-
Notifications
You must be signed in to change notification settings - Fork 1
/
data_preprocessing_unimodal_vs_oligomodal_including_stripmodes.py
213 lines (193 loc) · 9.99 KB
/
data_preprocessing_unimodal_vs_oligomodal_including_stripmodes.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
"Ryan van Mastrigt, 29.07.2022"
"This script preprocesses the pentodal metamaterial data for neural network training for classification (ii)"
import numpy as np
SEED = 0
np.random.seed(SEED)
# tf.random.set_seed(SEED)
# os.environ['PYTHONHASHSEED'] = str(SEED)
def unison_shuffled_copies(a, b):
assert len(a) == len(b)
p = np.random.permutation(len(a))
return a[p], b[p]
def translate_config(config, k):
"input: config(2k, 2k) & k, translate configuration in x and y direction"
configtile = np.tile(config, (2, 2))
translconfs = np.zeros((k*k, np.shape(config)[0], np.shape(config)[1]), dtype=int)
for x in range(k):
for y in range(k):
index = x*k + y
translconfs[index] = configtile[2*x:2*x+2*k, 2*y:2*y+2*k]
return translconfs[1::]
def transrotconfigs(pixreps, resultstotal, k, extended=False):
# pixreps = np.loadtxt('results\\modescaling\\PixelRep_{:d}x{:d}.txt'.format(k, k), delimiter=',')
# resultstotal = np.loadtxt('results\\modescaling\\results_analysis_i_Scen_slope_offset_M1k_{:d}x{:d}_fixn4.txt'.format(k, k), delimiter=',')
indB = np.argwhere(resultstotal[:, 1]==1)[:, 0]
pixreps = np.reshape(pixreps, (-1, 2*k, 2*k))
pixrepappend = pixreps.copy()
restotalappend = resultstotal.copy()
for i, ind in enumerate(indB):
config = pixreps[ind]
tconfs = translate_config(config, k)
pixrepappend = np.append(pixrepappend, tconfs, axis=0)
restotalappend = np.append(restotalappend, np.tile(restotalappend[ind], (np.shape(tconfs)[0], 1)), axis=0)
"rotate the configs"
indBapp = np.argwhere(restotalappend[:, 1]==1)[:, 0]
# rotconfs = pixrepappend[indBapp]
for r in range(1, 4):
rotconfs = np.rot90(pixrepappend[indBapp], k=r, axes=(1, 2))
pixrepappend = np.append(pixrepappend, rotconfs, axis=0)
restotalappend = np.append(restotalappend, restotalappend[indBapp], axis=0)
"remove duplicate configs"
pixrepappend = np.reshape(pixrepappend, (-1, 2*k*2*k))
pixrepfinal, indices = np.unique(pixrepappend, axis=0, return_index=True)
resfinal = restotalappend[indices]
if extended:
np.save('cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\PixelRep_{:d}x{:d}_xrest_rottrans_extended.npy'.format(k, k, k, k)
, pixrepfinal)
if k == 5:
np.save(
'cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\results_analysis_i_Scen_slope_offset_M1k_{:d}x{:d}_xrest_fixn4_'
'rottrans_extended.npy'.format(
k, k, k, k), resfinal)
else:
np.save(
'cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\results_analysis_i_Scen_slope_M1k_{:d}x{:d}_xrest_rottrans_'
'extended.npy'.format(
k, k, k, k), resfinal)
else:
np.save('cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\PixelRep_{:d}x{:d}_xrest_rottrans.npy'.format(k, k, k, k),
pixrepfinal)
if k==5:
np.save('cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\results_analysis_i_Scen_slope_offset_M1k_{:d}x{:d}_xrest_fixn4_rottrans.npy'.format(k, k, k, k), resfinal)
else:
np.save(
'cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\results_analysis_i_Scen_slope_M1k_{:d}x{:d}_xrest_rottrans.npy'.format(
k, k, k, k), resfinal)
return pixrepfinal, resfinal
extended=False
k=5
filter_size = (2, 2)
stride = (2, 2)
if k < 5:
datastring = r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\PixelRep_{:d}x{:d}.txt'.format(k, k)
else:
datastring = r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\PixelRep_{:d}x{:d}.npy'.format(k, k)
datastring_extended = r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\PixelRep_{:d}x{:d}_extended.npy'.\
format(k, k)
if k==5:
resultsstring=r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\results_analysis_unimodal_vs_oligomodal_vs_plurimodal_i_Scen_slope_modes_M1k_5x5_fixn4.txt'
elif k>6:
resultsstring = r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\results_analysis_unimodal_vs_oligomodal_vs_plurimodal_i_Scen_slope_modes_M1k_{:d}x{:d}_fixn4.txt'.format(
k, k)
resultsstring_extended = r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\results_analysis_unimodal_vs_oligomodal_vs_plurimodal_i_' \
r'Scen_slope_modes_M1k_{:d}x{:d}_extended.txt'.format(
k, k)
else:
resultsstring = r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\results_analysis_unimodal_vs_' \
r'oligomodal_vs_plurimodal_i_Scen_slope_modes_M1k_{:d}x{:d}_fixn4.txt'.format(k, k)
# data = np.loadtxt(datastring, delimiter=',')
# results = np.loadtxt(resultsstring, delimiter=',')
# np.save(r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\PixelRep_5x5.npy', data)
# np.save(r'C:\Users\ryanv\PycharmProjects\MetaCombi\results\modescaling\results_analysis_i_Scen_slope_offset_M1k_5x5_fixn4.npy', results)
if k < 5:
data = np.loadtxt(datastring, delimiter=',')
else:
data = np.load(datastring)
if extended:
data = np.append(data, np.load(datastring_extended), axis=0)
if k>6:
results = np.loadtxt(resultsstring, delimiter=',')
if extended:
results = np.append(results, np.loadtxt(resultsstring_extended, delimiter=','), axis=0)
else:
results = np.loadtxt(resultsstring, delimiter=',')
data = np.reshape(data, (-1, 2*k, 2*k))
x_total = data
"select only unimodal vs oligomodal (include stripmodes)"
ind_U = np.argwhere(results[:, 3].astype(int) == 1)
ind_O = np.argwhere(results[:, 3].astype(int) > 1)
ind_UO = np.append(ind_U, ind_O, axis=0)
x_total = x_total[ind_UO[:, 0]]
y_total = np.full_like(ind_U[:, 0], 0)
y_total = np.append(y_total, np.full_like(ind_O[:, 0], 1))
# y_total = results[ind_UO[:, 0], 1].astype(int)
"create a split of the data"
test_frac = 0.15
rest_frac = 1-test_frac
"create test set from original data (no rot + transl append)"
test_ind = np.arange(np.shape(x_total)[0])
np.random.shuffle(test_ind)
test_ind = test_ind[:int(test_frac*np.shape(x_total)[0])]
test_set = x_total[test_ind]
test_y = y_total[test_ind].astype(int)
rest_set = np.delete(x_total, test_ind, axis=0)
rest_y = np.delete(y_total, test_ind, axis=0)
"remove the C configs"
# indCt = np.argwhere(test_y==2)[:, 0]
# indCr = np.argwhere(rest_results[:, 1]==2)[:, 0]
# test_set = np.delete(test_set, indCt, axis=0)
# test_y = np.delete(test_y, indCt, axis=0)
#
# rest_set = np.delete(rest_set, indCr, axis=0)
# rest_results = np.delete(rest_results, indCr, axis=0)
"add translations + rotations to rest set"
# rest_set, rest_results = transrotconfigs(rest_set, rest_results, k, extended=extended)
# # rest_set = np.load('.\\cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\PixelRep_{:d}x{:d}_xrest_rottrans_extended.npy'
# # .format(k, k, k, k))
# # rest_results = np.load('.\\cnniter_HP_GS_SKF_{:d}x{:d}_AB_5050\\results_analysis_i_Scen_slope_M1k_{:d}x{:d}_xrest'
# # '_rottrans_extended.npy'.format(k, k, k, k))
#
# rest_set = np.reshape(rest_set, (-1, 2*k, 2*k))
# rest_y = rest_results[:, 1].astype(int)
"create periodic padding (make sure to set conv2d padding to valid!)"
# x_rest = np.tile(rest_set, (1, 3, 3))
# x_test = np.tile(test_set, (1, 3, 3))
restshape = np.shape(rest_set)
testshape = np.shape(test_set)
ind_horizontal = np.arange(0, restshape[1])
ind_horizontal = np.tile(ind_horizontal, 3)
ind_vertical = np.arange(0, restshape[2])
ind_vertical = np.tile(ind_vertical, 3)
# x_rest = np.broadcast_to(rest_set, (restshape[0], 3*restshape[1], 3*testshape[2]))
# x_test = np.broadcast_to(test_set, (testshape[0], 3*restshape[1], 3*testshape[2]))
# check how far the filter reaches with given stride and filter size over standard increased image of original
# size + filter_size. Add this length to the image if the extra length is not a multiple of k
x_over = ((filter_size[0]+np.ceil((2*k)/stride[0])*stride[0]) % (2*k+filter_size[0])) % k
y_over = ((filter_size[1]+np.ceil((2*k)/stride[1])*stride[1]) % (2*k+filter_size[1])) % k
x_left = np.floor(2*k-0.5*filter_size[0]).astype(int)
x_right = np.ceil(4*k+0.5*filter_size[0]+x_over).astype(int)
y_down = np.floor(2*k-0.5*filter_size[0]).astype(int)
y_up = np.ceil(4*k + 0.5*filter_size[0]+y_over).astype(int)
ind_horizontal = ind_horizontal[x_left:x_right]
ind_vertical = ind_vertical[y_down:y_up]
# x_rest = x_rest[:, x_left:x_right, y_down:y_up]
# x_test = x_test[:, x_left:x_right, y_down:y_up]
x_rest = np.zeros((restshape[0], restshape[1]+2, restshape[2] + 2))
x_test = np.zeros((testshape[0], testshape[1] + 2, testshape[2] + 2))
for i, ind in enumerate(ind_horizontal):
x_rest[:, i, :] = rest_set[:, ind, ind_vertical]
x_test[:, i, :] = test_set[:, ind, ind_vertical]
# x_rest = rest_set[:, ind_horizontal, ind_vertical]
# x_test = test_set[:, ind_horizontal, ind_vertical]
rest_set = x_rest.copy()
"make 5050 distribution rest set"
x_rest, rest_y = unison_shuffled_copies(x_rest, rest_y)
ind_U = np.argwhere(rest_y == 0)
ind_O = np.argwhere(rest_y == 1)
ind_U_short = ind_U[:int(ind_O.shape[0]), 0]
ind_O_short = ind_O[:int(ind_U.shape[0]), 0]
indices = np.append(ind_U_short, ind_O_short)
x_rest, y_rest = x_rest[indices], rest_y[indices]
"add channel axis"
x_rest = x_rest[..., np.newaxis]
rest_set = rest_set[..., np.newaxis]
x_test = x_test[..., np.newaxis]
"save the data"
if extended:
np.savez('data_unimodal_vs_oligomodal_inc_stripmodes_train_trainraw_test_{:d}x{:d}_extended.npz'.format(k, k,
k, k)
, x_rest=x_rest, y_rest=y_rest, x_rest_raw=rest_set, y_rest_raw=rest_y, x_test=x_test, y_test=test_y)
else:
np.savez('data_unimodal_vs_oligomodal_inc_stripmodes_train_trainraw_test_{:d}x{:d}.npz'.format(k, k, k, k),
x_rest=x_rest, y_rest=y_rest, x_rest_raw = rest_set, y_rest_raw = rest_y, x_test = x_test, y_test=test_y)
# np.savez(os.path.join(masterdir, 'data_prek_xy_train_trainraw_test.npz'), x_rest=x_rest, y_rest=y_rest, x_rest_raw=x_rest_raw, y_rest_raw=y_rest_raw, x_test=x_test, y_test=y_test)