph$input$name, possibly different order. output_map (character) graph's outputs use. Must subset graph$output$name. list_output (logical(1)) Whether output list tensors. FALSE (default), length(output_map) must 1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Graph Network — nn_graph","text":"nn_graph","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Graph Network — nn_graph","text":"","code":"graph = mlr3pipelines::Graph$new() graph$add_pipeop(po(\"module_1\", module = nn_linear(10, 20)), clone = FALSE) graph$add_pipeop(po(\"module_2\", module = nn_relu()), clone = FALSE) graph$add_pipeop(po(\"module_3\", module = nn_linear(20, 1)), clone = FALSE) graph$add_edge(\"module_1\", \"module_2\") graph$add_edge(\"module_2\", \"module_3\")  network = nn_graph(graph, shapes_in = list(module_1.input = c(NA, 10)))  x = torch_randn(16, 10)  network(module_1.input = x) #> torch_tensor #>  0.4208 #> -0.3031 #> -0.4180 #> -0.3396 #>  0.0821 #>  0.2099 #>  0.3070 #>  0.3691 #>  0.1470 #>  0.4891 #>  0.5106 #> -0.3846 #> -0.2878 #> -0.4239 #> -0.0854 #>  0.0169 #> [ CPUFloatType{16,1} ][ grad_fn = <AddmmBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_cat.html","id":null,"dir":"Reference","previous_headings":"","what":"Concatenates multiple tensors — nn_merge_cat","title":"Concatenates multiple tensors — nn_merge_cat","text":"Concatenates multiple tensors given dimension. broadcasting rules applied , must reshape tensors shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_cat.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Concatenates multiple tensors — nn_merge_cat","text":"","code":"nn_merge_cat(dim = -1)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_cat.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Concatenates multiple tensors — nn_merge_cat","text":"dim (integer(1)) dimension concatenation.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_prod.html","id":null,"dir":"Reference","previous_headings":"","what":"Product of multiple tensors — nn_merge_prod","title":"Product of multiple tensors — nn_merge_prod","text":"Calculates product input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_prod.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Product of multiple tensors — nn_merge_prod","text":"","code":"nn_merge_prod()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_sum.html","id":null,"dir":"Reference","previous_headings":"","what":"Sum of multiple tensors — nn_merge_sum","title":"Sum of multiple tensors — nn_merge_sum","text":"Calculates sum input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_sum.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Sum of multiple tensors — nn_merge_sum","text":"","code":"nn_merge_sum()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_reshape.html","id":null,"dir":"Reference","previous_headings":"","what":"Reshape — nn_reshape","title":"Reshape — nn_reshape","text":"Reshape tensor given shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_reshape.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Reshape — nn_reshape","text":"","code":"nn_reshape(shape)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_reshape.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Reshape — nn_reshape","text":"shape (integer()) desired output shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_squeeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Squeeze — nn_squeeze","title":"Squeeze — nn_squeeze","text":"Squeezes tensor calling torch::torch_squeeze() given dimension dim.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_squeeze.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Squeeze — nn_squeeze","text":"","code":"nn_squeeze(dim)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_squeeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Squeeze — nn_squeeze","text":"dim (integer()) dimension squeeze.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_unsqueeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Unsqueeze — nn_unsqueeze","title":"Unsqueeze — nn_unsqueeze","text":"Unsqueezes tensor calling torch::torch_unsqueeze() given dimension dim.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_unsqueeze.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Unsqueeze — nn_unsqueeze","text":"","code":"nn_unsqueeze(dim)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_unsqueeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Unsqueeze — nn_unsqueeze","text":"dim (integer(1)) dimension unsqueeze.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"Function create objects class PipeOpTaskPreprocTorch convenient way. Start reading documentation PipeOpTaskPreprocTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"","code":"pipeop_preproc_torch(   id,   fn,   shapes_out = NULL,   param_set = NULL,   packages = character(0),   rowwise = FALSE,   parent_env = parent.frame(),   stages_init = NULL,   tags = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"id (character(1)) id new object. fn (function) preprocessing function. shapes_out (function NULL \"infer\") private .shapes_out(shapes_in, param_vals, task) method PipeOpTaskPreprocTorch (see section Inheriting). Special values NULL infer: NULL, output shapes unknown. \"infer\", output shape function inferred calculates output shapes follows: input shape (NA, ...) meta-tensor shape (1, ...) created preprocessing function applied. Afterwards batch dimension (1) replaced NA shape returned. first dimension NA, output shape applying preprocessing function returned. Method \"infer\" correct cases, might fail edge cases. param_set (ParamSet NULL) parameter set. left NULL (default) parameter set inferred following way: parameters first ... fn set untyped parameters tags 'train' default value tagged 'required' well. Default values annotated. packages (character()) R packages object depends . rowwise (logical(1)) Whether preprocessing applied row-wise. parent_env (environment) parent environment R6 class. stages_init (character(1)) Initial value stages parameter. NULL (default), set \"\" case id starts \"trafo\" \"train\" starts \"augment\". Otherwise must specified. tags (character()) Tags pipeop","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"R6Class instance inheriting PipeOpTaskPreprocTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"","code":"PipeOpPreprocExample = pipeop_preproc_torch(\"preproc_example\", function(x, a) x + a) po_example = PipeOpPreprocExample$new() po_example$param_set #> <ParamSet(3)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:              a ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:         stages ParamFct    NA    NA       3 <NoDefault[0]> [NULL] #> 3: affect_columns ParamUty    NA    NA     Inf  <Selector[1]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/replace_head.html","id":null,"dir":"Reference","previous_headings":"","what":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","title":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","text":"Replace head network Replaces head network linear layer d_out classes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/replace_head.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","text":"","code":"replace_head(network, d_out)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/replace_head.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","text":"network (torch::nn_module) network d_out (integer(1)) number output classes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":null,"dir":"Reference","previous_headings":"","what":"Sugar Function for Torch Callback — t_clbk","title":"Sugar Function for Torch Callback — t_clbk","text":"Retrieves one TorchCallbacks mlr3torch_callbacks. Works like mlr3::lrn() mlr3::lrns().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Sugar Function for Torch Callback — t_clbk","text":"","code":"t_clbk(.key, ...)  t_clbks(.keys)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Sugar Function for Torch Callback — t_clbk","text":".key (character(1)) key torch callback. ... () See description dictionary_sugar_get(). .keys (character()) keys callbacks.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Sugar Function for Torch Callback — t_clbk","text":"TorchCallback list() TorchCallbacks","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Sugar Function for Torch Callback — t_clbk","text":"","code":"t_clbk(\"progress\") #> <TorchCallback:progress> Progress #> * Generator: CallbackSetProgress #> * Parameters: list() #> * Packages: progress,mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":null,"dir":"Reference","previous_headings":"","what":"Loss Function Quick Access — t_loss","title":"Loss Function Quick Access — t_loss","text":"Retrieve one TorchLosses mlr3torch_losses. Works like mlr3::lrn() mlr3::lrns().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Function Quick Access — t_loss","text":"","code":"t_loss(.key, ...)  t_losses(.keys, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Loss Function Quick Access — t_loss","text":".key (character(1)) Key object retrieve. ... () See description dictionary_sugar_get. .keys (character()) keys losses.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Loss Function Quick Access — t_loss","text":"TorchLoss","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Loss Function Quick Access — t_loss","text":"","code":"t_loss(\"mse\", reduction = \"mean\") #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: reduction=mean #> * Packages: torch,mlr3torch #> * Task Types: regr # get the dictionary t_loss() #> <DictionaryMlr3torchLosses> with 3 stored values #> Keys: cross_entropy, l1, mse t_losses(c(\"mse\", \"l1\")) #> $mse #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr #>  #> $l1 #> <TorchLoss:l1> Absolute Error #> * Generator: nn_l1_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr #>  # get the dictionary t_losses() #> <DictionaryMlr3torchLosses> with 3 stored values #> Keys: cross_entropy, l1, mse"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":null,"dir":"Reference","previous_headings":"","what":"Optimizers Quick Access — t_opt","title":"Optimizers Quick Access — t_opt","text":"Retrieves one TorchOptimizers mlr3torch_optimizers. Works like mlr3::lrn() mlr3::lrns().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizers Quick Access — t_opt","text":"","code":"t_opt(.key, ...)  t_opts(.keys, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Optimizers Quick Access — t_opt","text":".key (character(1)) Key object retrieve. ... () See description dictionary_sugar_get. .keys (character()) keys optimizers.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Optimizers Quick Access — t_opt","text":"TorchOptimizer","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Optimizers Quick Access — t_opt","text":"","code":"t_opt(\"adam\", lr = 0.1) #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: lr=0.1 #> * Packages: torch,mlr3torch # get the dictionary t_opt() #> <DictionaryMlr3torchOptimizers> with 7 stored values #> Keys: adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd t_opts(c(\"adam\", \"sgd\")) #> $adam #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch #>  #> $sgd #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: list() #> * Packages: torch,mlr3torch #>  # get the dictionary t_opts() #> <DictionaryMlr3torchOptimizers> with 7 stored values #> Keys: adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Dataset from a Task — task_dataset","title":"Create a Dataset from a Task — task_dataset","text":"Creates torch dataset mlr3 Task. resulting dataset's $.get_batch() method returns list elements x, y index: x list tensors, whose content defined parameter feature_ingress_tokens. y target variable content defined parameter target_batchgetter. .index index batch task's data. data returned device specified parameter device.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Dataset from a Task — task_dataset","text":"","code":"task_dataset(task, feature_ingress_tokens, target_batchgetter = NULL, device)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Dataset from a Task — task_dataset","text":"task (Task) task build dataset. feature_ingress_tokens (named list() TorchIngressToken) ingress token defines one item $x value batch corresponding names. target_batchgetter (function(data, device)) function taking arguments data, data.table containing target variable, device. must return target torch tensor selected device. device (character()) device, e.g. \"cuda\" \"cpu\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Dataset from a Task — task_dataset","text":"torch::dataset","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a Dataset from a Task — task_dataset","text":"","code":"task = tsk(\"iris\") sepal_ingress = TorchIngressToken(   features = c(\"Sepal.Length\", \"Sepal.Width\"),   batchgetter = batchgetter_num,   shape = c(NA, 2) ) petal_ingress = TorchIngressToken(   features = c(\"Petal.Length\", \"Petal.Width\"),   batchgetter = batchgetter_num,   shape = c(NA, 2) ) ingress_tokens = list(sepal = sepal_ingress, petal = petal_ingress)  target_batchgetter = function(data, device) {   torch_tensor(data = data[[1L]], dtype = torch_float32(), device)$unsqueeze(2) } dataset = task_dataset(task, ingress_tokens, target_batchgetter, \"cpu\") batch = dataset$.getbatch(1:10) batch #> $x #> $x$sepal #> torch_tensor #>  5.1000  3.5000 #>  4.9000  3.0000 #>  4.7000  3.2000 #>  4.6000  3.1000 #>  5.0000  3.6000 #>  5.4000  3.9000 #>  4.6000  3.4000 #>  5.0000  3.4000 #>  4.4000  2.9000 #>  4.9000  3.1000 #> [ CPUFloatType{10,2} ] #>  #> $x$petal #> torch_tensor #>  1.4000  0.2000 #>  1.4000  0.2000 #>  1.3000  0.2000 #>  1.5000  0.2000 #>  1.4000  0.2000 #>  1.7000  0.4000 #>  1.4000  0.3000 #>  1.5000  0.2000 #>  1.4000  0.2000 #>  1.5000  0.1000 #> [ CPUFloatType{10,2} ] #>  #>  #> $.index #> torch_tensor #>   1 #>   2 #>   3 #>   4 #>   5 #>   6 #>   7 #>   8 #>   9 #>  10 #> [ CPULongType{10} ] #>  #> $y #> torch_tensor #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #> [ CPUFloatType{10,1} ] #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Callback Desctiptor — torch_callback","title":"Create a Callback Desctiptor — torch_callback","text":"Convenience function create custom TorchCallback. arguments available callback_set() also available . information correctly implement new callback, see CallbackSet.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Callback Desctiptor — torch_callback","text":"","code":"torch_callback(   id,   classname = paste0(\"CallbackSet\", capitalize(id)),   param_set = NULL,   packages = NULL,   label = capitalize(id),   man = NULL,   on_begin = NULL,   on_end = NULL,   on_exit = NULL,   on_epoch_begin = NULL,   on_before_valid = NULL,   on_epoch_end = NULL,   on_batch_begin = NULL,   on_batch_end = NULL,   on_after_backward = NULL,   on_batch_valid_begin = NULL,   on_batch_valid_end = NULL,   on_valid_end = NULL,   state_dict = NULL,   load_state_dict = NULL,   initialize = NULL,   public = NULL,   private = NULL,   active = NULL,   parent_env = parent.frame(),   inherit = CallbackSet,   lock_objects = FALSE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Callback Desctiptor — torch_callback","text":"id (character(1))` id torch callback. classname (character(1)) class name. param_set (ParamSet) parameter set, present inferred $initialize() method. packages (character())packages callback depends . Default isNULL`. label (character(1)) label torch callback. Defaults capitalized id. man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help(). default NULL. on_begin, on_end, on_epoch_begin, on_before_valid, on_epoch_end, on_batch_begin, on_batch_end, on_after_backward, on_batch_valid_begin, on_batch_valid_end, on_valid_end, on_exit (function) Function execute given stage, see section Stages. state_dict (function()) function retrieves state dict callback. available learner training. load_state_dict (function(state_dict)) Function loads callback state. initialize (function()) initialization method callback. public, private, active (list()) Additional public, private, active fields add callback. parent_env (environment()) parent environment R6Class. inherit (R6ClassGenerator) class inherit. class must either CallbackSet (default) inherit . lock_objects (logical(1)) Whether lock objects resulting R6Class. FALSE (default), values can freely assigned self without declaring class definition.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Callback Desctiptor — torch_callback","text":"TorchCallback","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Create a Callback Desctiptor — torch_callback","text":"first creates R6 class inheriting CallbackSet (using callback_set()) wraps generator TorchCallback can passed torch learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"stages","dir":"Reference","previous_headings":"","what":"Stages","title":"Create a Callback Desctiptor — torch_callback","text":"begin :: Run training loop begins. epoch_begin :: Run beginning epoch. batch_begin :: Run forward call. after_backward :: Run backward call. batch_end :: Run optimizer step. batch_valid_begin :: Run forward call validation loop. batch_valid_end :: Run forward call validation loop. valid_end :: Run end validation. epoch_end :: Run end epoch. end :: Run last epoch. exit :: Run last, using .exit().","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a Callback Desctiptor — torch_callback","text":"","code":"custom_tcb = torch_callback(\"custom\",   initialize = function(name) {     self$name = name   },   on_begin = function() {     cat(\"Hello\", self$name, \", we will train for \", self$ctx$total_epochs, \"epochs.\\n\")   },   on_end = function() {     cat(\"Training is done.\")   } )  learner = lrn(\"classif.torch_featureless\",   batch_size = 16,   epochs = 1,   callbacks = custom_tcb,   cb.custom.name = \"Marie\",   device = \"cpu\" ) task = tsk(\"iris\") learner$train(task) #> Hello Marie , we will train for  1 epochs. #> Training is done."},{"path":"https://mlr3torch.mlr-org.com/dev/news/index.html","id":"mlr3torch-012","dir":"Changelog","previous_headings":"","what":"mlr3torch 0.1.2","title":"mlr3torch 0.1.2","text":"CRAN release: 2024-10-15 Don’t use deprecated data_formats anymore Added CallbackSetTB, allows logging can viewed TensorBoard.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/news/index.html","id":"mlr3torch-011","dir":"Changelog","previous_headings":"","what":"mlr3torch 0.1.1","title":"mlr3torch 0.1.1","text":"CRAN release: 2024-09-12 fix(preprocessing): regarding construction PipeOps po(\"trafo_resize\") failed cases. fix(ci): tests run CI fix(learner): LearnerTabResnet now works correctly Fix tests run CI feat: added nn() helper function simplify creation neural network layers","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/news/index.html","id":"mlr3torch-010","dir":"Changelog","previous_headings":"","what":"mlr3torch 0.1.0","title":"mlr3torch 0.1.0","text":"CRAN release: 2024-07-08 Initial CRAN release","code":""}]
+[{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":null,"dir":"","previous_headings":"","what":"GNU Lesser General Public License","title":"GNU Lesser General Public License","text":"Version 3, 29 June 2007Copyright © 2007 Free Software Foundation, Inc. <http://fsf.org/> Everyone permitted copy distribute verbatim copies license document, changing allowed. version GNU Lesser General Public License incorporates terms conditions version 3 GNU General Public License, supplemented additional permissions listed .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_0-additional-definitions","dir":"","previous_headings":"","what":"0. Additional Definitions","title":"GNU Lesser General Public License","text":"used herein, “License” refers version 3 GNU Lesser General Public License, “GNU GPL” refers version 3 GNU General Public License. “Library” refers covered work governed License, Application Combined Work defined . “Application” work makes use interface provided Library, otherwise based Library. Defining subclass class defined Library deemed mode using interface provided Library. “Combined Work” work produced combining linking Application Library. particular version Library Combined Work made also called “Linked Version”. “Minimal Corresponding Source” Combined Work means Corresponding Source Combined Work, excluding source code portions Combined Work , considered isolation, based Application, Linked Version. “Corresponding Application Code” Combined Work means object code /source code Application, including data utility programs needed reproducing Combined Work Application, excluding System Libraries Combined Work.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_1-exception-to-section-3-of-the-gnu-gpl","dir":"","previous_headings":"","what":"1. Exception to Section 3 of the GNU GPL","title":"GNU Lesser General Public License","text":"may convey covered work sections 3 4 License without bound section 3 GNU GPL.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_2-conveying-modified-versions","dir":"","previous_headings":"","what":"2. Conveying Modified Versions","title":"GNU Lesser General Public License","text":"modify copy Library, , modifications, facility refers function data supplied Application uses facility (argument passed facility invoked), may convey copy modified version: ) License, provided make good faith effort ensure , event Application supply function data, facility still operates, performs whatever part purpose remains meaningful, b) GNU GPL, none additional permissions License applicable copy.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_3-object-code-incorporating-material-from-library-header-files","dir":"","previous_headings":"","what":"3. Object Code Incorporating Material from Library Header Files","title":"GNU Lesser General Public License","text":"object code form Application may incorporate material header file part Library. may convey object code terms choice, provided , incorporated material limited numerical parameters, data structure layouts accessors, small macros, inline functions templates (ten fewer lines length), following: ) Give prominent notice copy object code Library used Library use covered License. b) Accompany object code copy GNU GPL license document.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_4-combined-works","dir":"","previous_headings":"","what":"4. Combined Works","title":"GNU Lesser General Public License","text":"may convey Combined Work terms choice , taken together, effectively restrict modification portions Library contained Combined Work reverse engineering debugging modifications, also following: ) Give prominent notice copy Combined Work Library used Library use covered License. b) Accompany Combined Work copy GNU GPL license document. c) Combined Work displays copyright notices execution, include copyright notice Library among notices, well reference directing user copies GNU GPL license document. d) one following: 0) Convey Minimal Corresponding Source terms License, Corresponding Application Code form suitable , terms permit, user recombine relink Application modified version Linked Version produce modified Combined Work, manner specified section 6 GNU GPL conveying Corresponding Source. 1) Use suitable shared library mechanism linking Library. suitable mechanism one () uses run time copy Library already present user’s computer system, (b) operate properly modified version Library interface-compatible Linked Version. e) Provide Installation Information, otherwise required provide information section 6 GNU GPL, extent information necessary install execute modified version Combined Work produced recombining relinking Application modified version Linked Version. (use option 4d0, Installation Information must accompany Minimal Corresponding Source Corresponding Application Code. use option 4d1, must provide Installation Information manner specified section 6 GNU GPL conveying Corresponding Source.)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_5-combined-libraries","dir":"","previous_headings":"","what":"5. Combined Libraries","title":"GNU Lesser General Public License","text":"may place library facilities work based Library side side single library together library facilities Applications covered License, convey combined library terms choice, following: ) Accompany combined library copy work based Library, uncombined library facilities, conveyed terms License. b) Give prominent notice combined library part work based Library, explaining find accompanying uncombined form work.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/LICENSE.html","id":"id_6-revised-versions-of-the-gnu-lesser-general-public-license","dir":"","previous_headings":"","what":"6. Revised Versions of the GNU Lesser General Public License","title":"GNU Lesser General Public License","text":"Free Software Foundation may publish revised /new versions GNU Lesser General Public License time time. new versions similar spirit present version, may differ detail address new problems concerns. version given distinguishing version number. Library received specifies certain numbered version GNU Lesser General Public License “later version” applies , option following terms conditions either published version later version published Free Software Foundation. Library received specify version number GNU Lesser General Public License, may choose version GNU Lesser General Public License ever published Free Software Foundation. Library received specifies proxy can decide whether future versions GNU Lesser General Public License shall apply, proxy’s public statement acceptance version permanent authorization choose version Library.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/articles/callbacks.html","id":"building-blocks","dir":"Articles","previous_headings":"","what":"Building Blocks","title":"Custom Callbacks","text":"heart callback mechanism three R6 classes: CallbackSet contains methods executed different stages training loop TorchCallback wraps CallbackSet annotates meta information, including ParamSet ContextTorch defines information training process CallbackSet access using predefined callbacks, one usually interacts TorchCallback class, constructed using t_clbk(<id>): wrapped CallbackSet accessible via field $generator, can create new instance calling $new() Within training loop torch model, different stages exist callback can execute code. , list stages available, also described documentation CallbackSet. history callback, see runs code beginning, validation starts, end epoch.","code":"tc_hist = t_clbk(\"history\") tc_hist ## <TorchCallback:history> History ## * Generator: CallbackSetHistory ## * Parameters: list() ## * Packages: mlr3torch,torch cbs_hist = tc_hist$generator$new() mlr_reflections$torch$callback_stages ##  [1] \"on_begin\"             \"on_epoch_begin\"       \"on_batch_begin\"       ##  [4] \"on_after_backward\"    \"on_batch_end\"         \"on_before_valid\"      ##  [7] \"on_batch_valid_begin\" \"on_batch_valid_end\"   \"on_valid_end\"         ## [10] \"on_epoch_end\"         \"on_end\"               \"on_exit\" cbs_hist ## <CallbackSetHistory> ## * Stages: on_begin, on_before_valid, on_epoch_end"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/callbacks.html","id":"writing-a-custom-logger","dir":"Articles","previous_headings":"","what":"Writing a Custom Logger","title":"Custom Callbacks","text":"order define custom callback, can use helper torch_callback() helper function. example, create custom logger keeps track exponential moving average train loss prints end every epoch. callback takes one argument alpha smoothing parameter store moving average self$moving_loss. value alpha can later configured Learner. , implement main logic callback using two stages: on_batch_end(), called optimizer updates network parameters using mini-batch. , access last loss via ContextTorch’s (accessible via self$ctx) $last_loss field update value self$moving_loss. on_before_valid(), run validation loop. point simply print exponential moving average training loss. Finally, order make final value moving average accessible Learner training finishes, implement state_dict() method return value load_state_dict(state_dict) method, takes previously retrieved state dict sets callback. Note access final $moving_loss via learner, otherwise implement methods. created TorchCallback object associated CallbackSet R6-class us:","code":"custom_logger = torch_callback(\"custom_logger\",   initialize = function(alpha = 0.1) {     self$alpha = alpha     self$moving_loss = NULL   },   on_batch_end = function() {     if (is.null(self$moving_training_loss)) {       self$moving_loss = self$ctx$last_loss     } else {       self$moving_loss = self$alpha * self$last_loss + (1 - self$alpha) * self$moving_loss     }   },   on_before_valid = function() {     cat(sprintf(\"Epoch %s: %.2f\\n\", self$ctx$epoch, self$moving_loss))   },   state_dict = function() {     self$moving_loss   },   load_state_dict = function(state_dict) {     self$moving_loss = state_dict   } ) custom_logger ## <TorchCallback:custom_logger> Custom_logger ## * Generator: CallbackSetCustom_logger ## * Parameters: list() ## * Packages: mlr3torch,torch custom_logger$generator ## <CallbackSetCustom_logger> object generator ##   Inherits from: <inherit> ##   Public: ##     initialize: function (alpha = 0.1)  ##     state_dict: function ()  ##     load_state_dict: function (state_dict)  ##     on_before_valid: function ()  ##     on_batch_end: function ()  ##   Parent env: <environment: 0x55f6ed50fb38> ##   Locked objects: FALSE ##   Locked class: FALSE ##   Portable: TRUE"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/callbacks.html","id":"using-the-custom-logger","dir":"Articles","previous_headings":"","what":"Using the Custom Logger","title":"Custom Callbacks","text":"order showcase effects, create train simple multi layer perceptron pass callback. set smoothing parameter alpha 0.05. information returnede state_dict() now accessible via Learner’s $model-slot:","code":"task = tsk(\"iris\") mlp = lrn(\"classif.mlp\",   callbacks = custom_logger, cb.custom_logger.alpha = 0.05,   epochs = 5, batch_size = 64, neurons = 20)  mlp$train(task) ## Epoch 1: 1.22 ## Epoch 2: 1.27 ## Epoch 3: 1.48 ## Epoch 4: 1.07 ## Epoch 5: 1.09 mlp$model$callbacks$custom_logger ## [1] 1.090038"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"quickstart","dir":"Articles","previous_headings":"","what":"Quickstart","title":"Get Started","text":"vignette show get started mlr3torch training simple neural network tabular regression problem. assume familiar mlr3 framework, see e.g. mlr3 book. first example, train simple multi-layer perceptron (MLP) well-known “mtcars” task, goal predict miles per galleon (‘mpg’) cars. architecture comes predfined learner mlr3torch, can also easily create new network architectures, see Neural Networks Graphs vignette detailed introduoduion. first set seed reproducibility, load library construct task. Learners mlr3torch work similary mlr3 learners. , construct simple multi layer perceptron regression. usual calling lrn() configuring parameters: use two hidden layers 50 neurons, training, set batch size 32, number training epochs 30 device \"cpu\". complete description available parameters see ?mlr3torch::LearnerTorchMLP. can use learner training prediction just like regression learner. , split observations training test set, train learner training set create predictions test set. Finally, compute mean squared error predictions.","code":"set.seed(314) library(mlr3torch) task = tsk(\"mtcars\") task$head() #>      mpg    am  carb   cyl  disp  drat  gear    hp  qsec    vs    wt #>    <num> <num> <num> <num> <num> <num> <num> <num> <num> <num> <num> #> 1:  21.0     1     4     6   160  3.90     4   110 16.46     0 2.620 #> 2:  21.0     1     4     6   160  3.90     4   110 17.02     0 2.875 #> 3:  22.8     1     1     4   108  3.85     4    93 18.61     1 2.320 #> 4:  21.4     0     1     6   258  3.08     3   110 19.44     1 3.215 #> 5:  18.7     0     2     8   360  3.15     3   175 17.02     0 3.440 #> 6:  18.1     0     1     6   225  2.76     3   105 20.22     1 3.460 mlp = lrn(\"regr.mlp\",   # architecture parameters   neurons = c(50, 50),   # training arguments   batch_size = 32, epochs = 30, device = \"cpu\" ) mlp #> <LearnerTorchMLP[regr]:regr.mlp>: My Little Powny #> * Model: - #> * Parameters: epochs=30, device=cpu, num_threads=1, #>   num_interop_threads=1, seed=random, eval_freq=1, #>   measures_train=<list>, measures_valid=<list>, patience=0, #>   min_delta=0, batch_size=32, neurons=50,50, p=0.5, #>   activation=<nn_relu>, activation_args=<list> #> * Validate: NULL #> * Packages: mlr3, mlr3torch, torch #> * Predict Types:  [response] #> * Feature Types: integer, numeric, lazy_tensor #> * Properties: internal_tuning, marshal, validation #> * Optimizer: adam #> * Loss: mse #> * Callbacks: - # Split the obersevations into training and test set splits = partition(task) # Train the learner on the train set mlp$train(task, row_ids = splits$train) # Predict the test set prediction = mlp$predict(task, row_ids = splits$test) # Compute the mse prediction$score(msr(\"regr.mse\")) #> regr.mse  #>  283.838"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"configuring-a-learner","dir":"Articles","previous_headings":"","what":"Configuring a Learner","title":"Get Started","text":"Although torch learners quite like mlr3 learners, differences. One LearnerTorch classes construction arguments, .e. torch learners modular learners. learners free implement construction arguments, common torch learners, namely loss, optimizer callbacks. object can parameters included LearnerTorch’s parameter set. previous example, specify explicitly used default values, Adam optimizer, MSE loss callbacks. now show configure three aspects learner mlr3torch::TorchOptimizer, mlr3torch::TorchLoss, mlr3torch::TorchCallback classes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"loss","dir":"Articles","previous_headings":"Configuring a Learner","what":"Loss","title":"Get Started","text":"loss function, also known objective function cost function, measures discrepancy predicted output true output. quantifies well model performing training. R package torch, underpins mlr3torch framework, already provides number predefined loss functions Mean Squared Error (nn_mse_loss), Mean Absolute Error (nn_l1_loss), cross entropy loss (nn_cross_entropy_loss). mlr3torch, represent loss functions using mlr3torch::TorchLoss class. provides thin wrapper around torch loss functions annotates meta information, importantly paradox::ParamSet allows configure loss function. object can constructed using t_loss(<key>). , construct L1 loss function, also known Mean Absolute Error (MAE). printed output informs us wrapped loss function (nn_l1_loss), configured parameters, packages depends task types can used. ParamSet contains one parameter, namely reduction, specifies loss reduced batch. wrapped loss module generator accessible slot $generator. can pass TorchLoss argument loss initialization learner. parameters loss added learner’s ParamSet, prefixed \"loss.\". predefined loss functions stored mlr3torch_losses dictionary, can retrieved using t_loss(<key>).","code":"l1 = t_loss(\"l1\") l1 #> <TorchLoss:l1> Absolute Error #> * Generator: nn_l1_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # the paradox::ParamSet of the loss l1$param_set #> <ParamSet(1)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <num> <num>   <num>  <list> <list> #> 1: reduction ParamFct    NA    NA       2    mean [NULL] l1$generator #> <nn_l1_loss> object generator #>   Inherits from: <inherit> #>   Public: #>     .classes: nn_l1_loss nn_loss nn_module #>     initialize: function (reduction = \"mean\")  #>     forward: function (input, target)  #>     clone: function (deep = FALSE, ..., replace_values = TRUE)  #>   Private: #>     .__clone_r6__: function (deep = FALSE)  #>   Parent env: <environment: 0x556c58224810> #>   Locked objects: FALSE #>   Locked class: FALSE #>   Portable: TRUE mlp_l1 = lrn(\"regr.mlp\", loss = l1) mlp_l1$param_set$values$loss.reduction #> NULL mlr3torch_losses #> <DictionaryMlr3torchLosses> with 3 stored values #> Keys: cross_entropy, l1, mse"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"optimizer","dir":"Articles","previous_headings":"Configuring a Learner","what":"Optimizer","title":"Get Started","text":"optimizer determines model’s weights updated based calculated loss. adjusts parameters model minimize loss function, optimizing model’s performance. Optimizers work analogous loss functions, .e. mlr3torch provides thin wrapper – TorchOptimizer class – around optimizers Adam (optim_adam) SGD (optim_sgd). TorchLoss objects can constructed using t_opt(<key>). optimizers, associated ParamSet interesting see : wrapped torch optimizer can accessed slot generator. Parameters TorchOptimizer (also TorchLoss TorchCallback) can set usual mlr3 way, .e. either construction, afterwards using $set_values() method parameter set. see optimizer’s parameters added learner’s ParamSet (prefixed \"opt.\") values set values specified. exposing optimizer’s parameters, can conveniently tuned using mlr3tuning. available optimizers stored mlr3torch_optimizers dictionary.","code":"sgd = t_opt(\"sgd\") sgd #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: list() #> * Packages: torch,mlr3torch  sgd$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamDbl     0   Inf     Inf <NoDefault[0]> [NULL] #> 2:     momentum ParamDbl     0     1     Inf              0 [NULL] #> 3:    dampening ParamDbl     0     1     Inf              0 [NULL] #> 4: weight_decay ParamDbl     0     1     Inf              0 [NULL] #> 5:     nesterov ParamLgl    NA    NA       2          FALSE [NULL] sgd$param_set$set_values(   lr = 0.5, # increase learning rate   nesterov = FALSE # no nesterov momentum ) mlp_sgd = lrn(\"regr.mlp\", optimizer = sgd) as.data.table(mlp_sgd$param_set)[   startsWith(id, \"opt.\")][[1L]] #> [1] \"opt.lr\"           \"opt.momentum\"     \"opt.dampening\"    \"opt.weight_decay\" #> [5] \"opt.nesterov\" mlp_sgd$param_set$values[c(\"opt.lr\", \"opt.nesterov\")] #> $opt.lr #> [1] 0.5 #>  #> $opt.nesterov #> [1] FALSE mlr3torch_optimizers #> <DictionaryMlr3torchOptimizers> with 7 stored values #> Keys: adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"callbacks","dir":"Articles","previous_headings":"Configuring a Learner","what":"Callbacks","title":"Get Started","text":"third important configuration option callbacks allow customize training process. allows saving model checkpoints, logging metrics, implementing custom functionality specific training scenarios. tutorial implement custom callback, see Custom Callbacks vignette. , show use predefined callbacks. , retrieve \"history\" callback using t_clbk(), parameters merely saves training validation history learner can accessed afterwards. wanted learn callback , can access help page wrapped object using $help() method. Note also possible loss optimizer. predefined callbacks stored mlr3torch_callbacks dictionary.","code":"history = t_clbk(\"history\") history #> <TorchCallback:history> History #> * Generator: CallbackSetHistory #> * Parameters: list() #> * Packages: mlr3torch,torch history$help() mlr3torch_callbacks #> <DictionaryMlr3torchCallbacks> with 4 stored values #> Keys: checkpoint, history, progress, tb"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/get_started.html","id":"putting-it-together","dir":"Articles","previous_headings":"Configuring a Learner","what":"Putting it Together","title":"Get Started","text":"now define customized MLP learner using loss, optimizer callback just covered. keep track performance, use 30% training data validation evaluate using MAE Measure. Note mearures_valid measures_train parameters LearnerTorch take common mlr3::Measures, whereas loss function must TorchLoss. now train learner “mtcars” task use train-test split . make predictions unseen test data compare scores. directly optimized L1 (aka MAE) loss tweaked learning rate, configured mlp_custom learner lower MAE default mlp learner. configured learner use history callback, can find validation history $model slot: plot shows epochs 6 30.  important information stored Learner’s model $network, underlying nn_module. full description model, see ?LearnerTorch.","code":"mlp_custom = lrn(\"regr.mlp\",   # construction arguments   optimizer = sgd, loss = l1, callbacks = history,   # scores to keep track of   measures_valid = msr(\"regr.mae\"),   # other parameters are left as-is:   # architecture   neurons = c(50, 50),   # training arguments   batch_size = 32, epochs = 30, device = \"cpu\",   # validation proportion   validate = 0.3 )  mlp_custom #> <LearnerTorchMLP[regr]:regr.mlp>: My Little Powny #> * Model: - #> * Parameters: epochs=30, device=cpu, num_threads=1, #>   num_interop_threads=1, seed=random, eval_freq=1, #>   measures_train=<list>, measures_valid=<MeasureRegrSimple>, #>   patience=0, min_delta=0, batch_size=32, neurons=50,50, p=0.5, #>   activation=<nn_relu>, activation_args=<list>, opt.lr=0.5, #>   opt.nesterov=FALSE #> * Validate: 0.3 #> * Packages: mlr3, mlr3torch, torch #> * Predict Types:  [response] #> * Feature Types: integer, numeric, lazy_tensor #> * Properties: internal_tuning, marshal, validation #> * Optimizer: sgd #> * Loss: l1 #> * Callbacks: history mlp_custom$train(task, row_ids = splits$train) prediction_custom = mlp_custom$predict(task, row_ids = splits$test) prediction_custom$score(msr(\"regr.mae\")) #> regr.mae  #> 7.122375 prediction$score(msr(\"regr.mae\")) #> regr.mae  #> 15.27983 head(mlp_custom$model$callbacks$history$valid) #>    epoch     regr.mae #>    <num>        <num> #> 1:     1 1.777395e+04 #> 2:     2 3.955504e+08 #> 3:     3 1.143863e+04 #> 4:     4 1.792927e+01 #> 5:     5 1.759594e+01 #> 6:     6 1.726260e+01"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"a-torch-primer","dir":"Articles","previous_headings":"","what":"A torch Primer","title":"Internals","text":"start sampling input tensor: 2 batches 3 features: nn_module constructed nn_module_generator. nn_linear one simpler generators: Applying module gives 2-batch 4 units: neural network one (4-unit) hidden layer two outputs needs following ingredients can pipe tensor layers follows. now continue showing neural network can represented mlr3torch.","code":"input = torch_randn(2, 3) input #> torch_tensor #>  0.9197  0.6295 -0.9055 #> -2.5884  0.7595  1.2294 #> [ CPUFloatType{2,3} ] module_1 = nn_linear(in_features = 3, out_features = 4, bias = TRUE) output = module_1(input) output #> torch_tensor #>  0.6356 -0.0022  0.3491 -1.0918 #> -1.5718  1.8213 -1.6298  0.2341 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ] activation = nn_sigmoid() module_2 = nn_linear(4, 3, bias = TRUE) softmax = nn_softmax(2) output = module_1(input) output = activation(output) output = module_2(output) output = softmax(output) output #> torch_tensor #>  0.2569  0.4082  0.3350 #>  0.3488  0.3890  0.2623 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"neural-networks-as-graphs","dir":"Articles","previous_headings":"","what":"Neural Networks as Graphs","title":"Internals","text":"mlr3torch, nn_modules wrapped PipeOpModule. advantage network structure can represented mlr3pipelines::Graph object made explicit (can plotted, can extended manipulated), compared e.g. writing function pipes input series modules. PipeOpModule can used wrap module directly, usually constructed PipeOpTorch (see later). typically single input single output, although multiple inputs possible (module called multiple arguments), multiple outputs possible module-function returns list. input output channels must explicitly declared construction. now continue recreate network using PipeOpModules. can wrap linear module_1 layers like : Note po(\"module_1\") equivalent po(\"module\", id = \"module_1\"). mechanism convenient avoid ID clashes graphs contain PipeOp multiple times. can use generated PipeOp familiar way: Note use $train(), since torch modules anything maps state (filled empty list). single hidden layer neural network can constructed Graph, can training . can now use graph’s $train() method pipe tensor whole Graph. object allows easily perform forward pass, inherit nn_module, useful various reasons. Instead class inherits nn_module Graph (work R6, since multiple inheritance available), class inherits nn_module contains Graph member slot composition. class nn_graph. constructed Graph, well information shape(s) torch_tensor(s) expects inputs. Shape info communicated integer-valued numeric vector; dimensions arbitrary, e.g. batch-size, given NA. network expects input shape c(NA, 3), since first layer created nn_linear(in_features = 3, ...). Graph multiple outputs, also possible select subset outputs use, change output order, giving output_map argument. module gives us convenience torch nn_module objects, e.g.: can used transform tensors just torch::nn_module:","code":"library(mlr3torch) po_module_1 = po(\"module_1\", module = module_1) output = po_module_1$train(list(input))[[1]] output #> torch_tensor #>  0.6356 -0.0022  0.3491 -1.0918 #> -1.5718  1.8213 -1.6298  0.2341 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ] po_activation = po(\"module\", id = \"activation\", activation) po_module_2 = po(\"module_2\", module = module_2) po_softmax = po(\"module\", id = \"softmax\", module = softmax)  module_graph = po_module_1 %>>%   po_activation %>>%   po_module_2 %>>%   po_softmax  module_graph$plot(html = TRUE) output = module_graph$train(input)[[1]] output #> torch_tensor #>  0.2569  0.4082  0.3350 #>  0.3488  0.3890  0.2623 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ] # the name of the single input is: module_graph$input #>              name        train predict    op.id channel.name #>            <char>       <char>  <char>   <char>       <char> #> 1: module_1.input torch_tensor    NULL module_1        input  graph_module = nn_graph(   module_graph,   shapes_in = list(module_1.input = c(NA, 3)) ) graph_module$children #> $module_list #> An `nn_module` containing 31 parameters. #>  #> ── Modules ───────────────────────────────────────────────────────────────────── #> • 0: <nn_linear> #16 parameters #> • 1: <nn_sigmoid> #0 parameters #> • 2: <nn_linear> #15 parameters #> • 3: <nn_softmax> #0 parameters graph_module(input) #> torch_tensor #>  0.2569  0.4082  0.3350 #>  0.3488  0.3890  0.2623 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ]"},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"modeldescriptor","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"ModelDescriptor","title":"Internals","text":"PipeOpModule represents nn_module fixed specific tensor shape hyperparameters. constructing neural network using operators, one take care output shape operations match input shapes following operations. complete Graph matching PipeOpModules can constructed using operators mostly inherit PipeOpTorch, making use ModelDescriptor class. ModelDescriptor class contains Graph (mostly) PipeOpModules information. PipeOpTorch transforms ModelDescriptor adds PipeOpModules Graph. ModelDescriptors always build Graph specific Task. easiest way initialize proper ModelDescriptor use appropriate PipeOpTorchIngress given datatype. use PipeOpTorchIngressNumeric, used numeric data. ModelDescriptor S3 object contains Graph, information generate data ($ingress $task), tags build model unrelated architecture ($optimizer, $loss $callbacks) well information necessary extend graph along given output ($pointer $pointer_shape). $pointer identifies output $graph PipeOpTorch extend. Piping ModelDescriptor PipeOpTorchLinear, example, adds PipeOpModule wrapping torch::nn_linear. $pointer now updated identify output PipeOpModule, $pointer_shape shows shape changed 4 units (3 input ). model_descriptor_to_module() function converts nn_graph, functional torch::nn_module.","code":"task = tsk(\"iris\")$select(colnames(iris)[1:3])  po_torch_in = po(\"torch_ingress_num\") md = po_torch_in$train(list(task))[[1]]  md #> <ModelDescriptor: 1 ops> #> * Ingress:  torch_ingress_num.input: [(NA,3)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  N/A #> * Loss:  N/A #> * pointer:  torch_ingress_num.output [(NA,3)] unclass(md) #> $graph #> Graph with 1 PipeOps: #>                 ID         State sccssors prdcssors #>             <char>        <char>   <char>    <char> #>  torch_ingress_num <<UNTRAINED>>                    #>  #> $ingress #> $ingress$torch_ingress_num.input #> Ingress: Task[Petal.Length,Sepal.Length,Sepal.Width] --> Tensor(NA, 3) #>  #>  #> $task #> <TaskClassif:iris> (150 x 4): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (3): #>   - dbl (3): Petal.Length, Sepal.Length, Sepal.Width #>  #> $optimizer #> NULL #>  #> $loss #> NULL #>  #> $callbacks #> named list() #>  #> $pointer #> [1] \"torch_ingress_num\" \"output\"            #>  #> $pointer_shape #> [1] NA  3 po_torch_linear = po(\"nn_linear\", out_features = 4) md = po_torch_linear$train(list(md))[[1]]  md$graph #> Graph with 2 PipeOps: #>                 ID         State  sccssors         prdcssors #>             <char>        <char>    <char>            <char> #>  torch_ingress_num <<UNTRAINED>> nn_linear                   #>          nn_linear <<UNTRAINED>>           torch_ingress_num md$pointer #> [1] \"nn_linear\" \"output\" md$pointer_shape #> [1] NA  4 small_module = model_descriptor_to_module(md, list(md$pointer))  small_module(input) #> torch_tensor #>  0.2935  0.0055  0.1483  0.6004 #>  0.3216  2.2073 -0.5215 -0.5771 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"using-modeldescriptor-to-get-data","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"Using ModelDescriptor to get Data","title":"Internals","text":"ModelDescriptor represent Graph nn_module created, also way Task processed get input batches. torch::dataset can created calling task_dataset(); task feature_ingress_tokens arguments can retrieved ModelDescriptor. target_batchgetter needs created extra (necessary), since depends ultimate machine learning model, looked far. Use $.getbatch() method get batch can given nn_module. Note $x $y slot, latter used, account possible target batches. $x slot also list, since able handle NNs multiple inputs (see ).","code":"td = task_dataset(   task = md$task,   feature_ingress_tokens = md$ingress,   device = \"cpu\" )  td #> <dataset> #>   Public: #>     .getbatch: function (index)  #>     .getitem: function (index)  #>     .length: function ()  #>     all_features: Petal.Length Sepal.Length Sepal.Width Species #>     cache_lazy_tensors: FALSE #>     clone: function (deep = FALSE)  #>     device: cpu #>     feature_ingress_tokens: list #>     initialize: function (task, feature_ingress_tokens, target_batchgetter = NULL,  #>     load_state_dict: function (x, ..., .refer_to_state_dict = FALSE)  #>     state_dict: function ()  #>     target_batchgetter: NULL #>     task: TaskClassif, TaskSupervised, Task, R6 batch = td$.getbatch(1:3) batch #> $x #> $x$torch_ingress_num.input #> torch_tensor #>  1.4000  5.1000  3.5000 #>  1.4000  4.9000  3.0000 #>  1.3000  4.7000  3.2000 #> [ CPUFloatType{3,3} ] #>  #>  #> $.index #> torch_tensor #>  1 #>  2 #>  3 #> [ CPULongType{3} ]  small_module(batch$x[[1]]) #> torch_tensor #>  1.7664  4.3770 -3.4375 -0.1343 #>  1.7621  4.0245 -3.1681 -0.1304 #>  1.6213  4.0500 -3.1483 -0.0796 #> [ CPUFloatType{3,4} ][ grad_fn = <AddmmBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"building-sequential-nns","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"Building sequential NNs","title":"Internals","text":"sequential NN can easily implemented follows: Note second nn_linear need informed output dimension first nn_linear, since ModelDescriptor passed along Graph edges knows info ($pointer_shape slot).","code":"graph_generator = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 4, id = \"linear1\") %>>%     po(\"nn_sigmoid\") %>>%   po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%   po(\"nn_softmax\", dim = 2) md_sequential = graph_generator$train(task)[[1]]  graph_module = model_descriptor_to_module(md_sequential, list(md_sequential$pointer))  graph_module(input) #> torch_tensor #>  0.2308  0.4097  0.3595 #>  0.1968  0.3787  0.4245 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"building-more-interesting-nns","dir":"Articles","previous_headings":"Building Torch Models for Tasks using PipeOpTorch","what":"Building more interesting NNs","title":"Internals","text":"One main features mlr3pipelines ability easily represent computational Graphs. ModelDescriptor / PipeOpTorch setup built make full use functionality. possible multiple inputs NN using multiple PipeOpTorchIngress inputs, possible parallel alternative path branching, possible multiple outputs. Consider following (bit nonsensical) network operates differently \"Petal\" \"Sepal\" features tsk(\"iris\") manually split task , shown wholly integrated mlr3pipelines pipeline can automatically. can use create neural network iris tasks created . set $keep_results debug flag can inspection happening: make multiple observations : can observe ModelDescriptor grows passed along edges graph_iris. Note $graph slot ModelDescriptor often updated -reference, time inspect intermediate results, may contain complete graph. However, see $ingress, $pointer $pointer_shape ModelDescriptors take sepal.-path differ ones take petal.-path: po(\"nn_merge_cat\") unites two ModelDescriptors contains common ingress. pointer_shape now reflects output “cat”-operation: 2nd dimension added : Multiple ModelDescriptors created, since graph_iris multiple outpus. makes possible create neural network multiple outputs. need unite outputs graph_iris using model_descriptor_union() can pass model_descriptor_to_module(). need collect output_pointers separately. parameter list_output must set TRUE since module multiple outputs. PipeOpBranch disappears resulting Graph PipeOpModule iris_module. PipeOpTorchs graph_iris add anything ModelDescriptors. branch interpeted graph_iris runs, nn_relu path followed. iris_module therefore contains Graph “relu” activation: ModelDescriptor’s $task slot contains Task features used create input data NN inputs. can given task_dataset(), along $ingress, create torch dataset creates batches. , output graph_iris can used: resulting module multiple inputs multiple outputs. call first two rows iris, set debug $keep_results flag can inspect happening nn_module’s $graph: first linear layer takes “Sepal” input (\"linear1\") creates 2x4 tensor (batch size 2, 4 units), \"linear3\" layer 2x5 output: observe po(\"nn_merge_cat\") concatenates , expected:","code":"iris_petal = tsk(\"iris\")$select(c(\"Petal.Length\", \"Petal.Width\")) iris_sepal = tsk(\"iris\")$select(c(\"Sepal.Length\", \"Sepal.Width\")) graph_sepal = po(\"torch_ingress_num\", id = \"sepal.in\") %>>%   po(\"nn_linear\", out_features = 4, id = \"linear1\")  graph_petal = po(\"torch_ingress_num\", id = \"petal.in\") %>>%   po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%   po(\"nn_tanh\") %>>%   po(\"nn_linear\", out_features = 5, id = \"linear3\")  graph_common = ppl(\"branch\", graphs = list(     sigmoid = po(\"nn_sigmoid\"),     relu = po(\"nn_relu\")   )) %>>%   gunion(list(     po(\"nn_linear\", out_features = 1, id = \"lin_out\"),     po(\"nn_linear\", out_features = 3, id = \"cat_out\") %>>%       po(\"nn_softmax\", dim = 2)   ))   graph_iris = gunion(list(graph_sepal, graph_petal)) %>>%   po(\"nn_merge_cat\") %>>%   graph_common  graph_iris$plot(html = TRUE) graph_iris$param_set$values$branch.selection = \"relu\"  graph_iris$keep_results = TRUE  iris_mds = graph_iris$train(   input = list(sepal.in.input = iris_sepal, petal.in.input = iris_petal),   single_input = FALSE )  iris_mds #> $lin_out.output #> <ModelDescriptor: 11 ops> #> * Ingress:  sepal.in.input: [(NA,2)], petal.in.input: [(NA,2)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  N/A #> * Loss:  N/A #> * pointer:  lin_out.output [(NA,1)] #>  #> $nn_softmax.output #> <ModelDescriptor: 11 ops> #> * Ingress:  sepal.in.input: [(NA,2)], petal.in.input: [(NA,2)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  N/A #> * Loss:  N/A #> * pointer:  nn_softmax.output [(NA,3)] # sepal.in path graph_iris$pipeops$linear1$.result[[1]]$ingress #> $sepal.in.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) graph_iris$pipeops$linear1$.result[[1]]$pointer #> [1] \"linear1\" \"output\" graph_iris$pipeops$linear1$.result[[1]]$pointer_shape #> [1] NA  4  # petal.in path graph_iris$pipeops$linear3$.result[[1]]$ingress #> $petal.in.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) graph_iris$pipeops$linear3$.result[[1]]$pointer #> [1] \"linear3\" \"output\" graph_iris$pipeops$linear3$.result[[1]]$pointer_shape #> [1] NA  5 graph_iris$pipeops$nn_merge_cat$.result[[1]]$ingress #> $sepal.in.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) #>  #> $petal.in.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) graph_iris$pipeops$nn_merge_cat$.result[[1]]$pointer_shape #> [1] NA  9 iris_mds_union = model_descriptor_union(iris_mds[[1]], iris_mds[[2]]) output_pointers = list(iris_mds[[1]]$pointer, iris_mds[[2]]$pointer) output_pointers #> [[1]] #> [1] \"lin_out\" \"output\"  #>  #> [[2]] #> [1] \"nn_softmax\" \"output\" iris_module = model_descriptor_to_module(iris_mds_union, output_pointers, list_output = TRUE) iris_module$graph$plot(html = TRUE) iris_mds_union$task  # contains all features #> <TaskClassif:iris> (150 x 5): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (4): #>   - dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Width  iris_td = task_dataset(   task = iris_mds_union$task,   feature_ingress_tokens = iris_mds_union$ingress,   device = \"cpu\" )  batch = iris_td$.getbatch(1:2) batch #> $x #> $x$sepal.in.input #> torch_tensor #>  5.1000  3.5000 #>  4.9000  3.0000 #> [ CPUFloatType{2,2} ] #>  #> $x$petal.in.input #> torch_tensor #>  1.4000  0.2000 #>  1.4000  0.2000 #> [ CPUFloatType{2,2} ] #>  #>  #> $.index #> torch_tensor #>  1 #>  2 #> [ CPULongType{2} ] iris_module$graph$keep_results = TRUE  iris_module(   sepal.in.input = batch$x$sepal.in.input,   petal.in.input = batch$x$petal.in.input ) #> $lin_out.output #> torch_tensor #> 0.01 * #> -8.6465 #>  -9.1034 #> [ CPUFloatType{2,1} ][ grad_fn = <AddmmBackward0> ] #>  #> $nn_softmax.output #> torch_tensor #>  0.3961  0.4251  0.1788 #>  0.3951  0.4239  0.1810 #> [ CPUFloatType{2,3} ][ grad_fn = <SoftmaxBackward0> ] iris_module$graph$pipeops$linear1$.result #> $output #> torch_tensor #>  2.1381 -1.4562 -2.3068 -3.0385 #>  2.0602 -1.6112 -2.1434 -2.9874 #> [ CPUFloatType{2,4} ][ grad_fn = <AddmmBackward0> ] iris_module$graph$pipeops$linear3$.result #> $output #> torch_tensor #> -0.4947 -0.6350  0.1133 -0.4768  0.0452 #> -0.4947 -0.6350  0.1133 -0.4768  0.0452 #> [ CPUFloatType{2,5} ][ grad_fn = <AddmmBackward0> ] iris_module$graph$pipeops$nn_merge_cat$.result #> $output #> torch_tensor #>  2.1381 -1.4562 -2.3068 -3.0385 -0.4947 -0.6350  0.1133 -0.4768  0.0452 #>  2.0602 -1.6112 -2.1434 -2.9874 -0.4947 -0.6350  0.1133 -0.4768  0.0452 #> [ CPUFloatType{2,9} ][ grad_fn = <CatBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"building-torch-learners","dir":"Articles","previous_headings":"","what":"Building Torch Learners","title":"Internals","text":"now seen NN Graphs PipeOpModule created turned nn_modules. Using PipeOpTorch even creates ModelDescriptor objects contain additional info batch tensors extracted Tasks. complete Learner, still necessary define loss-function used optimization, optimizer, optionally callbacks. already covered class representations – TorchLoss, TorchOptimizer, TorchCallbacks, Get Started vignette. use adam optimizer, cross-entropy loss function, history callback.","code":"adam = t_opt(\"adam\", lr = 0.02) adam #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: lr=0.02 #> * Packages: torch,mlr3torch  xe = t_loss(\"cross_entropy\") xe #> <TorchLoss:cross_entropy> Cross Entropy #> * Generator: nn_cross_entropy_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: classif  history = t_clbk(\"history\") history #> <TorchCallback:history> History #> * Generator: CallbackSetHistory #> * Parameters: list() #> * Packages: mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"learnertorchmodel","dir":"Articles","previous_headings":"Building Torch Learners","what":"LearnerTorchModel","title":"Internals","text":"LearnerTorchModel represents supervised model (regression classification) using torch NNs. needs nn_module, well list TorchIngressToken define batches created Task. TorchIngressToken hard-code column-names Task used data-input, Learner created like therefore works specific Task created. (Generally full mlr3pipelines-UI used problem, see .) following uses sequential NN : training model, set hyperparameters. following calls $predict_newdata function plot response surface along Sepal.Width = mean(Sepal.Width) plane, along ground-truth values:","code":"lr_sequential = lrn(\"classif.torch_model\",   task_type = \"classif\",   network = model_descriptor_to_module(md_sequential, list(md_sequential$pointer)),   ingress_tokens = md_sequential$ingress,   optimizer = adam,   callbacks = history,   loss = xe )  lr_sequential #> <LearnerTorchModel[classif]:classif.model>: Torch Model #> * Model: - #> * Parameters: device=auto, num_threads=1, num_interop_threads=1, #>   seed=random, eval_freq=1, measures_train=<list>, #>   measures_valid=<list>, patience=0, min_delta=0, opt.lr=0.02 #> * Validate: NULL #> * Packages: mlr3, mlr3torch, torch #> * Predict Types:  [response], prob #> * Feature Types: logical, integer, numeric, character, factor, ordered, #>   POSIXct, lazy_tensor #> * Properties: featureless, hotstart_backward, hotstart_forward, #>   importance, internal_tuning, loglik, marshal, missings, multiclass, #>   oob_error, selected_features, twoclass, validation, weights #> * Optimizer: adam #> * Loss: cross_entropy #> * Callbacks: history lr_sequential$param_set$set_values(   batch_size = 50,   epochs = 100,   measures_train = msrs(c(\"classif.logloss\", \"classif.ce\")) ) # This is required to evaluate the logloss during training lr_sequential$predict_type = \"prob\"  lr_sequential$train(md_sequential$task) library(data.table) library(ggplot2)  newdata = cbind(data.table(Sepal.Width = mean(iris$Sepal.Width)), CJ(   Sepal.Length = seq(min(iris$Sepal.Length), max(iris$Sepal.Length), length.out = 30),   Petal.Length = seq(min(iris$Petal.Length), max(iris$Petal.Length), length.out = 30) ))  predictions = lr_sequential$predict_newdata(newdata)  plot_predictions = function(predictions) {   ggplot(cbind(newdata, Species = predictions$response),       aes(x = Sepal.Length, y = Petal.Length, fill = Species)) +     geom_tile(alpha = .3) +     geom_tile(alpha = .3) +     geom_point(data = iris,       aes(x = Sepal.Length, y = Petal.Length, fill = Species),       color = \"black\", pch = 21, size = 3) +     theme_bw() } plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"torch-learner-pipelines","dir":"Articles","previous_headings":"","what":"Torch Learner Pipelines","title":"Internals","text":"model shown constructed using ModelDescriptor generated Graph PipeOpTorch operators. ModelDescriptor furthermore contains Task pertains. makes possible use create NN model gets trained right away, using PipeOpTorchModelClassif. missing prerequisite now add desired TorchOptimizer TorchLoss information ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"adding-optimizer-loss-and-callback-meta-info-to-modeldescriptor","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"Adding Optimizer, Loss and Callback Meta-Info to ModelDescriptor","title":"Internals","text":"Remember ModelDescriptor $optimizer, $loss $callbacks slots necessary build complete Learner NN. can set corresponding PipeOpTorch operators. po(\"torch_optimizer\") used set $optimizer slot ModelDescriptor; takes desired TorchOptimizer object construction exports ParamSet. works analogously loss-function. also callbacks:","code":"po_adam = po(\"torch_optimizer\", optimizer = adam)  # hyperparameters are made available and can be changed: po_adam$param_set$values #> $lr #> [1] 0.02  md_sequential = po_adam$train(list(md_sequential))[[1]]  md_sequential$optimizer #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: lr=0.02 #> * Packages: torch,mlr3torch po_xe = po(\"torch_loss\", loss = xe) md_sequential = po_xe$train(list(md_sequential))[[1]]  md_sequential$loss #> <TorchLoss:cross_entropy> Cross Entropy #> * Generator: nn_cross_entropy_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: classif po_history = po(\"torch_callbacks\", callbacks = t_clbk(\"history\")) md_sequential = po_history$train(list(md_sequential))[[1]]  md_sequential$callbacks #> $history #> <TorchCallback:history> History #> * Generator: CallbackSetHistory #> * Parameters: list() #> * Packages: mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"combined-instantiation-and-training-of-learnertorchmodel","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"Combined Instantiation and Training of LearnerTorchModel","title":"Internals","text":"ModelDescriptor can now given po(\"torch_model_classif\"). po(\"torch_model_classif\") behaves similarly PipeOpLearner: returns NULL training, prediction $predict(). po(\"torch_model_classif\") behaves similarly PipeOpLearner: returns NULL training, prediction $predict().","code":"po_model = po(\"torch_model_classif\", batch_size = 50, epochs = 50)  po_model$train(list(md_sequential)) #> $output #> NULL newtask = TaskClassif$new(\"newdata\", cbind(newdata, Species = factor(NA, levels = levels(iris$Species))), target = \"Species\")  predictions = po_model$predict(list(newtask))[[1]]  plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"the-whole-pipeline","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"The whole Pipeline","title":"Internals","text":"Remember md_sequential created using Graph initial Task piped . combine Graph PipeOpTorchModelClassif, get Graph behaves like Graph ends PipeOpLearner, can therefore wrapped GraphLearner. following uses one hidden layer : Compare resulting Graph Graph trained model: Predictions, (can use predict_newdata ):","code":"graph_sequential_full = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 4, id = \"linear1\") %>>%     po(\"nn_sigmoid\") %>>%     po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%     po(\"nn_softmax\", dim = 2, id = \"softmax\") %>>%     po(\"nn_linear\", out_features = 3, id = \"linear3\") %>>%     po(\"nn_softmax\", dim = 2, id = \"softmax2\") %>>%     po(\"torch_optimizer\", optimizer = adam) %>>%     po(\"torch_loss\", loss = xe) %>>%     po(\"torch_callbacks\", callbacks = history) %>>%     po(\"torch_model_classif\", batch_size = 50, epochs = 100) lr_sequential_full = as_learner(graph_sequential_full)  lr_sequential_full$train(task) graph_sequential_full$plot(html = TRUE) model = lr_sequential_full$graph_model$state$torch_model_classif$model model$network$graph$plot(html = TRUE) predictions = lr_sequential_full$predict_newdata(newdata)  plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/internals_pipeop_torch.html","id":"mixed-pipelines","dir":"Articles","previous_headings":"Torch Learner Pipelines","what":"Mixed Pipelines","title":"Internals","text":"just limited PipeOpTorch kinds Graphs, also limited single PipeOpTorchIngress. following pipeline, example, removes Petal.Length columns Task fits model:  using Petal.Length Sepal.Length separately first?  examples hopefully demonstrated possibilities come representation neural network layers PipeOps. Even though vignette quite technical, hope given -depth understanding underlying mechanisms.","code":"gr = po(\"select\", selector = selector_name(\"Petal.Length\")) %>>%   po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 5, id = \"linear1\") %>>%   po(\"nn_relu\") %>>%   po(\"nn_linear\", out_features = 3, id = \"linear2\") %>>%   po(\"nn_softmax\", dim = 2) %>>%   po(\"torch_optimizer\", optimizer = adam) %>>%   po(\"torch_loss\", loss = xe) %>>%   po(\"torch_model_classif\", batch_size = 50, epochs = 50)  gr$plot(html = TRUE) lr = as_learner(gr) lr$train(task) predictions = lr$predict_newdata(newdata)  plot_predictions(predictions) gr = gunion(list(       po(\"select\", selector = selector_name(\"Petal.Length\"), id = \"sel1\") %>>%         po(\"torch_ingress_num\", id = \"ingress.petal\") %>>%         po(\"nn_linear\", out_features = 3, id = \"linear1\"),       po(\"select\", selector = selector_name(\"Sepal.Length\"), id = \"sel2\") %>>%         po(\"torch_ingress_num\", id = \"ingress.sepal\") %>>%         po(\"nn_linear\", out_features = 3, id = \"linear2\")     )) %>>%     po(\"nn_merge_cat\")  %>>%     po(\"nn_relu\", id = \"act1\") %>>%     po(\"nn_linear\", out_features = 3, id = \"linear3\") %>>%     po(\"nn_softmax\", dim = 2, id = \"act3\") %>>%     po(\"torch_optimizer\", optimizer = adam, lr = 0.1) %>>%     po(\"torch_loss\", loss = xe) %>>%     po(\"torch_model_classif\", batch_size = 50, epochs = 50)  gr$plot(html = TRUE) lr = as_learner(gr) lr$train(task) predictions = lr$predict_newdata(newdata)  plot_predictions(predictions)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/lazy_tensor.html","id":"creating-a-lazy-tensor","dir":"Articles","previous_headings":"","what":"Creating a Lazy Tensor","title":"Lazy Tensor","text":"Every lazy_tensor built top torch::dataset, assume familiar . information create torch::datasets, recommend reading torch package documentation. additional restriction impose dataset must .getitem .getbatch method returns list named tensors. example, create lazy_tensor length 1000, whose elements drawn uniform distribution \\([0, 1]\\). data stored -memory example, necessary $.getitem() method can e.g. load images disk. order create lazy_tensor mydata, annotate returned shapes dataset passing named list dataset_shapes. first dimension must NA, batch dimension. can also set shape NULL indicate unknown, .e. varies elements. Note case, implemented .getbatch method, even omitted specifying dataset_shapes auto-inferred. can convert vector torch_tensor just like : added preprocessing, calling $.getbatch() method mydata selecting element x. continue creating example task lt, relationship x y variable polynomial. Note target variable, classification regression, lazy_tensor, must factor numeric respectively. , plot data:  next section, create custom PipeOp fit polynomial regression model.","code":"mydata = dataset(   initialize = function() {     self$x = runif(1000, -1, 1)   },   .getbatch = function(i) list(x = torch_tensor(self$x[i])$unsqueeze(2)),   .length = function() 1000 )() lt = as_lazy_tensor(mydata, dataset_shapes = list(x = c(NA, 1))) lt[1:5] #> <ltnsr[5]> #> [1] <tnsr[1]> <tnsr[1]> <tnsr[1]> <tnsr[1]> <tnsr[1]> materialize(lt[1:5], rbind = TRUE) #> torch_tensor #> -0.9852 #> -0.0672 #> -0.0044 #> -0.4205 #>  0.4658 #> [ CPUFloatType{5,1} ] torch_equal(   materialize(lt[1], rbind = TRUE),   mydata$.getbatch(1)$x ) #> [1] TRUE library(data.table) x = mydata$x y = 0.2 + 0.1 * x - 0.1 * x^2 - 0.3 * x^3 + 0.5 * x^4 + 0.5 * x^7 + 0.6 * x^11 + rnorm(length(mydata)) * 0.1 dt = data.table(y = y, x = lt) task_poly = as_task_regr(dt, target = \"y\", id = \"poly\") task_poly #> <TaskRegr:poly> (1000 x 2) #> * Target: y #> * Properties: - #> * Features (1): #>   - lt (1): x library(ggplot2) ggplot(data = data.frame(x = x, y = y)) +   geom_point(aes(x = x, y = y), alpha = 0.5)"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/lazy_tensor.html","id":"custom-preprocessing","dir":"Articles","previous_headings":"","what":"Custom Preprocessing","title":"Lazy Tensor","text":"order create custom preprocessing operator lazy tensor, create new PipeOp class. make convenient possible, mlr3torch offters pipeop_preproc_torch() function recommend using purpose. important arguments : id - Used default identifier PipeOp fn - preprocessing function. default, first argument assumed torch_tensor remaining arguments part PipeOp’s parameter set. shapes_out - function returns shapes output tensors given input shapes. can also set NULL unknown shape \"infer\" auto-inference, see ?pipeop_preproc_torch information. , create PipeOp, transforms vector x matrix \\((x^{d_1} ..., x^{d_n})\\), \\(d\\) degrees parameter PipeOp. can now create new instance PipeOp calling $new(), set parameter degrees degrees used simulating data . , set parameter stages, always available, \"\", means preprocessing applied training prediction. data augmentation can set \"train\". create polynomial regression learner, combine polynomial preprocessor lrn(\"regr.mlp\") hidden layer (.e. linear model) train learner task. , visualize predictions see model captured non-linear relationship reasonably:  next section, briefly cover implementation details lazy_tensor, necessary work data-type, feel free skip part.","code":"PipeOpPreprocTorchPoly = pipeop_preproc_torch(\"poly\",   fn = function(x, degrees) {     torch_cat(lapply(degrees, function(d) torch_pow(x, d)), dim = 2L)   },   shapes_out = \"infer\" ) po_poly = PipeOpPreprocTorchPoly$new()  po_poly$param_set$set_values(   degrees = c(0, 1, 2, 3, 4, 7, 11),   stages = \"both\" ) lrn_poly = as_learner(   po_poly %>>% lrn(\"regr.mlp\", batch_size = 256, epochs = 100,   neurons = integer(0)) )  lrn_poly$train(task_poly) pred = lrn_poly$predict(task_poly) dt = melt(data.table(   truth = pred$truth,   response = pred$response,   x = x),   id.vars = \"x\", measure.vars = c(\"truth\", \"response\") ) dt$variable = factor(dt$variable, levels = c(\"truth\", \"response\"))  ggplot(data = dt) +   geom_point(aes(x = x, y = value, color = variable))"},{"path":"https://mlr3torch.mlr-org.com/dev/articles/lazy_tensor.html","id":"digging-into-internals","dir":"Articles","previous_headings":"","what":"Digging Into Internals","title":"Lazy Tensor","text":"Internally, lazy_tensor vector uses DataDescriptor class represent (possibly) preprocessed data. similar ModelDescriptor class used build neural nerworks using PipeOpTorch objects. DataDescriptor stores torch::dataset, mlr3pipelines::Graph metadata. Per default, preprocessing graph contains single PipOpNop nothing. printed output data descriptor informs us : number PipeOps contained preprocessing graph output shapes dataset input map, .e. data passed preprocessing graph, important multiple inputs pointer, points specific channel output PipeOp. output channel tensor represented DataDescriptor. Note id input po(\"nop\") randomly generated, needed prevent id clashes one input preprocessing graph. shape, shape tensor position pointer lazy tensor can constructed integer vector DataDescriptor. integer vector specifies element DataDescriptor lazy_tensor contains. , first two elements lazy_tensor vector represent element DataDescriptor, third element represents different element. Note indices refer DataDescriptor. Internally, lazy tensor represented list lists, element containing id DataDescriptor Currently, can single DataDescriptor lazy_tensor vector. happens materialize(lt[1]) following: Preprocessing lazy_tensor vector adds new PipeOps preprocessing graph updates metainformation like pointer output shape. show , create simple example task, using lt vector feature. see $graph new pipeop id \"poly.x\" output pointer points poly.x. Also see shape tensor now c(NA, 7) c(NA, 1) , can verify calling materialize() :","code":"desc = DataDescriptor$new(   dataset = mydata,   dataset_shapes = list(x = c(NA, 1)) ) desc #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,1)] #> * input_map: (x) -> Graph #> * pointer: nop.563486.x.output #> * shape: [(NA,1)] lt = lazy_tensor(desc, ids = c(1, 1, 2)) materialize(lt, rbind = TRUE) #> torch_tensor #> -0.9852 #> -0.9852 #> -0.0672 #> [ CPUFloatType{3,1} ] unclass(lt[[1]]) #> [[1]] #> [1] 1 #>  #> [[2]] #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,1)] #> * input_map: (x) -> Graph #> * pointer: nop.563486.x.output #> * shape: [(NA,1)] # get index and data descriptor desc = lt[[1]][[2]] id = lt[[1]][[1]]  # retrieve the batch <id> from the datast dataset_output = desc$dataset$.getbatch(id)  # batch is reorganized according to the input map graph_input = dataset_output[desc$input_map] names(graph_input) = names(desc$graph$input$name)  # the reorganized batch is fed into the preprocessing graph graph_output = desc$graph$train(graph_input, single_input = FALSE)  # the output pointed to by the pointer is returned tensor = graph_output[[paste0(desc$pointer, collapse = \".\")]] tensor #> torch_tensor #> -0.9852 #> [ CPUFloatType{1,1} ] taskin = as_task_regr(data.table(x = lt, y = 1:3), target = \"y\")  taskout = po_poly$train(list(taskin))[[1L]]  lt_out = taskout$data(cols = \"x\")$x  descout = lt_out[[1]][[2]]  descout #> <DataDescriptor: 2 ops> #> * dataset_shapes: [x: (NA,1)] #> * input_map: (x) -> Graph #> * pointer: poly.x.output #> * shape: [(NA,7)]  descout$graph #> Graph with 2 PipeOps: #>            ID         State sccssors    prdcssors #>        <char>        <char>   <char>       <char> #>  nop.563486.x        <list>   poly.x              #>        poly.x <<UNTRAINED>>          nop.563486.x materialize(lt_out[1:2], rbind = TRUE) #> torch_tensor #>  1.0000 -0.9852  0.9706 -0.9563  0.9421 -0.9009 -0.8487 #>  1.0000 -0.9852  0.9706 -0.9563  0.9421 -0.9009 -0.8487 #> [ CPUFloatType{2,7} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/authors.html","id":null,"dir":"","previous_headings":"","what":"Authors","title":"Authors and Citation","text":"Sebastian Fischer. Maintainer, author. Bernd Bischl. Contributor. Lukas Burk. Contributor. Martin Binder. Author. Florian Pfisterer. Contributor. Carson Zhang. Contributor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/authors.html","id":"citation","dir":"","previous_headings":"","what":"Citation","title":"Authors and Citation","text":"Fischer S, Binder M (2024). mlr3torch: Deep Learning 'mlr3'. R package version 0.1.2-9000, https://github.com/mlr-org/mlr3torch/, https://mlr3torch.mlr-org.com/.","code":"@Manual{,   title = {mlr3torch: Deep Learning with 'mlr3'},   author = {Sebastian Fischer and Martin Binder},   year = {2024},   note = {R package version 0.1.2-9000, https://github.com/mlr-org/mlr3torch/},   url = {https://mlr3torch.mlr-org.com/}, }"},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"mlr3torch-","dir":"","previous_headings":"","what":"Deep Learning with mlr3","title":"Deep Learning with mlr3","text":"Package website: release | dev Deep Learning torch mlr3.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"installation","dir":"","previous_headings":"","what":"Installation","title":"Deep Learning with mlr3","text":"Afterwards, also need run command : information installing torch can found .","code":"# Install from CRAN install.packages(\"mlr3torch\") # Install the development version from GitHub: pak::pak(\"mlr-org/mlr3torch\") torch::install_torch()"},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"what-is-mlr3torch","dir":"","previous_headings":"","what":"What is mlr3torch?","title":"Deep Learning with mlr3","text":"mlr3torch deep learning framework mlr3 ecosystem built top torch. allows easily build, train evaluate deep learning models lines codes, without needing worry low-level details. --shelf learners readily available, custom architectures can defined connecting PipeOpTorch operators mlr3pipelines::Graph. Using predefined learners simple multi layer perceptron (MLP) works just like mlr3 Learner. , train learner sonar example task: Next, construct architecture using PipeOpTorch objects. first pipeop – PipeOpTorchIngress – defines entrypoint network. subsequent pipeops define neural network layers. turn learner, configure loss, optimizer, callbacks well training arguments. work generic tensors, lazy_tensor type can used. wraps torch::dataset, allows preprocess data (lazily) using PipeOp objects. , flatten MNIST task, can train multi-layer perceptron . Note transform data -memory, applied data actually loaded. actually access tensors, can call materialize(). show slice resulting tensor readability: , define complex architecture one single input lazy_tensor. , first define single residual block: Next, create neural network takes input lazy_tensor (po(\"torch_ingress_ltnsr\")). first applies linear layer repeats layer using special PipeOpTorchBlock, followed network’s head. , configure loss, optimizer training parameters. Note po(\"nn_linear_0\") equivalent po(\"nn_linear\", id = \"nn_linear_0\") need avoid ID clashes linear layer po(\"nn_block\"). Next, prepend preprocessing step flattens images can directly apply learner unflattened MNIST task. order keep track performance training, use 20% data evaluate using classification accuracy. left train learner:","code":"library(mlr3torch) learner_mlp = lrn(\"classif.mlp\",   # defining network parameters   activation     = nn_relu,   neurons        = c(20, 20),   # training parameters   batch_size     = 16,   epochs         = 50,   device         = \"cpu\",   # Proportion of data to use for validation   validate = 0.3,   # Defining the optimizer, loss, and callbacks   optimizer      = t_opt(\"adam\", lr = 0.1),   loss           = t_loss(\"cross_entropy\"),   callbacks      = t_clbk(\"history\"), # this saves the history in the learner   # Measures to track   measures_valid = msrs(c(\"classif.logloss\", \"classif.ce\")),   measures_train = msrs(c(\"classif.acc\")),   # predict type (required by logloss)   predict_type = \"prob\" ) learner_mlp$train(tsk(\"sonar\")) architecture = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 20) %>>%   po(\"nn_relu\") %>>%   po(\"nn_head\") graph_mlp = architecture %>>%   po(\"torch_loss\", loss = t_loss(\"cross_entropy\")) %>>%   po(\"torch_optimizer\", optimizer = t_opt(\"adam\", lr = 0.1)) %>>%   po(\"torch_callbacks\", callbacks = t_clbk(\"history\")) %>>%   po(\"torch_model_classif\",     batch_size = 16, epochs = 50, device = \"cpu\")  graph_lrn = as_learner(graph_mlp) # load the predefined mnist task mnist = tsk(\"mnist\") mnist$head(3L) #>     label           image #>    <fctr>   <lazy_tensor> #> 1:      5 <tnsr[1x28x28]> #> 2:      0 <tnsr[1x28x28]> #> 3:      4 <tnsr[1x28x28]>  # Flatten the images flattener = po(\"trafo_reshape\", shape = c(-1, 28 * 28)) mnist_flat = flattener$train(list(mnist))[[1L]]  mnist_flat$head(3L) #>     label         image #>    <fctr> <lazy_tensor> #> 1:      5   <tnsr[784]> #> 2:      0   <tnsr[784]> #> 3:      4   <tnsr[784]> materialize(   mnist_flat$data(1:2, cols = \"image\")[[1L]],   rbind = TRUE )[1:2, 1:4] #> torch_tensor #>  0  0  0  0 #>  0  0  0  0 #> [ CPUFloatType{2,4} ] layer = list(   po(\"nop\"),   po(\"nn_linear\", out_features = 50L) %>>%     po(\"nn_dropout\") %>>% po(\"nn_relu\") ) %>>% po(\"nn_merge_sum\") deep_network = po(\"torch_ingress_ltnsr\") %>>%   po(\"nn_linear_0\", out_features = 50L) %>>%   po(\"nn_block\", layer, n_blocks = 5L) %>>%   po(\"nn_head\") %>>%   po(\"torch_loss\", loss = t_loss(\"cross_entropy\")) %>>%   po(\"torch_optimizer\", optimizer = t_opt(\"adam\")) %>>%   po(\"torch_model_classif\",     epochs = 100L, batch_size = 32   ) deep_learner = as_learner(   flattener %>>% deep_network ) deep_learner$id = \"deep_network\" set_validate(deep_learner, 0.2) deep_learner$param_set$set_values(   torch_model_classif.measures_valid = msr(\"classif.acc\") ) deep_learner$train(mnist)"},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"feature-overview","dir":"","previous_headings":"","what":"Feature Overview","title":"Deep Learning with mlr3","text":"--shelf architectures readily available mlr3::Learners. Currently, supervised regression classification supported. Custom learners can defined using Graph language mlr3pipelines. package supports tabular data, well generic tensors via lazy_tensor type. Multi-modal data can handled conveniently, lazy_tensor objects can stored alongside tabular data. possible customize training process via (predefined custom) callbacks. package fully integrated mlr3 ecosystem. Neural network architectures, well hyperparameters can easily tuned via mlr3tuning friends.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"documentation","dir":"","previous_headings":"","what":"Documentation","title":"Deep Learning with mlr3","text":"Start reading one vignettes package website!","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"contributing","dir":"","previous_headings":"","what":"Contributing:","title":"Deep Learning with mlr3","text":"run tests one needs set environment variable TEST_TORCH = 1, e.g. adding .Renviron.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"acknowledgements","dir":"","previous_headings":"","what":"Acknowledgements","title":"Deep Learning with mlr3","text":"Without great R package torch none possible. names callback stages taken luz, another high-level deep learning framework R torch. Building neural networks using PipeOpTorch operators inspired keras.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/index.html","id":"bugs-questions-feedback","dir":"","previous_headings":"","what":"Bugs, Questions, Feedback","title":"Deep Learning with mlr3","text":"mlr3torch free open source software project encourages participation feedback. issues, questions, suggestions feedback, please hesitate open “issue” GitHub page! case problems / bugs, often helpful provide “minimum working example” showcases behaviour (don’t worry bug obvious). Please understand resources project limited: response may sometimes delayed days, feature suggestions may rejected deemed tangential vision behind project.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Data Descriptor — DataDescriptor","title":"Data Descriptor — DataDescriptor","text":"data descriptor rather internal data structure used lazy_tensor data type. essence annotated torch::dataset preprocessing graph (consisting mosty PipeOpModule operators). additional meta data (e.g. pointer, shapes) allows preprocess lazy_tensors mlr3pipelines::Graph just like (non-lazy) data types. preprocessing applied materialize() called lazy_tensor. create data descriptor, can also use as_data_descriptor() function.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Data Descriptor — DataDescriptor","text":"natural define S3 class, opted R6 class avoid usual trouble serializing S3 objects. row contained DataDescriptor S3 class, copy object serializing.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Data Descriptor — DataDescriptor","text":"dataset (torch::dataset) dataset. graph (Graph) preprocessing graph. dataset_shapes (named list() (integer() NULL)) shapes output. input_map (character()) input map dataset preprocessing graph. pointer (character(2)) output pointer. pointer_shape (integer() | NULL) shape output indicated pointer. dataset_hash (character(1)) Hash wrapped dataset. hash (character(1)) Hash data descriptor. graph_input (character()) input channels preprocessing graph (cached save time). pointer_shape_predict (integer() NULL) Internal use .","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Data Descriptor — DataDescriptor","text":"DataDescriptor$new() DataDescriptor$print() DataDescriptor$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Data Descriptor — DataDescriptor","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Data Descriptor — DataDescriptor","text":"","code":"DataDescriptor$new(   dataset,   dataset_shapes = NULL,   graph = NULL,   input_map = NULL,   pointer = NULL,   pointer_shape = NULL,   pointer_shape_predict = NULL,   clone_graph = TRUE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Data Descriptor — DataDescriptor","text":"dataset (torch::dataset) torch dataset. return named list() torch_tensor objects. dataset_shapes (named list() (integer() NULL)) shapes output. Names elements list returned dataset. shape NULL (unknown, e.g. images different sizes) first dimension must NA indicate batch dimension. graph (Graph) preprocessing graph. left NULL, preprocessing applied data input_map, pointer, pointer_shape, pointer_shape_predict inferred case dataset returns one element. input_map (character()) Character vector must length input graph. Specifies data dataset fed preprocessing graph. pointer (character(2) | NULL) Points output channel within graph: Element 1 PipeOp's id element 2 PipeOp's output channel. pointer_shape (integer() | NULL) Shape output indicated pointer. pointer_shape_predict (integer() NULL) Internal use . Used Graph anticipate possible mismatches train predict shapes. clone_graph (logical(1)) Whether clone preprocessing graph.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Data Descriptor — DataDescriptor","text":"Prints object","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Data Descriptor — DataDescriptor","text":"","code":"DataDescriptor$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Data Descriptor — DataDescriptor","text":"... () Unused","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Data Descriptor — DataDescriptor","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Data Descriptor — DataDescriptor","text":"","code":"DataDescriptor$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Data Descriptor — DataDescriptor","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/DataDescriptor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Data Descriptor — DataDescriptor","text":"","code":"# Create a dataset ds = dataset(   initialize = function() self$x = torch_randn(10, 3, 3),   .getitem = function(i) list(x = self$x[i, ]),   .length = function() nrow(self$x) )() dd = DataDescriptor$new(ds, list(x = c(NA, 3, 3))) dd #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,3,3)] #> * input_map: (x) -> Graph #> * pointer: nop.6c4b31.x.output #> * shape: [(NA,3,3)] # is the same as using the converter: as_data_descriptor(ds, list(x = c(NA, 3, 3))) #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,3,3)] #> * input_map: (x) -> Graph #> * pointer: nop.6c4b31.x.output #> * shape: [(NA,3,3)]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Represent a Model with Meta-Info — ModelDescriptor","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"Represents model; possibly complete model, possibly one process built . model takes input tensors shapes shapes_in pipes graph. Input shapes get mapped input channels graph. Output shapes named output channels graph; also possible represent -ops tensors, case names input output identical. ModelDescriptor objects typically represent partial models built , case pointer slot indicates specific point graph produces tensor shape pointer_shape, graph extended. allowed graph structure modified -reference different parts code. However, modifications may never add edges elements Graph destination. particular, element graph$input may removed reference, e.g. adding edge Graph input channel PipeOp previously without parent destination. cases better create specific ModelDescriptor training Graph consisting (mostly) operators PipeOpTorchIngress, PipeOpTorch, PipeOpTorchLoss, PipeOpTorchOptimizer, PipeOpTorchCallbacks. ModelDescriptor can converted nn_graph via model_descriptor_to_module.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"","code":"ModelDescriptor(   graph,   ingress,   task,   optimizer = NULL,   loss = NULL,   callbacks = NULL,   pointer = NULL,   pointer_shape = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"graph (Graph)Graph PipeOpModule PipeOpNOP operators. ingress (uniquely named list TorchIngressToken) List inputs go graph. Names must subset graph$input$name. task (Task) (Training)-Task model built. May necessary aspects loss use etc. optimizer (TorchOptimizer | NULL) Additional info: optimizer use. loss (TorchLoss | NULL) Additional info: loss use. callbacks (list CallbackSet NULL) Additional info: callbacks use. pointer (character(2) | NULL) Indicating element model . Points output channel within graph: Element 1 PipeOp's id element 2 PipeOp's output channel. pointer_shape (integer | NULL) Shape output indicated pointer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/ModelDescriptor.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Represent a Model with Meta-Info — ModelDescriptor","text":"(ModelDescriptor)","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCenterCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentCenterCrop — mlr_pipeops_augment_center_crop","title":"PipeOpPreprocTorchAugmentCenterCrop — mlr_pipeops_augment_center_crop","text":"Calls torchvision::transform_center_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCenterCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentCenterCrop — mlr_pipeops_augment_center_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentColorJitter.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentColorJitter — mlr_pipeops_augment_color_jitter","title":"PipeOpPreprocTorchAugmentColorJitter — mlr_pipeops_augment_color_jitter","text":"Calls torchvision::transform_color_jitter, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentColorJitter.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentColorJitter — mlr_pipeops_augment_color_jitter","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentCrop — mlr_pipeops_augment_crop","title":"PipeOpPreprocTorchAugmentCrop — mlr_pipeops_augment_crop","text":"Calls torchvision::transform_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentCrop — mlr_pipeops_augment_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentHflip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentHflip — mlr_pipeops_augment_hflip","title":"PipeOpPreprocTorchAugmentHflip — mlr_pipeops_augment_hflip","text":"Calls torchvision::transform_hflip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentHflip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentHflip — mlr_pipeops_augment_hflip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomAffine.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomAffine — mlr_pipeops_augment_random_affine","title":"PipeOpPreprocTorchAugmentRandomAffine — mlr_pipeops_augment_random_affine","text":"Calls torchvision::transform_random_affine, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomAffine.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomAffine — mlr_pipeops_augment_random_affine","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomChoice.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomChoice — mlr_pipeops_augment_random_choice","title":"PipeOpPreprocTorchAugmentRandomChoice — mlr_pipeops_augment_random_choice","text":"Calls torchvision::transform_random_choice, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomChoice.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomChoice — mlr_pipeops_augment_random_choice","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomCrop — mlr_pipeops_augment_random_crop","title":"PipeOpPreprocTorchAugmentRandomCrop — mlr_pipeops_augment_random_crop","text":"Calls torchvision::transform_random_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomCrop — mlr_pipeops_augment_random_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomHorizontalFlip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomHorizontalFlip — mlr_pipeops_augment_random_horizontal_flip","title":"PipeOpPreprocTorchAugmentRandomHorizontalFlip — mlr_pipeops_augment_random_horizontal_flip","text":"Calls torchvision::transform_random_horizontal_flip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomHorizontalFlip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomHorizontalFlip — mlr_pipeops_augment_random_horizontal_flip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomOrder.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomOrder — mlr_pipeops_augment_random_order","title":"PipeOpPreprocTorchAugmentRandomOrder — mlr_pipeops_augment_random_order","text":"Calls torchvision::transform_random_order, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomOrder.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomOrder — mlr_pipeops_augment_random_order","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomResizedCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomResizedCrop — mlr_pipeops_augment_random_resized_crop","title":"PipeOpPreprocTorchAugmentRandomResizedCrop — mlr_pipeops_augment_random_resized_crop","text":"Calls torchvision::transform_random_resized_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomResizedCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomResizedCrop — mlr_pipeops_augment_random_resized_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomVerticalFlip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRandomVerticalFlip — mlr_pipeops_augment_random_vertical_flip","title":"PipeOpPreprocTorchAugmentRandomVerticalFlip — mlr_pipeops_augment_random_vertical_flip","text":"Calls torchvision::transform_random_vertical_flip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRandomVerticalFlip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRandomVerticalFlip — mlr_pipeops_augment_random_vertical_flip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentResizedCrop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentResizedCrop — mlr_pipeops_augment_resized_crop","title":"PipeOpPreprocTorchAugmentResizedCrop — mlr_pipeops_augment_resized_crop","text":"Calls torchvision::transform_resized_crop, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentResizedCrop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentResizedCrop — mlr_pipeops_augment_resized_crop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRotate.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentRotate — mlr_pipeops_augment_rotate","title":"PipeOpPreprocTorchAugmentRotate — mlr_pipeops_augment_rotate","text":"Calls torchvision::transform_rotate, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentRotate.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentRotate — mlr_pipeops_augment_rotate","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentVflip.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchAugmentVflip — mlr_pipeops_augment_vflip","title":"PipeOpPreprocTorchAugmentVflip — mlr_pipeops_augment_vflip","text":"Calls torchvision::transform_vflip, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchAugmentVflip.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchAugmentVflip — mlr_pipeops_augment_vflip","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustBrightness.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustBrightness — mlr_pipeops_trafo_adjust_brightness","title":"PipeOpPreprocTorchTrafoAdjustBrightness — mlr_pipeops_trafo_adjust_brightness","text":"Calls torchvision::transform_adjust_brightness, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustBrightness.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustBrightness — mlr_pipeops_trafo_adjust_brightness","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustGamma.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustGamma — mlr_pipeops_trafo_adjust_gamma","title":"PipeOpPreprocTorchTrafoAdjustGamma — mlr_pipeops_trafo_adjust_gamma","text":"Calls torchvision::transform_adjust_gamma, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustGamma.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustGamma — mlr_pipeops_trafo_adjust_gamma","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustHue.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustHue — mlr_pipeops_trafo_adjust_hue","title":"PipeOpPreprocTorchTrafoAdjustHue — mlr_pipeops_trafo_adjust_hue","text":"Calls torchvision::transform_adjust_hue, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustHue.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustHue — mlr_pipeops_trafo_adjust_hue","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustSaturation.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoAdjustSaturation — mlr_pipeops_trafo_adjust_saturation","title":"PipeOpPreprocTorchTrafoAdjustSaturation — mlr_pipeops_trafo_adjust_saturation","text":"Calls torchvision::transform_adjust_saturation, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoAdjustSaturation.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoAdjustSaturation — mlr_pipeops_trafo_adjust_saturation","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoGrayscale.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoGrayscale — mlr_pipeops_trafo_grayscale","title":"PipeOpPreprocTorchTrafoGrayscale — mlr_pipeops_trafo_grayscale","text":"Calls torchvision::transform_grayscale, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoGrayscale.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoGrayscale — mlr_pipeops_trafo_grayscale","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNop.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoNop — mlr_pipeops_trafo_nop","title":"PipeOpPreprocTorchTrafoNop — mlr_pipeops_trafo_nop","text":"nothing.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNop.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoNop — mlr_pipeops_trafo_nop","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNormalize.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoNormalize — mlr_pipeops_trafo_normalize","title":"PipeOpPreprocTorchTrafoNormalize — mlr_pipeops_trafo_normalize","text":"Calls torchvision::transform_normalize, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoNormalize.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoNormalize — mlr_pipeops_trafo_normalize","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoPad.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoPad — mlr_pipeops_trafo_pad","title":"PipeOpPreprocTorchTrafoPad — mlr_pipeops_trafo_pad","text":"Calls torchvision::transform_pad, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoPad.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoPad — mlr_pipeops_trafo_pad","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoReshape.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","title":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","text":"Reshapes tensor according parameter shape, calling torch_reshape(). preprocessing function applied batch-wise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoReshape.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoReshape.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PipeOpPreprocTorchTrafoReshape — mlr_pipeops_trafo_reshape","text":"shape :: integer() desired output shape. first dimension batch dimension usually -1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoResize.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoResize — mlr_pipeops_trafo_resize","title":"PipeOpPreprocTorchTrafoResize — mlr_pipeops_trafo_resize","text":"Calls torchvision::transform_resize, see information parameters. preprocessing applied whole batch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoResize.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoResize — mlr_pipeops_trafo_resize","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoRgbToGrayscale.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOpPreprocTorchTrafoRgbToGrayscale — mlr_pipeops_trafo_rgb_to_grayscale","title":"PipeOpPreprocTorchTrafoRgbToGrayscale — mlr_pipeops_trafo_rgb_to_grayscale","text":"Calls torchvision::transform_rgb_to_grayscale, see information parameters. preprocessing applied row wise (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/PipeOpPreprocTorchTrafoRgbToGrayscale.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"PipeOpPreprocTorchTrafoRgbToGrayscale — mlr_pipeops_trafo_rgb_to_grayscale","text":"R6Class inheriting PipeOpTaskPreprocTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Callback — TorchCallback","title":"Torch Callback — TorchCallback","text":"wraps CallbackSet annotates metadata, importantly ParamSet. callback created given parameter values calling $generate() method. class usually used configure callback torch learner, e.g. constructing learner ModelDescriptor. list available callbacks, see mlr3torch_callbacks. conveniently retrieve TorchCallback, use t_clbk().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Callback — TorchCallback","text":"Defined constructor argument param_set. parameter set provided construction, parameter set constructed creating parameter argument wrapped loss function, parametes type ParamUty.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Torch Callback — TorchCallback","text":"mlr3torch::TorchDescriptor -> TorchCallback","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Callback — TorchCallback","text":"mlr3torch::TorchDescriptor$generate() mlr3torch::TorchDescriptor$help() mlr3torch::TorchDescriptor$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Callback — TorchCallback","text":"TorchCallback$new() TorchCallback$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Callback — TorchCallback","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Callback — TorchCallback","text":"","code":"TorchCallback$new(   callback_generator,   param_set = NULL,   id = NULL,   label = NULL,   packages = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Callback — TorchCallback","text":"callback_generator (R6ClassGenerator) class generator callback wrapped. param_set (ParamSet NULL) parameter set. NULL (default) inferred callback_generator. id (character(1)) id new object. label (character(1)) Label new instance. packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Callback — TorchCallback","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Callback — TorchCallback","text":"","code":"TorchCallback$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Callback — TorchCallback","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchCallback.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Callback — TorchCallback","text":"","code":"# Create a new torch callback from an existing callback set torch_callback = TorchCallback$new(CallbackSetCheckpoint) # The parameters are inferred torch_callback$param_set #> <ParamSet(3)> #>           id    class lower upper nlevels        default  value #>       <char>   <char> <num> <num>   <num>         <list> <list> #> 1:      path ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:      freq ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 3: freq_type ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]  # Retrieve a torch callback from the dictionary torch_callback = t_clbk(\"checkpoint\",   path = tempfile(), freq = 1 ) torch_callback #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: path=/tmp/Rtmpmt7gXp/file657f35f35d4b, freq=1 #> * Packages: mlr3torch,torch torch_callback$label #> [1] \"Checkpoint\" torch_callback$id #> [1] \"checkpoint\"  # open the help page of the wrapped callback set # torch_callback$help()  # Create the callback set callback = torch_callback$generate() callback #> <CallbackSetCheckpoint> #> * Stages: on_batch_end, on_epoch_end, on_exit # is the same as CallbackSetCheckpoint$new(   path = tempfile(), freq = 1 ) #> <CallbackSetCheckpoint> #> * Stages: on_batch_end, on_epoch_end, on_exit  # Use in a learner learner = lrn(\"regr.mlp\", callbacks = t_clbk(\"checkpoint\")) # the parameters of the callback are added to the learner's parameter set learner$param_set #> <ParamSetCollection(36)> #>                          id    class lower upper nlevels        default #>                      <char>   <char> <num> <num>   <num>         <list> #>  1:                  epochs ParamInt 0e+00   Inf     Inf <NoDefault[0]> #>  2:                  device ParamFct    NA    NA      12 <NoDefault[0]> #>  3:             num_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  4:     num_interop_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  5:                    seed ParamInt  -Inf   Inf     Inf <NoDefault[0]> #>  6:               eval_freq ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  7:          measures_train ParamUty    NA    NA     Inf <NoDefault[0]> #>  8:          measures_valid ParamUty    NA    NA     Inf <NoDefault[0]> #>  9:                patience ParamInt 0e+00   Inf     Inf <NoDefault[0]> #> 10:               min_delta ParamDbl 0e+00   Inf     Inf <NoDefault[0]> #> 11:              batch_size ParamInt 1e+00   Inf     Inf <NoDefault[0]> #> 12:                 shuffle ParamLgl    NA    NA       2          FALSE #> 13:                 sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 14:           batch_sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 15:             num_workers ParamInt 0e+00   Inf     Inf              0 #> 16:              collate_fn ParamUty    NA    NA     Inf         [NULL] #> 17:              pin_memory ParamLgl    NA    NA       2          FALSE #> 18:               drop_last ParamLgl    NA    NA       2          FALSE #> 19:                 timeout ParamDbl  -Inf   Inf     Inf             -1 #> 20:          worker_init_fn ParamUty    NA    NA     Inf <NoDefault[0]> #> 21:          worker_globals ParamUty    NA    NA     Inf <NoDefault[0]> #> 22:         worker_packages ParamUty    NA    NA     Inf <NoDefault[0]> #> 23:                 neurons ParamUty    NA    NA     Inf <NoDefault[0]> #> 24:                       p ParamDbl 0e+00 1e+00     Inf <NoDefault[0]> #> 25:              activation ParamUty    NA    NA     Inf <NoDefault[0]> #> 26:         activation_args ParamUty    NA    NA     Inf <NoDefault[0]> #> 27:                   shape ParamUty    NA    NA     Inf <NoDefault[0]> #> 28:                  opt.lr ParamDbl 0e+00   Inf     Inf          0.001 #> 29:               opt.betas ParamUty    NA    NA     Inf    0.900,0.999 #> 30:                 opt.eps ParamDbl 1e-16 1e-04     Inf          1e-08 #> 31:        opt.weight_decay ParamDbl 0e+00 1e+00     Inf              0 #> 32:             opt.amsgrad ParamLgl    NA    NA       2          FALSE #> 33:          loss.reduction ParamFct    NA    NA       2           mean #> 34:      cb.checkpoint.path ParamUty    NA    NA     Inf <NoDefault[0]> #> 35:      cb.checkpoint.freq ParamInt 1e+00   Inf     Inf <NoDefault[0]> #> 36: cb.checkpoint.freq_type ParamFct    NA    NA       2          epoch #>                          id    class lower upper nlevels        default #>            value #>           <list> #>  1:       [NULL] #>  2:         auto #>  3:            1 #>  4:            1 #>  5:       random #>  6:            1 #>  7:    <list[0]> #>  8:    <list[0]> #>  9:            0 #> 10:            0 #> 11:       [NULL] #> 12:       [NULL] #> 13:       [NULL] #> 14:       [NULL] #> 15:       [NULL] #> 16:       [NULL] #> 17:       [NULL] #> 18:       [NULL] #> 19:       [NULL] #> 20:       [NULL] #> 21:       [NULL] #> 22:       [NULL] #> 23:              #> 24:          0.5 #> 25: <nn_relu[1]> #> 26:    <list[0]> #> 27:       [NULL] #> 28:       [NULL] #> 29:       [NULL] #> 30:       [NULL] #> 31:       [NULL] #> 32:       [NULL] #> 33:       [NULL] #> 34:       [NULL] #> 35:       [NULL] #> 36:       [NULL] #>            value"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Torch Descriptors — TorchDescriptor","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Abstract Base Class TorchLoss, TorchOptimizer, TorchCallback inherit. class wraps generator (R6Class Generator torch version generator) annotates metadata ParamSet, label, ID, packages, manual page. parameters construction arguments wrapped generator parameter $values passed generator calling public method $generate().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Defined constructor argument param_set. parameters tagged \"train\", done automatically initialize.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"label (character(1)) Label object. Can used tables, plot text output instead ID. param_set (ParamSet) Set hyperparameters. packages (character(1)) Set required packages. packages loaded, attached. id (character(1)) Identifier object. Used tables, plot text output. generator wrapped generator described. man (character(1)) String format [pkg]::[topic] pointing manual page object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"phash (character(1)) Hash (unique identifier) partial object, excluding components varied systematically (e.g. parameter values).","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"TorchDescriptor$new() TorchDescriptor$print() TorchDescriptor$generate() TorchDescriptor$help() TorchDescriptor$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$new(   generator,   id = NULL,   param_set = NULL,   packages = NULL,   label = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"generator wrapped generator described. id (character(1)) id new object. param_set (ParamSet) parameter set. packages (character()) R packages object depends . label (character(1)) Label new instance. man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Prints object","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"... ","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-generate-","dir":"Reference","previous_headings":"","what":"Method generate()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Calls generator given parameter values.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$generate()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-help-","dir":"Reference","previous_headings":"","what":"Method help()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"Displays help file wrapped object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$help()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"","code":"TorchDescriptor$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchDescriptor.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Descriptors — TorchDescriptor","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Ingress Token — TorchIngressToken","title":"Torch Ingress Token — TorchIngressToken","text":"function creates S3 class class \"TorchIngressToken\", internal data structure. contains (meta-)information batch generated Task fed entry point neural network. stored ingress field ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Ingress Token — TorchIngressToken","text":"","code":"TorchIngressToken(features, batchgetter, shape)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Ingress Token — TorchIngressToken","text":"features (character) Features batchgetter operate. batchgetter (function) Function two arguments: data device. function given output Task$data(rows = batch_indices, cols = features) produce tensor shape shape_out. shape (integer) Shape batchgetter produce. Batch-dimension included NA.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Torch Ingress Token — TorchIngressToken","text":"TorchIngressToken object.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchIngressToken.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Ingress Token — TorchIngressToken","text":"","code":"# Define a task for which we want to define an ingress token task = tsk(\"iris\")  # We create an ingress token for two feature Sepal.Length and Petal.Length: # We have to specify the features, the batchgetter and the shape features = c(\"Sepal.Length\", \"Petal.Length\") # As a batchgetter we use batchgetter_num  batch_dt = task$data(rows = 1:10, cols =features) batch_dt #>     Sepal.Length Petal.Length #>            <num>        <num> #>  1:          5.1          1.4 #>  2:          4.9          1.4 #>  3:          4.7          1.3 #>  4:          4.6          1.5 #>  5:          5.0          1.4 #>  6:          5.4          1.7 #>  7:          4.6          1.4 #>  8:          5.0          1.5 #>  9:          4.4          1.4 #> 10:          4.9          1.5 batch_tensor = batchgetter_num(batch_dt, \"cpu\") batch_tensor #> torch_tensor #>  5.1000  1.4000 #>  4.9000  1.4000 #>  4.7000  1.3000 #>  4.6000  1.5000 #>  5.0000  1.4000 #>  5.4000  1.7000 #>  4.6000  1.4000 #>  5.0000  1.5000 #>  4.4000  1.4000 #>  4.9000  1.5000 #> [ CPUFloatType{10,2} ]  # The shape is unknown in the first dimension (batch dimension)  ingress_token = TorchIngressToken(   features = features,   batchgetter = batchgetter_num,   shape = c(NA, 2) ) ingress_token #> Ingress: Task[Sepal.Length,Petal.Length] --> Tensor(NA, 2)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Loss — TorchLoss","title":"Torch Loss — TorchLoss","text":"wraps torch::nn_loss annotates metadata, importantly ParamSet. loss function created given parameter values calling $generate() method. class usually used configure loss function torch learner, e.g. construcing learner ModelDescriptor. list available losses, see mlr3torch_losses. Items dictionary can retrieved using t_loss().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Loss — TorchLoss","text":"Defined constructor argument param_set. parameter set provided construction, parameter set constructed creating parameter argument wrapped loss function, parametes type ParamUty.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Torch Loss — TorchLoss","text":"mlr3torch::TorchDescriptor -> TorchLoss","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Torch Loss — TorchLoss","text":"task_types (character()) task types loss supports.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Loss — TorchLoss","text":"mlr3torch::TorchDescriptor$generate() mlr3torch::TorchDescriptor$help()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Loss — TorchLoss","text":"TorchLoss$new() TorchLoss$print() TorchLoss$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Loss — TorchLoss","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Loss — TorchLoss","text":"","code":"TorchLoss$new(   torch_loss,   task_types = NULL,   param_set = NULL,   id = NULL,   label = NULL,   packages = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Loss — TorchLoss","text":"torch_loss (nn_loss) loss module. task_types (character()) task types supported loss. param_set (ParamSet NULL) parameter set. NULL (default) inferred torch_loss. id (character(1)) id new object. label (character(1)) Label new instance. packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Torch Loss — TorchLoss","text":"Prints object","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Loss — TorchLoss","text":"","code":"TorchLoss$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Loss — TorchLoss","text":"... ","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Loss — TorchLoss","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Loss — TorchLoss","text":"","code":"TorchLoss$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Loss — TorchLoss","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchLoss.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Loss — TorchLoss","text":"","code":"# Create a new torch loss torch_loss = TorchLoss$new(torch_loss = nn_mse_loss, task_types = \"regr\") torch_loss #> <TorchLoss:nn_mse_loss> nn_mse_loss #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # the parameters are inferred torch_loss$param_set #> <ParamSet(1)> #>           id    class lower upper nlevels        default  value #>       <char>   <char> <num> <num>   <num>         <list> <list> #> 1: reduction ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]  # Retrieve a loss from the dictionary: torch_loss = t_loss(\"mse\", reduction = \"mean\") # is the same as torch_loss #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: reduction=mean #> * Packages: torch,mlr3torch #> * Task Types: regr torch_loss$param_set #> <ParamSet(1)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <num> <num>   <num>  <list> <list> #> 1: reduction ParamFct    NA    NA       2    mean   mean torch_loss$label #> [1] \"Mean Squared Error\" torch_loss$task_types #> [1] \"regr\" torch_loss$id #> [1] \"mse\"  # Create the loss function loss_fn = torch_loss$generate() loss_fn #> An `nn_module` containing 0 parameters. # Is the same as nn_mse_loss(reduction = \"mean\") #> An `nn_module` containing 0 parameters.  # open the help page of the wrapped loss function # torch_loss$help()  # Use in a learner learner = lrn(\"regr.mlp\", loss = t_loss(\"mse\")) # The parameters of the loss are added to the learner's parameter set learner$param_set #> <ParamSetCollection(33)> #>                      id    class lower upper nlevels        default #>                  <char>   <char> <num> <num>   <num>         <list> #>  1:              epochs ParamInt 0e+00   Inf     Inf <NoDefault[0]> #>  2:              device ParamFct    NA    NA      12 <NoDefault[0]> #>  3:         num_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  4: num_interop_threads ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  5:                seed ParamInt  -Inf   Inf     Inf <NoDefault[0]> #>  6:           eval_freq ParamInt 1e+00   Inf     Inf <NoDefault[0]> #>  7:      measures_train ParamUty    NA    NA     Inf <NoDefault[0]> #>  8:      measures_valid ParamUty    NA    NA     Inf <NoDefault[0]> #>  9:            patience ParamInt 0e+00   Inf     Inf <NoDefault[0]> #> 10:           min_delta ParamDbl 0e+00   Inf     Inf <NoDefault[0]> #> 11:          batch_size ParamInt 1e+00   Inf     Inf <NoDefault[0]> #> 12:             shuffle ParamLgl    NA    NA       2          FALSE #> 13:             sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 14:       batch_sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 15:         num_workers ParamInt 0e+00   Inf     Inf              0 #> 16:          collate_fn ParamUty    NA    NA     Inf         [NULL] #> 17:          pin_memory ParamLgl    NA    NA       2          FALSE #> 18:           drop_last ParamLgl    NA    NA       2          FALSE #> 19:             timeout ParamDbl  -Inf   Inf     Inf             -1 #> 20:      worker_init_fn ParamUty    NA    NA     Inf <NoDefault[0]> #> 21:      worker_globals ParamUty    NA    NA     Inf <NoDefault[0]> #> 22:     worker_packages ParamUty    NA    NA     Inf <NoDefault[0]> #> 23:             neurons ParamUty    NA    NA     Inf <NoDefault[0]> #> 24:                   p ParamDbl 0e+00 1e+00     Inf <NoDefault[0]> #> 25:          activation ParamUty    NA    NA     Inf <NoDefault[0]> #> 26:     activation_args ParamUty    NA    NA     Inf <NoDefault[0]> #> 27:               shape ParamUty    NA    NA     Inf <NoDefault[0]> #> 28:              opt.lr ParamDbl 0e+00   Inf     Inf          0.001 #> 29:           opt.betas ParamUty    NA    NA     Inf    0.900,0.999 #> 30:             opt.eps ParamDbl 1e-16 1e-04     Inf          1e-08 #> 31:    opt.weight_decay ParamDbl 0e+00 1e+00     Inf              0 #> 32:         opt.amsgrad ParamLgl    NA    NA       2          FALSE #> 33:      loss.reduction ParamFct    NA    NA       2           mean #>                      id    class lower upper nlevels        default #>            value #>           <list> #>  1:       [NULL] #>  2:         auto #>  3:            1 #>  4:            1 #>  5:       random #>  6:            1 #>  7:    <list[0]> #>  8:    <list[0]> #>  9:            0 #> 10:            0 #> 11:       [NULL] #> 12:       [NULL] #> 13:       [NULL] #> 14:       [NULL] #> 15:       [NULL] #> 16:       [NULL] #> 17:       [NULL] #> 18:       [NULL] #> 19:       [NULL] #> 20:       [NULL] #> 21:       [NULL] #> 22:       [NULL] #> 23:              #> 24:          0.5 #> 25: <nn_relu[1]> #> 26:    <list[0]> #> 27:       [NULL] #> 28:       [NULL] #> 29:       [NULL] #> 30:       [NULL] #> 31:       [NULL] #> 32:       [NULL] #> 33:       [NULL] #>            value"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Optimizer — TorchOptimizer","title":"Torch Optimizer — TorchOptimizer","text":"wraps torch::torch_optimizer_generatora annotates metadata, importantly ParamSet. optimizer created given parameter values calling $generate() method. class usually used configure optimizer torch learner, e.g. construcing learner ModelDescriptor. list available optimizers, see mlr3torch_optimizers. Items dictionary can retrieved using t_opt().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Optimizer — TorchOptimizer","text":"Defined constructor argument param_set. parameter set provided construction, parameter set constructed creating parameter argument wrapped loss function, parametes type ParamUty.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Torch Optimizer — TorchOptimizer","text":"mlr3torch::TorchDescriptor -> TorchOptimizer","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Optimizer — TorchOptimizer","text":"mlr3torch::TorchDescriptor$help() mlr3torch::TorchDescriptor$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Optimizer — TorchOptimizer","text":"TorchOptimizer$new() TorchOptimizer$generate() TorchOptimizer$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Optimizer — TorchOptimizer","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Optimizer — TorchOptimizer","text":"","code":"TorchOptimizer$new(   torch_optimizer,   param_set = NULL,   id = NULL,   label = NULL,   packages = NULL,   man = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Optimizer — TorchOptimizer","text":"torch_optimizer (torch_optimizer_generator) torch optimizer. param_set (ParamSet NULL) parameter set. NULL (default) inferred torch_optimizer. id (character(1)) id new object. label (character(1)) Label new instance. packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"method-generate-","dir":"Reference","previous_headings":"","what":"Method generate()","title":"Torch Optimizer — TorchOptimizer","text":"Instantiates optimizer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Optimizer — TorchOptimizer","text":"","code":"TorchOptimizer$generate(params)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Optimizer — TorchOptimizer","text":"params (named list() torch_tensors) parameters network.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Torch Optimizer — TorchOptimizer","text":"torch_optimizer","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Optimizer — TorchOptimizer","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Optimizer — TorchOptimizer","text":"","code":"TorchOptimizer$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Optimizer — TorchOptimizer","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/TorchOptimizer.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Optimizer — TorchOptimizer","text":"","code":"# Create a new torch loss torch_opt = TorchOptimizer$new(optim_adam, label = \"adam\") torch_opt #> <TorchOptimizer:optim_adam> adam #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch # If the param set is not specified, parameters are inferred but are of class ParamUty torch_opt$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:        betas ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 3:          eps ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 4: weight_decay ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 5:      amsgrad ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]  # open the help page of the wrapped optimizer # torch_opt$help()  # Retrieve an optimizer from the dictionary torch_opt = t_opt(\"sgd\", lr = 0.1) torch_opt #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: lr=0.1 #> * Packages: torch,mlr3torch torch_opt$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamDbl     0   Inf     Inf <NoDefault[0]>    0.1 #> 2:     momentum ParamDbl     0     1     Inf              0 [NULL] #> 3:    dampening ParamDbl     0     1     Inf              0 [NULL] #> 4: weight_decay ParamDbl     0     1     Inf              0 [NULL] #> 5:     nesterov ParamLgl    NA    NA       2          FALSE [NULL] torch_opt$label #> [1] \"Stochastic Gradient Descent\" torch_opt$id #> [1] \"sgd\"  # Create the optimizer for a network net = nn_linear(10, 1) opt = torch_opt$generate(net$parameters)  # is the same as optim_sgd(net$parameters, lr = 0.1) #> <optim_sgd> #>   Inherits from: <torch_optimizer> #>   Public: #>     add_param_group: function (param_group)  #>     clone: function (deep = FALSE)  #>     defaults: list #>     initialize: function (params, lr = optim_required(), momentum = 0, dampening = 0,  #>     load_state_dict: function (state_dict, ..., .refer_to_state_dict = FALSE)  #>     param_groups: list #>     state: State, R6 #>     state_dict: function ()  #>     step: function (closure = NULL)  #>     zero_grad: function ()  #>   Private: #>     step_helper: function (closure, loop_fun)   # Use in a learner learner = lrn(\"regr.mlp\", optimizer = t_opt(\"sgd\")) # The parameters of the optimizer are added to the learner's parameter set learner$param_set #> <ParamSetCollection(33)> #>                      id    class lower upper nlevels        default #>                  <char>   <char> <num> <num>   <num>         <list> #>  1:              epochs ParamInt     0   Inf     Inf <NoDefault[0]> #>  2:              device ParamFct    NA    NA      12 <NoDefault[0]> #>  3:         num_threads ParamInt     1   Inf     Inf <NoDefault[0]> #>  4: num_interop_threads ParamInt     1   Inf     Inf <NoDefault[0]> #>  5:                seed ParamInt  -Inf   Inf     Inf <NoDefault[0]> #>  6:           eval_freq ParamInt     1   Inf     Inf <NoDefault[0]> #>  7:      measures_train ParamUty    NA    NA     Inf <NoDefault[0]> #>  8:      measures_valid ParamUty    NA    NA     Inf <NoDefault[0]> #>  9:            patience ParamInt     0   Inf     Inf <NoDefault[0]> #> 10:           min_delta ParamDbl     0   Inf     Inf <NoDefault[0]> #> 11:          batch_size ParamInt     1   Inf     Inf <NoDefault[0]> #> 12:             shuffle ParamLgl    NA    NA       2          FALSE #> 13:             sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 14:       batch_sampler ParamUty    NA    NA     Inf <NoDefault[0]> #> 15:         num_workers ParamInt     0   Inf     Inf              0 #> 16:          collate_fn ParamUty    NA    NA     Inf         [NULL] #> 17:          pin_memory ParamLgl    NA    NA       2          FALSE #> 18:           drop_last ParamLgl    NA    NA       2          FALSE #> 19:             timeout ParamDbl  -Inf   Inf     Inf             -1 #> 20:      worker_init_fn ParamUty    NA    NA     Inf <NoDefault[0]> #> 21:      worker_globals ParamUty    NA    NA     Inf <NoDefault[0]> #> 22:     worker_packages ParamUty    NA    NA     Inf <NoDefault[0]> #> 23:             neurons ParamUty    NA    NA     Inf <NoDefault[0]> #> 24:                   p ParamDbl     0     1     Inf <NoDefault[0]> #> 25:          activation ParamUty    NA    NA     Inf <NoDefault[0]> #> 26:     activation_args ParamUty    NA    NA     Inf <NoDefault[0]> #> 27:               shape ParamUty    NA    NA     Inf <NoDefault[0]> #> 28:              opt.lr ParamDbl     0   Inf     Inf <NoDefault[0]> #> 29:        opt.momentum ParamDbl     0     1     Inf              0 #> 30:       opt.dampening ParamDbl     0     1     Inf              0 #> 31:    opt.weight_decay ParamDbl     0     1     Inf              0 #> 32:        opt.nesterov ParamLgl    NA    NA       2          FALSE #> 33:      loss.reduction ParamFct    NA    NA       2           mean #>                      id    class lower upper nlevels        default #>            value #>           <list> #>  1:       [NULL] #>  2:         auto #>  3:            1 #>  4:            1 #>  5:       random #>  6:            1 #>  7:    <list[0]> #>  8:    <list[0]> #>  9:            0 #> 10:            0 #> 11:       [NULL] #> 12:       [NULL] #> 13:       [NULL] #> 14:       [NULL] #> 15:       [NULL] #> 16:       [NULL] #> 17:       [NULL] #> 18:       [NULL] #> 19:       [NULL] #> 20:       [NULL] #> 21:       [NULL] #> 22:       [NULL] #> 23:              #> 24:          0.5 #> 25: <nn_relu[1]> #> 26:    <list[0]> #> 27:       [NULL] #> 28:       [NULL] #> 29:       [NULL] #> 30:       [NULL] #> 31:       [NULL] #> 32:       [NULL] #> 33:       [NULL] #>            value"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to Data Descriptor — as_data_descriptor","title":"Convert to Data Descriptor — as_data_descriptor","text":"Converts input DataDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to Data Descriptor — as_data_descriptor","text":"","code":"as_data_descriptor(x, dataset_shapes, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to Data Descriptor — as_data_descriptor","text":"x () Object convert. dataset_shapes (named list() (integer() NULL)) shapes output. Names elements list returned dataset. shape NULL (unknown, e.g. images different sizes) first dimension must NA indicate batch dimension. ... () arguments passed DataDescriptor constructor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_data_descriptor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Convert to Data Descriptor — as_data_descriptor","text":"","code":"ds = dataset(\"example\",   initialize = function() self$iris = iris[, -5],   .getitem = function(i) list(x = torch_tensor(as.numeric(self$iris[i, ]))),   .length = function() nrow(self$iris) )() as_data_descriptor(ds, list(x = c(NA, 4L))) #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,4)] #> * input_map: (x) -> Graph #> * pointer: nop.fe081b.x.output #> * shape: [(NA,4)]  # if the dataset has a .getbatch method, the shapes are inferred ds2 = dataset(\"example\",   initialize = function() self$iris = iris[, -5],   .getbatch = function(i) list(x = torch_tensor(as.matrix(self$iris[i, ]))),   .length = function() nrow(self$iris) )() as_data_descriptor(ds2) #> <DataDescriptor: 1 ops> #> * dataset_shapes: [x: (NA,4)] #> * input_map: (x) -> Graph #> * pointer: nop.d2a8ff.x.output #> * shape: [(NA,4)]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to Lazy Tensor — as_lazy_tensor","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"Convert object lazy_tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"","code":"as_lazy_tensor(x, ...)  # S3 method for class 'dataset' as_lazy_tensor(x, dataset_shapes = NULL, ids = NULL, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"x () Object convert lazy_tensor ... () Additional arguments passed method. dataset_shapes (named list() (integer() NULL)) shapes output. Names elements list returned dataset. shape NULL (unknown, e.g. images different sizes) first dimension must NA indicate batch dimension. ids (integer()) ids include lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_lazy_tensor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Convert to Lazy Tensor — as_lazy_tensor","text":"","code":"iris_ds = dataset(\"iris\",   initialize = function() {     self$iris = iris[, -5]   },   .getbatch = function(i) {     list(x = torch_tensor(as.matrix(self$iris[i, ])))   },   .length = function() nrow(self$iris) )() # no need to specify the dataset shapes as they can be inferred from the .getbatch method # only first 5 observations as_lazy_tensor(iris_ds, ids = 1:5) #> <ltnsr[5]> #> [1] <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> # all observations head(as_lazy_tensor(iris_ds)) #> <ltnsr[6]> #> [1] <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]>  iris_ds2 = dataset(\"iris\",   initialize = function() self$iris = iris[, -5],   .getitem = function(i) list(x = torch_tensor(as.numeric(self$iris[i, ]))),   .length = function() nrow(self$iris) )() # if .getitem is implemented we cannot infer the shapes as they might vary, # so we have to annotate them explicitly as_lazy_tensor(iris_ds2, dataset_shapes = list(x = c(NA, 4L)))[1:5] #> <ltnsr[5]> #> [1] <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]> <tnsr[4]>  # Convert a matrix lt = as_lazy_tensor(matrix(rnorm(100), nrow = 20)) materialize(lt[1:5], rbind = TRUE) #> torch_tensor #> -1.4000  0.4682  0.0700  1.0743  1.9243 #>  0.2553  0.3630 -0.6391 -0.6651  1.2984 #> -2.4373 -1.3045 -0.0500  1.1140  0.7488 #> -0.0056  0.7378 -0.2515 -0.2459  0.5562 #>  0.6216  1.8885  0.4448 -1.1776 -0.5483 #> [ CPUFloatType{5,5} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to a TorchCallback — as_torch_callback","title":"Convert to a TorchCallback — as_torch_callback","text":"Converts object TorchCallback.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to a TorchCallback — as_torch_callback","text":"","code":"as_torch_callback(x, clone = FALSE, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to a TorchCallback — as_torch_callback","text":"x () Object converted. clone (logical(1)) Whether make deep clone. ... () Additional arguments","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callback.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to a TorchCallback — as_torch_callback","text":"TorchCallback.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to a list of Torch Callbacks — as_torch_callbacks","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"Converts object list TorchCallback.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"","code":"as_torch_callbacks(x, clone, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"x () Object convert. clone (logical(1)) Whether create deep clone. ... () Additional arguments.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_callbacks.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to a list of Torch Callbacks — as_torch_callbacks","text":"list() TorchCallbacks","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to TorchLoss — as_torch_loss","title":"Convert to TorchLoss — as_torch_loss","text":"Converts object TorchLoss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to TorchLoss — as_torch_loss","text":"","code":"as_torch_loss(x, clone = FALSE, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to TorchLoss — as_torch_loss","text":"x () Object convert TorchLoss. clone (logical(1)) Whether make deep clone. ... () Additional arguments. Currently used pass additional constructor arguments TorchLoss objects type nn_loss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_loss.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to TorchLoss — as_torch_loss","text":"TorchLoss.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":null,"dir":"Reference","previous_headings":"","what":"Convert to TorchOptimizer — as_torch_optimizer","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"Converts object TorchOptimizer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"","code":"as_torch_optimizer(x, clone = FALSE, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"x () Object convert TorchOptimizer. clone (logical(1)) Whether make deep clone. Default FALSE. ... () Additional arguments. Currently used pass additional constructor arguments TorchOptimizer objects type torch_optimizer_generator.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/as_torch_optimizer.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Convert to TorchOptimizer — as_torch_optimizer","text":"TorchOptimizer","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/assert_lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Assert Lazy Tensor — assert_lazy_tensor","title":"Assert Lazy Tensor — assert_lazy_tensor","text":"Asserts whether something lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/assert_lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Assert Lazy Tensor — assert_lazy_tensor","text":"","code":"assert_lazy_tensor(x)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/assert_lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Assert Lazy Tensor — assert_lazy_tensor","text":"x () Object check.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/auto_device.html","id":null,"dir":"Reference","previous_headings":"","what":"Auto Device — auto_device","title":"Auto Device — auto_device","text":"First tries cuda, cpu.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/auto_device.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Auto Device — auto_device","text":"","code":"auto_device(device = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/auto_device.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Auto Device — auto_device","text":"device (character(1)) device. NULL, returned .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_categ.html","id":null,"dir":"Reference","previous_headings":"","what":"Batchgetter for Categorical data — batchgetter_categ","title":"Batchgetter for Categorical data — batchgetter_categ","text":"Converts data frame categorical data long tensor converting data integers. input checks performed.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_categ.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Batchgetter for Categorical data — batchgetter_categ","text":"","code":"batchgetter_categ(data, device, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_categ.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Batchgetter for Categorical data — batchgetter_categ","text":"data (data.table)data.table converted tensor. device (character(1)) device. ... () Unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_num.html","id":null,"dir":"Reference","previous_headings":"","what":"Batchgetter for Numeric Data — batchgetter_num","title":"Batchgetter for Numeric Data — batchgetter_num","text":"Converts data frame numeric data float tensor calling .matrix(). input checks performed","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_num.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Batchgetter for Numeric Data — batchgetter_num","text":"","code":"batchgetter_num(data, device, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/batchgetter_num.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Batchgetter for Numeric Data — batchgetter_num","text":"data (data.table())data.table converted tensor. device (character(1)) device tensor created. ... () Unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Set of Callbacks for Torch — callback_set","title":"Create a Set of Callbacks for Torch — callback_set","text":"Creates R6ClassGenerator inheriting CallbackSet. Additionally performs checks stages accidentally misspelled. create TorchCallback use torch_callback(). order resulting class cloneable, private method $deep_clone() must provided.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Set of Callbacks for Torch — callback_set","text":"","code":"callback_set(   classname,   on_begin = NULL,   on_end = NULL,   on_exit = NULL,   on_epoch_begin = NULL,   on_before_valid = NULL,   on_epoch_end = NULL,   on_batch_begin = NULL,   on_batch_end = NULL,   on_after_backward = NULL,   on_batch_valid_begin = NULL,   on_batch_valid_end = NULL,   on_valid_end = NULL,   state_dict = NULL,   load_state_dict = NULL,   initialize = NULL,   public = NULL,   private = NULL,   active = NULL,   parent_env = parent.frame(),   inherit = CallbackSet,   lock_objects = FALSE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Set of Callbacks for Torch — callback_set","text":"classname (character(1)) class name. on_begin, on_end, on_epoch_begin, on_before_valid, on_epoch_end, on_batch_begin, on_batch_end, on_after_backward, on_batch_valid_begin, on_batch_valid_end, on_valid_end, on_exit (function) Function execute given stage, see section Stages. state_dict (function()) function retrieves state dict callback. available learner training. load_state_dict (function(state_dict)) Function loads callback state. initialize (function()) initialization method callback. public, private, active (list()) Additional public, private, active fields add callback. parent_env (environment()) parent environment R6Class. inherit (R6ClassGenerator) class inherit. class must either CallbackSet (default) inherit . lock_objects (logical(1)) Whether lock objects resulting R6Class. FALSE (default), values can freely assigned self without declaring class definition.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/callback_set.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Set of Callbacks for Torch — callback_set","text":"CallbackSet","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/equals-.lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Compare lazy tensors — ==.lazy_tensor","title":"Compare lazy tensors — ==.lazy_tensor","text":"Compares lazy tensors using indices data descriptor's hash. means two lazy_tensors: equal: mateterialize tensors. unequal: might materialize tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/equals-.lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Compare lazy tensors — ==.lazy_tensor","text":"","code":"# S3 method for class 'lazy_tensor' x == y"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/equals-.lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Compare lazy tensors — ==.lazy_tensor","text":"x, y (lazy_tensor) Values compare.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/is_lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Check for lazy tensor — is_lazy_tensor","title":"Check for lazy tensor — is_lazy_tensor","text":"Checks whether object lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/is_lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Check for lazy tensor — is_lazy_tensor","text":"","code":"is_lazy_tensor(x)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/is_lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Check for lazy tensor — is_lazy_tensor","text":"x () Object check.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a lazy tensor — lazy_tensor","title":"Create a lazy tensor — lazy_tensor","text":"Create lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a lazy tensor — lazy_tensor","text":"","code":"lazy_tensor(data_descriptor = NULL, ids = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a lazy tensor — lazy_tensor","text":"data_descriptor (DataDescriptor NULL) data descriptor NULL lazy tensor length 0. ids (integer()) elements data_descriptor included lazy tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/lazy_tensor.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a lazy tensor — lazy_tensor","text":"","code":"ds = dataset(\"example\",   initialize = function() self$iris = iris[, -5],   .getitem = function(i) list(x = torch_tensor(as.numeric(self$iris[i, ]))),   .length = function() nrow(self$iris) )() dd = as_data_descriptor(ds, list(x = c(NA, 4L))) lt = as_lazy_tensor(dd)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":null,"dir":"Reference","previous_headings":"","what":"Materialize Lazy Tensor Columns — materialize","title":"Materialize Lazy Tensor Columns — materialize","text":"materialize lazy_tensor() data.frame() / list() containing – among things – lazy_tensor() columns. .e. data described underlying DataDescriptors loaded indices lazy_tensor(), preprocessed put unto specified device. elements lazy tensor must shape, list tensors returned default. elements shape, tensors can also rbinded single tensor (parameter rbind).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Materialize Lazy Tensor Columns — materialize","text":"","code":"materialize(x, device = \"cpu\", rbind = FALSE, ...)  # S3 method for class 'list' materialize(x, device = \"cpu\", rbind = FALSE, cache = \"auto\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Materialize Lazy Tensor Columns — materialize","text":"x () object materialize. Either lazy_tensor list() / data.frame() containing lazy_tensor columns. device (character(1)) torch device. rbind (logical(1)) Whether rbind lazy tensor columns (TRUE) return list tensors (FALSE). second case, batch dimension. ... () Additional arguments. cache (character(1) environment() NULL) Optional cache (intermediate) materialization results. Per default, caching enabled dataset data descriptor (different output pointer) used one lazy tensor column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Materialize Lazy Tensor Columns — materialize","text":"(list() lazy_tensors lazy_tensor)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Materialize Lazy Tensor Columns — materialize","text":"Materializing lazy tensor consists : Loading data internal dataset DataDescriptor. Processing batches preprocessing Graphs. Returning result PipeOp pointed DataDescriptor (pointer). multiple lazy_tensor columns can benefit caching : ) Output(s) dataset might input multiple graphs. b) Different lazy tensors might outputs graph. reason possible provide cache environment. hash key ) hash indices dataset. hash key b) hash indices, dataset preprocessing graph.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Materialize Lazy Tensor Columns — materialize","text":"","code":"lt1 = as_lazy_tensor(torch_randn(10, 3)) materialize(lt1, rbind = TRUE) #> torch_tensor #> -1.3420 -0.7548 -0.2027 #>  0.8642 -0.6203 -1.1163 #> -0.1052  0.5600 -0.3974 #> -0.3943 -0.0705 -1.7041 #>  1.5337  0.1588 -0.8624 #>  1.4815 -0.1104 -0.3846 #>  2.3301  1.1064  0.2867 #> -1.1371 -1.4983  0.6197 #>  0.3968  1.8850  0.5676 #>  1.4147  1.3346 -1.3396 #> [ CPUFloatType{10,3} ] materialize(lt1, rbind = FALSE) #> [[1]] #> torch_tensor #> -1.3420 #> -0.7548 #> -0.2027 #> [ CPUFloatType{3} ] #>  #> [[2]] #> torch_tensor #>  0.8642 #> -0.6203 #> -1.1163 #> [ CPUFloatType{3} ] #>  #> [[3]] #> torch_tensor #> -0.1052 #>  0.5600 #> -0.3974 #> [ CPUFloatType{3} ] #>  #> [[4]] #> torch_tensor #> -0.3943 #> -0.0705 #> -1.7041 #> [ CPUFloatType{3} ] #>  #> [[5]] #> torch_tensor #>  1.5337 #>  0.1588 #> -0.8624 #> [ CPUFloatType{3} ] #>  #> [[6]] #> torch_tensor #>  1.4815 #> -0.1104 #> -0.3846 #> [ CPUFloatType{3} ] #>  #> [[7]] #> torch_tensor #>  2.3301 #>  1.1064 #>  0.2867 #> [ CPUFloatType{3} ] #>  #> [[8]] #> torch_tensor #> -1.1371 #> -1.4983 #>  0.6197 #> [ CPUFloatType{3} ] #>  #> [[9]] #> torch_tensor #>  0.3968 #>  1.8850 #>  0.5676 #> [ CPUFloatType{3} ] #>  #> [[10]] #> torch_tensor #>  1.4147 #>  1.3346 #> -1.3396 #> [ CPUFloatType{3} ] #>  lt2 = as_lazy_tensor(torch_randn(10, 4)) d = data.table::data.table(lt1 = lt1, lt2 = lt2) materialize(d, rbind = TRUE) #> $lt1 #> torch_tensor #> -1.3420 -0.7548 -0.2027 #>  0.8642 -0.6203 -1.1163 #> -0.1052  0.5600 -0.3974 #> -0.3943 -0.0705 -1.7041 #>  1.5337  0.1588 -0.8624 #>  1.4815 -0.1104 -0.3846 #>  2.3301  1.1064  0.2867 #> -1.1371 -1.4983  0.6197 #>  0.3968  1.8850  0.5676 #>  1.4147  1.3346 -1.3396 #> [ CPUFloatType{10,3} ] #>  #> $lt2 #> torch_tensor #> -0.1006  1.3168  0.1388  0.5232 #>  0.2336 -0.0717 -1.0036  0.6683 #>  0.7921  1.4544  0.1958 -0.2346 #>  0.6385 -0.0318  0.6482 -0.1380 #>  1.3797  0.7916 -0.2373 -1.0011 #>  0.3849 -0.2163 -0.0237 -0.0863 #>  0.4032  0.1983 -1.6133 -2.2197 #>  0.2452  0.7073  0.3545  0.0833 #> -0.8829  0.2502 -0.3007 -0.1676 #>  1.9140 -0.7256 -0.7844 -1.4341 #> [ CPUFloatType{10,4} ] #>  materialize(d, rbind = FALSE) #> $lt1 #> $lt1[[1]] #> torch_tensor #> -1.3420 #> -0.7548 #> -0.2027 #> [ CPUFloatType{3} ] #>  #> $lt1[[2]] #> torch_tensor #>  0.8642 #> -0.6203 #> -1.1163 #> [ CPUFloatType{3} ] #>  #> $lt1[[3]] #> torch_tensor #> -0.1052 #>  0.5600 #> -0.3974 #> [ CPUFloatType{3} ] #>  #> $lt1[[4]] #> torch_tensor #> -0.3943 #> -0.0705 #> -1.7041 #> [ CPUFloatType{3} ] #>  #> $lt1[[5]] #> torch_tensor #>  1.5337 #>  0.1588 #> -0.8624 #> [ CPUFloatType{3} ] #>  #> $lt1[[6]] #> torch_tensor #>  1.4815 #> -0.1104 #> -0.3846 #> [ CPUFloatType{3} ] #>  #> $lt1[[7]] #> torch_tensor #>  2.3301 #>  1.1064 #>  0.2867 #> [ CPUFloatType{3} ] #>  #> $lt1[[8]] #> torch_tensor #> -1.1371 #> -1.4983 #>  0.6197 #> [ CPUFloatType{3} ] #>  #> $lt1[[9]] #> torch_tensor #>  0.3968 #>  1.8850 #>  0.5676 #> [ CPUFloatType{3} ] #>  #> $lt1[[10]] #> torch_tensor #>  1.4147 #>  1.3346 #> -1.3396 #> [ CPUFloatType{3} ] #>  #>  #> $lt2 #> $lt2[[1]] #> torch_tensor #> -0.1006 #>  1.3168 #>  0.1388 #>  0.5232 #> [ CPUFloatType{4} ] #>  #> $lt2[[2]] #> torch_tensor #>  0.2336 #> -0.0717 #> -1.0036 #>  0.6683 #> [ CPUFloatType{4} ] #>  #> $lt2[[3]] #> torch_tensor #>  0.7921 #>  1.4544 #>  0.1958 #> -0.2346 #> [ CPUFloatType{4} ] #>  #> $lt2[[4]] #> torch_tensor #>  0.6385 #> -0.0318 #>  0.6482 #> -0.1380 #> [ CPUFloatType{4} ] #>  #> $lt2[[5]] #> torch_tensor #>  1.3797 #>  0.7916 #> -0.2373 #> -1.0011 #> [ CPUFloatType{4} ] #>  #> $lt2[[6]] #> torch_tensor #>  0.3849 #> -0.2163 #> -0.0237 #> -0.0863 #> [ CPUFloatType{4} ] #>  #> $lt2[[7]] #> torch_tensor #>  0.4032 #>  0.1983 #> -1.6133 #> -2.2197 #> [ CPUFloatType{4} ] #>  #> $lt2[[8]] #> torch_tensor #>  0.2452 #>  0.7073 #>  0.3545 #>  0.0833 #> [ CPUFloatType{4} ] #>  #> $lt2[[9]] #> torch_tensor #> -0.8829 #>  0.2502 #> -0.3007 #> -0.1676 #> [ CPUFloatType{4} ] #>  #> $lt2[[10]] #> torch_tensor #>  1.9140 #> -0.7256 #> -0.7844 #> -1.4341 #> [ CPUFloatType{4} ] #>  #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":null,"dir":"Reference","previous_headings":"","what":"Materialize a Lazy Tensor — materialize_internal","title":"Materialize a Lazy Tensor — materialize_internal","text":"Convert lazy_tensor torch_tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Materialize a Lazy Tensor — materialize_internal","text":"","code":"materialize_internal(x, device = \"cpu\", cache = NULL, rbind)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Materialize a Lazy Tensor — materialize_internal","text":"x (lazy_tensor()) lazy tensor materialize. device (character(1L)) device put materialized tensor (running preprocessing graph). cache (NULL environment()) Whether cache (intermediate) results materialization. can make data loading faster multiple lazy_tensors reference dataset graph. rbind (logical(1)) Whtether rbind resulting tensors (TRUE) return list tensors (FALSE).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Materialize a Lazy Tensor — materialize_internal","text":"lazy_tensor()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/materialize_internal.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Materialize a Lazy Tensor — materialize_internal","text":"Materializing lazy tensor consists : Loading data internal dataset DataDescriptor. Processing batches preprocessing Graphs. Returning result PipeOp pointed DataDescriptor (pointer). materializing multiple lazy_tensor columns, caching can useful : ) Output(s) dataset might input multiple graphs. (task_dataset shoudl rarely case try merge ). b) Different lazy tensors might outputs graph. reason possible provide cache environment. hash key ) hash indices dataset. hash key b) hash indices dataset preprocessing graph.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch-package.html","id":null,"dir":"Reference","previous_headings":"","what":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","title":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","text":"Deep Learning library extends mlr3 framework building upon 'torch' package. allows conveniently build, train, evaluate deep learning models without worry low level details. Custom architectures can created using graph language defined 'mlr3pipelines'.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch-package.html","id":"options","dir":"Reference","previous_headings":"","what":"Options","title":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","text":"mlr3torch.cache: Whether cache downloaded data (TRUE) (FALSE, default). can also set specific folder file system used cache directory.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch-package.html","id":"author","dir":"Reference","previous_headings":"","what":"Author","title":"mlr3torch: Deep Learning with 'mlr3' — mlr3torch-package","text":"Maintainer: Sebastian Fischer sebf.fischer@gmail.com (ORCID) Authors: Martin Binder mlr.developer@mb706.com contributors: Bernd Bischl bernd_bischl@gmx.net (ORCID) [contributor] Lukas Burk github@quantenbrot.de (ORCID) [contributor] Florian Pfisterer pfistererf@googlemail.com (ORCID) [contributor] Carson Zhang carsonzhang4@gmail.com [contributor]","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":null,"dir":"Reference","previous_headings":"","what":"Dictionary of Torch Callbacks — mlr3torch_callbacks","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"mlr3misc::Dictionary torch callbacks. Use t_clbk() conveniently retrieve callbacks. Can converted data.table using .data.table.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"","code":"mlr3torch_callbacks"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"object class DictionaryMlr3torchCallbacks (inherits Dictionary, R6) length 13.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_callbacks.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Dictionary of Torch Callbacks — mlr3torch_callbacks","text":"","code":"mlr3torch_callbacks$get(\"checkpoint\") #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch # is the same as t_clbk(\"checkpoint\") #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch # convert to a data.table as.data.table(mlr3torch_callbacks) #> Key: <key> #>           key       label                 packages #>        <char>      <char>                   <char> #> 1: checkpoint  Checkpoint          mlr3torch,torch #> 2:    history     History          mlr3torch,torch #> 3:   progress    Progress progress,mlr3torch,torch #> 4:         tb TensorBoard tfevents,mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":null,"dir":"Reference","previous_headings":"","what":"Loss Functions — mlr3torch_losses","title":"Loss Functions — mlr3torch_losses","text":"Dictionary torch loss descriptors. See t_loss() conveniently retrieving loss function. Can converted data.table using .data.table.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Functions — mlr3torch_losses","text":"","code":"mlr3torch_losses"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Loss Functions — mlr3torch_losses","text":"object class DictionaryMlr3torchLosses (inherits Dictionary, R6) length 13.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"available-loss-functions","dir":"Reference","previous_headings":"","what":"Available Loss Functions","title":"Loss Functions — mlr3torch_losses","text":"cross_entropy, l1, mse","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_losses.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Loss Functions — mlr3torch_losses","text":"","code":"mlr3torch_losses$get(\"mse\") #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # is equivalent to t_loss(\"mse\") #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr # convert to a data.table as.data.table(mlr3torch_losses) #> Key: <key> #>              key              label task_types        packages #>           <char>             <char>     <list>          <char> #> 1: cross_entropy      Cross Entropy    classif torch,mlr3torch #> 2:            l1     Absolute Error       regr torch,mlr3torch #> 3:           mse Mean Squared Error       regr torch,mlr3torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":null,"dir":"Reference","previous_headings":"","what":"Optimizers — mlr3torch_optimizers","title":"Optimizers — mlr3torch_optimizers","text":"Dictionary torch optimizers. Use t_opt conveniently retrieving optimizers. Can converted data.table using .data.table.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizers — mlr3torch_optimizers","text":"","code":"mlr3torch_optimizers"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Optimizers — mlr3torch_optimizers","text":"object class DictionaryMlr3torchOptimizers (inherits Dictionary, R6) length 13.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"available-optimizers","dir":"Reference","previous_headings":"","what":"Available Optimizers","title":"Optimizers — mlr3torch_optimizers","text":"adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr3torch_optimizers.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Optimizers — mlr3torch_optimizers","text":"","code":"mlr3torch_optimizers$get(\"adam\") #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch # is equivalent to t_opt(\"adam\") #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch # convert to a data.table as.data.table(mlr3torch_optimizers) #> Key: <key> #>         key                         label        packages #>      <char>                        <char>          <char> #> 1: adadelta Adaptive Learning Rate Method torch,mlr3torch #> 2:  adagrad   Adaptive Gradient algorithm torch,mlr3torch #> 3:     adam    Adaptive Moment Estimation torch,mlr3torch #> 4:     asgd  SGD with Adaptive Batch Size torch,mlr3torch #> 5:  rmsprop  Root Mean Square Propagation torch,mlr3torch #> 6:    rprop     Resilient Backpropagation torch,mlr3torch #> 7:      sgd   Stochastic Gradient Descent torch,mlr3torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":null,"dir":"Reference","previous_headings":"","what":"Lazy Data Backend — mlr_backends_lazy","title":"Lazy Data Backend — mlr_backends_lazy","text":"lazy data backend wraps constructor lazily creates another backend, e.g. downloading (caching) data internet. backend used, metadata backend known advance accessible downloading actual data. backend first constructed, verified provided metadata correct, otherwise informative error message thrown. construction lazily constructed backend, calls like $data(), $missings(), $distinct(), $hash() redirected . Information available backend constructed : nrow - number rows (set length rownames). ncol - number columns (provided via id column col_info). colnames - column names. rownames - row names. col_info - column information, can obtained via mlr3::col_info(). Beware accessing backend's hash also contructs backend. Note cases data contains lazy_tensor columns, necessary naming class nothing lazy_tensor data type. Important constructor generates factor() variables important ordering levels data corresponds ordering levels col_info argument.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Lazy Data Backend — mlr_backends_lazy","text":"mlr3::DataBackend -> DataBackendLazy","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Lazy Data Backend — mlr_backends_lazy","text":"backend (DataBackend) wrapped backend lazily constructed first accessed. nrow (integer(1)) Number rows (observations). ncol (integer(1)) Number columns (variables), including primary key column. rownames (integer()) Returns vector distinct row identifiers, .e. contents primary key column. colnames (character()) Returns vector column names, including primary key column. is_constructed (logical(1)) Whether backend already constructed.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Lazy Data Backend — mlr_backends_lazy","text":"mlr3::DataBackend$format()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Lazy Data Backend — mlr_backends_lazy","text":"DataBackendLazy$new() DataBackendLazy$data() DataBackendLazy$head() DataBackendLazy$distinct() DataBackendLazy$missings() DataBackendLazy$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$new(constructor, rownames, col_info, primary_key)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"constructor (function) function argument backend (lazy backend), whose return value must actual backend. function called first time field $backend accessed. rownames (integer()) row names. Must permutation rownames lazily constructed backend. col_info (data.table::data.table()) data.table columns id, type levels containing column id, type levels. Note levels must provided correct order. primary_key (character(1)) Name primary key column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-data-","dir":"Reference","previous_headings":"","what":"Method data()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Returns slice data specified format. rows must addressed vector primary key values, columns must referred via column names. Queries rows matching row id queries columns matching column name silently ignored. Rows guaranteed returned order rows, columns may returned arbitrary order. Duplicated row ids result duplicated rows, duplicated column names lead exception. Accessing data triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$data(rows, cols)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"rows (integer()) Row indices. cols (character()) Column names.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-head-","dir":"Reference","previous_headings":"","what":"Method head()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Retrieve first n rows. triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$head(n = 6L)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"n (integer(1)) Number rows.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Lazy Data Backend — mlr_backends_lazy","text":"data.table::data.table() first n rows.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-distinct-","dir":"Reference","previous_headings":"","what":"Method distinct()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Returns named list vectors distinct values column specified. na_rm TRUE, missing values removed returned vectors distinct values. Non-existing rows columns silently ignored. triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$distinct(rows, cols, na_rm = TRUE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-3","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"rows (integer()) Row indices. cols (character()) Column names. na_rm (logical(1)) Whether remove NAs .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"returns-1","dir":"Reference","previous_headings":"","what":"Returns","title":"Lazy Data Backend — mlr_backends_lazy","text":"Named list() distinct values.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-missings-","dir":"Reference","previous_headings":"","what":"Method missings()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Returns number missing values per column specified slice data. Non-existing rows columns silently ignored. triggers construction backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$missings(rows, cols)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"arguments-4","dir":"Reference","previous_headings":"","what":"Arguments","title":"Lazy Data Backend — mlr_backends_lazy","text":"rows (integer()) Row indices. cols (character()) Column names.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"returns-2","dir":"Reference","previous_headings":"","what":"Returns","title":"Lazy Data Backend — mlr_backends_lazy","text":"Total missing values per column (named numeric()).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Lazy Data Backend — mlr_backends_lazy","text":"Printer.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"DataBackendLazy$print()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_backends_lazy.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Lazy Data Backend — mlr_backends_lazy","text":"","code":"# We first define a backend constructor constructor = function(backend) {   cat(\"Data is constructed!\\n\")   DataBackendDataTable$new(     data.table(x = rnorm(10), y = rnorm(10), row_id = 1:10),     primary_key = \"row_id\"   ) }  # to wrap this backend constructor in a lazy backend, we need to provide the correct metadata for it column_info = data.table(   id = c(\"x\", \"y\", \"row_id\"),   type = c(\"numeric\", \"numeric\", \"integer\"),   levels = list(NULL, NULL, NULL) ) backend_lazy = DataBackendLazy$new(   constructor = constructor,   rownames = 1:10,   col_info = column_info,   primary_key = \"row_id\" )  # Note that the constructor is not called for the calls below # as they can be read from the metadata backend_lazy$nrow #> [1] 10 backend_lazy$rownames #>  [1]  1  2  3  4  5  6  7  8  9 10 backend_lazy$ncol #> [1] 3 backend_lazy$colnames #> [1] \"x\"      \"y\"      \"row_id\" col_info(backend_lazy) #>        id    type levels #>    <char>  <char> <list> #> 1:      x numeric [NULL] #> 2:      y numeric [NULL] #> 3: row_id integer [NULL]  # Only now the backend is constructed backend_lazy$data(1, \"x\") #> Data is constructed! #>             x #>         <num> #> 1: -0.3872136 # Is the same as: backend_lazy$backend$data(1, \"x\") #>             x #>         <num> #> 1: -0.3872136"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":null,"dir":"Reference","previous_headings":"","what":"Checkpoint Callback — mlr_callback_set.checkpoint","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves optimizer network states training. final network optimizer always stored.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saving learner callback trained model impossible, model slot set last callback step executed.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"mlr3torch::CallbackSet -> CallbackSetCheckpoint","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"mlr3torch::CallbackSet$load_state_dict() mlr3torch::CallbackSet$print() mlr3torch::CallbackSet$state_dict()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"CallbackSetCheckpoint$new() CallbackSetCheckpoint$on_epoch_end() CallbackSetCheckpoint$on_batch_end() CallbackSetCheckpoint$on_exit() CallbackSetCheckpoint$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$new(path, freq, freq_type = \"epoch\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"path (character(1)) path folder models saved. freq (integer(1)) frequency often model saved. Frequency either per step epoch, can configured freq_type parameter. freq_type (character(1)) Can either \"epoch\" (default) \"step\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves network optimizer state dict. nothing freq_type freq met.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-on-batch-end-","dir":"Reference","previous_headings":"","what":"Method on_batch_end()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves selected objects defined save. nothing freq_type freq met.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$on_batch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-on-exit-","dir":"Reference","previous_headings":"","what":"Method on_exit()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"Saves learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$on_exit()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"","code":"CallbackSetCheckpoint$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.checkpoint.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Checkpoint Callback — mlr_callback_set.checkpoint","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":null,"dir":"Reference","previous_headings":"","what":"History Callback — mlr_callback_set.history","title":"History Callback — mlr_callback_set.history","text":"Saves training validation history training. history saved data.table $train $valid slots. first column always epoch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"History Callback — mlr_callback_set.history","text":"mlr3torch::CallbackSet -> CallbackSetHistory","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"History Callback — mlr_callback_set.history","text":"mlr3torch::CallbackSet$print()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"History Callback — mlr_callback_set.history","text":"CallbackSetHistory$on_begin() CallbackSetHistory$state_dict() CallbackSetHistory$load_state_dict() CallbackSetHistory$on_before_valid() CallbackSetHistory$on_epoch_end() CallbackSetHistory$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-on-begin-","dir":"Reference","previous_headings":"","what":"Method on_begin()","title":"History Callback — mlr_callback_set.history","text":"Initializes lists train validation metrics stored.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$on_begin()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-state-dict-","dir":"Reference","previous_headings":"","what":"Method state_dict()","title":"History Callback — mlr_callback_set.history","text":"Converts lists data.tables.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$state_dict()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-load-state-dict-","dir":"Reference","previous_headings":"","what":"Method load_state_dict()","title":"History Callback — mlr_callback_set.history","text":"Sets field $train $valid contained state dict.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$load_state_dict(state_dict)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"History Callback — mlr_callback_set.history","text":"state_dict (callback_state_history) state dict retrieved via $state_dict().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-on-before-valid-","dir":"Reference","previous_headings":"","what":"Method on_before_valid()","title":"History Callback — mlr_callback_set.history","text":"Add latest training scores history.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$on_before_valid()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"History Callback — mlr_callback_set.history","text":"Add latest validation scores history.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"History Callback — mlr_callback_set.history","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"History Callback — mlr_callback_set.history","text":"","code":"CallbackSetHistory$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.history.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"History Callback — mlr_callback_set.history","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Callbacks — mlr_callback_set","title":"Base Class for Callbacks — mlr_callback_set","text":"Base class callbacks inherit (see section Inheriting). callback set collection functions executed different stages training loop. can used gain control training process neural network without write everything scratch. used torch learner, CallbackSet wrapped TorchCallback. latters parameter set represents arguments CallbackSet's $initialize() method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Callbacks — mlr_callback_set","text":"available stage (see section Stages) public method $on_<stage>() can defined. evaluation context (ContextTorch) can accessed via self$ctx, contains current state training loop. context assigned beginning training loop removed afterwards. Different stages callback can communicate assigning values $self. State: able store information $model slot LearnerTorch, callbacks support state API. can overload $state_dict() public method define stored learner$model$callbacks$<id> training finishes. also requires implement $load_state_dict(state_dict) method defines load previously saved callback state different callback. Note $state_dict() include parameter values used initialize callback. creating custom callbacks, function torch_callback() recommended, creates CallbackSet wraps TorchCallback. create CallbackSet convenience function callback_set() can used. functions perform checks stages accidentally misspelled.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"stages","dir":"Reference","previous_headings":"","what":"Stages","title":"Base Class for Callbacks — mlr_callback_set","text":"begin :: Run training loop begins. epoch_begin :: Run beginning epoch. batch_begin :: Run forward call. after_backward :: Run backward call. batch_end :: Run optimizer step. batch_valid_begin :: Run forward call validation loop. batch_valid_end :: Run forward call validation loop. valid_end :: Run end validation. epoch_end :: Run end epoch. end :: Run last epoch. exit :: Run last, using .exit().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"terminate-training","dir":"Reference","previous_headings":"","what":"Terminate Training","title":"Base Class for Callbacks — mlr_callback_set","text":"training stopped, possible set field $terminate ContextTorch. end every epoch field checked TRUE, training stops. can example used implement custom early stopping.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Base Class for Callbacks — mlr_callback_set","text":"ctx (ContextTorch NULL) evaluation context callback. field always NULL except $train() call torch learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Callbacks — mlr_callback_set","text":"stages (character()) active stages callback set.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Callbacks — mlr_callback_set","text":"CallbackSet$print() CallbackSet$state_dict() CallbackSet$load_state_dict() CallbackSet$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Base Class for Callbacks — mlr_callback_set","text":"Prints object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Callbacks — mlr_callback_set","text":"... () Currently unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-state-dict-","dir":"Reference","previous_headings":"","what":"Method state_dict()","title":"Base Class for Callbacks — mlr_callback_set","text":"Returns information kept LearnerTorch's state training. information loadable callback using $load_state_dict() able continue training. returns NULL default.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$state_dict()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-load-state-dict-","dir":"Reference","previous_headings":"","what":"Method load_state_dict()","title":"Base Class for Callbacks — mlr_callback_set","text":"Loads state dict callback continue training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$load_state_dict(state_dict)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Callbacks — mlr_callback_set","text":"state_dict () state dict retrieved via $state_dict().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Callbacks — mlr_callback_set","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Callbacks — mlr_callback_set","text":"","code":"CallbackSet$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Callbacks — mlr_callback_set","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":null,"dir":"Reference","previous_headings":"","what":"Progress Callback — mlr_callback_set.progress","title":"Progress Callback — mlr_callback_set.progress","text":"Prints progress bar metrics training validation.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Progress Callback — mlr_callback_set.progress","text":"mlr3torch::CallbackSet -> CallbackSetProgress","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Progress Callback — mlr_callback_set.progress","text":"mlr3torch::CallbackSet$load_state_dict() mlr3torch::CallbackSet$print() mlr3torch::CallbackSet$state_dict()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Progress Callback — mlr_callback_set.progress","text":"CallbackSetProgress$on_epoch_begin() CallbackSetProgress$on_batch_end() CallbackSetProgress$on_before_valid() CallbackSetProgress$on_batch_valid_end() CallbackSetProgress$on_epoch_end() CallbackSetProgress$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-epoch-begin-","dir":"Reference","previous_headings":"","what":"Method on_epoch_begin()","title":"Progress Callback — mlr_callback_set.progress","text":"Initializes progress bar training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_epoch_begin()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-batch-end-","dir":"Reference","previous_headings":"","what":"Method on_batch_end()","title":"Progress Callback — mlr_callback_set.progress","text":"Increments training progress bar.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_batch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-before-valid-","dir":"Reference","previous_headings":"","what":"Method on_before_valid()","title":"Progress Callback — mlr_callback_set.progress","text":"Creates progress bar validation.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_before_valid()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-batch-valid-end-","dir":"Reference","previous_headings":"","what":"Method on_batch_valid_end()","title":"Progress Callback — mlr_callback_set.progress","text":"Increments validation progress bar.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_batch_valid_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"Progress Callback — mlr_callback_set.progress","text":"Prints summary training validation process.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Progress Callback — mlr_callback_set.progress","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"Progress Callback — mlr_callback_set.progress","text":"","code":"CallbackSetProgress$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.progress.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Progress Callback — mlr_callback_set.progress","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":null,"dir":"Reference","previous_headings":"","what":"TensorBoard Logging Callback — mlr_callback_set.tb","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Logs training loss, training measures, validation measures events. view , use TensorBoard tensorflow::tensorboard() (requires tensorflow) CLI.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Logs events every epoch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"mlr3torch::CallbackSet -> CallbackSetTB","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"mlr3torch::CallbackSet$load_state_dict() mlr3torch::CallbackSet$print() mlr3torch::CallbackSet$state_dict()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"CallbackSetTB$new() CallbackSetTB$on_epoch_end() CallbackSetTB$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"","code":"CallbackSetTB$new(path, log_train_loss)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"path (character(1)) path folder events logged. Point TensorBoard folder view . log_train_loss (logical(1)) Whether log training loss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"method-on-epoch-end-","dir":"Reference","previous_headings":"","what":"Method on_epoch_end()","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"Logs training loss, training measures, validation measures TensorBoard events.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"","code":"CallbackSetTB$on_epoch_end()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"","code":"CallbackSetTB$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_callback_set.tb.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"TensorBoard Logging Callback — mlr_callback_set.tb","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Context for Torch Learner — mlr_context_torch","title":"Context for Torch Learner — mlr_context_torch","text":"Context training torch learner. - mostly read-- information callbacks access argument ctx. information callbacks, see CallbackSet.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Context for Torch Learner — mlr_context_torch","text":"learner (Learner) torch learner. task_train (Task) training task. task_valid (Task NULL) validation task. loader_train (torch::dataloader) data loader training. loader_valid (torch::dataloader) data loader validation. measures_train (list() Measures) Measures used training. measures_valid (list() Measures) Measures used validation. network (torch::nn_module) torch network. optimizer (torch::optimizer) optimizer. loss_fn (torch::nn_module) loss function. total_epochs (integer(1)) total number epochs learner trained . last_scores_train (named list() NULL) scores last training batch. Names ids training measures. LearnerTorch sets eval_freq different 1, NULL epochs evaluate model. last_scores_valid (list()) scores last validation batch. Names ids validation measures. LearnerTorch sets eval_freq different 1, NULL epochs evaluate model. epoch (integer(1)) current epoch. step (integer(1)) current iteration. prediction_encoder (function()) learner's prediction encoder. batch (named list() torch_tensors) current batch. terminate (logical(1)) field set TRUE end epoch, training stops.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Context for Torch Learner — mlr_context_torch","text":"ContextTorch$new() ContextTorch$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Context for Torch Learner — mlr_context_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Context for Torch Learner — mlr_context_torch","text":"","code":"ContextTorch$new(   learner,   task_train,   task_valid = NULL,   loader_train,   loader_valid = NULL,   measures_train = NULL,   measures_valid = NULL,   network,   optimizer,   loss_fn,   total_epochs,   prediction_encoder,   eval_freq = 1L )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Context for Torch Learner — mlr_context_torch","text":"learner (Learner) torch learner. task_train (Task) training task. task_valid (Task NULL) validation task. loader_train (torch::dataloader) data loader training. loader_valid (torch::dataloader NULL) data loader validation. measures_train (list() Measures NULL) Measures used training. Default NULL. measures_valid (list() Measures NULL) Measures used validation. network (torch::nn_module) torch network. optimizer (torch::optimizer) optimizer. loss_fn (torch::nn_module) loss function. total_epochs (integer(1)) total number epochs learner trained . prediction_encoder (function()) learner's prediction encoder. eval_freq (integer(1)) evaluation frequency.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Context for Torch Learner — mlr_context_torch","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Context for Torch Learner — mlr_context_torch","text":"","code":"ContextTorch$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_context_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Context for Torch Learner — mlr_context_torch","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":null,"dir":"Reference","previous_headings":"","what":"My Little Pony — mlr_learners.mlp","title":"My Little Pony — mlr_learners.mlp","text":"Fully connected feed forward network dropout activation function. features can either single lazy_tensor one numeric columns ().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"dictionary","dir":"Reference","previous_headings":"","what":"Dictionary","title":"My Little Pony — mlr_learners.mlp","text":"Learner can instantiated using sugar function lrn():","code":"lrn(\"classif.mlp\", ...) lrn(\"regr.mlp\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"My Little Pony — mlr_learners.mlp","text":"Supported task types: 'classif', 'regr' Predict Types: classif: 'response', 'prob' regr: 'response' Feature Types: “integer”, “numeric”, “lazy_tensor” Required Packages: mlr3, mlr3torch, torch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"My Little Pony — mlr_learners.mlp","text":"Parameters LearnerTorch, well : activation :: [nn_module] activation function. initialized nn_relu. activation_args :: named list() named list initialization arguments activation function. intialized empty list. neurons :: integer() number neurons per hidden layer. default hidden layer. Setting c(10, 20) first hidden layer 10 neurons second 20. p :: numeric(1) dropout probability. initialized 0.5. shape :: integer() NULL input shape length 2, e.g. c(NA, 5). needs present lazy tensor input unknown shape (NULL). Otherwise input shape inferred number numeric features.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"My Little Pony — mlr_learners.mlp","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchMLP","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"My Little Pony — mlr_learners.mlp","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"My Little Pony — mlr_learners.mlp","text":"LearnerTorchMLP$new() LearnerTorchMLP$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"My Little Pony — mlr_learners.mlp","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"My Little Pony — mlr_learners.mlp","text":"","code":"LearnerTorchMLP$new(   task_type,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"My Little Pony — mlr_learners.mlp","text":"task_type (character(1)) task type, either \"classif\" \"regr\". optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"My Little Pony — mlr_learners.mlp","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"My Little Pony — mlr_learners.mlp","text":"","code":"LearnerTorchMLP$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"My Little Pony — mlr_learners.mlp","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.mlp.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"My Little Pony — mlr_learners.mlp","text":"","code":"# Define the Learner and set parameter values learner = lrn(\"classif.mlp\") learner$param_set$set_values(   epochs = 1, batch_size = 16, device = \"cpu\",   neurons = 10 )  # Define a Task task = tsk(\"iris\")  # Create train and test set ids = partition(task)  # Train the learner on the training ids learner$train(task, row_ids = ids$train)  # Make predictions for the test rows predictions = learner$predict(task, row_ids = ids$test)  # Score the predictions predictions$score() #> classif.ce  #>       0.68"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":null,"dir":"Reference","previous_headings":"","what":"Tabular ResNet — mlr_learners.tab_resnet","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Tabular resnet.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"dictionary","dir":"Reference","previous_headings":"","what":"Dictionary","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Learner can instantiated using sugar function lrn():","code":"lrn(\"classif.tab_resnet\", ...) lrn(\"regr.tab_resnet\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Supported task types: 'classif', 'regr' Predict Types: classif: 'response', 'prob' regr: 'response' Feature Types: “integer”, “numeric” Required Packages: mlr3, mlr3torch, torch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Parameters LearnerTorch, well : n_blocks :: integer(1) number blocks. d_block :: integer(1) input output dimension block. d_hidden :: integer(1) latent dimension block. d_hidden_multiplier :: integer(1) Alternative way specify latent dimension d_block * d_hidden_multiplier. dropout1 :: numeric(1) First dropout ratio. dropout2 :: numeric(1) Second dropout ratio.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Gorishniy Y, Rubachev , Khrulkov V, Babenko (2021). “Revisiting Deep Learning  Tabular Data.” arXiv, 2106.11959.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchTabResNet","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"LearnerTorchTabResNet$new() LearnerTorchTabResNet$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"","code":"LearnerTorchTabResNet$new(   task_type,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"task_type (character(1)) task type, either \"classif\" \"regr\". optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"","code":"LearnerTorchTabResNet$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.tab_resnet.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tabular ResNet — mlr_learners.tab_resnet","text":"","code":"# Define the Learner and set parameter values learner = lrn(\"classif.tab_resnet\") learner$param_set$set_values(   epochs = 1, batch_size = 16, device = \"cpu\",   n_blocks = 2, d_block = 10, d_hidden = 20, dropout1 = 0.3, dropout2 = 0.3 )  # Define a Task task = tsk(\"iris\")  # Create train and test set ids = partition(task)  # Train the learner on the training ids learner$train(task, row_ids = ids$train)  # Make predictions for the test rows predictions = learner$predict(task, row_ids = ids$test)  # Score the predictions predictions$score() #> classif.ce  #>        0.9"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":null,"dir":"Reference","previous_headings":"","what":"Featureless Torch Learner — mlr_learners.torch_featureless","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Featureless torch learner. Output constant weight learned training. classification, (asymptoptically) result majority class prediction using standard cross-entropy loss. regression, result median L1 loss mean L2 loss.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"dictionary","dir":"Reference","previous_headings":"","what":"Dictionary","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Learner can instantiated using sugar function lrn():","code":"lrn(\"classif.torch_featureless\", ...) lrn(\"regr.torch_featureless\", ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Supported task types: 'classif', 'regr' Predict Types: classif: 'response', 'prob' regr: 'response' Feature Types: “logical”, “integer”, “numeric”, “character”, “factor”, “ordered”, “POSIXct”, “lazy_tensor” Required Packages: mlr3, mlr3torch, torch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"LearnerTorch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchFeatureless","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"LearnerTorchFeatureless$new() LearnerTorchFeatureless$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"","code":"LearnerTorchFeatureless$new(   task_type,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"task_type (character(1)) task type, either \"classif\" \"regr\". optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"","code":"LearnerTorchFeatureless$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torch_featureless.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Featureless Torch Learner — mlr_learners.torch_featureless","text":"","code":"# Define the Learner and set parameter values learner = lrn(\"classif.torch_featureless\") learner$param_set$set_values(   epochs = 1, batch_size = 16, device = \"cpu\"    )  # Define a Task task = tsk(\"iris\")  # Create train and test set ids = partition(task)  # Train the learner on the training ids learner$train(task, row_ids = ids$train)  # Make predictions for the test rows predictions = learner$predict(task, row_ids = ids$test)  # Score the predictions predictions$score() #> classif.ce  #>       0.64"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":null,"dir":"Reference","previous_headings":"","what":"AlexNet Image Classifier — mlr_learners.torchvision","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Classic image classification networks torchvision.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Parameters LearnerTorchImage pretrained :: logical(1) Whether use pretrained model. final linear layer replaced new nn_linear number classes inferred Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Supported task types: \"classif\" Predict Types: \"response\" \"prob\" Feature Types: \"lazy_tensor\" Required packages: \"mlr3torch\", \"torch\", \"torchvision\"","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> mlr3torch::LearnerTorchImage -> LearnerTorchVision","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"LearnerTorchVision$new() LearnerTorchVision$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"","code":"LearnerTorchVision$new(   name,   module_generator,   label,   optimizer = NULL,   loss = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"name (character(1)) name network. module_generator (function(pretrained, num_classes)) Function generates network. label (character(1)) label network. #' @references Krizhevsky, Alex, Sutskever, Ilya, Hinton, E. G (2017). “Imagenet classification deep convolutional neural networks.” Communications ACM, 60(6), 84–90. Sandler, Mark, Howard, Andrew, Zhu, Menglong, Zhmoginov, Andrey, Chen, Liang-Chieh (2018). “Mobilenetv2: Inverted residuals linear bottlenecks.” Proceedings IEEE conference computer vision pattern recognition, 4510–4520. , Kaiming, Zhang, Xiangyu, Ren, Shaoqing, Sun, Jian (2016). “Deep residual learning image recognition.” Proceedings IEEE conference computer vision pattern recognition, 770–778. Simonyan, Karen, Zisserman, Andrew (2014). “deep convolutional networks large-scale image recognition.” arXiv preprint arXiv:1409.1556. optimizer (TorchOptimizer) optimizer use training. Per default, adam used. loss (TorchLoss) loss used train network. Per default, mse used regression cross_entropy classification. callbacks (list() TorchCallbacks) callbacks. Must unique ids.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"","code":"LearnerTorchVision$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners.torchvision.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"AlexNet Image Classifier — mlr_learners.torchvision","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Torch Learners — mlr_learners_torch","title":"Base Class for Torch Learners — mlr_learners_torch","text":"base class provides basic functionality training prediction neural network. torch learners inherit class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"validation","dir":"Reference","previous_headings":"","what":"Validation","title":"Base Class for Torch Learners — mlr_learners_torch","text":"specify validation data, can set $validate field Learner, can set : NULL: validation ratio: proportion 1 - ratio task used training ratio used validation. \"test\" means \"test\" task resampling used possible calling $train() manually. \"predefined\": use predefined $internal_valid_task mlr3::Task. validation data can also used early stopping, see description Learner's parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"saving-a-learner","dir":"Reference","previous_headings":"","what":"Saving a Learner","title":"Base Class for Torch Learners — mlr_learners_torch","text":"order save LearnerTorch later usage, necessary call $marshal() method Learner writing disk, object otherwise saved correctly. loading marshaled LearnerTorch R , need call $unmarshal() transform useable state.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"early-stopping-and-tuning","dir":"Reference","previous_headings":"","what":"Early Stopping and Tuning","title":"Base Class for Torch Learners — mlr_learners_torch","text":"order prevent overfitting, LearnerTorch class allows use early stopping via patience min_delta parameters, see Learner's parameters. tuning LearnerTorch also possible combine explicit tuning via mlr3tuning LearnerTorch's internal tuning epochs via early stopping. , just need include epochs = to_tune(upper = <upper>, internal = TRUE) search space, <upper> maximally allowed number epochs, configure early stopping.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"model","dir":"Reference","previous_headings":"","what":"Model","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Model list class \"learner_torch_model\" following elements: network :: trained network. optimizer :: $state_dict() optimizer used train network. loss_fn :: $state_dict() loss used train network. callbacks :: callbacks used train network. seed :: seed / used training prediction. epochs :: many epochs model trained (early stopping). task_col_info :: data.table() containing information train-task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Torch Learners — mlr_learners_torch","text":"General: parameters optimizer, loss callbacks, prefixed \"opt.\", \"loss.\" \"cb.<callback id>.\" respectively, well : epochs :: integer(1) number epochs. device :: character(1) device. One \"auto\", \"cpu\", \"cuda\" values defined mlr_reflections$torch$devices. value initialized \"auto\", select \"cuda\" possible, try \"mps\" otherwise fall back \"cpu\". num_threads :: integer(1) number threads intraop pararallelization (device \"cpu\"). value initialized 1. num_interop_threads :: integer(1) number threads intraop interop pararallelization (device \"cpu\"). value initialized 1. Note can set session changing value within R session raise warning. seed :: integer(1) \"random\" NULL torch seed used training prediction. value initialized \"random\", means random seed sampled beginning training phase. seed (either set randomly sampled) available via $model$seed training used prediction. Note setting seed training phase mean default (.e. seed \"random\"), clones learner use different seed. set NULL, seeding done. Evaluation: measures_train :: Measure list() Measures. Measures evaluated training. measures_valid :: Measure list() Measures. Measures evaluated validation. eval_freq :: integer(1) often train / validation predictions evaluated using measures_train / measures_valid. initialized 1. Note final model always evaluated. Early Stopping: patience :: integer(1) activates early stopping using validation scores. performance model improve patience evaluation steps, training ended. Note final model stored learner, best model. initialized 0, means early stopping. first entry measures_valid used metric. also requires specify $validate field Learner, well measures_valid. min_delta :: double(1) minimum improvement threshold (>) early stopping. initialized 0. Dataloader: batch_size :: integer(1) batch size (required). shuffle :: logical(1) Whether shuffle instances dataset. Default FALSE. impact validation. sampler :: torch::sampler Object defines dataloader draw samples. batch_sampler :: torch::sampler Object defines dataloader draws batches. num_workers :: integer(1) number workers data loading (batches loaded parallel). default 0, means data loaded main process. collate_fn :: function merge list samples form batch. pin_memory :: logical(1) Whether dataloader copies tensors CUDA pinned memory returning . drop_last :: logical(1) Whether drop last training batch epoch training. Default FALSE. timeout :: numeric(1) timeout value collecting batch workers. Negative values mean timeout default -1. worker_init_fn :: function(id) function receives worker id ([1, num_workers]) exectued seeding worker data loading. worker_globals :: list() | character() loading data parallel, allows export globals workers. character vector, objects global environment names copied workers. worker_packages :: character() packages load workers. Also see torch::dataloder information.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Torch Learners — mlr_learners_torch","text":"seperate classes classification regression inherit . Instead, task_type must specified  construction argument. Currently, classification regression supported. inheriting class, one overload two private methods: .network(task, param_vals) (Task, list()) -> nn_module Construct torch::nn_module object given task parameter values, .e. neural network trained learner. classification, output network expected scores application final softmax layer. .dataset(task, param_vals) (Task, list()) -> torch::dataset Create dataset task. Must respect parameter value device. Moreover, one needs pay attention respect row ids provided task. also possible overwrite private .dataloader() method instead .dataset() method. Per default, dataloader constructed using output .dataset() method. However, respect dataloader parameters ParamSet. .dataloader(task, param_vals) (Task, list()) -> torch::dataloader Create dataloader task. Needs respect least batch_size shuffle (otherwise predictions can permuted). change predict types, private .encode_prediction() method can overwritten: .encode_prediction(predict_tensor, task, param_vals) (torch_tensor, Task, list()) -> list() Take raw predictions self$network (predict_tensor) encode format can converted valid mlr3 predictions using mlr3::as_prediction_data(). method must take self$predict_type account. possible add parameters specifying param_set construction argument, currently possible remove existing parameters, .e. listed section Parameters. None parameters provided param_set can id starts \"loss.\", \"opt.\", \"cb.\"`, preserved dynamically constructed parameters optimizer, loss function, callbacks. perform additional input checks task, private .verify_train_task(task, param_vals) .verify_predict_task(task, param_vals) can overwritten. learners construction arguments change hash learner, required implement private $.additional_phash_input().","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Base Class for Torch Learners — mlr_learners_torch","text":"mlr3::Learner -> LearnerTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Torch Learners — mlr_learners_torch","text":"validate construct internal validation data. parameter can either NULL, ratio $(0, 1)$, \"test\", \"predefined\". loss (TorchLoss) torch loss. optimizer (TorchOptimizer) torch optimizer. callbacks (list() TorchCallbacks) List torch callbacks. ids set names. internal_valid_scores Retrieves internal validation scores named list(). Specify $validate field measures_valid parameter configure . Returns NULL learner trained yet. internal_tuned_values early stopping activate, returns named list early-stopped epochs, otherwise empty list returned. Returns NULL learner trained yet. marshaled (logical(1)) Whether learner marshaled. network (nn_module()) Shortcut learner$model$network. param_set (ParamSet) parameter set hash (character(1)) Hash (unique identifier) object. phash (character(1)) Hash (unique identifier) partial object, excluding components varied systematically tuning (parameter values).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Base Class for Torch Learners — mlr_learners_torch","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Torch Learners — mlr_learners_torch","text":"LearnerTorch$new() LearnerTorch$format() LearnerTorch$print() LearnerTorch$marshal() LearnerTorch$unmarshal() LearnerTorch$dataset() LearnerTorch$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$new(   id,   task_type,   param_set,   properties,   man,   label,   feature_types,   optimizer = NULL,   loss = NULL,   packages = character(),   predict_types = NULL,   callbacks = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"id (character(1)) id new object. task_type (character(1)) task type. param_set (ParamSet alist()) Either parameter set, alist() containing different values self, e.g. alist(private$.param_set1, private$.param_set2), ParamSet collection created. properties (character()) properties object. See mlr_reflections$learner_properties available values. man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help(). label (character(1)) Label new instance. feature_types (character()) feature types. See mlr_reflections$task_feature_types available values, Additionally, \"lazy_tensor\" supported. optimizer (NULL TorchOptimizer) optimizer use training. Defaults adam. loss (NULL TorchLoss) loss use training. Defaults MSE regression cross entropy classification. packages (character()) R packages object depends . predict_types (character()) predict types. See mlr_reflections$learner_predict_types available values. regression, default \"response\". classification, defaults \"response\" \"prob\". deviate defaults, necessary overwrite private $.encode_prediction() method, see section Inheriting. callbacks (list() TorchCallbacks) callbacks use training. Defaults empty list(), .e. callbacks.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-format-","dir":"Reference","previous_headings":"","what":"Method format()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Helper print outputs.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$format(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... (ignored).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-print-","dir":"Reference","previous_headings":"","what":"Method print()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Prints object.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$print(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... () Currently unused.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-marshal-","dir":"Reference","previous_headings":"","what":"Method marshal()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Marshal learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-3","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$marshal(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-3","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... () Additional parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Learners — mlr_learners_torch","text":"self","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-unmarshal-","dir":"Reference","previous_headings":"","what":"Method unmarshal()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Unmarshal learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-4","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$unmarshal(...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-4","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"... () Additional parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"returns-1","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Learners — mlr_learners_torch","text":"self","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-dataset-","dir":"Reference","previous_headings":"","what":"Method dataset()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"Create dataset task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-5","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$dataset(task)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-5","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"task Task task","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"returns-2","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Learners — mlr_learners_torch","text":"dataset","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Torch Learners — mlr_learners_torch","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"usage-6","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Learners — mlr_learners_torch","text":"","code":"LearnerTorch$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch.html","id":"arguments-6","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Learners — mlr_learners_torch","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":null,"dir":"Reference","previous_headings":"","what":"Image Learner — mlr_learners_torch_image","title":"Image Learner — mlr_learners_torch_image","text":"Base Class Image Learners. features assumed single lazy_tensor column RGB format.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Image Learner — mlr_learners_torch_image","text":"Parameters include inherited LearnerTorch param_set construction argument.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Image Learner — mlr_learners_torch_image","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchImage","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Image Learner — mlr_learners_torch_image","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Image Learner — mlr_learners_torch_image","text":"LearnerTorchImage$new() LearnerTorchImage$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Image Learner — mlr_learners_torch_image","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Image Learner — mlr_learners_torch_image","text":"","code":"LearnerTorchImage$new(   id,   task_type,   param_set = ps(),   label,   optimizer = NULL,   loss = NULL,   callbacks = list(),   packages = c(\"torchvision\", \"magick\"),   man,   properties = NULL,   predict_types = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Image Learner — mlr_learners_torch_image","text":"id (character(1)) id new object. task_type (character(1)) task type. param_set (ParamSet) parameter set. label (character(1)) Label new instance. optimizer (TorchOptimizer) torch optimizer. loss (TorchLoss) loss use training. callbacks (list() TorchCallbacks) callbacks used training. Must unique ids. executed order provided packages (character()) R packages object depends . man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help(). properties (character()) properties object. See mlr_reflections$learner_properties available values. predict_types (character()) predict types. See mlr_reflections$learner_predict_types available values.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Image Learner — mlr_learners_torch_image","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Image Learner — mlr_learners_torch_image","text":"","code":"LearnerTorchImage$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_image.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Image Learner — mlr_learners_torch_image","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":null,"dir":"Reference","previous_headings":"","what":"Learner Torch Model — mlr_learners_torch_model","title":"Learner Torch Model — mlr_learners_torch_model","text":"Create torch learner instantiated nn_module(). classification, output network must scores (softmax).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Learner Torch Model — mlr_learners_torch_model","text":"See LearnerTorch","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Learner Torch Model — mlr_learners_torch_model","text":"mlr3::Learner -> mlr3torch::LearnerTorch -> LearnerTorchModel","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Learner Torch Model — mlr_learners_torch_model","text":"network_stored (nn_module NULL) network trained. calling $train(), NULL. ingress_tokens (named list() TorchIngressToken NULL) ingress tokens. Must non-NULL calling $train().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Learner Torch Model — mlr_learners_torch_model","text":"mlr3::Learner$base_learner() mlr3::Learner$encapsulate() mlr3::Learner$help() mlr3::Learner$predict() mlr3::Learner$predict_newdata() mlr3::Learner$reset() mlr3::Learner$train() mlr3torch::LearnerTorch$dataset() mlr3torch::LearnerTorch$format() mlr3torch::LearnerTorch$marshal() mlr3torch::LearnerTorch$print() mlr3torch::LearnerTorch$unmarshal()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Learner Torch Model — mlr_learners_torch_model","text":"LearnerTorchModel$new() LearnerTorchModel$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Learner Torch Model — mlr_learners_torch_model","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Learner Torch Model — mlr_learners_torch_model","text":"","code":"LearnerTorchModel$new(   network = NULL,   ingress_tokens = NULL,   task_type,   properties = NULL,   optimizer = NULL,   loss = NULL,   callbacks = list(),   packages = character(0),   feature_types = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Learner Torch Model — mlr_learners_torch_model","text":"network (nn_module) instantiated nn_module. cloned construction. classification, outputs must scores (softmax). ingress_tokens (list TorchIngressToken()) list ingress tokens defines dataloader defined. task_type (character(1)) task type. properties (NULL character()) properties learner. Defaults available properties given task type. optimizer (TorchOptimizer) torch optimizer. loss (TorchLoss) loss use training. callbacks (list() TorchCallbacks) callbacks used training. Must unique ids. executed order provided packages (character()) R packages object depends . feature_types (NULL character()) feature types. Defaults available feature types.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Learner Torch Model — mlr_learners_torch_model","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Learner Torch Model — mlr_learners_torch_model","text":"","code":"LearnerTorchModel$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Learner Torch Model — mlr_learners_torch_model","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_learners_torch_model.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Learner Torch Model — mlr_learners_torch_model","text":"","code":"# We show the learner using a classification task  # The iris task has 4 features and 3 classes network = nn_linear(4, 3) task = tsk(\"iris\")  # This defines the dataloader. # It loads all 4 features, which are also numeric. # The shape is (NA, 4) because the batch dimension is generally NA ingress_tokens = list(   input = TorchIngressToken(task$feature_names, batchgetter_num, c(NA, 4)) )  # Creating the learner and setting required parameters learner = lrn(\"classif.torch_model\",   network = network,   ingress_tokens = ingress_tokens,   batch_size = 16,   epochs = 1,   device = \"cpu\" )  # A simple train-predict ids = partition(task) learner$train(task, ids$train) learner$predict(task, ids$test) #> <PredictionClassif> for 50 observations: #>  row_ids     truth response #>        3    setosa   setosa #>        9    setosa   setosa #>       10    setosa   setosa #>      ---       ---      --- #>      147 virginica   setosa #>      149 virginica   setosa #>      150 virginica   setosa"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":null,"dir":"Reference","previous_headings":"","what":"Class for Torch Module Wrappers — mlr_pipeops_module","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"PipeOpModule wraps nn_module function called train phase mlr3pipelines::PipeOp. , allows assemble PipeOpModules computational mlr3pipelines::Graph represents either neural network preprocessing graph lazy_tensor. cases easier create network creating graph generates graph. cases easier create network creating structurally related graph consisting nodes class PipeOpTorchIngress PipeOpTorch. graph generate graph consisting PipeOpModules part ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"number names input output channels can set construction. input output \"torch_tensor\" training, NULL prediction prediction phase currently serves meaningful purpose.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"training, wrapped nn_module / function called provided inputs order channels defined. Arguments matched name.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"mlr3pipelines::PipeOp -> PipeOpModule","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"module (nn_module) torch module called training phase.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"PipeOpModule$new() PipeOpModule$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"","code":"PipeOpModule$new(   id = \"module\",   module = nn_identity(),   inname = \"input\",   outname = \"output\",   param_vals = list(),   packages = character(0) )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"id (character(1)) id new object. module (nn_module function()) torch module function wrapped. inname (character()) names input channels. outname (character()) names output channels. parameter length 1, parameter module must return tensor. Otherwise must return list() tensors corresponding length. param_vals (named list()) Parameter values set construction. packages (character()) R packages object depends .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"","code":"PipeOpModule$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_module.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Class for Torch Module Wrappers — mlr_pipeops_module","text":"","code":"## creating an PipeOpModule manually  # one input and output channel po_module = po(\"module\",   id = \"linear\",   module = torch::nn_linear(10, 20),   inname = \"input\",   outname = \"output\" ) x = torch::torch_randn(16, 10) # This calls the forward function of the wrapped module. y = po_module$train(list(input = x)) str(y) #> List of 1 #>  $ output:Float [1:16, 1:20]  # multiple input and output channels nn_custom = torch::nn_module(\"nn_custom\",   initialize = function(in_features, out_features) {     self$lin1 = torch::nn_linear(in_features, out_features)     self$lin2 = torch::nn_linear(in_features, out_features)   },   forward = function(x, z) {     list(out1 = self$lin1(x), out2 = torch::nnf_relu(self$lin2(z)))   } )  module = nn_custom(3, 2) po_module = po(\"module\",   id = \"custom\",   module = module,   inname = c(\"x\", \"z\"),   outname = c(\"out1\", \"out2\") ) x = torch::torch_randn(1, 3) z = torch::torch_randn(1, 3) out = po_module$train(list(x = x, z = z)) str(out) #> List of 2 #>  $ out1:Float [1:1, 1:2] #>  $ out2:Float [1:1, 1:2]  # How such a PipeOpModule is usually generated graph = po(\"torch_ingress_num\") %>>% po(\"nn_linear\", out_features = 10L) result = graph$train(tsk(\"iris\")) # The PipeOpTorchLinear generates a PipeOpModule and adds it to a new (module) graph result[[1]]$graph #> Graph with 2 PipeOps: #>                 ID         State  sccssors         prdcssors #>             <char>        <char>    <char>            <char> #>  torch_ingress_num <<UNTRAINED>> nn_linear                   #>          nn_linear <<UNTRAINED>>           torch_ingress_num linear_module = result[[1L]]$graph$pipeops$nn_linear linear_module #> PipeOp: <nn_linear> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [torch_tensor,NULL] #> Output channels <name [train type, predict type]>: #>   output [torch_tensor,NULL] formalArgs(linear_module$module) #> [1] \"input\" linear_module$input$name #> [1] \"input\"  # Constructing a PipeOpModule using a simple function po_add1 = po(\"module\",   id = \"add_one\",   module = function(x) x + 1 ) input = list(torch_tensor(1)) po_add1$train(input)$output #> torch_tensor #>  2 #> [ CPUFloatType{1} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Applies 1D adaptive average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"output_size :: integer() target output size. single number.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Calls nn_adaptive_avg_pool1d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAdaptiveAvgPool -> PipeOpTorchAdaptiveAvgPool1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"PipeOpTorchAdaptiveAvgPool1D$new() PipeOpTorchAdaptiveAvgPool1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"","code":"PipeOpTorchAdaptiveAvgPool1D$new(   id = \"nn_adaptive_avg_pool1d\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"","code":"PipeOpTorchAdaptiveAvgPool1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_adaptive_avg_pool1d\") pipeop #> PipeOp: <nn_adaptive_avg_pool1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: output_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Applies 2D adaptive average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"output_size :: integer() target output size. Can single number vector.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Calls nn_adaptive_avg_pool2d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAdaptiveAvgPool -> PipeOpTorchAdaptiveAvgPool2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"PipeOpTorchAdaptiveAvgPool2D$new() PipeOpTorchAdaptiveAvgPool2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"","code":"PipeOpTorchAdaptiveAvgPool2D$new(   id = \"nn_adaptive_avg_pool2d\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"","code":"PipeOpTorchAdaptiveAvgPool2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_adaptive_avg_pool2d\") pipeop #> PipeOp: <nn_adaptive_avg_pool2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: output_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Applies 3D adaptive average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"output_size :: integer() target output size. Can single number vector.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Calls nn_adaptive_avg_pool3d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAdaptiveAvgPool -> PipeOpTorchAdaptiveAvgPool3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"PipeOpTorchAdaptiveAvgPool3D$new() PipeOpTorchAdaptiveAvgPool3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"","code":"PipeOpTorchAdaptiveAvgPool3D$new(   id = \"nn_adaptive_avg_pool3d\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"","code":"PipeOpTorchAdaptiveAvgPool3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_adaptive_avg_pool3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Adaptive Average Pooling — mlr_pipeops_nn_adaptive_avg_pool3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_adaptive_avg_pool3d\") pipeop #> PipeOp: <nn_adaptive_avg_pool3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: output_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Applies 1D average pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"kernel_size :: (integer()) size window. Can single number vector. stride :: integer() stride window. Can single number vector. Default: kernel_size. padding :: integer() Implicit zero paddings sides input. Can single number vector. Default: 0. ceil_mode :: integer() TRUE, use ceil instead floor compute output shape. Default: FALSE. count_include_pad :: logical(1) TRUE, include zero-padding averaging calculation. Default: TRUE. divisor_override :: logical(1) specified, used divisor, otherwise size pooling region used. Default: NULL. available dimension greater 1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Calls nn_avg_pool1d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAvgPool -> PipeOpTorchAvgPool1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"PipeOpTorchAvgPool1D$new() PipeOpTorchAvgPool1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"","code":"PipeOpTorchAvgPool1D$new(id = \"nn_avg_pool1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"","code":"PipeOpTorchAvgPool1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Average Pooling — mlr_pipeops_nn_avg_pool1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_avg_pool1d\") pipeop #> PipeOp: <nn_avg_pool1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>                   id    class lower upper nlevels        default  value #>               <char>   <char> <num> <num>   <num>         <list> <list> #> 1:       kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:            stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 3:           padding ParamUty    NA    NA     Inf              0 [NULL] #> 4:         ceil_mode ParamLgl    NA    NA       2          FALSE [NULL] #> 5: count_include_pad ParamLgl    NA    NA       2           TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Applies 2D average-pooling operation \\(kH * kW\\) regions step size \\(sH * sW\\) steps. number output features equal number input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Calls nn_avg_pool2d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"kernel_size :: (integer()) size window. Can single number vector. stride :: integer() stride window. Can single number vector. Default: kernel_size. padding :: integer() Implicit zero paddings sides input. Can single number vector. Default: 0. ceil_mode :: integer() TRUE, use ceil instead floor compute output shape. Default: FALSE. count_include_pad :: logical(1) TRUE, include zero-padding averaging calculation. Default: TRUE. divisor_override :: logical(1) specified, used divisor, otherwise size pooling region used. Default: NULL. available dimension greater 1.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAvgPool -> PipeOpTorchAvgPool2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"PipeOpTorchAvgPool2D$new() PipeOpTorchAvgPool2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"","code":"PipeOpTorchAvgPool2D$new(id = \"nn_avg_pool2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"","code":"PipeOpTorchAvgPool2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Average Pooling — mlr_pipeops_nn_avg_pool2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_avg_pool2d\") pipeop #> PipeOp: <nn_avg_pool2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(6)> #>                   id    class lower upper nlevels        default  value #>               <char>   <char> <num> <num>   <num>         <list> <list> #> 1:       kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:            stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 3:           padding ParamUty    NA    NA     Inf              0 [NULL] #> 4:         ceil_mode ParamLgl    NA    NA       2          FALSE [NULL] #> 5: count_include_pad ParamLgl    NA    NA       2           TRUE [NULL] #> 6:  divisor_override ParamDbl     0   Inf     Inf         [NULL] [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Applies 3D average-pooling operation \\(kT * kH * kW\\) regions step size \\(sT * sH * sW\\) steps. number output features equal \\(\\lfloor \\frac{ \\mbox{input planes} }{sT} \\rfloor\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Calls nn_avg_pool3d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"kernel_size :: (integer()) size window. Can single number vector. stride :: integer() stride window. Can single number vector. Default: kernel_size. padding :: integer() Implicit zero paddings sides input. Can single number vector. Default: 0. ceil_mode :: integer() TRUE, use ceil instead floor compute output shape. Default: FALSE. count_include_pad :: logical(1) TRUE, include zero-padding averaging calculation. Default: TRUE. divisor_override :: logical(1) specified, used divisor, otherwise size pooling region used. Default: NULL. available dimension greater 1.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchAvgPool -> PipeOpTorchAvgPool3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"PipeOpTorchAvgPool3D$new() PipeOpTorchAvgPool3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"","code":"PipeOpTorchAvgPool3D$new(id = \"nn_avg_pool3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"","code":"PipeOpTorchAvgPool3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_avg_pool3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Average Pooling — mlr_pipeops_nn_avg_pool3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_avg_pool3d\") pipeop #> PipeOp: <nn_avg_pool3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(6)> #>                   id    class lower upper nlevels        default  value #>               <char>   <char> <num> <num>   <num>         <list> <list> #> 1:       kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:            stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 3:           padding ParamUty    NA    NA     Inf              0 [NULL] #> 4:         ceil_mode ParamLgl    NA    NA       2          FALSE [NULL] #> 5: count_include_pad ParamLgl    NA    NA       2           TRUE [NULL] #> 6:  divisor_override ParamDbl     0   Inf     Inf         [NULL] [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Applies Batch Normalization channel across batch data.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"eps :: numeric(1) value added denominator numerical stability. Default: 1e-5. momentum :: numeric(1) value used running_mean running_var computation. Can set NULL cumulative moving average (.e. simple average). Default: 0.1 affine :: logical(1) boolean value set TRUE, module learnable affine parameters. Default: TRUE track_running_stats :: logical(1) boolean value set TRUE, module tracks running mean variance, set FALSE, module track statistics always uses batch statistics training eval modes. Default: TRUE","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Calls torch::nn_batch_norm1d(). parameter num_features inferred second dimension input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchBatchNorm -> PipeOpTorchBatchNorm1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"PipeOpTorchBatchNorm1D$new() PipeOpTorchBatchNorm1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"","code":"PipeOpTorchBatchNorm1D$new(id = \"nn_batch_norm1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"","code":"PipeOpTorchBatchNorm1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Batch Normalization — mlr_pipeops_nn_batch_norm1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_batch_norm1d\") pipeop #> PipeOp: <nn_batch_norm1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(4)> #>                     id    class lower upper nlevels default  value #>                 <char>   <char> <num> <num>   <num>  <list> <list> #> 1:                 eps ParamDbl     0   Inf     Inf   1e-05 [NULL] #> 2:            momentum ParamDbl     0   Inf     Inf     0.1 [NULL] #> 3:              affine ParamLgl    NA    NA       2    TRUE [NULL] #> 4: track_running_stats ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Applies Batch Normalization channel across batch data.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Calls torch::nn_batch_norm2d(). parameter num_features inferred second dimension input shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"eps :: numeric(1) value added denominator numerical stability. Default: 1e-5. momentum :: numeric(1) value used running_mean running_var computation. Can set NULL cumulative moving average (.e. simple average). Default: 0.1 affine :: logical(1) boolean value set TRUE, module learnable affine parameters. Default: TRUE track_running_stats :: logical(1) boolean value set TRUE, module tracks running mean variance, set FALSE, module track statistics always uses batch statistics training eval modes. Default: TRUE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchBatchNorm -> PipeOpTorchBatchNorm2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"PipeOpTorchBatchNorm2D$new() PipeOpTorchBatchNorm2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"","code":"PipeOpTorchBatchNorm2D$new(id = \"nn_batch_norm2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"","code":"PipeOpTorchBatchNorm2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Batch Normalization — mlr_pipeops_nn_batch_norm2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_batch_norm2d\") pipeop #> PipeOp: <nn_batch_norm2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(4)> #>                     id    class lower upper nlevels default  value #>                 <char>   <char> <num> <num>   <num>  <list> <list> #> 1:                 eps ParamDbl     0   Inf     Inf   1e-05 [NULL] #> 2:            momentum ParamDbl     0   Inf     Inf     0.1 [NULL] #> 3:              affine ParamLgl    NA    NA       2    TRUE [NULL] #> 4: track_running_stats ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Applies Batch Normalization channel across batch data.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Calls torch::nn_batch_norm3d(). parameter num_features inferred second dimension input shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"eps :: numeric(1) value added denominator numerical stability. Default: 1e-5. momentum :: numeric(1) value used running_mean running_var computation. Can set NULL cumulative moving average (.e. simple average). Default: 0.1 affine :: logical(1) boolean value set TRUE, module learnable affine parameters. Default: TRUE track_running_stats :: logical(1) boolean value set TRUE, module tracks running mean variance, set FALSE, module track statistics always uses batch statistics training eval modes. Default: TRUE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchBatchNorm -> PipeOpTorchBatchNorm3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"PipeOpTorchBatchNorm3D$new() PipeOpTorchBatchNorm3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"","code":"PipeOpTorchBatchNorm3D$new(id = \"nn_batch_norm3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"","code":"PipeOpTorchBatchNorm3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_batch_norm3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Batch Normalization — mlr_pipeops_nn_batch_norm3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_batch_norm3d\") pipeop #> PipeOp: <nn_batch_norm3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(4)> #>                     id    class lower upper nlevels default  value #>                 <char>   <char> <num> <num>   <num>  <list> <list> #> 1:                 eps ParamDbl     0   Inf     Inf   1e-05 [NULL] #> 2:            momentum ParamDbl     0   Inf     Inf     0.1 [NULL] #> 3:              affine ParamLgl    NA    NA       2    TRUE [NULL] #> 4: track_running_stats ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":null,"dir":"Reference","previous_headings":"","what":"Block Repetition — mlr_pipeops_nn_block","title":"Block Repetition — mlr_pipeops_nn_block","text":"Repeat block n_blocks times.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Block Repetition — mlr_pipeops_nn_block","text":"parameters available block , well n_blocks :: integer(1) often repeat block.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Block Repetition — mlr_pipeops_nn_block","text":"PipeOp sets input output channels block (Graph) received construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Block Repetition — mlr_pipeops_nn_block","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Block Repetition — mlr_pipeops_nn_block","text":"Part documentation copied adapted documentation torch.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Block Repetition — mlr_pipeops_nn_block","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchBlock","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Block Repetition — mlr_pipeops_nn_block","text":"block (Graph) neural network segment repeated PipeOp.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Block Repetition — mlr_pipeops_nn_block","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Block Repetition — mlr_pipeops_nn_block","text":"PipeOpTorchBlock$new() PipeOpTorchBlock$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Block Repetition — mlr_pipeops_nn_block","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Block Repetition — mlr_pipeops_nn_block","text":"","code":"PipeOpTorchBlock$new(block, id = \"nn_block\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Block Repetition — mlr_pipeops_nn_block","text":"block (Graph) graph consisting primarily PipeOpTorch objects repeated. id (character(1)) id new object. param_vals (named list()) Parameter values set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Block Repetition — mlr_pipeops_nn_block","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Block Repetition — mlr_pipeops_nn_block","text":"","code":"PipeOpTorchBlock$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Block Repetition — mlr_pipeops_nn_block","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_block.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Block Repetition — mlr_pipeops_nn_block","text":"","code":"block = po(\"nn_linear\") %>>% po(\"nn_relu\") po_block = po(\"nn_block\", block, nn_linear.out_features = 10L, n_blocks = 3) network = po(\"torch_ingress_num\") %>>% po_block %>>% po(\"nn_head\") %>>% po(\"torch_loss\", t_loss(\"cross_entropy\")) %>>% po(\"torch_optimizer\", t_opt(\"adam\")) %>>% po(\"torch_model_classif\",   batch_size = 50,   epochs = 3)  task = tsk(\"iris\") network$train(task) #> $torch_model_classif.output #> NULL #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":null,"dir":"Reference","previous_headings":"","what":"CELU Activation Function — mlr_pipeops_nn_celu","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Applies element-wise, \\(CELU(x) = max(0,x) + min(0, \\alpha * (exp(x \\alpha) - 1))\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"alpha :: numeric(1) alpha value ELU formulation. Default: 1.0 inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Calls torch::nn_celu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchCELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"PipeOpTorchCELU$new() PipeOpTorchCELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"","code":"PipeOpTorchCELU$new(id = \"nn_celu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"","code":"PipeOpTorchCELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_celu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"CELU Activation Function — mlr_pipeops_nn_celu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_celu\") pipeop #> PipeOp: <nn_celu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1:   alpha ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 2: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Convolution — mlr_pipeops_nn_conv1d","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Applies 1D convolution input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"out_channels :: integer(1) Number channels produced convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. default 1. padding :: integer() ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘TRUE’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Calls torch::nn_conv1d() trained. paramter in_channels inferred second dimension input tensor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConv -> PipeOpTorchConv1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"PipeOpTorchConv1D$new() PipeOpTorchConv1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"","code":"PipeOpTorchConv1D$new(id = \"nn_conv1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"","code":"PipeOpTorchConv1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Convolution — mlr_pipeops_nn_conv1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv1d\", kernel_size = 10, out_channels = 1) pipeop #> PipeOp: <nn_conv1d> (not trained) #> values: <out_channels=1, kernel_size=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(8)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2:  kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>     10 #> 3:       stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:      padding ParamUty    NA    NA     Inf              0 [NULL] #> 5:     dilation ParamUty    NA    NA     Inf              1 [NULL] #> 6:       groups ParamInt     1   Inf     Inf              1 [NULL] #> 7:         bias ParamLgl    NA    NA       2           TRUE [NULL] #> 8: padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Convolution — mlr_pipeops_nn_conv2d","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Applies 2D convolution input image composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Calls torch::nn_conv2d() trained. paramter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"out_channels :: integer(1) Number channels produced convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. default 1. padding :: integer() ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘TRUE’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConv -> PipeOpTorchConv2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"PipeOpTorchConv2D$new() PipeOpTorchConv2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"","code":"PipeOpTorchConv2D$new(id = \"nn_conv2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"","code":"PipeOpTorchConv2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Convolution — mlr_pipeops_nn_conv2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv2d\", kernel_size = 10, out_channels = 1) pipeop #> PipeOp: <nn_conv2d> (not trained) #> values: <out_channels=1, kernel_size=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(8)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2:  kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>     10 #> 3:       stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:      padding ParamUty    NA    NA     Inf              0 [NULL] #> 5:     dilation ParamUty    NA    NA     Inf              1 [NULL] #> 6:       groups ParamInt     1   Inf     Inf              1 [NULL] #> 7:         bias ParamLgl    NA    NA       2           TRUE [NULL] #> 8: padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Convolution — mlr_pipeops_nn_conv3d","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Applies 3D convolution input image composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Calls torch::nn_conv3d() trained. paramter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"out_channels :: integer(1) Number channels produced convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. default 1. padding :: integer() ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘TRUE’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConv -> PipeOpTorchConv3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"PipeOpTorchConv3D$new() PipeOpTorchConv3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"","code":"PipeOpTorchConv3D$new(id = \"nn_conv3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"","code":"PipeOpTorchConv3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Convolution — mlr_pipeops_nn_conv3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv3d\", kernel_size = 10, out_channels = 1) pipeop #> PipeOp: <nn_conv3d> (not trained) #> values: <out_channels=1, kernel_size=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(8)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2:  kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>     10 #> 3:       stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:      padding ParamUty    NA    NA     Inf              0 [NULL] #> 5:     dilation ParamUty    NA    NA     Inf              1 [NULL] #> 6:       groups ParamInt     1   Inf     Inf              1 [NULL] #> 7:         bias ParamLgl    NA    NA       2           TRUE [NULL] #> 8: padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":null,"dir":"Reference","previous_headings":"","what":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Transpose 1D Convolution Transpose 1D Convolution","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"out_channels :: integer(1) Number output channels produce convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. Default: 1. padding ::  integer()` ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. output_padding ::integer() Additional size added one side output shape. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘True’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Calls nn_conv_transpose1d. parameter in_channels inferred second dimension input tensor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConvTranspose -> PipeOpTorchConvTranspose1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"PipeOpTorchConvTranspose1D$new() PipeOpTorchConvTranspose1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"","code":"PipeOpTorchConvTranspose1D$new(id = \"nn_conv_transpose1d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"","code":"PipeOpTorchConvTranspose1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Transpose 1D Convolution — mlr_pipeops_nn_conv_transpose1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv_transpose1d\", kernel_size = 3, out_channels = 2) pipeop #> PipeOp: <nn_conv_transpose1d> (not trained) #> values: <out_channels=2, kernel_size=3> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(9)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:   out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      2 #> 2:    kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>      3 #> 3:         stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:        padding ParamUty    NA    NA     Inf              0 [NULL] #> 5: output_padding ParamUty    NA    NA     Inf              0 [NULL] #> 6:       dilation ParamUty    NA    NA     Inf              1 [NULL] #> 7:         groups ParamInt     1   Inf     Inf              1 [NULL] #> 8:           bias ParamLgl    NA    NA       2           TRUE [NULL] #> 9:   padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":null,"dir":"Reference","previous_headings":"","what":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Applies 2D transposed convolution operator input image composed several input planes, sometimes also called \"deconvolution\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Calls nn_conv_transpose2d. parameter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"out_channels :: integer(1) Number output channels produce convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. Default: 1. padding ::  integer()` ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. output_padding ::integer() Additional size added one side output shape. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘True’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConvTranspose -> PipeOpTorchConvTranspose2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"PipeOpTorchConvTranspose2D$new() PipeOpTorchConvTranspose2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"","code":"PipeOpTorchConvTranspose2D$new(id = \"nn_conv_transpose2d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"","code":"PipeOpTorchConvTranspose2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Transpose 2D Convolution — mlr_pipeops_nn_conv_transpose2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv_transpose2d\", kernel_size = 3, out_channels = 2) pipeop #> PipeOp: <nn_conv_transpose2d> (not trained) #> values: <out_channels=2, kernel_size=3> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(9)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:   out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      2 #> 2:    kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>      3 #> 3:         stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:        padding ParamUty    NA    NA     Inf              0 [NULL] #> 5: output_padding ParamUty    NA    NA     Inf              0 [NULL] #> 6:       dilation ParamUty    NA    NA     Inf              1 [NULL] #> 7:         groups ParamInt     1   Inf     Inf              1 [NULL] #> 8:           bias ParamLgl    NA    NA       2           TRUE [NULL] #> 9:   padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":null,"dir":"Reference","previous_headings":"","what":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Applies 3D transposed convolution operator input image composed several input planes, sometimes also called \"deconvolution\"","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Calls nn_conv_transpose3d. parameter in_channels inferred second dimension input tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"out_channels :: integer(1) Number output channels produce convolution. kernel_size :: integer() Size convolving kernel. stride :: integer() Stride convolution. Default: 1. padding ::  integer()` ‘dilation * (kernel_size - 1) - padding’ zero-padding added sides input. Default: 0. output_padding ::integer() Additional size added one side output shape. Default: 0. groups :: integer() Number blocked connections input channels output channels. Default: 1 bias :: logical(1) ‘True’, adds learnable bias output. Default: ‘TRUE’. dilation :: integer() Spacing kernel elements. Default: 1. padding_mode :: character(1) padding mode. One \"zeros\", \"reflect\", \"replicate\", \"circular\". Default \"zeros\".","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchConvTranspose -> PipeOpTorchConvTranspose3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"PipeOpTorchConvTranspose3D$new() PipeOpTorchConvTranspose3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"","code":"PipeOpTorchConvTranspose3D$new(id = \"nn_conv_transpose3d\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"","code":"PipeOpTorchConvTranspose3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_conv_transpose3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Transpose 3D Convolution — mlr_pipeops_nn_conv_transpose3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_conv_transpose3d\", kernel_size = 3, out_channels = 2) pipeop #> PipeOp: <nn_conv_transpose3d> (not trained) #> values: <out_channels=2, kernel_size=3> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(9)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:   out_channels ParamInt     1   Inf     Inf <NoDefault[0]>      2 #> 2:    kernel_size ParamUty    NA    NA     Inf <NoDefault[0]>      3 #> 3:         stride ParamUty    NA    NA     Inf              1 [NULL] #> 4:        padding ParamUty    NA    NA     Inf              0 [NULL] #> 5: output_padding ParamUty    NA    NA     Inf              0 [NULL] #> 6:       dilation ParamUty    NA    NA     Inf              1 [NULL] #> 7:         groups ParamInt     1   Inf     Inf              1 [NULL] #> 8:           bias ParamLgl    NA    NA       2           TRUE [NULL] #> 9:   padding_mode ParamFct    NA    NA       4          zeros [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":null,"dir":"Reference","previous_headings":"","what":"Dropout — mlr_pipeops_nn_dropout","title":"Dropout — mlr_pipeops_nn_dropout","text":"training, randomly zeroes elements input tensor probability p using samples Bernoulli distribution.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Dropout — mlr_pipeops_nn_dropout","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Dropout — mlr_pipeops_nn_dropout","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Dropout — mlr_pipeops_nn_dropout","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Dropout — mlr_pipeops_nn_dropout","text":"p :: numeric(1) Probability element zeroed. Default: 0.5 inplace inplace :: logical(1) set TRUE, operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Dropout — mlr_pipeops_nn_dropout","text":"Calls torch::nn_dropout() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Dropout — mlr_pipeops_nn_dropout","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchDropout","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Dropout — mlr_pipeops_nn_dropout","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Dropout — mlr_pipeops_nn_dropout","text":"PipeOpTorchDropout$new() PipeOpTorchDropout$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Dropout — mlr_pipeops_nn_dropout","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Dropout — mlr_pipeops_nn_dropout","text":"","code":"PipeOpTorchDropout$new(id = \"nn_dropout\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Dropout — mlr_pipeops_nn_dropout","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Dropout — mlr_pipeops_nn_dropout","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Dropout — mlr_pipeops_nn_dropout","text":"","code":"PipeOpTorchDropout$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Dropout — mlr_pipeops_nn_dropout","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_dropout.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Dropout — mlr_pipeops_nn_dropout","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_dropout\") pipeop #> PipeOp: <nn_dropout> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1:       p ParamDbl     0     1     Inf     0.5 [NULL] #> 2: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":null,"dir":"Reference","previous_headings":"","what":"ELU Activation Function — mlr_pipeops_nn_elu","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Applies element-wise, $$ELU(x) = max(0,x) + min(0, \\alpha * (exp(x) - 1))$$.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"alpha :: numeric(1) alpha value ELU formulation. Default: 1.0 inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Calls torch::nn_elu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"PipeOpTorchELU$new() PipeOpTorchELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"","code":"PipeOpTorchELU$new(id = \"nn_elu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"","code":"PipeOpTorchELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_elu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"ELU Activation Function — mlr_pipeops_nn_elu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_elu\") pipeop #> PipeOp: <nn_elu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1:   alpha ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 2: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":null,"dir":"Reference","previous_headings":"","what":"Flattens a Tensor — mlr_pipeops_nn_flatten","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"use nn_sequential.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"start_dim :: integer(1) wich dimension start flattening. Default 2. end_dim :: integer(1) wich dimension stop flattening. Default -1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"Calls torch::nn_flatten() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchFlatten","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"PipeOpTorchFlatten$new() PipeOpTorchFlatten$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"","code":"PipeOpTorchFlatten$new(id = \"nn_flatten\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"","code":"PipeOpTorchFlatten$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_flatten.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Flattens a Tensor — mlr_pipeops_nn_flatten","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_flatten\") pipeop #> PipeOp: <nn_flatten> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <int> <num>   <num>  <list> <list> #> 1: start_dim ParamInt     1   Inf     Inf       2 [NULL] #> 2:   end_dim ParamInt     1   Inf     Inf      -1 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":null,"dir":"Reference","previous_headings":"","what":"GELU Activation Function — mlr_pipeops_nn_gelu","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Gelu","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"approximate :: character(1) Whether use approximation algorithm. Default \"none\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Calls torch::nn_gelu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchGELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"PipeOpTorchGELU$new() PipeOpTorchGELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"","code":"PipeOpTorchGELU$new(id = \"nn_gelu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"","code":"PipeOpTorchGELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_gelu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"GELU Activation Function — mlr_pipeops_nn_gelu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_gelu\") pipeop #> PipeOp: <nn_gelu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>             id    class lower upper nlevels default  value #>         <char>   <char> <num> <num>   <num>  <list> <list> #> 1: approximate ParamFct    NA    NA       2    none [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":null,"dir":"Reference","previous_headings":"","what":"GLU Activation Function — mlr_pipeops_nn_glu","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"gated linear unit. Computes:","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"dim :: integer(1) Dimension split input. Default: -1","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"Calls torch::nn_glu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchGLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"PipeOpTorchGLU$new() PipeOpTorchGLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"","code":"PipeOpTorchGLU$new(id = \"nn_glu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"","code":"PipeOpTorchGLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_glu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"GLU Activation Function — mlr_pipeops_nn_glu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_glu\") pipeop #> PipeOp: <nn_glu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <int> <num>   <num>  <list> <list> #> 1:    dim ParamInt     1   Inf     Inf      -1 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":null,"dir":"Reference","previous_headings":"","what":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Applies hard shrinkage function element-wise","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"lambd :: numeric(1) lambda value Hardshrink formulation formulation. Default 0.5.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Calls torch::nn_hardshrink() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHardShrink","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"PipeOpTorchHardShrink$new() PipeOpTorchHardShrink$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"","code":"PipeOpTorchHardShrink$new(id = \"nn_hardshrink\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"","code":"PipeOpTorchHardShrink$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardshrink.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Hard Shrink Activation Function — mlr_pipeops_nn_hardshrink","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_hardshrink\") pipeop #> PipeOp: <nn_hardshrink> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:  lambd ParamDbl  -Inf   Inf     Inf     0.5 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":null,"dir":"Reference","previous_headings":"","what":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Applies element-wise function \\(\\mbox{Hardsigmoid}(x) = \\frac{ReLU6(x + 3)}{6}\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Calls torch::nn_hardsigmoid() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHardSigmoid","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"PipeOpTorchHardSigmoid$new() PipeOpTorchHardSigmoid$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"","code":"PipeOpTorchHardSigmoid$new(id = \"nn_hardsigmoid\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"","code":"PipeOpTorchHardSigmoid$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardsigmoid.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Hard Sigmoid Activation Function — mlr_pipeops_nn_hardsigmoid","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_hardsigmoid\") pipeop #> PipeOp: <nn_hardsigmoid> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":null,"dir":"Reference","previous_headings":"","what":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Applies HardTanh function element-wise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"min_val :: numeric(1) Minimum value linear region range. Default: -1. max_val :: numeric(1) Maximum value linear region range. Default: 1. inplace :: logical(1) Can optionally operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Calls torch::nn_hardtanh() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHardTanh","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"PipeOpTorchHardTanh$new() PipeOpTorchHardTanh$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"","code":"PipeOpTorchHardTanh$new(id = \"nn_hardtanh\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"","code":"PipeOpTorchHardTanh$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_hardtanh.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Hard Tanh Activation Function — mlr_pipeops_nn_hardtanh","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_hardtanh\") pipeop #> PipeOp: <nn_hardtanh> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: min_val ParamDbl  -Inf   Inf     Inf      -1 [NULL] #> 2: max_val ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 3: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":null,"dir":"Reference","previous_headings":"","what":"Output Head — mlr_pipeops_nn_head","title":"Output Head — mlr_pipeops_nn_head","text":"Output head classification regresssion. NOTE method $shapes_out() access task, returns c(NA, NA). PipeOp trained however, model descriptor correct output shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Output Head — mlr_pipeops_nn_head","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Output Head — mlr_pipeops_nn_head","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Output Head — mlr_pipeops_nn_head","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Output Head — mlr_pipeops_nn_head","text":"bias :: logical(1) Whether use bias. Default TRUE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Output Head — mlr_pipeops_nn_head","text":"Calls torch::nn_linear() input output features inferred input shape / task.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Output Head — mlr_pipeops_nn_head","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchHead","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Output Head — mlr_pipeops_nn_head","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Output Head — mlr_pipeops_nn_head","text":"PipeOpTorchHead$new() PipeOpTorchHead$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Output Head — mlr_pipeops_nn_head","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Output Head — mlr_pipeops_nn_head","text":"","code":"PipeOpTorchHead$new(id = \"nn_head\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Output Head — mlr_pipeops_nn_head","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Output Head — mlr_pipeops_nn_head","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Output Head — mlr_pipeops_nn_head","text":"","code":"PipeOpTorchHead$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Output Head — mlr_pipeops_nn_head","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_head.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Output Head — mlr_pipeops_nn_head","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_head\") pipeop #> PipeOp: <nn_head> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:   bias ParamLgl    NA    NA       2    TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":null,"dir":"Reference","previous_headings":"","what":"Layer Normalization — mlr_pipeops_nn_layer_norm","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Applies Layer Normalization last certain number dimensions.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"dims :: integer(1) number dimensions normalized (starting last dimension). elementwise_affine :: logical(1) Whether learn affine-linear parameters initialized 1 weights 0 biases. default TRUE. eps :: numeric(1) value added denominator numerical stability.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Calls torch::nn_layer_norm() trained. parameter normalized_shape inferred dimensions last dims dimensions input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLayerNorm","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"PipeOpTorchLayerNorm$new() PipeOpTorchLayerNorm$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"","code":"PipeOpTorchLayerNorm$new(id = \"nn_layer_norm\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"","code":"PipeOpTorchLayerNorm$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_layer_norm.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Layer Normalization — mlr_pipeops_nn_layer_norm","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_layer_norm\", dims = 1) pipeop #> PipeOp: <nn_layer_norm> (not trained) #> values: <dims=1> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>                    id    class lower upper nlevels        default  value #>                <char>   <char> <num> <num>   <num>         <list> <list> #> 1:               dims ParamInt     1   Inf     Inf <NoDefault[0]>      1 #> 2: elementwise_affine ParamLgl    NA    NA       2           TRUE [NULL] #> 3:                eps ParamDbl     0   Inf     Inf          1e-05 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":null,"dir":"Reference","previous_headings":"","what":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Applies element-wise, \\(LeakyReLU(x) = max(0, x) + negative_slope * min(0, x)\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"negative_slope :: numeric(1) Controls angle negative slope. Default: 1e-2. inplace :: logical(1) Can optionally operation -place. Default: ‘FALSE’.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Calls torch::nn_hardswish() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLeakyReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"PipeOpTorchLeakyReLU$new() PipeOpTorchLeakyReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"","code":"PipeOpTorchLeakyReLU$new(id = \"nn_leaky_relu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"","code":"PipeOpTorchLeakyReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_leaky_relu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Leaky ReLU Activation Function — mlr_pipeops_nn_leaky_relu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_leaky_relu\") pipeop #> PipeOp: <nn_leaky_relu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>                id    class lower upper nlevels default  value #>            <char>   <char> <num> <num>   <num>  <list> <list> #> 1: negative_slope ParamDbl  -Inf   Inf     Inf    0.01 [NULL] #> 2:        inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":null,"dir":"Reference","previous_headings":"","what":"Linear Layer — mlr_pipeops_nn_linear","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Applies linear transformation incoming data: \\(y = xA^T + b\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Linear Layer — mlr_pipeops_nn_linear","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Linear Layer — mlr_pipeops_nn_linear","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Linear Layer — mlr_pipeops_nn_linear","text":"out_features :: integer(1) output features linear layer. bias :: logical(1) Whether use bias. Default TRUE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Calls torch::nn_linear() trained parameter in_features inferred second last dimension input tensor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Linear Layer — mlr_pipeops_nn_linear","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLinear","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Linear Layer — mlr_pipeops_nn_linear","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Linear Layer — mlr_pipeops_nn_linear","text":"PipeOpTorchLinear$new() PipeOpTorchLinear$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Linear Layer — mlr_pipeops_nn_linear","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Linear Layer — mlr_pipeops_nn_linear","text":"","code":"PipeOpTorchLinear$new(id = \"nn_linear\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Linear Layer — mlr_pipeops_nn_linear","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Linear Layer — mlr_pipeops_nn_linear","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Linear Layer — mlr_pipeops_nn_linear","text":"","code":"PipeOpTorchLinear$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Linear Layer — mlr_pipeops_nn_linear","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_linear.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Linear Layer — mlr_pipeops_nn_linear","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_linear\", out_features = 10) pipeop #> PipeOp: <nn_linear> (not trained) #> values: <out_features=10> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1: out_features ParamInt     1   Inf     Inf <NoDefault[0]>     10 #> 2:         bias ParamLgl    NA    NA       2           TRUE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":null,"dir":"Reference","previous_headings":"","what":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Applies element-wise \\(LogSigmoid(x_i) = log(\\frac{1}{1 + exp(-x_i)})\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Calls torch::nn_log_sigmoid() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchLogSigmoid","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"PipeOpTorchLogSigmoid$new() PipeOpTorchLogSigmoid$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"","code":"PipeOpTorchLogSigmoid$new(id = \"nn_log_sigmoid\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"","code":"PipeOpTorchLogSigmoid$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_log_sigmoid.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Log Sigmoid Activation Function — mlr_pipeops_nn_log_sigmoid","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_log_sigmoid\") pipeop #> PipeOp: <nn_log_sigmoid> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":null,"dir":"Reference","previous_headings":"","what":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Applies 1D max pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"return_indices FALSE construction, one input channel 'input' one output channel 'output'. return_indices TRUE, two output channels 'output' 'indices'. explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"kernel_size :: integer() size window. Can single number vector. stride :: (integer(1)) stride window. Can single number vector. Default: kernel_size padding :: integer() Implicit zero paddings sides input. Can single number tuple (padW,). Default: 0 dilation :: integer() Controls spacing kernel points; also known à trous algorithm. Default: 1 ceil_mode :: logical(1) True, use ceil instead floor compute output shape. Default: FALSE","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Calls torch::nn_max_pool1d() training.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMaxPool -> PipeOpTorchMaxPool1D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"PipeOpTorchMaxPool1D$new() PipeOpTorchMaxPool1D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"","code":"PipeOpTorchMaxPool1D$new(   id = \"nn_max_pool1d\",   return_indices = FALSE,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"id (character(1)) Identifier resulting  object. return_indices (logical(1)) Whether return indices. TRUE, two output channels \"output\" \"indices\". param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"","code":"PipeOpTorchMaxPool1D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool1d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"1D Max Pooling — mlr_pipeops_nn_max_pool1d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_max_pool1d\") pipeop #> PipeOp: <nn_max_pool1d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:     padding ParamUty    NA    NA     Inf              0 [NULL] #> 3:      stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 4:    dilation ParamInt  -Inf   Inf     Inf              1 [NULL] #> 5:   ceil_mode ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":null,"dir":"Reference","previous_headings":"","what":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Applies 2D max pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Calls torch::nn_max_pool2d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"return_indices FALSE construction, one input channel 'input' one output channel 'output'. return_indices TRUE, two output channels 'output' 'indices'. explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"kernel_size :: integer() size window. Can single number vector. stride :: (integer(1)) stride window. Can single number vector. Default: kernel_size padding :: integer() Implicit zero paddings sides input. Can single number tuple (padW,). Default: 0 dilation :: integer() Controls spacing kernel points; also known à trous algorithm. Default: 1 ceil_mode :: logical(1) True, use ceil instead floor compute output shape. Default: FALSE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMaxPool -> PipeOpTorchMaxPool2D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"PipeOpTorchMaxPool2D$new() PipeOpTorchMaxPool2D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"","code":"PipeOpTorchMaxPool2D$new(   id = \"nn_max_pool2d\",   return_indices = FALSE,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"id (character(1)) Identifier resulting  object. return_indices (logical(1)) Whether return indices. TRUE, two output channels \"output\" \"indices\". param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"","code":"PipeOpTorchMaxPool2D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool2d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"2D Max Pooling — mlr_pipeops_nn_max_pool2d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_max_pool2d\") pipeop #> PipeOp: <nn_max_pool2d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:     padding ParamUty    NA    NA     Inf              0 [NULL] #> 3:      stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 4:    dilation ParamInt  -Inf   Inf     Inf              1 [NULL] #> 5:   ceil_mode ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":null,"dir":"Reference","previous_headings":"","what":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Applies 3D max pooling input signal composed several input planes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Calls torch::nn_max_pool3d() training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"return_indices FALSE construction, one input channel 'input' one output channel 'output'. return_indices TRUE, two output channels 'output' 'indices'. explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"kernel_size :: integer() size window. Can single number vector. stride :: (integer(1)) stride window. Can single number vector. Default: kernel_size padding :: integer() Implicit zero paddings sides input. Can single number tuple (padW,). Default: 0 dilation :: integer() Controls spacing kernel points; also known à trous algorithm. Default: 1 ceil_mode :: logical(1) True, use ceil instead floor compute output shape. Default: FALSE","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMaxPool -> PipeOpTorchMaxPool3D","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"PipeOpTorchMaxPool3D$new() PipeOpTorchMaxPool3D$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"","code":"PipeOpTorchMaxPool3D$new(   id = \"nn_max_pool3d\",   return_indices = FALSE,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"id (character(1)) Identifier resulting  object. return_indices (logical(1)) Whether return indices. TRUE, two output channels \"output\" \"indices\". param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"","code":"PipeOpTorchMaxPool3D$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_max_pool3d.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"3D Max Pooling — mlr_pipeops_nn_max_pool3d","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_max_pool3d\") pipeop #> PipeOp: <nn_max_pool3d> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(5)> #>             id    class lower upper nlevels        default  value #>         <char>   <char> <num> <num>   <num>         <list> <list> #> 1: kernel_size ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:     padding ParamUty    NA    NA     Inf              0 [NULL] #> 3:      stride ParamUty    NA    NA     Inf         [NULL] [NULL] #> 4:    dilation ParamInt  -Inf   Inf     Inf              1 [NULL] #> 5:   ceil_mode ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge Operation — mlr_pipeops_nn_merge","title":"Merge Operation — mlr_pipeops_nn_merge","text":"Base class merge operations addition (PipeOpTorchMergeSum), multiplication (PipeOpTorchMergeProd concatenation (PipeOpTorchMergeCat).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge Operation — mlr_pipeops_nn_merge","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge Operation — mlr_pipeops_nn_merge","text":"PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge Operation — mlr_pipeops_nn_merge","text":"See respective child class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge Operation — mlr_pipeops_nn_merge","text":"Per default, private$.shapes_out() method outputs broadcasted tensors. two things aware: NAs assumed batch (almost always batch size first dimension). Tensors expected number dimensions, .e. missing dimensions filled 1s. reason first dimension batch dimension. private method can overwritten PipeOpTorchs inheriting class.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge Operation — mlr_pipeops_nn_merge","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchMerge","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge Operation — mlr_pipeops_nn_merge","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge Operation — mlr_pipeops_nn_merge","text":"PipeOpTorchMerge$new() PipeOpTorchMerge$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge Operation — mlr_pipeops_nn_merge","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge Operation — mlr_pipeops_nn_merge","text":"","code":"PipeOpTorchMerge$new(   id,   module_generator,   param_set = ps(),   innum = 0,   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge Operation — mlr_pipeops_nn_merge","text":"id (character(1)) Identifier resulting  object. module_generator (nn_module_generator) torch module generator. param_set (ParamSet) parameter set. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge Operation — mlr_pipeops_nn_merge","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge Operation — mlr_pipeops_nn_merge","text":"","code":"PipeOpTorchMerge$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge Operation — mlr_pipeops_nn_merge","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Concatenates multiple tensors given dimension. broadcasting rules applied , must reshape tensors shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch. PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"dim :: integer(1) dimension along concatenate tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Calls nn_merge_cat() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMerge -> PipeOpTorchMergeCat","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"PipeOpTorchMergeCat$new() PipeOpTorchMergeCat$speak() PipeOpTorchMergeCat$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"PipeOpTorchMergeCat$new(id = \"nn_merge_cat\", innum = 0, param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"id (character(1)) Identifier resulting  object. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"method-speak-","dir":"Reference","previous_headings":"","what":"Method speak()","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"cat say?","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"PipeOpTorchMergeCat$speak()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"PipeOpTorchMergeCat$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_cat.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Merge by Concatenation — mlr_pipeops_nn_merge_cat","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_merge_cat\") pipeop #> PipeOp: <nn_merge_cat> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   ... [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:    dim ParamInt  -Inf   Inf     Inf      -1 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge by Product — mlr_pipeops_nn_merge_prod","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Calculates product input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch. PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Calls nn_merge_prod() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMerge -> PipeOpTorchMergeProd","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"PipeOpTorchMergeProd$new() PipeOpTorchMergeProd$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"","code":"PipeOpTorchMergeProd$new(id = \"nn_merge_prod\", innum = 0, param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"id (character(1)) Identifier resulting  object. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"","code":"PipeOpTorchMergeProd$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_prod.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Merge by Product — mlr_pipeops_nn_merge_prod","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_merge_prod\") pipeop #> PipeOp: <nn_merge_prod> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   ... [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":null,"dir":"Reference","previous_headings":"","what":"Merge by Summation — mlr_pipeops_nn_merge_sum","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Calculates sum input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch. PipeOpTorchMerges either vararg input channel constructor argument innum set, input channels \"input1\", ..., \"input<innum>\". one output channel \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Calls nn_merge_sum() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> mlr3torch::PipeOpTorchMerge -> PipeOpTorchMergeSum","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"PipeOpTorchMergeSum$new() PipeOpTorchMergeSum$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"","code":"PipeOpTorchMergeSum$new(id = \"nn_merge_sum\", innum = 0, param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"id (character(1)) Identifier resulting  object. innum (integer(1)) number inputs. Default 0 means one vararg input channel. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"","code":"PipeOpTorchMergeSum$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_merge_sum.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Merge by Summation — mlr_pipeops_nn_merge_sum","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_merge_sum\") pipeop #> PipeOp: <nn_merge_sum> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   ... [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":null,"dir":"Reference","previous_headings":"","what":"PReLU Activation Function — mlr_pipeops_nn_prelu","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Applies element-wise function \\(PReLU(x) = max(0,x) + weight * min(0,x)\\) weight learnable parameter.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"num_parameters :: integer(1): Number learn. Although takes int input, two values legitimate: 1, number channels input. Default: 1. init :: numeric(1) T initial value . Default: 0.25.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Calls torch::nn_prelu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchPReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"PipeOpTorchPReLU$new() PipeOpTorchPReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"","code":"PipeOpTorchPReLU$new(id = \"nn_prelu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"","code":"PipeOpTorchPReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_prelu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"PReLU Activation Function — mlr_pipeops_nn_prelu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_prelu\") pipeop #> PipeOp: <nn_prelu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>                id    class lower upper nlevels default  value #>            <char>   <char> <num> <num>   <num>  <list> <list> #> 1: num_parameters ParamInt     1   Inf     Inf       1 [NULL] #> 2:           init ParamDbl  -Inf   Inf     Inf    0.25 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":null,"dir":"Reference","previous_headings":"","what":"ReLU Activation Function — mlr_pipeops_nn_relu","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Applies rectified linear unit function element-wise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Calls torch::nn_relu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"PipeOpTorchReLU$new() PipeOpTorchReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"","code":"PipeOpTorchReLU$new(id = \"nn_relu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"","code":"PipeOpTorchReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"ReLU Activation Function — mlr_pipeops_nn_relu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_relu\") pipeop #> PipeOp: <nn_relu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":null,"dir":"Reference","previous_headings":"","what":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Applies element-wise function \\(ReLU6(x) = min(max(0,x), 6)\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Calls torch::nn_relu6() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchReLU6","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"PipeOpTorchReLU6$new() PipeOpTorchReLU6$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"","code":"PipeOpTorchReLU6$new(id = \"nn_relu6\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"","code":"PipeOpTorchReLU6$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_relu6.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"ReLU6 Activation Function — mlr_pipeops_nn_relu6","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_relu6\") pipeop #> PipeOp: <nn_relu6> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":null,"dir":"Reference","previous_headings":"","what":"Reshape a Tensor — mlr_pipeops_nn_reshape","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Reshape tensor given shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"shape :: integer(1) desired output shape. Unknown dimension (one ) can either specified -1 NA.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Calls nn_reshape() trained. internally calls torch::torch_reshape() given shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchReshape","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"PipeOpTorchReshape$new() PipeOpTorchReshape$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"","code":"PipeOpTorchReshape$new(id = \"nn_reshape\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"","code":"PipeOpTorchReshape$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_reshape.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Reshape a Tensor — mlr_pipeops_nn_reshape","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_reshape\") pipeop #> PipeOp: <nn_reshape> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <num> <num>   <num>         <list> <list> #> 1:  shape ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":null,"dir":"Reference","previous_headings":"","what":"RReLU Activation Function — mlr_pipeops_nn_rrelu","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Randomized leaky ReLU.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"lower:: numeric(1) Lower bound uniform distribution. Default: 1/8. upper:: numeric(1) Upper bound uniform distribution. Default: 1/3. inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Calls torch::nn_rrelu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchRReLU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"PipeOpTorchRReLU$new() PipeOpTorchRReLU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"","code":"PipeOpTorchRReLU$new(id = \"nn_rrelu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"","code":"PipeOpTorchRReLU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_rrelu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"RReLU Activation Function — mlr_pipeops_nn_rrelu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_rrelu\") pipeop #> PipeOp: <nn_rrelu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>         id    class lower upper nlevels   default  value #>     <char>   <char> <num> <num>   <num>    <list> <list> #> 1:   lower ParamDbl  -Inf   Inf     Inf     0.125 [NULL] #> 2:   upper ParamDbl  -Inf   Inf     Inf 0.3333333 [NULL] #> 3: inplace ParamLgl    NA    NA       2     FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":null,"dir":"Reference","previous_headings":"","what":"SELU Activation Function — mlr_pipeops_nn_selu","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Applies element-wise, $$SELU(x) = scale * (max(0,x) + min(0, \\alpha * (exp(x) - 1)))$$, \\(\\alpha=1.6732632423543772848170429916717\\) \\(scale=1.0507009873554804934193349852946\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"inplace :: logical(1) Whether operation -place. Default: FALSE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Calls torch::nn_selu() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSELU","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"PipeOpTorchSELU$new() PipeOpTorchSELU$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"","code":"PipeOpTorchSELU$new(id = \"nn_selu\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"","code":"PipeOpTorchSELU$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_selu.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"SELU Activation Function — mlr_pipeops_nn_selu","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_selu\") pipeop #> PipeOp: <nn_selu> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>         id    class lower upper nlevels default  value #>     <char>   <char> <num> <num>   <num>  <list> <list> #> 1: inplace ParamLgl    NA    NA       2   FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":null,"dir":"Reference","previous_headings":"","what":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Applies element-wise \\(Sigmoid(x_i) = \\frac{1}{1 + exp(-x_i)}\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Calls torch::nn_sigmoid() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSigmoid","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"PipeOpTorchSigmoid$new() PipeOpTorchSigmoid$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"","code":"PipeOpTorchSigmoid$new(id = \"nn_sigmoid\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"","code":"PipeOpTorchSigmoid$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_sigmoid.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Sigmoid Activation Function — mlr_pipeops_nn_sigmoid","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_sigmoid\") pipeop #> PipeOp: <nn_sigmoid> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":null,"dir":"Reference","previous_headings":"","what":"Softmax — mlr_pipeops_nn_softmax","title":"Softmax — mlr_pipeops_nn_softmax","text":"Applies softmax function.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Softmax — mlr_pipeops_nn_softmax","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Softmax — mlr_pipeops_nn_softmax","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Softmax — mlr_pipeops_nn_softmax","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Softmax — mlr_pipeops_nn_softmax","text":"dim :: integer(1) dimension along Softmax computed (every slice along dim sum 1).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Softmax — mlr_pipeops_nn_softmax","text":"Calls torch::nn_softmax() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Softmax — mlr_pipeops_nn_softmax","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftmax","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Softmax — mlr_pipeops_nn_softmax","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Softmax — mlr_pipeops_nn_softmax","text":"PipeOpTorchSoftmax$new() PipeOpTorchSoftmax$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Softmax — mlr_pipeops_nn_softmax","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Softmax — mlr_pipeops_nn_softmax","text":"","code":"PipeOpTorchSoftmax$new(id = \"nn_softmax\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Softmax — mlr_pipeops_nn_softmax","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Softmax — mlr_pipeops_nn_softmax","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Softmax — mlr_pipeops_nn_softmax","text":"","code":"PipeOpTorchSoftmax$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Softmax — mlr_pipeops_nn_softmax","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softmax.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Softmax — mlr_pipeops_nn_softmax","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softmax\") pipeop #> PipeOp: <nn_softmax> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <int> <num>   <num>         <list> <list> #> 1:    dim ParamInt     1   Inf     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":null,"dir":"Reference","previous_headings":"","what":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Applies element-wise, function \\(Softplus(x) = 1/\\beta * log(1 + exp(\\beta * x))\\).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"beta :: numeric(1) beta value Softplus formulation. Default: 1 threshold :: numeric(1) Values revert linear function. Default: 20","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Calls torch::nn_softplus() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftPlus","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"PipeOpTorchSoftPlus$new() PipeOpTorchSoftPlus$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"","code":"PipeOpTorchSoftPlus$new(id = \"nn_softplus\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"","code":"PipeOpTorchSoftPlus$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softplus.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"SoftPlus Activation Function — mlr_pipeops_nn_softplus","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softplus\") pipeop #> PipeOp: <nn_softplus> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(2)> #>           id    class lower upper nlevels default  value #>       <char>   <char> <num> <num>   <num>  <list> <list> #> 1:      beta ParamDbl  -Inf   Inf     Inf       1 [NULL] #> 2: threshold ParamDbl  -Inf   Inf     Inf      20 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":null,"dir":"Reference","previous_headings":"","what":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Applies soft shrinkage function elementwise","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"lamd :: numeric(1) lambda (must less zero) value Softshrink formulation. Default: 0.5","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Calls torch::nn_softshrink() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftShrink","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"PipeOpTorchSoftShrink$new() PipeOpTorchSoftShrink$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"","code":"PipeOpTorchSoftShrink$new(id = \"nn_softshrink\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"","code":"PipeOpTorchSoftShrink$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softshrink.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Soft Shrink Activation Function — mlr_pipeops_nn_softshrink","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softshrink\") pipeop #> PipeOp: <nn_softshrink> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels default  value #>    <char>   <char> <num> <num>   <num>  <list> <list> #> 1:  lambd ParamDbl  -Inf     1     Inf     0.5 [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":null,"dir":"Reference","previous_headings":"","what":"SoftSign Activation Function — mlr_pipeops_nn_softsign","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Applies element-wise, function \\(SoftSign(x) = x/(1 + |x|\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Calls torch::nn_softsign() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSoftSign","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"PipeOpTorchSoftSign$new() PipeOpTorchSoftSign$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"","code":"PipeOpTorchSoftSign$new(id = \"nn_softsign\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"","code":"PipeOpTorchSoftSign$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_softsign.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"SoftSign Activation Function — mlr_pipeops_nn_softsign","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_softsign\") pipeop #> PipeOp: <nn_softsign> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Squeezes tensor calling torch::torch_squeeze() given dimension dim.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"dim :: integer(1) dimension squeeze. NULL, dimensions size 1 squeezed. Negative values interpreted downwards last dimension.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Calls nn_squeeze() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchSqueeze","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"PipeOpTorchSqueeze$new() PipeOpTorchSqueeze$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"","code":"PipeOpTorchSqueeze$new(id = \"nn_squeeze\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"","code":"PipeOpTorchSqueeze$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_squeeze.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Squeeze a Tensor — mlr_pipeops_nn_squeeze","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_squeeze\") pipeop #> PipeOp: <nn_squeeze> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <num> <num>   <num>         <list> <list> #> 1:    dim ParamUty    NA    NA     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":null,"dir":"Reference","previous_headings":"","what":"Tanh Activation Function — mlr_pipeops_nn_tanh","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Applies element-wise function:","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Calls torch::nn_tanh() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchTanh","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"PipeOpTorchTanh$new() PipeOpTorchTanh$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"","code":"PipeOpTorchTanh$new(id = \"nn_tanh\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"","code":"PipeOpTorchTanh$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanh.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tanh Activation Function — mlr_pipeops_nn_tanh","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_tanh\") pipeop #> PipeOp: <nn_tanh> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":null,"dir":"Reference","previous_headings":"","what":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Applies element-wise, \\(Tanhshrink(x) = x - Tanh(x)\\)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"parameters.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Calls torch::nn_tanhshrink() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchTanhShrink","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"PipeOpTorchTanhShrink$new() PipeOpTorchTanhShrink$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"","code":"PipeOpTorchTanhShrink$new(id = \"nn_tanhshrink\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"","code":"PipeOpTorchTanhShrink$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_tanhshrink.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tanh Shrink Activation Function — mlr_pipeops_nn_tanhshrink","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_tanhshrink\") pipeop #> PipeOp: <nn_tanhshrink> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(0)> #> Empty."},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":null,"dir":"Reference","previous_headings":"","what":"Treshold Activation Function — mlr_pipeops_nn_threshold","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Thresholds element input Tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"threshold :: numeric(1) value threshold . value :: numeric(1) value replace . inplace :: logical(1) Can optionally operation -place. Default: ‘FALSE’.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Calls torch::nn_threshold() trained.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchThreshold","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"PipeOpTorchThreshold$new() PipeOpTorchThreshold$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"","code":"PipeOpTorchThreshold$new(id = \"nn_threshold\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"","code":"PipeOpTorchThreshold$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_threshold.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Treshold Activation Function — mlr_pipeops_nn_threshold","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_threshold\", threshold = 1, value = 2) pipeop #> PipeOp: <nn_threshold> (not trained) #> values: <threshold=1, value=2> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(3)> #>           id    class lower upper nlevels        default  value #>       <char>   <char> <num> <num>   <num>         <list> <list> #> 1: threshold ParamDbl  -Inf   Inf     Inf <NoDefault[0]>      1 #> 2:     value ParamDbl  -Inf   Inf     Inf <NoDefault[0]>      2 #> 3:   inplace ParamLgl    NA    NA       2          FALSE [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Unqueeze Tensor Unqueeze Tensor","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"state value calculated public method $shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"credit","dir":"Reference","previous_headings":"","what":"Credit","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Part documentation copied adapted documentation torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"dim :: integer(1) dimension unsqueeze. Negative values interpreted downwards last dimension.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Calls nn_unsqueeze() trained. internally calls torch::torch_unsqueeze().","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorch -> PipeOpTorchUnsqueeze","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train() mlr3torch::PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"PipeOpTorchUnsqueeze$new() PipeOpTorchUnsqueeze$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"","code":"PipeOpTorchUnsqueeze$new(id = \"nn_unsqueeze\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"","code":"PipeOpTorchUnsqueeze$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_nn_unsqueeze.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Unqueeze a Tensor — mlr_pipeops_nn_unsqueeze","text":"","code":"# Construct the PipeOp pipeop = po(\"nn_unsqueeze\") pipeop #> PipeOp: <nn_unsqueeze> (not trained) #> values: <list()> #> Input channels <name [train type, predict type]>: #>   input [ModelDescriptor,Task] #> Output channels <name [train type, predict type]>: #>   output [ModelDescriptor,Task] # The available parameters pipeop$param_set #> <ParamSet(1)> #>        id    class lower upper nlevels        default  value #>    <char>   <char> <num> <num>   <num>         <list> <list> #> 1:    dim ParamInt  -Inf   Inf     Inf <NoDefault[0]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"PipeOp can used preprocess (one ) lazy_tensor columns contained mlr3::Task. preprocessing function specified construction argument fn additional arguments function can defined PipeOp's parameter set. preprocessing done per column, .e. number lazy tensor output columns equal number lazy tensor input columns. create custom preprocessing PipeOps can use pipeop_preproc_torch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"addition specifying construction arguments, can overwrite private .shapes_out() method. overwrite , output shapes assumed unknown (NULL). .shapes_out(shapes_in, param_vals, task) (list(), list(), TaskorNULL) -> list()\\cr private method calculates output shapes lazy tensor columns created applying preprocessing function provided parameter values (param_vals). taskis rarely needed, checked notNULL`. private method responsibility calculate output shapes one input column, .e. input shapes_in can assumed exactly one shape vector must calculate output shapes return list() length 1. can also assumed shape NULL (.e. unknown). Also, first dimension can NA, .e. unknown (batch dimension).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"See PipeOpTaskPreproc.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"addition state elements PipeOpTaskPreprocSimple, state also contains $param_vals set training.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"addition parameters inherited PipeOpTaskPreproc well specified construction argument param_set following parameters: stages :: character(1) stages apply preprocessing. Can one \"train\", \"predict\" \"\". initial value parameter set \"train\" PipeOp's id starts \"augment_\" \"\" otherwise. Note preprocessing applied $predict() uses parameters set $train() set performing prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"$train() / $predict(), PipeOpModule one input one output channel created. pipeop applies function fn input tensor additionally passing parameter values (minus stages affect_columns) fn. preprocessing graph lazy tensor columns shallowly cloned PipeOpModule added. done avoid modifying user input means identical PipeOpModules can part different preprocessing graphs. possible, created PipeOpModule stateless. later point graph, preprocessing graphs merged possible avoid unnecessary computation. best illustrated example: One lazy tensor column's preprocessing graph -> B. , two branches created B -> C B -> D, creating two preprocessing graphs -> B -> C -> B -> D. loading data, want run preprocessing , .e. want run -> B part twice. reason, task_dataset() try merge graphs cache results graphs. However, graphs using dataset can currently merged. Also, shapes created $train() $predict() might differ. avoid creation graphs predict shapes incompatible train shapes, hypothetical predict shapes already calculated $train() (parameters set train also used predict) PipeOpTorchModel check train predict shapes compatibility starting training. Otherwise, mechanism similar ModelDescriptor construct.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpTaskPreproc -> PipeOpTaskPreprocTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"fn preprocessing function. rowwise Whether preprocessing applied rowwise.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"PipeOpTaskPreprocTorch$new() PipeOpTaskPreprocTorch$shapes_out() PipeOpTaskPreprocTorch$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"PipeOpTaskPreprocTorch$new(   fn,   id = \"preproc_torch\",   param_vals = list(),   param_set = ps(),   packages = character(0),   rowwise = FALSE,   stages_init = NULL,   tags = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"fn (function character(2)) preprocessing function. Must modify input -place. character(2), first element namespace second element name. preprocessing function applied tensor, tensor passed position first argument. param_set inferred (left NULL) assumed first argument torch_tensor. id (character(1)) id new object. param_vals (named list()) Parameter values set construction. param_set (ParamSet) case function fn takes additional parameter besides torch_tensor can specfied parameters. None parameters can \"predict\" tag. tags include \"train\". packages (character()) packages preprocessing function depends . rowwise (logical(1)) Whether preprocessing function applied rowwise (concatenated row) directly whole tensor. first case batch dimension. stages_init (character(1)) Initial value stages parameter. tags (character()) Tags pipeop.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"method-shapes-out-","dir":"Reference","previous_headings":"","what":"Method shapes_out()","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"Calculates output shapes result applying preprocessing one lazy tensor columns provided shape. Names ignored order matters. uses parameter values currently set.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"PipeOpTaskPreprocTorch$shapes_out(shapes_in, stage = NULL, task = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"shapes_in (list() (integer() NULL)) input input shapes lazy tensors. NULL indicates shape unknown. First dimension must NA (NULL). stage (character(1)) stage: either \"train\" \"predict\". task (Task NULL) task, rarely needed.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"list() (integer() NULL)","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"usage-2","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"PipeOpTaskPreprocTorch$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"arguments-2","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_preproc_torch.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Base Class for Lazy Tensor Preprocessing — mlr_pipeops_preproc_torch","text":"","code":"# Creating a simple task d = data.table(   x1 = as_lazy_tensor(rnorm(10)),   x2 = as_lazy_tensor(rnorm(10)),   x3 = as_lazy_tensor(as.double(1:10)),   y = rnorm(10) )  taskin = as_task_regr(d, target = \"y\")  # Creating a simple preprocessing pipeop po_simple = po(\"preproc_torch\",   # get rid of environment baggage   fn = mlr3misc::crate(function(x, a) x + a),   param_set = paradox::ps(a = paradox::p_int(tags = c(\"train\", \"required\"))) )  po_simple$param_set$set_values(   a = 100,   affect_columns = selector_name(c(\"x1\", \"x2\")),   stages = \"both\" # use during train and predict )  taskout_train = po_simple$train(list(taskin))[[1L]] materialize(taskout_train$data(cols = c(\"x1\", \"x2\")), rbind = TRUE) #> $x1 #> torch_tensor #>  100.6242 #>   99.3046 #>   99.2164 #>   99.0469 #>  101.7928 #>  100.3490 #>  100.2591 #>   99.1940 #>  100.1057 #>   99.6664 #> [ CPUFloatType{10,1} ] #>  #> $x2 #> torch_tensor #>  101.6418 #>   99.3561 #>  100.5870 #>   99.8496 #>   98.2892 #>  101.4310 #>   97.3548 #>   98.9675 #>   99.2925 #>   99.2994 #> [ CPUFloatType{10,1} ] #>   taskout_predict_noaug = po_simple$predict(list(taskin))[[1L]] materialize(taskout_predict_noaug$data(cols = c(\"x1\", \"x2\")), rbind = TRUE) #> $x1 #> torch_tensor #>  100.6242 #>   99.3046 #>   99.2164 #>   99.0469 #>  101.7928 #>  100.3490 #>  100.2591 #>   99.1940 #>  100.1057 #>   99.6664 #> [ CPUFloatType{10,1} ] #>  #> $x2 #> torch_tensor #>  101.6418 #>   99.3561 #>  100.5870 #>   99.8496 #>   98.2892 #>  101.4310 #>   97.3548 #>   98.9675 #>   99.2925 #>   99.2994 #> [ CPUFloatType{10,1} ] #>   po_simple$param_set$set_values(   stages = \"train\" )  # transformation is not applied taskout_predict_aug = po_simple$predict(list(taskin))[[1L]] materialize(taskout_predict_aug$data(cols = c(\"x1\", \"x2\")), rbind = TRUE) #> $x1 #> torch_tensor #>  100.6242 #>   99.3046 #>   99.2164 #>   99.0469 #>  101.7928 #>  100.3490 #>  100.2591 #>   99.1940 #>  100.1057 #>   99.6664 #> [ CPUFloatType{10,1} ] #>  #> $x2 #> torch_tensor #>  101.6418 #>   99.3561 #>  100.5870 #>   99.8496 #>   98.2892 #>  101.4310 #>   97.3548 #>   98.9675 #>   99.2925 #>   99.2994 #> [ CPUFloatType{10,1} ] #>   # Creating a more complex preprocessing PipeOp PipeOpPreprocTorchPoly = R6::R6Class(\"PipeOpPreprocTorchPoly\",  inherit = PipeOpTaskPreprocTorch,  public = list(    initialize = function(id = \"preproc_poly\", param_vals = list()) {      param_set = paradox::ps(        n_degree = paradox::p_int(lower = 1L, tags = c(\"train\", \"required\"))      )      param_set$set_values(        n_degree = 1L      )      fn = mlr3misc::crate(function(x, n_degree) {        torch::torch_cat(          lapply(seq_len(n_degree), function(d) torch::torch_pow(x, d)),          dim = 2L        )      })       super$initialize(        fn = fn,        id = id,        packages = character(0),        param_vals = param_vals,        param_set = param_set,        stages_init = \"both\"      )    }  ),  private = list(    .shapes_out = function(shapes_in, param_vals, task) {      # shapes_in is a list of length 1 containing the shapes      checkmate::assert_true(length(shapes_in[[1L]]) == 2L)      if (shapes_in[[1L]][2L] != 1L) {        stop(\"Input shape must be (NA, 1)\")      }      list(c(NA, param_vals$n_degree))    }  ) )  po_poly = PipeOpPreprocTorchPoly$new(   param_vals = list(n_degree = 3L, affect_columns = selector_name(\"x3\")) )  po_poly$shapes_out(list(c(NA, 1L)), stage = \"train\") #> [[1]] #> [1] NA  3 #>   taskout = po_poly$train(list(taskin))[[1L]] materialize(taskout$data(cols = \"x3\"), rbind = TRUE) #> $x3 #> torch_tensor #>     1     1     1 #>     2     4     8 #>     3     9    27 #>     4    16    64 #>     5    25   125 #>     6    36   216 #>     7    49   343 #>     8    64   512 #>     9    81   729 #>    10   100  1000 #> [ CPUFloatType{10,3} ] #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"PipeOpTorch base class PipeOps represent neural network layers Graph. training, generates PipeOpModule wraps nn_module attaches architecture, also represented Graph consisting mostly PipeOpModules PipeOpNOPs. former Graph operates ModelDescriptors, latter operates tensors. relationship PipeOpTorch PipeOpModule similar relationshop nn_module_generator (like nn_linear) nn_module (like output nn_linear(...)). crucial difference PipeOpTorch infers auxiliary parameters (like in_features nn_linear) automatically intermediate tensor shapes communicated ModelDescriptor. prediction, PipeOpTorch takes Task channel outputs new Task resulting feature union channel. one input output channel, task simply piped .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"inheriting","dir":"Reference","previous_headings":"","what":"Inheriting","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"inheriting class, one overload either private$.shapes_out() private$.shape_dependent_params() methods, overload private$.make_module(). .make_module(shapes_in, param_vals, task) (list(), list()) -> nn_module private method called generated nn_module passed argument module PipeOpModule. must overwritten, module_generator provided. left , calls provided module_generator arguments obtained private method .shape_dependent_params(). .shapes_out(shapes_in, param_vals, task) (list(), list(), Task NULL) -> named list() private method gets list numeric vectors (shapes_in), parameter values (param_vals), well (optional) Task. shapes_in can assumed order input names PipeOp. output shapes must order output names PipeOp. case output shapes depends task (case PipeOpTorchHead), function return valid output shapes (possibly containing NAs) task argument provided . .shape_dependent_params(shapes_in, param_vals, task) (list(), list()) -> named list() private method inputs .shapes_out. .make_module() overwritten, constructs arguments passed module_generator. Usually means must infer auxiliary parameters can inferred input shapes add user-supplied parameter values (param_vals).","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"training, inputs outputs class ModelDescriptor. prediction, input output channels class Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"ParamSet specified child class inheriting PipeOpTorch. Usually parameters arguments wrapped nn_module minus auxiliary parameter can automatically inferred shapes input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"training, PipeOpTorch creates PipeOpModule given parameter specification input shapes incoming ModelDescriptors using private method .make_module(). input shapes provided slot pointer_shape incoming ModelDescriptors. channel names PipeOpModule identical channel names generating PipeOpTorch. model descriptor union incoming ModelDescriptors created. Note modifies graph first ModelDescriptor place efficiency. PipeOpModule added graph slot union edges connect sending PipeOpModules input channel PipeOpModule addeded graph. possible every incoming ModelDescriptor contains information id channel name sending PipeOp slot pointer. new graph model_descriptor_union represents current state neural network architecture. structurally similar subgraph consists pipeops class PipeOpTorch PipeOpTorchIngress ancestors PipeOpTorch. output, shallow copy ModelDescriptor created pointer pointer_shape updated accordingly. shallow copy means ModelDescriptors point Graph allows graph modified -reference different parts code.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"mlr3pipelines::PipeOp -> PipeOpTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"public-fields","dir":"Reference","previous_headings":"","what":"Public fields","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"module_generator (nn_module_generator NULL) module generator wrapped PipeOpTorch. NULL, private method private$.make_module(shapes_in, param_vals) must overwritte, see section 'Inheriting'. change construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"PipeOpTorch$new() PipeOpTorch$shapes_out()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"","code":"PipeOpTorch$new(   id,   module_generator,   param_set = ps(),   param_vals = list(),   inname = \"input\",   outname = \"output\",   packages = \"torch\",   tags = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"id (character(1)) Identifier resulting  object. module_generator (nn_module_generator) torch module generator. param_set (ParamSet) parameter set. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction. inname (character()) names PipeOp's input channels. input channels generated PipeOpModule. Unless wrapped module_generator's forward method (present) argument ..., inname must identical argument names order avoid ambiguity. forward method argument ..., order input channels determines tensors passed wrapped nn_module. left NULL (default), argument module_generator must given argument names modue_generator's forward function set inname. outname (character())  names output channels channels. ouput channels generated PipeOpModule therefore also names list returned $train(). case one output channel, nn_module constructed PipeOp training must return named list(), names list names output channels. default \"output\". packages (character()) R packages object depends . tags (character()) tags PipeOp. tags \"torch\" always added.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"method-shapes-out-","dir":"Reference","previous_headings":"","what":"Method shapes_out()","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"Calculates output shapes given input shapes, parameters task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"","code":"PipeOpTorch$shapes_out(shapes_in, task = NULL)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"shapes_in (list() integer()) input input shapes, must order input channel names PipeOp. task (Task NULL) task, rarely used (default NULL). exception PipeOpTorchHead.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"returns","dir":"Reference","previous_headings":"","what":"Returns","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"named list() containing output shapes. names names output channels PipeOp.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Base Class for Torch Module Constructor Wrappers — mlr_pipeops_torch","text":"","code":"## Creating a neural network # In torch  task = tsk(\"iris\")  network_generator = torch::nn_module(   initialize = function(task, d_hidden) {     d_in = length(task$feature_names)     self$linear = torch::nn_linear(d_in, d_hidden)     self$output = if (task$task_type == \"regr\") {       torch::nn_linear(d_hidden, 1)     } else if (task$task_type == \"classif\") {       torch::nn_linear(d_hidden, length(task$class_names))     }   },   forward = function(x) {     x = self$linear(x)     x = torch::nnf_relu(x)     self$output(x)   } )  network = network_generator(task, d_hidden = 50) x = torch::torch_tensor(as.matrix(task$data(1, task$feature_names))) y = torch::with_no_grad(network(x))   # In mlr3torch network_generator = po(\"torch_ingress_num\") %>>%   po(\"nn_linear\", out_features = 50) %>>%   po(\"nn_head\") md = network_generator$train(task)[[1L]] network = model_descriptor_to_module(md) y = torch::with_no_grad(network(torch_ingress_num.input = x))    ## Implementing a custom PipeOpTorch  # defining a custom module nn_custom = nn_module(\"nn_custom\",   initialize = function(d_in1, d_in2, d_out1, d_out2, bias = TRUE) {     self$linear1 = nn_linear(d_in1, d_out1, bias)     self$linear2 = nn_linear(d_in2, d_out2, bias)   },   forward = function(input1, input2) {     output1 = self$linear1(input1)     output2 = self$linear1(input2)      list(output1 = output1, output2 = output2)   } )  # wrapping the module into a custom PipeOpTorch  library(paradox)  PipeOpTorchCustom = R6::R6Class(\"PipeOpTorchCustom\",   inherit = PipeOpTorch,   public = list(     initialize = function(id = \"nn_custom\", param_vals = list()) {       param_set = ps(         d_out1 = p_int(lower = 1, tags = c(\"required\", \"train\")),         d_out2 = p_int(lower = 1, tags = c(\"required\", \"train\")),         bias = p_lgl(default = TRUE, tags = \"train\")       )       super$initialize(         id = id,         param_vals = param_vals,         param_set = param_set,         inname = c(\"input1\", \"input2\"),         outname = c(\"output1\", \"output2\"),         module_generator = nn_custom       )     }   ),   private = list(     .shape_dependent_params = function(shapes_in, param_vals, task) {       c(param_vals,         list(d_in1 = tail(shapes_in[[\"input1\"]], 1)), d_in2 = tail(shapes_in[[\"input2\"]], 1)       )     },     .shapes_out = function(shapes_in, param_vals, task) {       list(         input1 = c(head(shapes_in[[\"input1\"]], -1), param_vals$d_out1),         input2 = c(head(shapes_in[[\"input2\"]], -1), param_vals$d_out2)       )     }   ) )  ## Training  # generate input task = tsk(\"iris\") task1 = task$clone()$select(paste0(\"Sepal.\", c(\"Length\", \"Width\"))) task2 = task$clone()$select(paste0(\"Petal.\", c(\"Length\", \"Width\"))) graph = gunion(list(po(\"torch_ingress_num_1\"), po(\"torch_ingress_num_2\"))) mds_in = graph$train(list(task1, task2), single_input = FALSE)  mds_in[[1L]][c(\"graph\", \"task\", \"ingress\", \"pointer\", \"pointer_shape\")] #> $graph #> Graph with 1 PipeOps: #>                   ID         State sccssors prdcssors #>               <char>        <char>   <char>    <char> #>  torch_ingress_num_1 <<UNTRAINED>>                    #>  #> $task #> <TaskClassif:iris> (150 x 3): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (2): #>   - dbl (2): Sepal.Length, Sepal.Width #>  #> $ingress #> $ingress$torch_ingress_num_1.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) #>  #>  #> $pointer #> [1] \"torch_ingress_num_1\" \"output\"              #>  #> $pointer_shape #> [1] NA  2 #>  mds_in[[2L]][c(\"graph\", \"task\", \"ingress\", \"pointer\", \"pointer_shape\")] #> $graph #> Graph with 1 PipeOps: #>                   ID         State sccssors prdcssors #>               <char>        <char>   <char>    <char> #>  torch_ingress_num_2 <<UNTRAINED>>                    #>  #> $task #> <TaskClassif:iris> (150 x 3): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (2): #>   - dbl (2): Petal.Length, Petal.Width #>  #> $ingress #> $ingress$torch_ingress_num_2.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) #>  #>  #> $pointer #> [1] \"torch_ingress_num_2\" \"output\"              #>  #> $pointer_shape #> [1] NA  2 #>   # creating the PipeOpTorch and training it po_torch = PipeOpTorchCustom$new() po_torch$param_set$values = list(d_out1 = 10, d_out2 = 20) train_input = list(input1 = mds_in[[1L]], input2 = mds_in[[2L]]) mds_out = do.call(po_torch$train, args = list(input = train_input)) po_torch$state #> $output1 #> [1] NA 10 #>  #> $output2 #> [1] NA 20 #>   # the new model descriptors  # the resulting graphs are identical identical(mds_out[[1L]]$graph, mds_out[[2L]]$graph) #> [1] TRUE # not that as a side-effect, also one of the input graphs is modified in-place for efficiency mds_in[[1L]]$graph$edges #>                 src_id src_channel    dst_id dst_channel #>                 <char>      <char>    <char>      <char> #> 1: torch_ingress_num_1      output nn_custom      input1 #> 2: torch_ingress_num_2      output nn_custom      input2  # The new task has both Sepal and Petal features identical(mds_out[[1L]]$task, mds_out[[2L]]$task) #> [1] TRUE mds_out[[2L]]$task #> <TaskClassif:iris> (150 x 5): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (4): #>   - dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Width  # The new ingress slot contains all ingressors identical(mds_out[[1L]]$ingress, mds_out[[2L]]$ingress) #> [1] TRUE mds_out[[1L]]$ingress #> $torch_ingress_num_1.input #> Ingress: Task[Sepal.Length,Sepal.Width] --> Tensor(NA, 2) #>  #> $torch_ingress_num_2.input #> Ingress: Task[Petal.Length,Petal.Width] --> Tensor(NA, 2) #>   # The pointer and pointer_shape slots are different mds_out[[1L]]$pointer #> [1] \"nn_custom\" \"output1\"   mds_out[[2L]]$pointer #> [1] \"nn_custom\" \"output2\"    mds_out[[1L]]$pointer_shape #> [1] NA 10 mds_out[[2L]]$pointer_shape #> [1] NA 20  ## Prediction predict_input = list(input1 = task1, input2 = task2) tasks_out = do.call(po_torch$predict, args = list(input = predict_input)) identical(tasks_out[[1L]], tasks_out[[2L]]) #> [1] TRUE"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":null,"dir":"Reference","previous_headings":"","what":"Callback Configuration — mlr_pipeops_torch_callbacks","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"Configures callbacks deep learning model.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"one input channel \"input\" one output channel \"output\". training, channels class ModelDescriptor. prediction, channels class Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"parameters defined dynamically callbacks, id respective callbacks respective set id.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"training callbacks cloned added ModelDescriptor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"mlr3pipelines::PipeOp -> PipeOpTorchCallbacks","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"PipeOpTorchCallbacks$new() PipeOpTorchCallbacks$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"","code":"PipeOpTorchCallbacks$new(   callbacks = list(),   id = \"torch_callbacks\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"callbacks (list TorchCallbacks)  callbacks (something convertible via as_torch_callbacks()). Must unique ids. callbacks cloned construction. id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"","code":"PipeOpTorchCallbacks$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_callbacks.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Callback Configuration — mlr_pipeops_torch_callbacks","text":"","code":"po_cb = po(\"torch_callbacks\", \"checkpoint\") po_cb$param_set #> <ParamSetCollection(3)> #>                      id    class lower upper nlevels        default  value #>                  <char>   <char> <num> <num>   <num>         <list> <list> #> 1:      checkpoint.path ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:      checkpoint.freq ParamInt     1   Inf     Inf <NoDefault[0]> [NULL] #> 3: checkpoint.freq_type ParamFct    NA    NA       2          epoch [NULL] mdin = po(\"torch_ingress_num\")$train(list(tsk(\"iris\"))) mdin[[1L]]$callbacks #> named list() mdout = po_cb$train(mdin)[[1L]] mdout$callbacks #> $checkpoint #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch #>  # Can be called again po_cb1 = po(\"torch_callbacks\", t_clbk(\"progress\")) mdout1 = po_cb1$train(list(mdout))[[1L]] mdout1$callbacks #> $progress #> <TorchCallback:progress> Progress #> * Generator: CallbackSetProgress #> * Parameters: list() #> * Packages: progress,mlr3torch,torch #>  #> $checkpoint #> <TorchCallback:checkpoint> Checkpoint #> * Generator: CallbackSetCheckpoint #> * Parameters: list() #> * Packages: mlr3torch,torch #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":null,"dir":"Reference","previous_headings":"","what":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Use entry-point mlr3torch-networks. Unless advanced user, need use directly PipeOpTorchIngressNumeric, PipeOpTorchIngressCategorical PipeOpTorchIngressLazyTensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"state set input shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Defined construction argument param_set.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Creates object class TorchIngressToken given task. purpuse store information construct torch dataloader task entry point network.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"mlr3pipelines::PipeOp -> PipeOpTorchIngress","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"active-bindings","dir":"Reference","previous_headings":"","what":"Active bindings","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"feature_types (character(1)) features types can consumed PipeOpTorchIngress.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"PipeOpTorchIngress$new() PipeOpTorchIngress$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"","code":"PipeOpTorchIngress$new(   id,   param_set = ps(),   param_vals = list(),   packages = character(0),   feature_types )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"id (character(1)) Identifier resulting  object. param_set (ParamSet) parameter set. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction. packages (character()) R packages object depends . feature_types (character()) feature types. See mlr_reflections$task_feature_types available values, Additionally, \"lazy_tensor\" supported.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"","code":"PipeOpTorchIngress$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Entrypoint to Torch Network — mlr_pipeops_torch_ingress","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"Ingress PipeOp represents categorical (factor(), ordered() logical()) entry point torch network.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"select :: logical(1) Whether PipeOp selected supported feature types. Otherwise err receiving tasks unsupported feature types.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"Uses batchgetter_categ().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"state set input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorchIngress -> PipeOpTorchIngressCategorical","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"PipeOpTorchIngressCategorical$new() PipeOpTorchIngressCategorical$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"","code":"PipeOpTorchIngressCategorical$new(   id = \"torch_ingress_categ\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"","code":"PipeOpTorchIngressCategorical$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_categ.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Entry Point for Categorical Features — mlr_pipeops_torch_ingress_categ","text":"","code":"graph = po(\"select\", selector = selector_type(\"factor\")) %>>%   po(\"torch_ingress_categ\") task = tsk(\"german_credit\") # The output is a model descriptor md = graph$train(task)[[1L]] ingress = md$ingress[[1L]] ingress$batchgetter(task$data(1, ingress$features), \"cpu\") #> torch_tensor #>  5  5  1  2  3  1  3  1  3  1  4  5  1  2 #> [ CPULongType{1,14} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":null,"dir":"Reference","previous_headings":"","what":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"Ingress single lazy_tensor column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"shape :: integer() shape tensor, first dimension (batch) must NA. specified, lazy tensor input column needs known shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"returned batchgetter materializes lazy tensor column tensor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"state set input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorchIngress -> PipeOpTorchIngressLazyTensor","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"PipeOpTorchIngressLazyTensor$new() PipeOpTorchIngressLazyTensor$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"","code":"PipeOpTorchIngressLazyTensor$new(   id = \"torch_ingress_ltnsr\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"","code":"PipeOpTorchIngressLazyTensor$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_ltnsr.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Ingress for Lazy Tensor — mlr_pipeops_torch_ingress_ltnsr","text":"","code":"po_ingress = po(\"torch_ingress_ltnsr\") task = tsk(\"lazy_iris\")  md = po_ingress$train(list(task))[[1L]]  ingress = md$ingress x_batch = ingress[[1L]]$batchgetter(data = task$data(1, \"x\"), device = \"cpu\", cache = NULL) x_batch #> torch_tensor #>  5.1000  3.5000  1.4000  0.2000 #> [ CPUFloatType{1,4} ]  # Now we try a lazy tensor with unknown shape, i.e. the shapes between the rows can differ  ds = dataset(   initialize = function() self$x = list(torch_randn(3, 10, 10), torch_randn(3, 8, 8)),   .getitem = function(i) list(x = self$x[[i]]),   .length = function() 2)()  task_unknown = as_task_regr(data.table(   x = as_lazy_tensor(ds, dataset_shapes = list(x = NULL)),   y = rnorm(2) ), target = \"y\", id = \"example2\")  # this task (as it is) can NOT be processed by PipeOpTorchIngressLazyTensor # It therefore needs to be preprocessed po_resize = po(\"trafo_resize\", size = c(6, 6)) task_unknown_resize = po_resize$train(list(task_unknown))[[1L]]  # printing the transformed column still shows unknown shapes, # because the preprocessing pipeop cannot infer them, # however we know that the shape is now (3, 10, 10) for all rows task_unknown_resize$data(1:2, \"x\") #>                x #>    <lazy_tensor> #> 1:      <tnsr[]> #> 2:      <tnsr[]> po_ingress$param_set$set_values(shape = c(NA, 3, 6, 6))  md2 = po_ingress$train(list(task_unknown_resize))[[1L]]  ingress2 = md2$ingress x_batch2 = ingress2[[1L]]$batchgetter(   data = task_unknown_resize$data(1:2, \"x\"),   device = \"cpu\",   cache = NULL )  x_batch2 #> torch_tensor #> (1,1,.,.) =  #>   0.1683  0.9521 -0.3007 -0.6531 -0.5466 -0.2548 #>   0.8280  0.7439 -1.1441 -0.2175  0.6100 -1.8303 #>   0.6169  1.3783  0.2681 -0.2965 -0.6230  0.1022 #>  -0.4736  0.9308  0.0779  0.6800 -0.6026  0.1194 #>  -0.7146  0.7433 -0.7192  0.7677  0.5538  0.4407 #>   0.3748  0.1611 -0.8856 -0.3067 -0.1230  0.5929 #>  #> (2,1,.,.) =  #>  -0.1821 -0.2982  1.1122  0.6025 -0.0759 -0.8357 #>  -0.1558 -0.9088 -0.7307  0.6930 -0.1729  0.8293 #>   0.4416 -0.2946  0.4691  0.5564 -0.7042 -1.1576 #>   0.5102  0.1238 -0.2319  0.7806 -1.2194 -1.4314 #>   1.5757  0.0112 -1.0029  0.4919  0.4147 -0.0619 #>   1.1526  0.4710 -0.9762 -1.2955 -0.1328  0.0883 #>  #> (1,2,.,.) =  #>  -0.2915 -0.8742  0.2805 -0.3131 -0.0847  0.3623 #>   0.8970  1.4281  1.1198 -0.4452 -0.3298 -0.7621 #>  -0.6495  0.3189  0.3577  0.3509  0.4461  0.3858 #>  -0.2846 -1.6956 -0.5109  1.1558  0.6108  0.1983 #>   0.3220  0.4607  0.1320  0.0521 -1.6308 -0.7962 #>   0.9437  1.8788 -0.5748  0.0037 -0.1264 -0.8996 #>  #> (2,2,.,.) =  #>   0.2720  0.9749 -1.7862  0.7320 -0.4558 -0.2841 #>   0.6081  0.0487  0.1334 -0.1368 -0.8086 -0.4014 #>   0.7191  0.0231 -0.3324 -0.8849 -0.8758 -0.2848 #>  -1.2964  0.1189  0.2938  0.0605 -0.3840 -0.5428 #>   0.3190  0.1164 -0.5326  1.0797  0.4521 -0.6165 #> ... [the output was truncated (use n=-1 to disable)] #> [ CPUFloatType{2,3,6,6} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"Ingress PipeOp represents numeric (integer() numeric()) entry point torch network.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"Uses batchgetter_num().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"state set input shape.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"mlr3pipelines::PipeOp -> mlr3torch::PipeOpTorchIngress -> PipeOpTorchIngressNumeric","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"PipeOpTorchIngressNumeric$new() PipeOpTorchIngressNumeric$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"","code":"PipeOpTorchIngressNumeric$new(id = \"torch_ingress_num\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"","code":"PipeOpTorchIngressNumeric$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_ingress_num.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Entry Point for Numeric Features — mlr_pipeops_torch_ingress_num","text":"","code":"graph = po(\"select\", selector = selector_type(c(\"numeric\", \"integer\"))) %>>%   po(\"torch_ingress_num\") task = tsk(\"german_credit\") # The output is a model descriptor md = graph$train(task)[[1L]] ingress = md$ingress[[1L]] ingress$batchgetter(task$data(1:5, ingress$features), \"cpu\") #> torch_tensor #>    67  1169     6 #>    22  5951    48 #>    49  2096    12 #>    45  7882    42 #>    53  4870    24 #> [ CPUFloatType{5,3} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":null,"dir":"Reference","previous_headings":"","what":"Loss Configuration — mlr_pipeops_torch_loss","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"Configures loss deep learning model.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"One input channel called \"input\" one output channel called \"output\". explanation see PipeOpTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"parameters defined dynamically loss set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"training loss cloned added ModelDescriptor.","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"mlr3pipelines::PipeOp -> PipeOpTorchLoss","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"PipeOpTorchLoss$new() PipeOpTorchLoss$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"","code":"PipeOpTorchLoss$new(loss, id = \"torch_loss\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"loss (TorchLoss character(1) nn_loss) loss (something convertible via as_torch_loss()). id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"","code":"PipeOpTorchLoss$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_loss.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Loss Configuration — mlr_pipeops_torch_loss","text":"","code":"po_loss = po(\"torch_loss\", loss = t_loss(\"cross_entropy\")) po_loss$param_set #> <ParamSet(3)> #>              id    class lower upper nlevels default  value #>          <char>   <char> <num> <num>   <num>  <list> <list> #> 1:       weight ParamUty    NA    NA     Inf  [NULL] [NULL] #> 2: ignore_index ParamInt  -Inf   Inf     Inf    -100 [NULL] #> 3:    reduction ParamFct    NA    NA       2    mean [NULL] mdin = po(\"torch_ingress_num\")$train(list(tsk(\"iris\"))) mdin[[1L]]$loss #> NULL mdout = po_loss$train(mdin)[[1L]] mdout$loss #> <TorchLoss:cross_entropy> Cross Entropy #> * Generator: nn_cross_entropy_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: classif"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOp Torch Model — mlr_pipeops_torch_model","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"Builds Torch Learner ModelDescriptor trains given parameter specification. task type must specified construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"one input channel \"input\" takes ModelDescriptor traing Task specified task_type prediction. output NULL training Prediction given task_type prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"trained LearnerTorchModel.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"General: parameters optimizer, loss callbacks, prefixed \"opt.\", \"loss.\" \"cb.<callback id>.\" respectively, well : epochs :: integer(1) number epochs. device :: character(1) device. One \"auto\", \"cpu\", \"cuda\" values defined mlr_reflections$torch$devices. value initialized \"auto\", select \"cuda\" possible, try \"mps\" otherwise fall back \"cpu\". num_threads :: integer(1) number threads intraop pararallelization (device \"cpu\"). value initialized 1. num_interop_threads :: integer(1) number threads intraop interop pararallelization (device \"cpu\"). value initialized 1. Note can set session changing value within R session raise warning. seed :: integer(1) \"random\" NULL torch seed used training prediction. value initialized \"random\", means random seed sampled beginning training phase. seed (either set randomly sampled) available via $model$seed training used prediction. Note setting seed training phase mean default (.e. seed \"random\"), clones learner use different seed. set NULL, seeding done. Evaluation: measures_train :: Measure list() Measures. Measures evaluated training. measures_valid :: Measure list() Measures. Measures evaluated validation. eval_freq :: integer(1) often train / validation predictions evaluated using measures_train / measures_valid. initialized 1. Note final model always evaluated. Early Stopping: patience :: integer(1) activates early stopping using validation scores. performance model improve patience evaluation steps, training ended. Note final model stored learner, best model. initialized 0, means early stopping. first entry measures_valid used metric. also requires specify $validate field Learner, well measures_valid. min_delta :: double(1) minimum improvement threshold (>) early stopping. initialized 0. Dataloader: batch_size :: integer(1) batch size (required). shuffle :: logical(1) Whether shuffle instances dataset. Default FALSE. impact validation. sampler :: torch::sampler Object defines dataloader draw samples. batch_sampler :: torch::sampler Object defines dataloader draws batches. num_workers :: integer(1) number workers data loading (batches loaded parallel). default 0, means data loaded main process. collate_fn :: function merge list samples form batch. pin_memory :: logical(1) Whether dataloader copies tensors CUDA pinned memory returning . drop_last :: logical(1) Whether drop last training batch epoch training. Default FALSE. timeout :: numeric(1) timeout value collecting batch workers. Negative values mean timeout default -1. worker_init_fn :: function(id) function receives worker id ([1, num_workers]) exectued seeding worker data loading. worker_globals :: list() | character() loading data parallel, allows export globals workers. character vector, objects global environment names copied workers. worker_packages :: character() packages load workers. Also see torch::dataloder information.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"LearnerTorchModel created calling model_descriptor_to_learner() provided ModelDescriptor received input channel. parameters set according parameters specified PipeOpTorchModel '$train() method called [Task][mlr3::Task] stored [ModelDescriptor`].","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpLearner -> PipeOpTorchModel","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"PipeOpTorchModel$new() PipeOpTorchModel$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"","code":"PipeOpTorchModel$new(task_type, id = \"torch_model\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"task_type (character(1)) task type model. id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"","code":"PipeOpTorchModel$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Model — mlr_pipeops_torch_model","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":null,"dir":"Reference","previous_headings":"","what":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"Builds torch classifier trains .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"See LearnerTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"one input channel \"input\" takes ModelDescriptor traing Task specified task_type prediction. output NULL training Prediction given task_type prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"trained LearnerTorchModel.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"LearnerTorchModel created calling model_descriptor_to_learner() provided ModelDescriptor received input channel. parameters set according parameters specified PipeOpTorchModel '$train() method called [Task][mlr3::Task] stored [ModelDescriptor`].","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpLearner -> mlr3torch::PipeOpTorchModel -> PipeOpTorchModelClassif","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"PipeOpTorchModelClassif$new() PipeOpTorchModelClassif$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"","code":"PipeOpTorchModelClassif$new(id = \"torch_model_classif\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"","code":"PipeOpTorchModelClassif$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_classif.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"PipeOp Torch Classifier — mlr_pipeops_torch_model_classif","text":"","code":"# simple logistic regression  # configure the model descriptor md = as_graph(po(\"torch_ingress_num\") %>>%   po(\"nn_head\") %>>%   po(\"torch_loss\", \"cross_entropy\") %>>%   po(\"torch_optimizer\", \"adam\"))$train(tsk(\"iris\"))[[1L]]  print(md) #> <ModelDescriptor: 2 ops> #> * Ingress:  torch_ingress_num.input: [(NA,4)] #> * Task:  iris [classif] #> * Callbacks:  N/A #> * Optimizer:  Adaptive Moment Estimation #> * Loss:  Cross Entropy #> * pointer:  nn_head.output [(NA,3)]  # build the learner from the model descriptor and train it po_model = po(\"torch_model_classif\", batch_size = 50, epochs = 1) po_model$train(list(md)) #> $output #> NULL #>  po_model$state #> $model #> $network #> An `nn_module` containing 15 parameters. #>  #> ── Modules ───────────────────────────────────────────────────────────────────── #> • module_list: <nn_module_list> #15 parameters #>  #> $internal_valid_scores #> NULL #>  #> $loss_fn #> list() #>  #> $optimizer #> $optimizer$param_groups #> $optimizer$param_groups[[1]] #> $optimizer$param_groups[[1]]$params #> [1] 1 2 #>  #> $optimizer$param_groups[[1]]$lr #> [1] 0.001 #>  #> $optimizer$param_groups[[1]]$betas #> [1] 0.900 0.999 #>  #> $optimizer$param_groups[[1]]$eps #> [1] 1e-08 #>  #> $optimizer$param_groups[[1]]$weight_decay #> [1] 0 #>  #> $optimizer$param_groups[[1]]$amsgrad #> [1] FALSE #>  #>  #>  #> $optimizer$state #> $optimizer$state$`1` #> $optimizer$state$`1`$step #> [1] 3 #>  #> $optimizer$state$`1`$exp_avg #> torch_tensor #> -0.0813 -0.0115 -0.3071 -0.2139 #>  0.6080  0.2067  0.9018  0.4727 #> -0.5266 -0.1952 -0.5947 -0.2588 #> [ CPUFloatType{3,4} ] #>  #> $optimizer$state$`1`$exp_avg_sq #> torch_tensor #> 0.01 * #>  0.1392  0.0041  1.6147  0.7540 #>   3.0754  0.3983  5.4061  1.4248 #>   3.0210  0.4019  4.2777  0.8787 #> [ CPUFloatType{3,4} ] #>  #>  #> $optimizer$state$`2` #> $optimizer$state$`2`$step #> [1] 3 #>  #> $optimizer$state$`2`$exp_avg #> torch_tensor #> -0.0619 #>  0.1498 #> -0.0879 #> [ CPUFloatType{3} ] #>  #> $optimizer$state$`2`$exp_avg_sq #> torch_tensor #> 0.001 * #>  0.6434 #>  1.4496 #>  0.9899 #> [ CPUFloatType{3} ] #>  #>  #>  #>  #> $epochs #> [1] 1 #>  #> $callbacks #> named list() #>  #> $seed #> [1] 1285070263 #>  #> $task_col_info #> Key: <id> #>              id    type                      levels #>          <char>  <char>                      <list> #> 1: Petal.Length numeric                      [NULL] #> 2:  Petal.Width numeric                      [NULL] #> 3: Sepal.Length numeric                      [NULL] #> 4:  Sepal.Width numeric                      [NULL] #> 5:      Species  factor setosa,versicolor,virginica #>  #> attr(,\"class\") #> [1] \"learner_torch_model\" \"list\"                #>  #> $log #> Empty data.table (0 rows and 3 cols): stage,class,msg #>  #> $train_time #> [1] 0.053 #>  #> $param_vals #> $param_vals$epochs #> [1] 1 #>  #> $param_vals$device #> [1] \"auto\" #>  #> $param_vals$num_threads #> [1] 1 #>  #> $param_vals$num_interop_threads #> [1] 1 #>  #> $param_vals$seed #> [1] \"random\" #>  #> $param_vals$eval_freq #> [1] 1 #>  #> $param_vals$measures_train #> list() #>  #> $param_vals$measures_valid #> list() #>  #> $param_vals$patience #> [1] 0 #>  #> $param_vals$min_delta #> [1] 0 #>  #> $param_vals$batch_size #> [1] 50 #>  #>  #> $task_hash #> [1] \"34a6beb27c3181c3\" #>  #> $feature_names #> [1] \"Petal.Length\" \"Petal.Width\"  \"Sepal.Length\" \"Sepal.Width\"  #>  #> $validate #> NULL #>  #> $mlr3_version #> [1] ‘0.21.1’ #>  #> $internal_tuned_values #> named list() #>  #> $data_prototype #> Empty data.table (0 rows and 5 cols): Species,Petal.Length,Petal.Width,Sepal.Length,Sepal.Width #>  #> $task_prototype #> Empty data.table (0 rows and 5 cols): Species,Petal.Length,Petal.Width,Sepal.Length,Sepal.Width #>  #> $train_task #> <TaskClassif:iris> (150 x 5): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (4): #>   - dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Width #>  #> attr(,\"class\") #> [1] \"learner_state\" \"list\""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":null,"dir":"Reference","previous_headings":"","what":"Torch Regression Model — mlr_pipeops_torch_model_regr","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"Builds torch regression model trains .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"See LearnerTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"one input channel \"input\" takes ModelDescriptor traing Task specified task_type prediction. output NULL training Prediction given task_type prediction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"trained LearnerTorchModel.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"LearnerTorchModel created calling model_descriptor_to_learner() provided ModelDescriptor received input channel. parameters set according parameters specified PipeOpTorchModel '$train() method called [Task][mlr3::Task] stored [ModelDescriptor`].","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"super-classes","dir":"Reference","previous_headings":"","what":"Super classes","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"mlr3pipelines::PipeOp -> mlr3pipelines::PipeOpLearner -> mlr3torch::PipeOpTorchModel -> PipeOpTorchModelRegr","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"PipeOpTorchModelRegr$new() PipeOpTorchModelRegr$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"","code":"PipeOpTorchModelRegr$new(id = \"torch_model_regr\", param_vals = list())"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"","code":"PipeOpTorchModelRegr$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_model_regr.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Torch Regression Model — mlr_pipeops_torch_model_regr","text":"","code":"# simple linear regression  # build the model descriptor md = as_graph(po(\"torch_ingress_num\") %>>%   po(\"nn_head\") %>>%   po(\"torch_loss\", \"mse\") %>>%   po(\"torch_optimizer\", \"adam\"))$train(tsk(\"mtcars\"))[[1L]]  print(md) #> <ModelDescriptor: 2 ops> #> * Ingress:  torch_ingress_num.input: [(NA,10)] #> * Task:  mtcars [regr] #> * Callbacks:  N/A #> * Optimizer:  Adaptive Moment Estimation #> * Loss:  Mean Squared Error #> * pointer:  nn_head.output [(NA,1)]  # build the learner from the model descriptor and train it po_model = po(\"torch_model_regr\", batch_size = 20, epochs = 1) po_model$train(list(md)) #> $output #> NULL #>  po_model$state #> $model #> $network #> An `nn_module` containing 11 parameters. #>  #> ── Modules ───────────────────────────────────────────────────────────────────── #> • module_list: <nn_module_list> #11 parameters #>  #> $internal_valid_scores #> NULL #>  #> $loss_fn #> list() #>  #> $optimizer #> $optimizer$param_groups #> $optimizer$param_groups[[1]] #> $optimizer$param_groups[[1]]$params #> [1] 1 2 #>  #> $optimizer$param_groups[[1]]$lr #> [1] 0.001 #>  #> $optimizer$param_groups[[1]]$betas #> [1] 0.900 0.999 #>  #> $optimizer$param_groups[[1]]$eps #> [1] 1e-08 #>  #> $optimizer$param_groups[[1]]$weight_decay #> [1] 0 #>  #> $optimizer$param_groups[[1]]$amsgrad #> [1] FALSE #>  #>  #>  #> $optimizer$state #> $optimizer$state$`1` #> $optimizer$state$`1`$step #> [1] 2 #>  #> $optimizer$state$`1`$exp_avg #> torch_tensor #> -8.7515 -70.6545 -155.1045 -6165.2896 -82.2042 -85.4106 -3880.7810 -405.2719 -7.2561 -78.5860 #> [ CPUFloatType{1,10} ] #>  #> $optimizer$state$`1`$exp_avg_sq #> torch_tensor #> Columns 1 to 6 4.7060e+00  2.7571e+02  1.3329e+03  2.1144e+06  3.7363e+02  4.0293e+02 #>  #> Columns 7 to 10 8.3214e+05  9.1789e+03  3.2087e+00  3.4837e+02 #> [ CPUFloatType{1,10} ] #>  #>  #> $optimizer$state$`2` #> $optimizer$state$`2`$step #> [1] 2 #>  #> $optimizer$state$`2`$exp_avg #> torch_tensor #> -23.3342 #> [ CPUFloatType{1} ] #>  #> $optimizer$state$`2`$exp_avg_sq #> torch_tensor #>  30.1836 #> [ CPUFloatType{1} ] #>  #>  #>  #>  #> $epochs #> [1] 1 #>  #> $callbacks #> named list() #>  #> $seed #> [1] 1984753776 #>  #> $task_col_info #>         id    type levels #>     <char>  <char> <list> #>  1:     am numeric [NULL] #>  2:   carb numeric [NULL] #>  3:    cyl numeric [NULL] #>  4:   disp numeric [NULL] #>  5:   drat numeric [NULL] #>  6:   gear numeric [NULL] #>  7:     hp numeric [NULL] #>  8:   qsec numeric [NULL] #>  9:     vs numeric [NULL] #> 10:     wt numeric [NULL] #> 11:    mpg numeric [NULL] #>  #> attr(,\"class\") #> [1] \"learner_torch_model\" \"list\"                #>  #> $log #> Empty data.table (0 rows and 3 cols): stage,class,msg #>  #> $train_time #> [1] 0.039 #>  #> $param_vals #> $param_vals$epochs #> [1] 1 #>  #> $param_vals$device #> [1] \"auto\" #>  #> $param_vals$num_threads #> [1] 1 #>  #> $param_vals$num_interop_threads #> [1] 1 #>  #> $param_vals$seed #> [1] \"random\" #>  #> $param_vals$eval_freq #> [1] 1 #>  #> $param_vals$measures_train #> list() #>  #> $param_vals$measures_valid #> list() #>  #> $param_vals$patience #> [1] 0 #>  #> $param_vals$min_delta #> [1] 0 #>  #> $param_vals$batch_size #> [1] 20 #>  #>  #> $task_hash #> [1] \"6fbb93eed42adb31\" #>  #> $feature_names #>  [1] \"am\"   \"carb\" \"cyl\"  \"disp\" \"drat\" \"gear\" \"hp\"   \"qsec\" \"vs\"   \"wt\"   #>  #> $validate #> NULL #>  #> $mlr3_version #> [1] ‘0.21.1’ #>  #> $internal_tuned_values #> named list() #>  #> $data_prototype #> Empty data.table (0 rows and 11 cols): mpg,am,carb,cyl,disp,drat... #>  #> $task_prototype #> Empty data.table (0 rows and 11 cols): mpg,am,carb,cyl,disp,drat... #>  #> $train_task #> <TaskRegr:mtcars> (32 x 11): Motor Trends #> * Target: mpg #> * Properties: - #> * Features (10): #>   - dbl (10): am, carb, cyl, disp, drat, gear, hp, qsec, vs, wt #>  #> attr(,\"class\") #> [1] \"learner_state\" \"list\""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":null,"dir":"Reference","previous_headings":"","what":"Optimizer Configuration — mlr_pipeops_torch_optimizer","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"Configures optimizer deep learning model.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"input-and-output-channels","dir":"Reference","previous_headings":"","what":"Input and Output Channels","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"one input channel \"input\" one output channel \"output\". training, channels class ModelDescriptor. prediction, channels class Task.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"state","dir":"Reference","previous_headings":"","what":"State","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"state value calculated public method shapes_out().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"parameters","dir":"Reference","previous_headings":"","what":"Parameters","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"parameters defined dynamically optimizer set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"training, optimizer cloned added ModelDescriptor. Note parameter set stored TorchOptimizer reference-identical parameter set pipeop .","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"super-class","dir":"Reference","previous_headings":"","what":"Super class","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"mlr3pipelines::PipeOp -> PipeOpTorchOptimizer","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"methods","dir":"Reference","previous_headings":"","what":"Methods","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"mlr3pipelines::PipeOp$help() mlr3pipelines::PipeOp$predict() mlr3pipelines::PipeOp$print() mlr3pipelines::PipeOp$train()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"public-methods","dir":"Reference","previous_headings":"","what":"Public methods","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"PipeOpTorchOptimizer$new() PipeOpTorchOptimizer$clone()","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"method-new-","dir":"Reference","previous_headings":"","what":"Method new()","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"Creates new instance R6 class.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"","code":"PipeOpTorchOptimizer$new(   optimizer = t_opt(\"adam\"),   id = \"torch_optimizer\",   param_vals = list() )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"optimizer (TorchOptimizer character(1) torch_optimizer_generator) optimizer (something convertible via as_torch_optimizer()). id (character(1)) Identifier resulting  object. param_vals (list()) List hyperparameter settings, overwriting hyperparameter settings otherwise set construction.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"method-clone-","dir":"Reference","previous_headings":"","what":"Method clone()","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"objects class cloneable method.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"usage-1","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"","code":"PipeOpTorchOptimizer$clone(deep = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"arguments-1","dir":"Reference","previous_headings":"","what":"Arguments","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"deep Whether make deep clone.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_pipeops_torch_optimizer.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Optimizer Configuration — mlr_pipeops_torch_optimizer","text":"","code":"po_opt = po(\"torch_optimizer\", \"sgd\", lr = 0.01) po_opt$param_set #> <ParamSet(5)> #>              id    class lower upper nlevels        default  value #>          <char>   <char> <num> <num>   <num>         <list> <list> #> 1:           lr ParamDbl     0   Inf     Inf <NoDefault[0]>   0.01 #> 2:     momentum ParamDbl     0     1     Inf              0 [NULL] #> 3:    dampening ParamDbl     0     1     Inf              0 [NULL] #> 4: weight_decay ParamDbl     0     1     Inf              0 [NULL] #> 5:     nesterov ParamLgl    NA    NA       2          FALSE [NULL] mdin = po(\"torch_ingress_num\")$train(list(tsk(\"iris\"))) mdin[[1L]]$optimizer #> NULL mdout = po_opt$train(mdin) mdout[[1L]]$optimizer #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: lr=0.01 #> * Packages: torch,mlr3torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":null,"dir":"Reference","previous_headings":"","what":"Iris Classification Task — mlr_tasks_lazy_iris","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"classification task popular datasets::iris data set. Just like iris task, features represented one lazy tensor column.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"format","dir":"Reference","previous_headings":"","what":"Format","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"R6::R6Class inheriting mlr3::TaskClassif.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"https://en.wikipedia.org/wiki/Iris_flower_data_set","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"construction","dir":"Reference","previous_headings":"","what":"Construction","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"","code":"tsk(\"lazy_iris\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"Task type: “classif” Properties: “multiclass” Missings: Target: “Species” Features: “x” Data Dimension: 150x3","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"Anderson E (1936). “Species Problem Iris.” Annals Missouri Botanical Garden, 23(3), 457. doi:10.2307/2394164 .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_lazy_iris.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Iris Classification Task — mlr_tasks_lazy_iris","text":"","code":"task = tsk(\"lazy_iris\") task #> <TaskClassif:lazy_iris> (150 x 2): Iris Flowers #> * Target: Species #> * Properties: multiclass #> * Features (1): #>   - lt (1): x df = task$data() materialize(df$x[1:6], rbind = TRUE) #> torch_tensor #>  5.1000  3.5000  1.4000  0.2000 #>  4.9000  3.0000  1.4000  0.2000 #>  4.7000  3.2000  1.3000  0.2000 #>  4.6000  3.1000  1.5000  0.2000 #>  5.0000  3.6000  1.4000  0.2000 #>  5.4000  3.9000  1.7000  0.4000 #> [ CPUFloatType{6,4} ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":null,"dir":"Reference","previous_headings":"","what":"MNIST Image classification — mlr_tasks_mnist","title":"MNIST Image classification — mlr_tasks_mnist","text":"Classic MNIST image classification. underlying DataBackend contains columns \"label\", \"image\", \"row_id\", \"split\", last column indicates whether row belongs train test set. first 60000 rows belong training set, last 10000 rows test set.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"source","dir":"Reference","previous_headings":"","what":"Source","title":"MNIST Image classification — mlr_tasks_mnist","text":"https://torchvision.mlverse.org/reference/mnist_dataset.html","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"construction","dir":"Reference","previous_headings":"","what":"Construction","title":"MNIST Image classification — mlr_tasks_mnist","text":"","code":"tsk(\"mnist\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"download","dir":"Reference","previous_headings":"","what":"Download","title":"MNIST Image classification — mlr_tasks_mnist","text":"task's backend DataBackendLazy download data requested. meta-data already available . can cache datasets setting mlr3torch.cache option TRUE specific path used cache directory.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"MNIST Image classification — mlr_tasks_mnist","text":"Task type: “classif” Properties: “multiclass” Missings: Target: “label” Features: “image” Data Dimension: 70000x4","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"MNIST Image classification — mlr_tasks_mnist","text":"Lecun, Y., Bottou, L., Bengio, Y., Haffner, P. (1998). “Gradient-based learning applied document recognition.” Proceedings IEEE, 86(11), 2278-2324. doi:10.1109/5.726791 .","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_mnist.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"MNIST Image classification — mlr_tasks_mnist","text":"","code":"task = tsk(\"mnist\") task #> <TaskClassif:mnist> (70000 x 2): MNIST Digit Classification #> * Target: label #> * Properties: multiclass #> * Features (1): #>   - lt (1): image"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":null,"dir":"Reference","previous_headings":"","what":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"Subset famous ImageNet dataset. data obtained torchvision::tiny_imagenet_dataset(). underlying DataBackend contains columns \"class\", \"image\", \"..row_id\", \"split\", last column indicates whether row belongs train, validation test set defined provided torchvision. labels test rows, default, observations inactive, means task uses 110000 120000 observations defined underlying data backend.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"construction","dir":"Reference","previous_headings":"","what":"Construction","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"","code":"tsk(\"tiny_imagenet\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"download","dir":"Reference","previous_headings":"","what":"Download","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"task's backend DataBackendLazy download data requested. meta-data already available . can cache datasets setting mlr3torch.cache option TRUE specific path used cache directory.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"properties","dir":"Reference","previous_headings":"","what":"Properties","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"Task type: “classif” Properties: “multiclass” Missings: Target: “class” Features: “image” Data Dimension: 120000x4","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"references","dir":"Reference","previous_headings":"","what":"References","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"Deng, Jia, Dong, Wei, Socher, Richard, Li, Li-Jia, Li, Kai, Fei-Fei, Li (2009). “Imagenet: large-scale hierarchical image database.” 2009 IEEE conference computer vision pattern recognition, 248–255. IEEE.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/mlr_tasks_tiny_imagenet.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Tiny ImageNet Classification Task — mlr_tasks_tiny_imagenet","text":"","code":"task = tsk(\"tiny_imagenet\") task #> <TaskClassif:tiny_imagenet> (110000 x 2): ImageNet Subset #> * Target: class #> * Properties: multiclass #> * Features (1): #>   - lt (1): image"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"First nn_graph created using model_descriptor_to_module learner created module remaining information model descriptor, must include optimizer loss function optionally callbacks.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"","code":"model_descriptor_to_learner(model_descriptor)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"model_descriptor (ModelDescriptor) model descriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_learner.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Torch Learner from a ModelDescriptor — model_descriptor_to_learner","text":"Learner","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"Creates nn_graph ModelDescriptor. Mostly internal use, since ModelDescriptor circumstances harder use just creating nn_graph directly.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"","code":"model_descriptor_to_module(   model_descriptor,   output_pointers = NULL,   list_output = FALSE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"model_descriptor (ModelDescriptor) Model Descriptor. pointer ignored, instead output_pointer values used. $graph member modified -reference. output_pointers (list character) Collection pointers indicate part model_descriptor$graph used output. Entries format ModelDescriptor$pointer. list_output (logical(1)) Whether output list tensors. FALSE, length(output_pointers) must 1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_to_module.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a nn_graph from ModelDescriptor — model_descriptor_to_module","text":"nn_graph","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":null,"dir":"Reference","previous_headings":"","what":"Union of ModelDescriptors — model_descriptor_union","title":"Union of ModelDescriptors — model_descriptor_union","text":"mostly internal function used PipeOpTorchs multiple input channels. creates union multiple ModelDescriptors: graphs combinded (identical begin ). first entry's graph modified reference. PipeOps ID must identical. new input edges may added PipeOps. Drops pointer / pointer_shape entries. new task feature union two incoming tasks. optimizer loss ModelDescriptors must identical. Ingress tokens callbacks merged, objects \"id\" must identical.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Union of ModelDescriptors — model_descriptor_union","text":"","code":"model_descriptor_union(md1, md2)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Union of ModelDescriptors — model_descriptor_union","text":"md1 (ModelDescriptor) first ModelDescriptor. md2 (ModelDescriptor) second ModelDescriptor.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Union of ModelDescriptors — model_descriptor_union","text":"ModelDescriptor","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/model_descriptor_union.html","id":"details","dir":"Reference","previous_headings":"","what":"Details","title":"Union of ModelDescriptors — model_descriptor_union","text":"requirement new input edgedes may added PipeOps  theoretically necessary, since assume ModelDescriptor built beginning end (.e. PipeOps never get new ancestors) can make assumption simplify things. Otherwise need treat \"...\"-inputs special.)","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Neural Network Layer — nn","title":"Create a Neural Network Layer — nn","text":"Retrieve neural network layer mlr_pipeops dictionary.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Neural Network Layer — nn","text":"","code":"nn(.key, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Neural Network Layer — nn","text":".key (character(1)) ... () Additional parameters, constructor arguments fields.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a Neural Network Layer — nn","text":"","code":"po1 = po(\"nn_linear\", id = \"linear\") # is the same as: po2 = nn(\"linear\")"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":null,"dir":"Reference","previous_headings":"","what":"Graph Network — nn_graph","title":"Graph Network — nn_graph","text":"Represents neural network using Graph usually costains mostly PipeOpModules.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Graph Network — nn_graph","text":"","code":"nn_graph(graph, shapes_in, output_map = graph$output$name, list_output = FALSE)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Graph Network — nn_graph","text":"graph (Graph) Graph wrap. cloned. shapes_in (named integer) Shape info tensors go graph. Names must graph$input$name, possibly different order. output_map (character) graph's outputs use. Must subset graph$output$name. list_output (logical(1)) Whether output list tensors. FALSE (default), length(output_map) must 1.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Graph Network — nn_graph","text":"nn_graph","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_graph.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Graph Network — nn_graph","text":"","code":"graph = mlr3pipelines::Graph$new() graph$add_pipeop(po(\"module_1\", module = nn_linear(10, 20)), clone = FALSE) graph$add_pipeop(po(\"module_2\", module = nn_relu()), clone = FALSE) graph$add_pipeop(po(\"module_3\", module = nn_linear(20, 1)), clone = FALSE) graph$add_edge(\"module_1\", \"module_2\") graph$add_edge(\"module_2\", \"module_3\")  network = nn_graph(graph, shapes_in = list(module_1.input = c(NA, 10)))  x = torch_randn(16, 10)  network(module_1.input = x) #> torch_tensor #>  0.2515 #> -0.0529 #>  0.0081 #>  0.2387 #> -0.0763 #>  0.0844 #>  0.2905 #> -0.1252 #> -0.0067 #>  0.1168 #> -0.0734 #>  0.3246 #> -0.1031 #>  0.5290 #> -0.0672 #>  0.0823 #> [ CPUFloatType{16,1} ][ grad_fn = <AddmmBackward0> ]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_cat.html","id":null,"dir":"Reference","previous_headings":"","what":"Concatenates multiple tensors — nn_merge_cat","title":"Concatenates multiple tensors — nn_merge_cat","text":"Concatenates multiple tensors given dimension. broadcasting rules applied , must reshape tensors shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_cat.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Concatenates multiple tensors — nn_merge_cat","text":"","code":"nn_merge_cat(dim = -1)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_cat.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Concatenates multiple tensors — nn_merge_cat","text":"dim (integer(1)) dimension concatenation.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_prod.html","id":null,"dir":"Reference","previous_headings":"","what":"Product of multiple tensors — nn_merge_prod","title":"Product of multiple tensors — nn_merge_prod","text":"Calculates product input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_prod.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Product of multiple tensors — nn_merge_prod","text":"","code":"nn_merge_prod()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_sum.html","id":null,"dir":"Reference","previous_headings":"","what":"Sum of multiple tensors — nn_merge_sum","title":"Sum of multiple tensors — nn_merge_sum","text":"Calculates sum input tensors.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_merge_sum.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Sum of multiple tensors — nn_merge_sum","text":"","code":"nn_merge_sum()"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_reshape.html","id":null,"dir":"Reference","previous_headings":"","what":"Reshape — nn_reshape","title":"Reshape — nn_reshape","text":"Reshape tensor given shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_reshape.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Reshape — nn_reshape","text":"","code":"nn_reshape(shape)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_reshape.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Reshape — nn_reshape","text":"shape (integer()) desired output shape.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_squeeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Squeeze — nn_squeeze","title":"Squeeze — nn_squeeze","text":"Squeezes tensor calling torch::torch_squeeze() given dimension dim.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_squeeze.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Squeeze — nn_squeeze","text":"","code":"nn_squeeze(dim)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_squeeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Squeeze — nn_squeeze","text":"dim (integer()) dimension squeeze.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_unsqueeze.html","id":null,"dir":"Reference","previous_headings":"","what":"Unsqueeze — nn_unsqueeze","title":"Unsqueeze — nn_unsqueeze","text":"Unsqueezes tensor calling torch::torch_unsqueeze() given dimension dim.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_unsqueeze.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Unsqueeze — nn_unsqueeze","text":"","code":"nn_unsqueeze(dim)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/nn_unsqueeze.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Unsqueeze — nn_unsqueeze","text":"dim (integer(1)) dimension unsqueeze.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":null,"dir":"Reference","previous_headings":"","what":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"Function create objects class PipeOpTaskPreprocTorch convenient way. Start reading documentation PipeOpTaskPreprocTorch.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"","code":"pipeop_preproc_torch(   id,   fn,   shapes_out = NULL,   param_set = NULL,   packages = character(0),   rowwise = FALSE,   parent_env = parent.frame(),   stages_init = NULL,   tags = NULL )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"id (character(1)) id new object. fn (function) preprocessing function. shapes_out (function NULL \"infer\") private .shapes_out(shapes_in, param_vals, task) method PipeOpTaskPreprocTorch (see section Inheriting). Special values NULL infer: NULL, output shapes unknown. \"infer\", output shape function inferred calculates output shapes follows: input shape (NA, ...) meta-tensor shape (1, ...) created preprocessing function applied. Afterwards batch dimension (1) replaced NA shape returned. first dimension NA, output shape applying preprocessing function returned. Method \"infer\" correct cases, might fail edge cases. param_set (ParamSet NULL) parameter set. left NULL (default) parameter set inferred following way: parameters first ... fn set untyped parameters tags 'train' default value tagged 'required' well. Default values annotated. packages (character()) R packages object depends . rowwise (logical(1)) Whether preprocessing applied row-wise. parent_env (environment) parent environment R6 class. stages_init (character(1)) Initial value stages parameter. NULL (default), set \"\" case id starts \"trafo\" \"train\" starts \"augment\". Otherwise must specified. tags (character()) Tags pipeop","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"R6Class instance inheriting PipeOpTaskPreprocTorch","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/pipeop_preproc_torch.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create Torch Preprocessing PipeOps — pipeop_preproc_torch","text":"","code":"PipeOpPreprocExample = pipeop_preproc_torch(\"preproc_example\", function(x, a) x + a) po_example = PipeOpPreprocExample$new() po_example$param_set #> <ParamSet(3)> #>                id    class lower upper nlevels        default  value #>            <char>   <char> <num> <num>   <num>         <list> <list> #> 1:              a ParamUty    NA    NA     Inf <NoDefault[0]> [NULL] #> 2:         stages ParamFct    NA    NA       3 <NoDefault[0]> [NULL] #> 3: affect_columns ParamUty    NA    NA     Inf  <Selector[1]> [NULL]"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/replace_head.html","id":null,"dir":"Reference","previous_headings":"","what":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","title":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","text":"Replace head network Replaces head network linear layer d_out classes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/replace_head.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","text":"","code":"replace_head(network, d_out)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/replace_head.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Replace the head of a network Replaces the head of the network with a linear layer with d_out classes. — replace_head","text":"network (torch::nn_module) network d_out (integer(1)) number output classes.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":null,"dir":"Reference","previous_headings":"","what":"Sugar Function for Torch Callback — t_clbk","title":"Sugar Function for Torch Callback — t_clbk","text":"Retrieves one TorchCallbacks mlr3torch_callbacks. Works like mlr3::lrn() mlr3::lrns().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Sugar Function for Torch Callback — t_clbk","text":"","code":"t_clbk(.key, ...)  t_clbks(.keys)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Sugar Function for Torch Callback — t_clbk","text":".key (character(1)) key torch callback. ... () See description dictionary_sugar_get(). .keys (character()) keys callbacks.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Sugar Function for Torch Callback — t_clbk","text":"TorchCallback list() TorchCallbacks","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_clbk.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Sugar Function for Torch Callback — t_clbk","text":"","code":"t_clbk(\"progress\") #> <TorchCallback:progress> Progress #> * Generator: CallbackSetProgress #> * Parameters: list() #> * Packages: progress,mlr3torch,torch"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":null,"dir":"Reference","previous_headings":"","what":"Loss Function Quick Access — t_loss","title":"Loss Function Quick Access — t_loss","text":"Retrieve one TorchLosses mlr3torch_losses. Works like mlr3::lrn() mlr3::lrns().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Loss Function Quick Access — t_loss","text":"","code":"t_loss(.key, ...)  t_losses(.keys, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Loss Function Quick Access — t_loss","text":".key (character(1)) Key object retrieve. ... () See description dictionary_sugar_get. .keys (character()) keys losses.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Loss Function Quick Access — t_loss","text":"TorchLoss","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_loss.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Loss Function Quick Access — t_loss","text":"","code":"t_loss(\"mse\", reduction = \"mean\") #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: reduction=mean #> * Packages: torch,mlr3torch #> * Task Types: regr # get the dictionary t_loss() #> <DictionaryMlr3torchLosses> with 3 stored values #> Keys: cross_entropy, l1, mse t_losses(c(\"mse\", \"l1\")) #> $mse #> <TorchLoss:mse> Mean Squared Error #> * Generator: nn_mse_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr #>  #> $l1 #> <TorchLoss:l1> Absolute Error #> * Generator: nn_l1_loss #> * Parameters: list() #> * Packages: torch,mlr3torch #> * Task Types: regr #>  # get the dictionary t_losses() #> <DictionaryMlr3torchLosses> with 3 stored values #> Keys: cross_entropy, l1, mse"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":null,"dir":"Reference","previous_headings":"","what":"Optimizers Quick Access — t_opt","title":"Optimizers Quick Access — t_opt","text":"Retrieves one TorchOptimizers mlr3torch_optimizers. Works like mlr3::lrn() mlr3::lrns().","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Optimizers Quick Access — t_opt","text":"","code":"t_opt(.key, ...)  t_opts(.keys, ...)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Optimizers Quick Access — t_opt","text":".key (character(1)) Key object retrieve. ... () See description dictionary_sugar_get. .keys (character()) keys optimizers.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Optimizers Quick Access — t_opt","text":"TorchOptimizer","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/t_opt.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Optimizers Quick Access — t_opt","text":"","code":"t_opt(\"adam\", lr = 0.1) #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: lr=0.1 #> * Packages: torch,mlr3torch # get the dictionary t_opt() #> <DictionaryMlr3torchOptimizers> with 7 stored values #> Keys: adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd t_opts(c(\"adam\", \"sgd\")) #> $adam #> <TorchOptimizer:adam> Adaptive Moment Estimation #> * Generator: optim_adam #> * Parameters: list() #> * Packages: torch,mlr3torch #>  #> $sgd #> <TorchOptimizer:sgd> Stochastic Gradient Descent #> * Generator: optim_sgd #> * Parameters: list() #> * Packages: torch,mlr3torch #>  # get the dictionary t_opts() #> <DictionaryMlr3torchOptimizers> with 7 stored values #> Keys: adadelta, adagrad, adam, asgd, rmsprop, rprop, sgd"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Dataset from a Task — task_dataset","title":"Create a Dataset from a Task — task_dataset","text":"Creates torch dataset mlr3 Task. resulting dataset's $.get_batch() method returns list elements x, y index: x list tensors, whose content defined parameter feature_ingress_tokens. y target variable content defined parameter target_batchgetter. .index index batch task's data. data returned device specified parameter device.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Dataset from a Task — task_dataset","text":"","code":"task_dataset(task, feature_ingress_tokens, target_batchgetter = NULL, device)"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Dataset from a Task — task_dataset","text":"task (Task) task build dataset. feature_ingress_tokens (named list() TorchIngressToken) ingress token defines one item $x value batch corresponding names. target_batchgetter (function(data, device)) function taking arguments data, data.table containing target variable, device. must return target torch tensor selected device. device (character()) device, e.g. \"cuda\" \"cpu\".","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Dataset from a Task — task_dataset","text":"torch::dataset","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/task_dataset.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a Dataset from a Task — task_dataset","text":"","code":"task = tsk(\"iris\") sepal_ingress = TorchIngressToken(   features = c(\"Sepal.Length\", \"Sepal.Width\"),   batchgetter = batchgetter_num,   shape = c(NA, 2) ) petal_ingress = TorchIngressToken(   features = c(\"Petal.Length\", \"Petal.Width\"),   batchgetter = batchgetter_num,   shape = c(NA, 2) ) ingress_tokens = list(sepal = sepal_ingress, petal = petal_ingress)  target_batchgetter = function(data, device) {   torch_tensor(data = data[[1L]], dtype = torch_float32(), device)$unsqueeze(2) } dataset = task_dataset(task, ingress_tokens, target_batchgetter, \"cpu\") batch = dataset$.getbatch(1:10) batch #> $x #> $x$sepal #> torch_tensor #>  5.1000  3.5000 #>  4.9000  3.0000 #>  4.7000  3.2000 #>  4.6000  3.1000 #>  5.0000  3.6000 #>  5.4000  3.9000 #>  4.6000  3.4000 #>  5.0000  3.4000 #>  4.4000  2.9000 #>  4.9000  3.1000 #> [ CPUFloatType{10,2} ] #>  #> $x$petal #> torch_tensor #>  1.4000  0.2000 #>  1.4000  0.2000 #>  1.3000  0.2000 #>  1.5000  0.2000 #>  1.4000  0.2000 #>  1.7000  0.4000 #>  1.4000  0.3000 #>  1.5000  0.2000 #>  1.4000  0.2000 #>  1.5000  0.1000 #> [ CPUFloatType{10,2} ] #>  #>  #> $.index #> torch_tensor #>   1 #>   2 #>   3 #>   4 #>   5 #>   6 #>   7 #>   8 #>   9 #>  10 #> [ CPULongType{10} ] #>  #> $y #> torch_tensor #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #>  1 #> [ CPUFloatType{10,1} ] #>"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":null,"dir":"Reference","previous_headings":"","what":"Create a Callback Desctiptor — torch_callback","title":"Create a Callback Desctiptor — torch_callback","text":"Convenience function create custom TorchCallback. arguments available callback_set() also available . information correctly implement new callback, see CallbackSet.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"ref-usage","dir":"Reference","previous_headings":"","what":"Usage","title":"Create a Callback Desctiptor — torch_callback","text":"","code":"torch_callback(   id,   classname = paste0(\"CallbackSet\", capitalize(id)),   param_set = NULL,   packages = NULL,   label = capitalize(id),   man = NULL,   on_begin = NULL,   on_end = NULL,   on_exit = NULL,   on_epoch_begin = NULL,   on_before_valid = NULL,   on_epoch_end = NULL,   on_batch_begin = NULL,   on_batch_end = NULL,   on_after_backward = NULL,   on_batch_valid_begin = NULL,   on_batch_valid_end = NULL,   on_valid_end = NULL,   state_dict = NULL,   load_state_dict = NULL,   initialize = NULL,   public = NULL,   private = NULL,   active = NULL,   parent_env = parent.frame(),   inherit = CallbackSet,   lock_objects = FALSE )"},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"arguments","dir":"Reference","previous_headings":"","what":"Arguments","title":"Create a Callback Desctiptor — torch_callback","text":"id (character(1))` id torch callback. classname (character(1)) class name. param_set (ParamSet) parameter set, present inferred $initialize() method. packages (character())packages callback depends . Default isNULL`. label (character(1)) label torch callback. Defaults capitalized id. man (character(1)) String format [pkg]::[topic] pointing manual page object. referenced help package can opened via method $help(). default NULL. on_begin, on_end, on_epoch_begin, on_before_valid, on_epoch_end, on_batch_begin, on_batch_end, on_after_backward, on_batch_valid_begin, on_batch_valid_end, on_valid_end, on_exit (function) Function execute given stage, see section Stages. state_dict (function()) function retrieves state dict callback. available learner training. load_state_dict (function(state_dict)) Function loads callback state. initialize (function()) initialization method callback. public, private, active (list()) Additional public, private, active fields add callback. parent_env (environment()) parent environment R6Class. inherit (R6ClassGenerator) class inherit. class must either CallbackSet (default) inherit . lock_objects (logical(1)) Whether lock objects resulting R6Class. FALSE (default), values can freely assigned self without declaring class definition.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"value","dir":"Reference","previous_headings":"","what":"Value","title":"Create a Callback Desctiptor — torch_callback","text":"TorchCallback","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"internals","dir":"Reference","previous_headings":"","what":"Internals","title":"Create a Callback Desctiptor — torch_callback","text":"first creates R6 class inheriting CallbackSet (using callback_set()) wraps generator TorchCallback can passed torch learner.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"stages","dir":"Reference","previous_headings":"","what":"Stages","title":"Create a Callback Desctiptor — torch_callback","text":"begin :: Run training loop begins. epoch_begin :: Run beginning epoch. batch_begin :: Run forward call. after_backward :: Run backward call. batch_end :: Run optimizer step. batch_valid_begin :: Run forward call validation loop. batch_valid_end :: Run forward call validation loop. valid_end :: Run end validation. epoch_end :: Run end epoch. end :: Run last epoch. exit :: Run last, using .exit().","code":""},{"path":[]},{"path":"https://mlr3torch.mlr-org.com/dev/reference/torch_callback.html","id":"ref-examples","dir":"Reference","previous_headings":"","what":"Examples","title":"Create a Callback Desctiptor — torch_callback","text":"","code":"custom_tcb = torch_callback(\"custom\",   initialize = function(name) {     self$name = name   },   on_begin = function() {     cat(\"Hello\", self$name, \", we will train for \", self$ctx$total_epochs, \"epochs.\\n\")   },   on_end = function() {     cat(\"Training is done.\")   } )  learner = lrn(\"classif.torch_featureless\",   batch_size = 16,   epochs = 1,   callbacks = custom_tcb,   cb.custom.name = \"Marie\",   device = \"cpu\" ) task = tsk(\"iris\") learner$train(task) #> Hello Marie , we will train for  1 epochs. #> Training is done."},{"path":"https://mlr3torch.mlr-org.com/dev/news/index.html","id":"mlr3torch-012","dir":"Changelog","previous_headings":"","what":"mlr3torch 0.1.2","title":"mlr3torch 0.1.2","text":"CRAN release: 2024-10-15 Don’t use deprecated data_formats anymore Added CallbackSetTB, allows logging can viewed TensorBoard.","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/news/index.html","id":"mlr3torch-011","dir":"Changelog","previous_headings":"","what":"mlr3torch 0.1.1","title":"mlr3torch 0.1.1","text":"CRAN release: 2024-09-12 fix(preprocessing): regarding construction PipeOps po(\"trafo_resize\") failed cases. fix(ci): tests run CI fix(learner): LearnerTabResnet now works correctly Fix tests run CI feat: added nn() helper function simplify creation neural network layers","code":""},{"path":"https://mlr3torch.mlr-org.com/dev/news/index.html","id":"mlr3torch-010","dir":"Changelog","previous_headings":"","what":"mlr3torch 0.1.0","title":"mlr3torch 0.1.0","text":"CRAN release: 2024-07-08 Initial CRAN release","code":""}]