Additional updates to provide full set of runtime numbers and to fix …

…notebook descriptions, typos, etc.

cuml/README.md

@@ -8,7 +8,7 @@
 * `tsvd_demo`: notebook showcasing truncated singular value decomposition (tsvd) algorithm comparison between cuML and scikit-learn.
 * `linear_regression`: notebook showcasing linear regression comparison between cuML and scikit-learn.
 * `ridge_regression`: notebook showcasing ridge regression comparison between cuML and scikit-learn.
-* `sgd`: notebook showcasing ridge regression comparison between cuML and scikit-learn.
+* `sgd`: notebook showcasing stochastic gradient descent comparison between cuML and scikit-learn.
 * `umap`: notebook showcasing and evaluating cuML's UMAP dimension reduction technique.
 
 ## dbscan_demo
@@ -37,8 +37,6 @@ Note that the timing differences depend upon the exact dataset being used. Also,
 
 ## knn_demo
 
-The `knn_demo` notebook demonstrates how cuml establishes interoperability between `cudf` and `faiss-gpu`. There is native support for this demo with CUDA 9.2. With CUDA 10.0, the user must build `faiss-gpu` from source with [GPU support](https://github.com/facebookresearch/faiss/blob/master/INSTALL.md).
-
 Typical output of the cells processing knn looks like:
 
 - For scikit-learn knn:
@@ -111,114 +109,144 @@ compare tsvd: cuml vs sklearn transformed results equal
 Typical output of the cells processing linear_regression looks like:
 
 - For scikit-learn:
+
+Fit:
+
+```
+CPU times: user 1min 8s, sys: 21.2 s, total: 1min 30s
+Wall time: 6.06 s
+```
+
+Predict:
+
 ```
-CPU times: user  s, sys:  s, total:  s
-Wall time:  s
+CPU times: user 5.46 s, sys: 312 ms, total: 5.77 s
+Wall time: 471 ms
 ```
 
+
 - For cuML:
+
+Fit:
+
 ```
-CPU times: user  s, sys:  s, total:  s
-Wall time:  s
+CPU times: user 504 ms, sys: 347 ms, total: 851 ms
+Wall time: 1.08 s
 ```
 
+Predict:
+
+```
+CPU times: user 144 ms, sys: 7.71 ms, total: 152 ms
+Wall time: 145 ms
+```
+
+
 Final cell of the notebook should output:
 
 ```
 SKL MSE(y):
-3.4750864e-13
+5.6481553e-05
 CUML MSE(y):
-5.827862e-07
+7.246567e-07
 ```
 
 ## ridge_regression_demo
 
 Typical output of the cells processing ridge_regression looks like:
 
 - For scikit-learn:
-```
-CPU times: user  s, sys:  s, total:  s
-Wall time:  s
-```
 
-- For cuML:
+Fit:
+
 ```
-CPU times: user  s, sys:  s, total:  s
-Wall time:  s
+CPU times: user 1min 1s, sys: 1.99 s, total: 1min 3s
+Wall time: 5.02 s
 ```
 
-Final cell of the notebook should output:
+Predict:
 
 ```
-SKL MSE(y):
-1.8886121326984265e-08
-CUML MSE(y):
-1.9204549e-08
+CPU times: user 2.66 s, sys: 75.6 ms, total: 2.74 s
+Wall time: 180 ms
 ```
 
-## sgd_demo
 
-Typical output of the cells processing sgd looks like:
+- For cuML:
+
+Fit:
 
-- For scikit-learn:
 ```
-CPU times: user  s, sys:  s, total:  s
-Wall time:  s
+CPU times: user 518 ms, sys: 309 ms, total: 827 ms
+Wall time: 831 ms
 ```
 
-- For cuML:
+Predict:
+
 ```
-CPU times: user  s, sys:  s, total:  s
-Wall time:  s
+CPU times: user 146 ms, sys: 7.31 ms, total: 154 ms
+Wall time: 149 ms
 ```
 
 Final cell of the notebook should output:
 
 ```
 SKL MSE(y):
-1.1356839999498491e-07
+0.0204307456949534
 CUML MSE(y):
-1.04257616e-07
+0.00012496959
 ```
 
-## umap_demo
+## sgd_demo
 
-Typical output of the cells processing umap looks like:
+Typical output of the cells processing sgd looks like:
 
-- For blobs clustering test:
-```
-CPU times: user 15.4 s, sys: 1.71 s, total: 17.1 s
-Wall time: 4.53 s
-```
+- For scikit-learn:
 
-Final cell of blobs clustering test should output:
+Fit:
 
 ```
-Cluster demonstration completed successfully
+CPU times: user 10min 22s, sys: 417 ms, total: 10min 22s
+Wall time: 10min 20s
 ```
 
+Prefict:
 
-- For trustworthiness evaluation using random initialization:
 ```
-CPU times: user 9.84 s, sys: 380 ms, total: 10.2 s
-Wall time: 945 ms
+CPU times: user 146 ms, sys: 63 ms, total: 209 ms
+Wall time: 166 ms
 ```
 
-Final cell of trustworthiness evaluation with random initialization should output:
+
+- For cuML:
+
+Fit:
 
 ```
-Trustworthiness on random initialization passed successfully
+CPU times: user 2min 13s, sys: 8.1 s, total: 2min 21s
+Wall time: 2min 18s
 ```
 
-- For trustworthiness evaluation using spectral initialization:
+Prefict:
 
 ```
-CPU times: user 9.83 s, sys: 296 ms, total: 10.1 s
-Wall time: 832 ms
+CPU times: user 139 ms, sys: 10.9 ms, total: 150 ms
+Wall time: 142 ms
 ```
 
-Final cell of trustworthiness evaluation with spectral initialization should output:
+Final cell of the notebook should output:
 
 ```
-Trustworthiness on spectral initialization passed successfully
+SKL MSE(y):
+1.144686926876654e-07
+CUML MSE(y):
+1.0390148e-07
 ```
+
+## umap_demo
+
+This notebook currently performs assertions and does not print any output. It contains two types of tests that evaluate UMAP's ability to preserve local neighborhood structure.
+
+The first test verifies that when the input contains blobs generated from several different clusters, the UMAP output produces low-dimensional blobs with the same number of clusters.
+
+The second test demonstrates that the neighborhoods of the low-dimensional embeddings are similar to the neighborhoods of the inputs. A score, known as trustworthiness, and made popular by t-SNE, is used to evaluate the UMAP embeddings for both random and spectral initialization strategies.