-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
1467 lines (1152 loc) · 54.9 KB
/
utils.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
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import imageio
import matplotlib
matplotlib.use('Agg')
import torchvision
import torch
from datetime import datetime
import time
from sklearn.metrics import confusion_matrix, classification_report
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import average_precision_score
import numpy as np
from sklearn.preprocessing import label_binarize
from scipy.stats import pearsonr
import os
import seaborn as sns
import warnings
warnings.filterwarnings("ignore")
import pandas as pd
from sklearn import metrics
import csv
import altair as alt
from sklearn.metrics import roc_auc_score, roc_curve, accuracy_score, auc
from PIL import Image, ImageDraw
from altair_saver import save
import matplotlib.pyplot as plt
import json
def imshow(inp, marker, title=None):
inp = inp.numpy().transpose((1, 2, 0))
# plt.figure(figsize=(10, 10))
fig = plt.figure(figsize=(10, 10))
plt.axis('off')
plt.imshow(inp)
if title is not None:
plt.title(title)
now = datetime.now()
now = now.strftime("%d%m%Y%H%M%S")
fig.savefig('result/visual_result_' + '_' + now + '_' + marker + '.pdf')
plt.pause(0.001)
def show_databatch(class_names, inputs, classes, marker):
out = torchvision.utils.make_grid(inputs)
imshow(out, marker, title=[class_names[x] for x in classes])
def show_databatch_regression(inputs, classes, marker):
out = torchvision.utils.make_grid(inputs)
imshow(out, marker, title=[round(x, 2) for x in classes.numpy()])
def plot_confusion_matrix(cm, target_names, title='Confusion matrix', cmap=None, normalize=True):
"""
given a sklearn confusion matrix (cm), make a nice plot
Arguments
---------
cm: confusion matrix from sklearn.metrics.confusion_matrix
target_names: given classification classes such as [0, 1, 2]
the class names, for example: ['high', 'medium', 'low']
title: the text to display at the top of the matrix
cmap: the gradient of the values displayed from matplotlib.pyplot.cm
see http://matplotlib.org/examples/color/colormaps_reference.html
plt.get_cmap('jet') or plt.cm.Blues
normalize: If False, plot the raw numbers
If True, plot the proportions
Usage
-----
plot_confusion_matrix(cm = cm, # confusion matrix created by
# sklearn.metrics.confusion_matrix
normalize = True, # show proportions
target_names = y_labels_vals, # list of names of the classes
title = best_estimator_name) # title of graph
Citiation
---------
http://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html
"""
import matplotlib.pyplot as plt
import numpy as np
import itertools
font = {'family': 'Calibri',
'weight': 'normal',
'size': 18}
matplotlib.rc('font', **font)
accuracy = np.trace(cm) / float(np.sum(cm))
misclass = 1 - accuracy
# SMALL_SIZE = 8
# matplotlib.rc('font', size=SMALL_SIZE)
# matplotlib.rc('axes', titlesize=SMALL_SIZE)
if cmap is None:
cmap = plt.get_cmap('Blues')
fig = plt.figure(figsize=(8, 6))
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
plt.colorbar()
if target_names is not None:
tick_marks = np.arange(len(target_names))
plt.xticks(tick_marks, target_names, rotation=45)
plt.yticks(tick_marks, target_names)
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
thresh = cm.max() / 1.5 if normalize else cm.max() / 2
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
if normalize:
plt.text(j, i, "{:0.4f}".format(cm[i, j]),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
else:
plt.text(j, i, "{:,}".format(cm[i, j]),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout(pad=3)
plt.ylabel('True category')
# plt.xlabel('Predicted label\naccuracy={:0.4f}; misclass={:0.4f}'.format(accuracy, misclass))
plt.xlabel('Predicted category \nAccuracy={:0.4f}'.format(accuracy))
now = datetime.now()
now = now.strftime("%d%m%Y%H%M%S")
fig.savefig('result/conf_matrix_result_' + '_' + now + '.pdf')
plt.show()
def plot_precision_recall(n_classes, Y_test, y_score):
###############################################################################
# The average precision score in multi-label settings
# np_labels_bn= label_binarize(np_labels, classes=[0, 1, 2])
# np_preds_bn = label_binarize(filter_preds, classes=[0, 1, 2])
# For each class
precision = dict()
recall = dict()
average_precision = dict()
for i in range(n_classes):
precision[i], recall[i], _ = precision_recall_curve(Y_test[i], y_score[i])
average_precision[i] = average_precision_score(Y_test[i], y_score[i])
# A "micro-average": quantifying score on all classes jointly
precision["micro"], recall["micro"], _ = precision_recall_curve(Y_test.ravel(),
y_score.ravel())
average_precision["micro"] = average_precision_score(Y_test, y_score,
average="micro")
print('Average precision score, micro-averaged over all classes: {0:0.2f}'
.format(average_precision["micro"]))
###############################################################################
# Plot the micro-averaged Precision-Recall curve
# ...............................................
#
plt.figure()
plt.step(recall['micro'], precision['micro'], where='post')
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.ylim([0.0, 1.05])
plt.xlim([0.0, 1.0])
plt.title(
'Average precision score, micro-averaged over all classes: AP={0:0.2f}'
.format(average_precision["micro"]))
###############################################################################
# Plot Precision-Recall curve for each class and iso-f1 curves
# .............................................................
#
from itertools import cycle
# setup plot details
colors = cycle(['navy', 'turquoise', 'darkorange', 'cornflowerblue', 'teal'])
plt.figure(figsize=(7, 8))
f_scores = np.linspace(0.2, 0.8, num=4)
lines = []
labels = []
for f_score in f_scores:
x = np.linspace(0.01, 1)
y = f_score * x / (2 * x - f_score)
l, = plt.plot(x[y >= 0], y[y >= 0], color='gray', alpha=0.2)
plt.annotate('f1={0:0.1f}'.format(f_score), xy=(0.9, y[45] + 0.02))
lines.append(l)
labels.append('iso-f1 curves')
l, = plt.plot(recall["micro"], precision["micro"], color='gold', lw=2)
lines.append(l)
labels.append('micro-average Precision-recall (area = {0:0.2f})'
''.format(average_precision["micro"]))
for i, color in zip(range(n_classes), colors):
l, = plt.plot(recall[i], precision[i], color=color, lw=2)
lines.append(l)
labels.append('Precision-recall for class {0} (area = {1:0.2f})'
''.format(i, average_precision[i]))
fig = plt.gcf()
fig.subplots_adjust(bottom=0.25)
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.title('Extension of Precision-Recall curve to multi-class')
plt.legend(lines, labels, loc=(0, -.38), prop=dict(size=14))
plt.show()
def filter_array(np_labels, np_preds, n_classes):
results_labels = {}
results_preds = {}
for i in range(n_classes):
condition = (np_labels == i)
results_labels[i] = label_binarize(np.extract(condition, np_labels), classes=[0, 1, 2])
results_preds[i] = label_binarize(np.extract(condition, np_preds), classes=[0, 1, 2])
return results_labels, results_preds
def plot_correaltion(y, y_cv, score_cv_r, score_cv_r2, mse_cv, test_folder):
# Fit a line to the C vs response
z = np.polyfit(y, y_cv, 1)
sns.set()
fig, ax = plt.subplots(figsize=(6, 6))
ax.scatter(y, y_cv, c='red', edgecolors='k')
# Plot the best fit line
ax.plot(y, np.polyval(z, y), c='blue', linewidth=1)
plt.title(test_folder + '\n $R^{2}$: %5.4f' % score_cv_r2 + ' \n R: %5.4f' % score_cv_r + '; MSE: %5.4f' % mse_cv)
plt.ylabel('Predicted valence rating')
plt.xlabel('Measured valence rating')
# set axes range
plt.xlim(1, 9)
plt.ylim(1, 9)
plt.show()
now = datetime.now()
if not os.path.exists('./result/'):
os.makedirs('./result/')
fig.savefig('result/' + test_folder + '_regression_plot' + '_' + str(now) + '_' + 'R-%5.4f' % score_cv_r + '.pdf')
def plot_tuning_accuracy(strenths, accuracy, category, tuning_category, test_folder,
tuning_layer, result_folder, plot_extra_title=None):
sns.set()
fig, ax = plt.subplots(figsize=(8, 8))
ax.plot(strenths, accuracy, color='black', linewidth=1, marker='o', markerfacecolor='black', markersize=3)
if plot_extra_title is not None:
title = 'Accuracy of {} when tuning {} for {} layer on dataset {} \n {} '.format(category, tuning_category,
tuning_layer,
test_folder, plot_extra_title)
else:
title = 'Accuracy of {} when tuning {} for {} layer on dataset {} '.format(category,
tuning_category, tuning_layer,
test_folder)
plt.title(title)
plt.ylabel('Accuracy')
plt.xlabel('Tuning strenth')
plt.show()
now = datetime.now()
filename = tuning_layer + 'layer_' + tuning_category + '-tuning_category_' \
+ category + '_response_' + str(now) + '.pdf'
path_to_save_result = os.path.join(result_folder, filename)
fig.savefig(path_to_save_result)
# plot one layer each time
def plot_tuning_accuracy_changes(performance_changes, class_names, tuning_category, test_folder, plot_extra_title=None):
lists = sorted(performance_changes.items()) # sorted by key, return a list of tuples
layers, per_changes = zip(*lists) # unpack a list of pairs into two tuples
for cat in class_names:
scores_to_plot = [p[cat] for p in per_changes]
sns.set()
fig, ax = plt.subplots(figsize=(8, 8))
ax.plot(layers, scores_to_plot, color='black', linewidth=1, marker='o', markerfacecolor='black', markersize=3)
if plot_extra_title is not None:
title = 'Performance changes of {} when tuning {} on dataset {} \n {} '.format(cat, tuning_category,
test_folder,
plot_extra_title)
else:
title = 'Performance changes of {} when tuning {} on dataset {} '.format(cat, tuning_category,
test_folder
)
plt.title(title)
plt.ylabel('Performance changes (%pts)')
plt.xlabel('Tuning layers')
plt.xlim(min(layers), max(layers))
plt.show()
now = datetime.now()
result_folder = './result/tuning/'
if not os.path.exists(result_folder):
os.makedirs(result_folder)
fig.savefig(result_folder + 'performance_changes_all_layers_' + tuning_category + '-tuning_category_'
+ cat + '_response_' + str(now) + '.pdf')
# plot all layers
def plot_tuning_performance_changes_pygal(performance_changes, class_names, tuning_category, test_folder, result_folder,
plot_extra_title=None):
import pygal
from pygal.style import Style
custom_style = Style(
colors=('#0343df', '#e50000', '#ffff14', '#929591'),
font_family='Roboto,Helvetica,Arial,sans-serif',
background='transparent',
label_font_size=18,
)
if plot_extra_title is not None:
title = 'Performance changes of neutral, pleasant, unpleasant when tuning {} on dataset {} \n {} '.format(
tuning_category,
test_folder,
plot_extra_title)
else:
title = 'Performance changes of neutral, pleasant, unpleasant when tuning {} on dataset {} '.format(
tuning_category,
test_folder
)
c = pygal.Bar(
title=title,
style=custom_style,
y_title='Performance changes (%pts)',
x_title='Tuning layers',
width=1200,
x_label_rotation=270,
)
lists = sorted(performance_changes.items()) # sorted by key, return a list of tuples
layers, per_changes = zip(*lists) # unpack a list of pairs into two tuples
for cat in class_names:
scores_to_plot = [p[cat] for p in per_changes]
c.add(cat, scores_to_plot)
c.x_labels = layers
# c.render_to_file('pygal.svg')
now = datetime.now()
if not os.path.exists(result_folder):
os.makedirs(result_folder)
filename = 'performance_changes_all_layers_' + tuning_category + '-tuning_category_' + 'all_emotions_response_' + str(
now) + '.svg'
path_to_save_result = os.path.join(result_folder, filename)
c.render_to_file(path_to_save_result)
def plot_tuning_performance_changes_altair(data_to_plot, tuning_category, test_folder, result_folder,
plot_extra_title=None):
alt.renderers.enable('altair_viewer')
df = pd.DataFrame({'layer': data_to_plot['layerName'],
'per_ch': data_to_plot['performance_change'],
'category': data_to_plot['response_category']
})
res_categories = np.unique(data_to_plot['response_category'].values)
if res_categories.size == 3:
cmap = {
'neutral': '#0343df',
'pleasant': '#e50000',
'unpleasant': '#ffff14'
}
cat_names = res_categories[0], res_categories[1], \
res_categories[2]
else:
cmap = {
res_categories[0]: assign_color(res_categories[0]),
res_categories[1]: assign_color(res_categories[1]),
}
cat_names = res_categories[0], res_categories[1]
if plot_extra_title is not None:
title = 'Performance changes of {} when tuning {} on dataset {} \n {} '.format(cat_names, tuning_category,
test_folder,
plot_extra_title)
else:
title = 'Performance changes of {} when tuning {} on dataset {} '.format(cat_names, tuning_category,
test_folder
)
df['layer'] = df['layer'].astype(str)
# We're still assigning, e.g. 'party' to x, but now we've wrapped it
# in alt.X in order to specify its styling
chart = alt.Chart(df).mark_bar().encode(
x=alt.X('category', title=None),
y=alt.Y('per_ch', title='Performance changes (% pts)'),
column=alt.Column('layer', sort=list(df['category']), title='Layers'),
color=alt.Color('category', scale=alt.Scale(domain=list(cmap.keys()), range=list(cmap.values())))
).properties(title=title)
# c.render_to_file('pygal.svg')
now = datetime.now()
if not os.path.exists(result_folder):
os.makedirs(result_folder)
filename = 'performance_changes_all_layers_' + tuning_category + '-tuning_category_' \
+ 'all_emotions_response_' + str(now) + '.html'
path_to_save_result = os.path.join(result_folder, filename)
chart.save(path_to_save_result)
def plot_tuning_performance_changes_sns(data_to_plot, tuning_category, test_folder,
result_folder, is_lesoin=False, plot_extra_title=None):
sns.set(style="ticks", color_codes=True)
sns.set(font='Calibri')
sns.set(font_scale=1.5)
res_categories = np.unique(data_to_plot['response_category'].values)
if res_categories.size == 3:
cat_names = res_categories[0], res_categories[1], \
res_categories[2]
else:
cat_names = res_categories[0], res_categories[1]
if is_lesoin:
title_sub = 'Lesion'
else:
title_sub = 'Tuning'
if plot_extra_title is not None:
# title = 'Performance changes of {} when tuning {} on dataset {} \n {} '.format(cat_names, tuning_category,
# test_folder,
# plot_extra_title)
title = title_sub + ' {} \n {}'.format(tuning_category, plot_extra_title)
else:
title = title_sub + ' {}'.format(tuning_category)
# title = 'Performance changes of {} when tuning {} on dataset {} '.format(cat_names, tuning_category,
# test_folder
# )
g = sns.catplot(x="layerName", y="performance_change", hue="response_category", kind="bar",
legend_out=False, aspect=3, data=data_to_plot)
g.set_axis_labels("", "Performance changes (% pts)")
plt.title(title)
# g.set(ylim=(-100,0))
g._legend.set_title('')
now = datetime.now()
if not os.path.exists(result_folder):
os.makedirs(result_folder)
filename = 'performance_changes_all_layers_' + tuning_category + '-tuning_category_' \
+ 'all_emotions_response_' + str(now) + '_sns_.pdf'
path_to_save_result = os.path.join(result_folder, filename)
g.savefig(path_to_save_result)
def convert_layerID_to_convName(layer):
switcher = {
1: "conv1",
3: "conv2",
6: "conv3",
8: "conv4",
11: "conv5",
13: "conv6",
15: "conv7",
18: "conv8",
20: "conv9",
22: "conv10",
25: "conv11",
27: "conv12",
29: "conv13"
}
# get() method of dictionary data type returns
# value of passed argument if it is present
# in dictionary otherwise second argument will
# be assigned as default value of passed argument
return switcher.get(int(layer), "nothing")
def assign_color(category):
if 'unpleasant' == category:
color = '#ffff14'
elif 'pleasant' == category:
color = '#e50000'
elif 'non_unpleasant' == category:
color = '#AF7AC5'
elif 'non_pleasant' == category:
color = '#2ECC71'
elif 'neutral' == category:
color = '#0343df'
else:
color = '#566573'
return color
def plot_selectivity_channel_overlabp_altair(data_to_plot, result_folder, overlap_with, plot_extra_title=None):
alt.renderers.enable('altair_viewer')
df = pd.DataFrame({'layer': data_to_plot['layer'],
overlap_with: data_to_plot[overlap_with],
'category': data_to_plot['category']
})
res_categories = np.unique(data_to_plot['category'].values)
if res_categories.size == 3:
cmap = {
'neutral': '#0343df',
'pleasant': '#e50000',
'unpleasant': '#ffff14'
}
cat_names = res_categories[0], res_categories[1], \
res_categories[2]
elif res_categories.size == 2:
cmap = {
res_categories[0]: assign_color(res_categories[0]),
res_categories[1]: assign_color(res_categories[1]),
}
cat_names = res_categories[0], res_categories[1]
else:
cmap = {
res_categories[0]: assign_color(res_categories[0]),
}
cat_names = res_categories[0]
if plot_extra_title is not None:
title = 'Overlap with {} of selective channels for {} \n {}'.format(overlap_with, cat_names, plot_extra_title)
else:
title = 'Overlap with {} of selective channels for {} '.format(overlap_with, cat_names)
df['layer'] = df['layer'].astype(str)
# We're still assigning, e.g. 'party' to x, but now we've wrapped it
# in alt.X in order to specify its styling
chart = alt.Chart(df).mark_bar().encode(
x=alt.X('category', title=None),
y=alt.Y(overlap_with, title='Overlap (% pts)'),
column=alt.Column('layer', sort=list(df['category']), title='Layers'),
color=alt.Color('category', scale=alt.Scale(domain=list(cmap.keys()), range=list(cmap.values())))
).properties(title=title)
# c.render_to_file('pygal.svg')
now = datetime.now()
if not os.path.exists(result_folder):
os.makedirs(result_folder)
filename = overlap_with + '_selectivity_overlap_summary_' + str(now) + '.html'
path_to_save_result = os.path.join(result_folder, filename)
chart.save(path_to_save_result)
def plot_num_channel_within_layers(data_to_plot, result_folder, plot_extra_title=None):
sns.set(style="ticks", color_codes=True)
sns_plot = sns.catplot(x="layer", y="num_channels", hue="category", kind="bar", data=data_to_plot)
now = datetime.now()
if not os.path.exists(result_folder):
os.makedirs(result_folder)
filename = 'number_selectivity_channels_within_layers_' + str(now) + '.pdf'
path_to_save_result = os.path.join(result_folder, filename)
sns_plot.savefig(path_to_save_result)
# for classification
def cls_visualize_model(dataloaders, class_names, vgg, test_folder, num_images, device):
was_training = vgg.training
# Set model for evaluation
vgg.train(False)
vgg.eval()
images_so_far = 0
for i, data in enumerate(dataloaders[test_folder]):
inputs, labels = data
size = inputs.size()[0]
inputs = inputs.to(device)
labels = labels.to(device)
outputs = vgg(inputs)
_, preds = torch.max(outputs.data, 1)
predicted_labels = [preds[j] for j in range(inputs.size()[0])]
print("Ground truth:")
show_databatch(class_names, inputs.data.cpu(), labels.data.cpu(), 'Ground truth')
print("Prediction:")
show_databatch(class_names, inputs.data.cpu(), predicted_labels, 'Prediction')
del inputs, labels, outputs, preds, predicted_labels
torch.cuda.empty_cache()
images_so_far += size
if images_so_far >= num_images:
break
vgg.train(mode=was_training) # Revert model back to original training state
def cls_process_output(criterion, outputs, labels, loss_test, acc_test, labels_list, preds_list, i, preds_proba_list,
positive_class_idx=1, istuning=False, strenth=None):
probabilities = torch.nn.functional.softmax(outputs, dim=1)
_, preds = torch.max(outputs.data, 1)
loss = criterion(outputs, labels)
loss_test += loss.data
acc_test += torch.sum(preds == labels.data)
if istuning:
print(i + 1, strenth, 'loss: {:.4f}'.format(loss.cpu().item()))
else:
print(i + 1, 'loss: {:.4f}'.format(loss.cpu().item()))
labels_list.extend(labels.cpu().numpy())
preds_list.extend(preds.cpu().numpy())
#### keep probabilities for the positive outcome only how to find which dim is the positive class ????
preds_proba_list.extend(probabilities[:, 1].detach().cpu().numpy())
return loss_test, acc_test, labels_list, preds_list, preds_proba_list
def print_result(loss_test, acc_test, data_size, class_names, labels_list, preds_list, preds_proba_list):
# calculate accuracy
avg_loss = loss_test / data_size
avg_acc = acc_test.double() / data_size
# plot confusion matrix
conf_matr = confusion_matrix(labels_list, preds_list)
plot_confusion_matrix(conf_matr, normalize=False, target_names=class_names, title="Confusion Matrix")
# auc_score = roc_auc_score(labels_list, preds_list)
# acc_score = accuracy_score(labels_list, preds_list)
fpr, tpr, thresholds = roc_curve(labels_list, preds_proba_list, pos_label=2)
auc_score = auc(fpr, tpr)
print("Avg loss (test): {:.4f}".format(avg_loss))
print("Avg acc (test): {:.4f}".format(avg_acc))
print("ROC AUC (test):{:.4f}".format(auc_score))
print(conf_matr)
print('-' * 10)
report = classification_report(labels_list, preds_list, digits=4)
print(report)
print('-' * 10)
# # Print Area Under Curve
plt.figure(figsize=[8, 6])
plt.title('Receiver Operating Characteristic (ROC)')
plt.plot(fpr, tpr, 'b', label='ROC curve (area = %0.4f)' % auc_score)
plt.legend(loc='lower right')
plt.plot([0, 1], [0, 1], 'r--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.0])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')
now = datetime.now()
plt.savefig('./result/roc_' + str(now) + '.pdf')
plt.show()
# for classification
def cls_eval_model(dataloaders, dataset_sizes, class_names, test_folder, vgg, criterion, device,
istunning=False, tuningstrenths=None, allcategory=None,
allcategory_idx=None, tuning_category=None,
plot_extra_title=None, tuning_layer=None, result_folder=None, model_metric='AUC',
is_record_tuning_detail=True):
since = time.time()
loss_test = 0
acc_test = 0
labels_list = []
preds_list = []
preds_proba_list = []
if istunning:
tuning_labels_list = {}
tuning_preds_list = {}
tuning_loss_test = {}
tuning_acc_test = {}
tuning_preds_proba_list = {}
for strenth in tuningstrenths:
tuning_labels_list[strenth] = []
tuning_preds_list[strenth] = []
tuning_preds_proba_list[strenth] = []
tuning_loss_test[strenth] = 0
tuning_acc_test[strenth] = 0
test_batches = len(dataloaders[test_folder])
print("Evaluating model")
print('-' * 10)
for i, data in enumerate(dataloaders[test_folder]):
if i % 100 == 0:
print("\rTest batch {}/{} \n".format(i, test_batches), end='', flush=True)
vgg.train(False)
vgg.eval()
inputs, labels = data
inputs = inputs.to(device)
labels = labels.to(device)
positive_class_idx = 1
for c in class_names:
if 'non' not in c:
positive_class_idx = allcategory_idx[c]
break
if not istunning:
outputs = vgg(inputs)
loss_test, acc_test, labels_list, preds_list, preds_proba_list = cls_process_output(
criterion, outputs, labels, loss_test, acc_test, labels_list, preds_list, i, preds_proba_list,
positive_class_idx=positive_class_idx)
else:
labels_names = [class_names[x] for x in labels]
outputs = vgg(inputs, labels_names)
for strenth in tuningstrenths:
currrent_output = outputs[strenth]
tuning_loss_test[strenth], tuning_acc_test[strenth], tuning_labels_list[strenth], \
tuning_preds_list[strenth], tuning_preds_proba_list[strenth] = cls_process_output(
criterion, currrent_output, labels, tuning_loss_test[strenth],
tuning_acc_test[strenth], tuning_labels_list[strenth], tuning_preds_list[strenth], i,
tuning_preds_proba_list[strenth],
positive_class_idx=positive_class_idx, istuning=True, strenth=strenth)
del inputs, labels, outputs
torch.cuda.empty_cache()
elapsed_time = time.time() - since
print()
if not istunning:
print_result(loss_test, acc_test, dataset_sizes[test_folder], class_names, labels_list, preds_list,
preds_proba_list)
else:
performance_scores = {}
for strenth in tuningstrenths:
print('\n Tuning strenth:' + str(strenth))
print_result(tuning_loss_test[strenth], tuning_acc_test[strenth],
dataset_sizes[test_folder], class_names, tuning_labels_list[strenth],
tuning_preds_list[strenth], tuning_preds_proba_list[strenth])
if model_metric == 'AUC': # we may need to get the prediction probability (currently, it is binary available))
fpr, tpr, _ = roc_curve(tuning_labels_list[strenth], tuning_preds_proba_list[strenth])
clf_rep = auc(fpr, tpr)
performance_scores[strenth] = clf_rep
elif model_metric == "accuracy":
clf_rep = accuracy_score(tuning_labels_list[strenth], tuning_preds_list[strenth])
performance_scores[strenth] = clf_rep
else: # F1-score (can work multiple classes)
clf_rep = metrics.precision_recall_fscore_support(tuning_labels_list[strenth],
tuning_preds_list[strenth])
performance_scores[strenth] = clf_rep[2].round(4)
if is_record_tuning_detail:
details_csv_schema = ['layer', 'tuning_strength', 'tuning_category', 'response_category', 'new_performance']
if len(performance_scores) > 1:
lists = sorted(performance_scores.items()) # sorted by key, return a list of tuples
strenths, scores = zip(*lists) # unpack a list of pairs into two tuples
now = datetime.now()
details_tuning_csv_name = tuning_layer + 'layer_' + tuning_category + '-tuning_category_' + str(
now) + '.csv'
details_tuning_csv = os.path.join(result_folder, details_tuning_csv_name)
with open(details_tuning_csv, 'w', newline='') as details_file:
writer_detail = csv.writer(details_file)
# Gives the header name row into csv
writer_detail.writerow([g for g in details_csv_schema])
for cat in allcategory:
cat_idx = allcategory_idx[cat]
if model_metric == 'accuracy':
scores_to_plot = [s for s in scores]
elif model_metric == 'F1-score':
scores_to_plot = [s[cat_idx] for s in scores]
else: # AUC only available for binary classification
scores_to_plot = [s for s in scores]
plot_tuning_accuracy(strenths, scores_to_plot, cat,
tuning_category, test_folder, tuning_layer, result_folder,
plot_extra_title)
for i in range(len(strenths)):
writer_detail.writerow(
[tuning_layer, strenths[i], tuning_category, cat, scores_to_plot[i]])
else:
for cat in allcategory:
score = performance_scores[tuningstrenths[0]]
cat_idx = allcategory_idx[cat]
if model_metric == 'AUC':
print('Tuning {}: {} score {}'.format(tuning_category, cat, score))
elif model_metric == "accuracy":
print('Tuning {}: {} score {}'.format(tuning_category, cat, score))
else:
print('Tuning {}: {} score {}'.format(tuning_category, cat, score[cat_idx]))
print("\nEvaluation completed in {:.0f}m {:.0f}s".format(elapsed_time // 60, elapsed_time % 60))
if istunning and len(performance_scores) > 1:
return performance_scores
# for regression
def reg_visualize_model(dataloaders, vgg, test_folder, num_images, device):
was_training = vgg.training
# Set model for evaluation
vgg.train(False)
vgg.eval()
images_so_far = 0
for i, data in enumerate(dataloaders[test_folder]):
_, inputs, labels = data
size = inputs.size()[0]
inputs = inputs.to(device)
labels = labels.to(device)
outputs = vgg(inputs)
# _, preds = torch.max(outputs.data, 1)
predicted_labels = outputs.view(labels.size()) # [outputs[j] for j in range(inputs.size()[0])]
print("Ground truth:")
if num_images <= inputs.shape[0]:
show_databatch_regression(inputs.data[:num_images].cpu(), labels.data[:num_images].cpu(), 'Ground truth')
else:
show_databatch_regression(inputs.data.cpu(), labels.data.cpu(), 'Ground truth')
print("Prediction:")
if num_images <= inputs.shape[0]:
show_databatch_regression(inputs.data[:num_images].cpu(), predicted_labels[:num_images].detach().cpu(),
'Prediction')
else:
show_databatch_regression(inputs.data.cpu(), predicted_labels, 'Prediction')
del inputs, labels, outputs, predicted_labels
torch.cuda.empty_cache()
images_so_far += size
if images_so_far >= num_images:
break
vgg.train(mode=was_training) # Revert model back to original training state
# for regression
def reg_eval_model(dataloaders, dataset_sizes, test_folder, vgg, criterion, device):
since = time.time()
loss_test = 0
labels_list = []
preds_list = []
# test_batches = len(dataloaders[test_folder])
print("Evaluating model")
print('-' * 10)
for i, data in enumerate(dataloaders[test_folder]):
# if i % 100 == 0:
# print("\rTest batch {}/{} \n".format(i, test_batches), end='', flush=True)
# print(i, data['image'].size(), data['lable'].size())
_, inputs, labels = data
inputs = inputs.to(device)
labels = labels.to(device)
vgg.train(False)
vgg.eval()
outputs = vgg(inputs)
outputs = 1 + outputs * (
9 - 1) # normalize to 1-9 Because our network last layer is sigmoid, which gives 0-1 values to restrict the range of outputs to be [0-1] or [1-9]
loss = criterion(outputs.view(labels.size()), labels.float())
loss_test += loss.data
# print(i+1, '{:.4f}'.format(loss.cpu().item()))
# here taking lots of time to fix the GPU out of Memory issue: need to be numpy()
labels_list.extend(labels.cpu().numpy())
preds = outputs.reshape(labels.shape).cpu().detach().numpy()
preds_list.extend(preds)
# clean the cache
del inputs, labels, outputs, loss, preds
torch.cuda.empty_cache()
# calculate correalation R & mean valence
original_mean_valence = np.mean(labels_list)
decoded_mean_valence = np.mean(preds_list)
avg_loss = loss_test / dataset_sizes[test_folder]
score_c_r, _ = pearsonr(labels_list, preds_list)
score_cv_r2 = np.around(score_c_r ** 2, 2) # r2_score(labels_list, preds_list)
# mse_loss = mean_squared_error(labels_list, preds_list)
elapsed_time = time.time() - since
# print()
print("Evaluation completed in {:.0f}m {:.0f}s".format(elapsed_time // 60, elapsed_time % 60))
print("Mean value of original valence {:.4f}".format(original_mean_valence))
print("Mean value of model's decoded valence {:.4f}".format(decoded_mean_valence))
print("Avg MSE loss (test): {:.4f}".format(avg_loss))
print("R (test): {:.4f}".format(score_c_r))
print('R2 : %5.4f' % score_cv_r2)
# plot correatlion
plot_correaltion(labels_list, preds_list, score_c_r, score_cv_r2, avg_loss, test_folder)
print('-' * 10)
return score_c_r
def cond_eval_model(dataloaders, dataset_sizes, test_folder, vgg, criterion, device, is_gist=False, is_saliency=False):
since = time.time()
loss_test = 0
labels_list = []
preds_list = []
# test_batches = len(dataloaders[test_folder])
print("Evaluating model")
print('-' * 10)
for i, data in enumerate(dataloaders[test_folder]):
# if i % 100 == 0:
# print("\rTest batch {}/{} \n".format(i, test_batches), end='', flush=True)
# print(i, data['image'].size(), data['lable'].size())
_, inputs, labels = data
if not is_gist and not is_saliency:
inputs = inputs.to(device)
elif is_gist and is_saliency:
inputs[0] = inputs[0].to(device)
inputs[1] = inputs[1].to(device)
inputs[2] = inputs[2].to(device)
elif is_gist or is_saliency:
inputs[0] = inputs[0].to(device)
inputs[1] = inputs[1].to(device)
labels = labels.to(device)
vgg.train(False)
vgg.eval()
outputs = vgg(inputs)
outputs = 1 + outputs * (
9 - 1) # normalize to 1-9 Because our network last layer is sigmoid, which gives 0-1 values to restrict the range of outputs to be [0-1] or [1-9]
loss = criterion(outputs.view(labels.size()), labels.float())
loss_test += loss.data
# print(i+1, '{:.4f}'.format(loss.cpu().item()))
# here taking lots of time to fix the GPU out of Memory issue: need to be numpy()
labels_list.extend(labels.cpu().numpy())
preds = outputs.reshape(labels.shape).cpu().detach().numpy()
preds_list.extend(preds)
# clean the cache
del inputs, labels, outputs, loss, preds
torch.cuda.empty_cache()
# calculate correalation R & mean valence
original_mean_valence = np.mean(labels_list)
decoded_mean_valence = np.mean(preds_list)
avg_loss = loss_test / dataset_sizes[test_folder]
score_c_r, _ = pearsonr(labels_list, preds_list)
score_cv_r2 = np.around(score_c_r ** 2, 2) # r2_score(labels_list, preds_list)
# mse_loss = mean_squared_error(labels_list, preds_list)
elapsed_time = time.time() - since