-
Notifications
You must be signed in to change notification settings - Fork 3
/
selection.py
1824 lines (1515 loc) · 79.2 KB
/
selection.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 argparse
import os,sys
import numpy as np
import random
import csv, ast
import matplotlib.pyplot as plt
from PIL import Image
import copy
from tqdm import tqdm
import sklearn
import time
import torch
import torchvision
import torch.nn.functional as F
import torch.optim as optim
import torch.nn as nn
from torchvision.utils import save_image
from torchvision import datasets, transforms
import torchvision.models as models
import mymodels
from scipy.spatial import distance
import torch_geometric
from torch_geometric.nn import knn_graph
import torch_geometric.nn as pyg_nn
import torch_geometric.utils as pyg_utils
import torch_geometric.transforms as T
from torch_geometric.data import Data, DataLoader
from torch_cluster import knn
from typing import Optional
from sklearn.cluster import KMeans, MiniBatchKMeans
from sklearn.manifold import TSNE
from scipy.special import softmax
from scipy.spatial.distance import pdist, squareform
import pandas as pd
# import networkx as nx
from gragh import propogate
from utils import AverageMeter, RecorderMeter
import data
# device: gpu or cpu
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# device = torch.device("cpu")
# CUDA_LAUNCH_BLOCKING=1
model_names = sorted(name for name in mymodels.__dict__
if name.islower() and not name.startswith("__")
and callable(mymodels.__dict__[name]))
################# Options ##################################################
############################################################################
parser = argparse.ArgumentParser("Arguments for sample selection.")
parser.add_argument('--data_path',
default='../../../datasets/',
type=str,
help='Path to dataset')
parser.add_argument('--dataset',
type=str,
choices=['cifar10', 'cifar100', 'imagenet', 'tiny-imagenet-200', 'svhn', 'stl10', 'mnist', 'emnist', 'gtsrb'],
help='Choices: cifar10, cifar100, imagenet, svhn, stl10, mnist, gtsrb.')
parser.add_argument('--batch_size',
type=int,
default=128,
help='Batch size')
parser.add_argument('--num_workers',
type=int,
default=8,
help='Batch size')
# test model
parser.add_argument('--model2test_arch',
metavar='ARCH',
choices=model_names,
help='model architecture: ' + ' | '.join(model_names) +
' (default: resnext29_8_64)')
parser.add_argument("--model2test_path",
type=str,
default=None,
help=" model of test center (resnet18, resnet50)")
parser.add_argument("--model_number", type=int,
help='choose from three different settings: all ones, random 1, random 2')
# test config
# baseline
parser.add_argument("--baseline_gini",
action='store_true',
help='using gini technique to select test cases')
parser.add_argument("--baseline_mcp",
action='store_true',
help='using mcp technique to select test cases')
parser.add_argument("--baseline_dsa",
action='store_true',
help='using distance surprise adaquacy technique to select test cases')
parser.add_argument("--baseline_uncertainty",
action='store_true',
help='using dropout technique to estimate the uncertainty for selecting test cases')
parser.add_argument("--baseline_random",
action='store_true',
help='randomly select test cases, only implemented for retraining')
parser.add_argument("--sel_method",
type=str,
default=None,
help="[kmeans, random]")
parser.add_argument("--feature_extractor_id",
type=int, default=1, help="available choices: [0. unsupervised BYOL model, 1. model under test")
parser.add_argument("--graph_nn",
action="store_true",
help="use graph network to generate test selection")
parser.add_argument("--no_neighbors",
type=int,
default=100,
help='number of neighbors to construct the knn graph')
# direct and indirect feedback combine method
parser.add_argument("--bf_mixed",
action="store_true",
help="combine direct and indirect correlation with a brute force way (multiplication)")
parser.add_argument("--learn_mixed",
action="store_true",
help='learn to combine direct and indirect correlation')
parser.add_argument("--latent_space_plot",
action='store_true',
help='plot all latent space or not.')
parser.add_argument("--supervised_latents",
action='store_true',
help='supervised or unsupervised encoder')
parser.add_argument("--save_path",
type=str,
default='', help='the file to store the hit ratios')
# random seed
parser.add_argument('--manualSeed', type=int, default=None, help='manual seed')
##########################################################################
args = parser.parse_args()
# Give a random seed if no manual configuration
if args.manualSeed is None:
args.manualSeed = random.randint(1, 10000)
random.seed(args.manualSeed)
torch.manual_seed(args.manualSeed)
np.random.seed(args.manualSeed)
if device == 'cuda':
torch.cuda.manual_seed(args.manualSeed)
default_hyperparams = {
'custom': {'img_size':32,'num_classes':10, 'channel':3, 'gnn_hidden_dim': 32, 'gnn_train_epoch':600, 'retrain_epoch':10, 'feature_dim':512},
'svhn': {'img_size':32,'num_classes':10, 'channel':3, 'gnn_hidden_dim': 32, 'gnn_train_epoch':600, 'retrain_epoch':10, 'feature_dim':512},
'cifar10': {'img_size':32,'num_classes':10, 'channel':3, 'gnn_hidden_dim': 32, 'gnn_train_epoch':600, 'retrain_epoch':10, 'feature_dim':512},
'stl10': {'img_size':96,'num_classes':10, 'channel':3, 'gnn_hidden_dim': 32, 'gnn_train_epoch':600, 'retrain_epoch':10, 'feature_dim':512},
}
weight_per_class_1ist = [np.ones(10),
np.array([0.77132064, 0.02075195, 0.63364823, 0.74880388, 0.49850701,
0.22479665, 0.19806286, 0.76053071, 0.16911084, 0.08833981]),
np.array([0.68535982, 0.95339335, 0.00394827, 0.51219226, 0.81262096,
0.61252607, 0.72175532, 0.29187607, 0.91777412, 0.71457578])]
weight_per_class = weight_per_class_1ist[args.model_number]
retrain_epoch = default_hyperparams[args.dataset]['retrain_epoch']
no_neighbors = args.no_neighbors
###############################################################################
###############################################################################
def main():
global retrain_epoch
global no_neighbors
# Init logger6
if not os.path.isdir(args.save_path):
os.makedirs(args.save_path)
log = open(
os.path.join(args.save_path,
'log_seed_{}.txt'.format(args.manualSeed)), 'w')
state = {k: v for k, v in args._get_kwargs()}
print_log(state, log)
print_log("Random Seed: {}".format(args.manualSeed), log)
print_log("python version : {}".format(sys.version.replace('\n', ' ')),
log)
print_log("torch version : {}".format(torch.__version__), log)
print_log("cudnn version : {}".format(torch.backends.cudnn.version()),
log)
print_log(args, log)
# Init dataset
if not os.path.isdir(args.data_path):
os.makedirs(args.data_path)
if args.dataset == 'cifar10':
num_classes = 10
mean = [x / 255 for x in [125.3, 123.0, 113.9]]
std = [x / 255 for x in [63.0, 62.1, 66.7]]
elif args.dataset == 'cifar100':
num_classes = 100
mean = [x / 255 for x in [129.3, 124.1, 112.4]]
std = [x / 255 for x in [68.2, 65.4, 70.4]]
elif args.dataset == 'svhn':
num_classes = 10
mean = [0.5, 0.5, 0.5]
std = [0.5, 0.5, 0.5]
elif args.dataset == 'stl10':
num_classes = 10
mean = [0.4914, 0.4822, 0.4465]
std = [0.2471, 0.2435, 0.2616]
else:
assert False, "Unknow dataset : {}".format(args.dataset)
data_dir = os.path.join(args.data_path, args.dataset)
img_size = default_hyperparams[args.dataset]['img_size']
if args.dataset == 'svhn':
train_transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.RandomCrop(img_size, padding=4),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])
test_transform = transforms.Compose(
[transforms.Resize((img_size, img_size)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
train_set = torchvision.datasets.SVHN(root=data_dir,
split = 'train',
transform=train_transform,
download=True)
num_train = len(train_set)
train_idx = list(range(num_train))[:40000]
val_idx = list(range(num_train))[40000:50000]
sub_train_set = torch.utils.data.Subset(train_set, train_idx)
elif args.dataset == 'cifar10':
train_transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.RandomCrop(img_size, padding=4),
transforms.ToTensor(),
transforms.Normalize(mean, std)
])
test_transform = transforms.Compose(
[transforms.Resize((img_size, img_size)),
transforms.ToTensor(),
transforms.Normalize(mean, std)])
train_set = torchvision.datasets.CIFAR10(root=data_dir,
train=True,
transform=train_transform,
download=True)
num_train = len(train_set)
train_idx = list(range(num_train))[:15000]
val_idx = list(range(num_train))[15000:20000]
sub_train_set = torch.utils.data.Subset(train_set, train_idx)
elif args.dataset == 'stl10':
train_transform = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.RandomCrop(img_size, padding=4),
transforms.ToTensor(),
transforms.Normalize(mean, std)
])
test_transform = transforms.Compose(
[transforms.Resize((img_size, img_size)),
transforms.ToTensor(),
transforms.Normalize(mean, std)])
train_set = torchvision.datasets.STL10(root=data_dir,
split='train',
transform=train_transform,
download=True)
num_train = len(train_set)
train_idx = list(range(num_train))[:4000]
val_idx = list(range(num_train))[4000:5000]
sub_train_set = torch.utils.data.Subset(train_set, train_idx)
else:
raise ValueError('Invalid dataset')
#####################################################
# ---- Test Center: Receive models under test------
print_log("=> creating model 2 test'{}'".format(args.model2test_arch), log)
# Init model, criterion, and optimizer
model2test = mymodels.__dict__[args.model2test_arch](num_classes=num_classes, channels=default_hyperparams[args.dataset]['channel']).to(device)
print_log("=> network :\n {}".format(model2test), log)
# load weights
checkpoint = torch.load(args.model2test_path, map_location=device)
model2test.load_state_dict(checkpoint['net'])
# restore the trainset training the model
dataset_save_path = os.path.join('./checkpoint',
args.dataset,
'ckpt_bias',
'biased_dataset',
str(args.model_number))
sub_train_index = np.load(os.path.join(dataset_save_path, 'train.npy'))
sub_val_index = np.load(os.path.join(dataset_save_path, 'val.npy'))
# sub_train_index = np.array(sub_train_index, dtype=np.int32)
# sub_val_index = np.array(sub_val_index, dtype=np.int32)
# print (sub_train_index, sub_val_index)
model2test_trainset = torch.utils.data.Subset(sub_train_set, sub_train_index)
#####################################################
# ---- Test Center: Prepare Test Dataset------
# get testset and testloader
train_idx = np.array(train_idx)
val_idx = np.array(val_idx)
if args.dataset == 'cifar10':
# test data : total 30k + 10k = 40k
train_set = torchvision.datasets.CIFAR10(
root=data_dir,
train=True,
transform=test_transform,
download=False )
test_set = torchvision.datasets.CIFAR10(root=data_dir,
train=False,
transform=test_transform,
download=False)
mix_test_set = torch.utils.data.ConcatDataset([train_set, test_set])
indices = list(range(len(mix_test_set)))
# train_indices = np.array(indices[:20000])
T2_indices = np.array(indices[20000:21000]) # 1k images
T2_set = torch.utils.data.Subset(mix_test_set, T2_indices)
# labeled case number
rest_indices = np.array(indices[21000:])
test_set = torch.utils.data.Subset(mix_test_set, rest_indices)
#split test dataset into labeled and unlabeled parts
num_test = len(test_set)
indices = list(range(num_test))
labeled_indices = np.array(indices[:7800]) # 8k image
unlabeled_indices = np.array(indices[7800:])
print_log("dataset segmentation: labeled/unlabled/T2: {} {} {}".format(labeled_indices.size, unlabeled_indices.size, T2_indices.size), log)
elif args.dataset == 'stl10':
# test data :
train_set = torchvision.datasets.STL10(
root=data_dir,
split='train',
transform=test_transform,
download=False )
test_set = torchvision.datasets.STL10(root=data_dir,
split='test',
transform=test_transform,
download=False)
mix_test_set = torch.utils.data.ConcatDataset([train_set, test_set])
indices = list(range(len(mix_test_set)))
T2_indices = np.array(indices[5000:5500]) # 1k images
rest_indices = np.array(indices[5500:])
T2_set = torch.utils.data.Subset(mix_test_set, T2_indices)
test_set = torch.utils.data.Subset(mix_test_set, rest_indices)
# split test dataset into labeled and unlabeled parts
num_test = len(test_set)
indices = list(range(num_test))
labeled_indices = np.array(indices[:int(0.2*num_test)]) # 8k images
unlabeled_indices = np.array(indices[int(0.2*num_test):]) # 1w images
print_log("dataset segmentation: labeled/unlabled/T2: {} {} {}".format(labeled_indices.size, unlabeled_indices.size, T2_indices.size), log)
elif args.dataset == 'svhn':
# test data: unlabeled 3w, labeled 1w, hold-out test data 53w
train_set = torchvision.datasets.SVHN(root=data_dir,
split='train',
transform=test_transform,
download=True)
test_set = torchvision.datasets.SVHN(root=data_dir,
split='test',
transform=test_transform,
download=True)
T2_set = torchvision.datasets.SVHN(root=data_dir,
split='extra',
transform=test_transform,
download=True)
mix_test_set = torch.utils.data.ConcatDataset([train_set, test_set])
num_test = len(mix_test_set)
indices = list(range(num_test))
rest_indices = np.array(indices[50000:])
test_set = torch.utils.data.Subset(mix_test_set, rest_indices)
labeled_indices = np.arange(len(test_set))[:10000]
unlabeled_indices = np.arange(len(test_set))[10000:]
print_log("dataset segmentation: labeled/unlabled/T2: {} {} {}".format(labeled_indices.size, unlabeled_indices.size, len(T2_set)), log)
else:
print_log("not supported dataset", log)
exit(0)
####################### A pre-evaluation ############
# Accuracy before retrain
# T2: test cases in deployment
T2_acc = 0
# if args.retrain:
print_log ("# of hold-out test inputs: {}".format(len(T2_set)), log)
T2_acc = test_acc(T2_set, model2test)
print_log("ACC in Hold-Out dataset: {}".format(round(T2_acc, 2)), log)
####################### Test Selection ############
# percentage of budget over all test cases
p_budget_lst = [1, 2, 5, 10, 15, 20, 30, 40, 50, 60, 70, 80, 90, 100] # percentage of budget
budget_lst = []
pfd_lst = []
ideal_pfd_lst = []
budget_lst.insert(0, 0)
pfd_lst.insert(0, 0)
ideal_pfd_lst.insert(0, 0)
if args.baseline_random:
unlabeled_test_inputs = torch.utils.data.Subset(test_set, unlabeled_indices)
print_log ("# of unlabeled test inputs: {}".format(len(unlabeled_test_inputs)), log)
correct_array, logits = test(unlabeled_test_inputs, model2test,
num_classes=num_classes)
misclass_array = (correct_array==0).astype(int)
ranked_indexes = list(np.arange(len(unlabeled_test_inputs)))
random.shuffle(ranked_indexes)
# test
for p_budget in p_budget_lst:
print_log("\n ###### budget percent is {} ######".format(p_budget), log)
budget = int(p_budget*len(unlabeled_indices)/100.0)
model2test_temp = copy.deepcopy(model2test)
selected = set(ranked_indexes[:budget])
# write results and logging
out_file = os.path.join(args.save_path, 'random_result.csv')
print_log ("writing output to csv file: {}".format(out_file), log)
with open(out_file, 'a+') as f:
writer = csv.writer(f, delimiter=',')
writer.writerow([args.model2test_arch,
args.model2test_path,
budget,
p_budget,
args.sel_method])
print_log('success!', log)
log.close()
return
# baseline 1
if args.baseline_gini:
# DeepGini: Prioritizing Massive Tests to Enhance the Robustness of Deep Neural Networks
unlabeled_test_inputs = torch.utils.data.Subset(test_set, unlabeled_indices)
print_log ("# of unlabeled test inputs: {}".format(len(unlabeled_test_inputs)), log)
correct_array, logits = test(unlabeled_test_inputs, model2test,
num_classes=num_classes)
# get classification result rank
prob = softmax(logits, axis=1)
pred = np.sum(np.square(prob), axis=1)
ranked_indexes = np.argsort(pred)
# gt
misclass_array = (correct_array==0).astype(int)
# plot histogram: relation between score and correctness
bins = np.linspace(0, 1, 100)
plt.hist(pred, bins, alpha=0.3, label='total')
plt.hist(pred[correct_array.astype(bool)], bins, log=True, alpha=0.5, color='b', label='correct')
plt.hist(pred[misclass_array.astype(bool)], bins, log=True, alpha=0.5, color='r', label='misclassified')
plt.legend()
plt.savefig(os.path.join(args.save_path, 'confidence_label.pdf'))
# test
for p_budget in p_budget_lst:
print_log("\n ###### budget percent is {} ######".format(p_budget), log)
budget = int(p_budget*len(unlabeled_indices)/100.0)
model2test_temp = copy.deepcopy(model2test)
selected = ranked_indexes[:budget]
pos_count = misclass_array[selected].sum()
# store result
p_fault_detected = 100.0*pos_count/misclass_array.sum()
ideal_fault_detected = 100.0*budget/misclass_array.sum()
if ideal_fault_detected > 100.0:
ideal_fault_detected = 100.0
random_p_fault_detected = (100.0*budget/misclass_array.shape[0])
print_log("Percentage of fault detected: %s "%(p_fault_detected), log)
print_log("Percentage of fault detected (random): %s "%(random_p_fault_detected), log)
# write results and logging
out_file = os.path.join(args.save_path, 'gini_result.csv')
print_log ("writing output to csv file: {}".format(out_file), log)
budget_lst.append(p_budget/100.0)
pfd_lst.append(p_fault_detected/100.0)
ideal_pfd_lst.append(ideal_fault_detected/100.0)
apfd = 0
ideal_apfd = 0
if p_budget == 100:
print (pfd_lst, ideal_pfd_lst)
apfd = get_APFD(copy.copy(budget_lst), pfd_lst)
ideal_apfd = get_APFD(budget_lst, ideal_pfd_lst)
with open(out_file, 'a+') as f:
writer = csv.writer(f, delimiter=',')
writer.writerow([args.model2test_arch,
args.model2test_path,
budget,
p_budget,
args.sel_method,
'FaultDetected',
p_fault_detected,
ideal_fault_detected,
random_p_fault_detected,
'APFD',
apfd,
ideal_apfd,
'TRC',
p_fault_detected/ideal_fault_detected])
print_log('success!', log)
log.close()
return
if args.baseline_mcp:
import mcp
# Multiple-Boundary Clustering and Prioritization to Promote Neural Network Retraining
unlabeled_test_inputs = torch.utils.data.Subset(test_set, unlabeled_indices)
print_log ("# of unlabeled test inputs: {}".format(len(unlabeled_test_inputs)), log)
# gt
correct_array, _ = test(unlabeled_test_inputs, model2test,
num_classes=num_classes)
misclass_array = (correct_array==0).astype(int)
_, logits = test(unlabeled_test_inputs, model2test, num_classes=num_classes)
prob = softmax(logits, axis=1)
dicratio=[[] for i in range(num_classes*num_classes)]
dicindex=[[] for i in range(num_classes*num_classes)]
for i in range(len(prob)):
act=prob[i]
max_index,sec_index,ratio = mcp.get_boundary_priority(act)#max_index
dicratio[max_index*num_classes+sec_index].append(ratio)
dicindex[max_index*num_classes+sec_index].append(i)
for p_budget in p_budget_lst:
print_log("\n ###### budget percent is {} ######".format(p_budget), log)
model2test_temp = copy.deepcopy(model2test)
dicindex_temp = copy.deepcopy(dicindex)
dicratio_temp = copy.deepcopy(dicratio)
budget = int(p_budget*len(unlabeled_indices)/100.0)
if p_budget == 100:
selected = list(np.arange(len(unlabeled_indices)))
else:
selected = mcp.select_from_firstsec_dic(budget, dicratio_temp, dicindex_temp, num_classes=num_classes)
# test
pos_count = misclass_array[selected].sum()
# store result
p_fault_detected = 100.0*pos_count/misclass_array.sum()
ideal_fault_detected = 100.0*budget/misclass_array.sum()
if ideal_fault_detected > 100.0:
ideal_fault_detected = 100.0
random_p_fault_detected = (100.0*budget/misclass_array.shape[0])
print_log("Percentage of fault detected: %s "%(p_fault_detected), log)
print_log("Percentage of fault detected (random): %s "%(random_p_fault_detected), log)
# write results and logging
out_file = os.path.join(args.save_path, 'mcp_result.csv')
print_log ("writing output to csv file: {}".format(out_file), log)
budget_lst.append(p_budget/100.0)
pfd_lst.append(p_fault_detected/100.0)
ideal_pfd_lst.append(ideal_fault_detected/100.0)
apfd = 0
ideal_apfd = 0
if p_budget == 100:
# print (budget_lst, pfd_lst, ideal_pfd_lst)
apfd = get_APFD(copy.copy(budget_lst), pfd_lst)
ideal_apfd = get_APFD(budget_lst, ideal_pfd_lst)
with open(out_file, 'a+') as f:
writer = csv.writer(f, delimiter=',')
writer.writerow([args.model2test_arch,
args.model2test_path,
budget,
p_budget,
args.sel_method,
'FaultDetected',
p_fault_detected,
ideal_fault_detected,
random_p_fault_detected,
'APFD',
apfd,
ideal_apfd,
'TRC',
p_fault_detected/ideal_fault_detected])
print_log('success!', log)
log.close()
return
if args.baseline_dsa:
from keras import backend as K
from pytorch2keras import pytorch_to_keras
import dsa
from torch.autograd import Variable
os.environ['KERAS_BACKEND'] = 'tensorflow'
K.set_image_data_format('channels_first')
print_log('{} {}'.format(K.image_data_format(), K.backend()), log)
# implementation of surprise adaquacy as selection method
print_log("Select test case with surprise adaquacy", log)
unlabeled_test_inputs = torch.utils.data.Subset(test_set, unlabeled_indices)
print_log("# of unlabeled test inputs: {}".format(len(unlabeled_test_inputs)), log)
# gt
correct_array, _ = test(unlabeled_test_inputs, model2test,
num_classes=num_classes)
misclass_array = (correct_array==0).astype(int)
# preprare keras model, dataset, for dsa calculation
# dummy variable
channel = default_hyperparams[args.dataset]['channel']
image_size = default_hyperparams[args.dataset]['img_size']
print (channel, image_size)
input_np = np.random.uniform(0,1, (1, channel, image_size, image_size))
input_var = Variable(torch.FloatTensor(input_np))
pytorch_output = model2test(input_var.to(device))[0].cpu().detach().numpy()
k_model = pytorch_to_keras(model2test, input_var.to(device), [(channel, image_size, image_size,)],
verbose=True,
name_policy='short')
print_log(k_model.summary(), log)
keras_output = k_model.predict(input_np)[0]
# print (keras_output.shape, pytorch_output.shape)
error = np.max(pytorch_output - keras_output)
print_log('{} {} {}'.format(pytorch_output, '\n', keras_output), log)
print ("error, ", error)
# check again: to be deleted
input_np = np.random.uniform(0, 1, (1, channel, image_size, image_size))
input_var = Variable(torch.FloatTensor(input_np))
p_output = model2test(input_var.to(device))[0].cpu().detach().numpy()
k_output = k_model.predict(input_np)[0]
print (p_output, k_output)
print (np.sum(p_output - k_output))
# sda ranking
def dataset_convertion(input_data):
output_data = [input_data[i][0].numpy() for i in range(len(input_data))]
output_label = [input_data[i][1] for i in range(len(input_data))]
# print (output_data)
output_data = np.array(output_data)
output_label = np.array(output_label)
return output_data, output_label
keras_model_train_data = dataset_convertion(model2test_trainset)[0]
keras_unlabeled_data = dataset_convertion(unlabeled_test_inputs)[0]
args.num_classes = num_classes
if args.model2test_arch == 'resnet18':
layer_names = ['189']
else:
layer_names = ['output_1']
dsa_values = dsa.fetch_dsa(model=k_model,
x_train=keras_model_train_data,
x_target=keras_unlabeled_data,
target_name='', layer_names=layer_names, args=args)
# higher dsa values in the front
ranked_indexes = np.argsort(dsa_values)[::-1]
print (ranked_indexes)
print ("dsa values: ", dsa_values[ranked_indexes[0]], dsa_values[ranked_indexes[1]], dsa_values[ranked_indexes[2]])
# return
for p_budget in p_budget_lst:
print_log("\n ###### budget percent is {} ######".format(p_budget), log)
budget = int(p_budget*len(unlabeled_indices)/100.0)
model2test_temp = copy.deepcopy(model2test)
selected = ranked_indexes[:budget]
pos_count = misclass_array[selected].sum()
# store result
p_fault_detected = 100.0*pos_count/misclass_array.sum()
ideal_fault_detected = 100.0*budget/misclass_array.sum()
if ideal_fault_detected > 100.0:
ideal_fault_detected = 100.0
random_p_fault_detected = (100.0*budget/misclass_array.shape[0])
print_log("Percentage of fault detected: %s "%(p_fault_detected), log)
print_log("Percentage of fault detected (random): %s "%(random_p_fault_detected), log)
# write results and logging
out_file = os.path.join(args.save_path, 'dsa_result.csv')
print_log ("writing output to csv file: {}".format(out_file), log)
budget_lst.append(p_budget/100.0)
pfd_lst.append(p_fault_detected/100.0)
ideal_pfd_lst.append(ideal_fault_detected/100.0)
apfd = 0
ideal_apfd = 0
if p_budget == 100:
print (pfd_lst, ideal_pfd_lst)
apfd = get_APFD(copy.copy(budget_lst), pfd_lst)
ideal_apfd = get_APFD(budget_lst, ideal_pfd_lst)
with open(out_file, 'a+') as f:
writer = csv.writer(f, delimiter=',')
writer.writerow([args.model2test_arch,
args.model2test_path,
budget,
p_budget,
args.sel_method,
'FaultDetected',
p_fault_detected,
ideal_fault_detected,
random_p_fault_detected,
'APFD',
apfd,
ideal_apfd,
'TRC',
p_fault_detected/ideal_fault_detected])
print_log('success!', log)
log.close()
return
if args.baseline_uncertainty:
# implementation of dropout uncertainty
print_log ("Select test case with dropout uncertainty", log)
unlabeled_test_inputs = torch.utils.data.Subset(test_set, unlabeled_indices)
print_log ("# of unlabeled test inputs: {}".format(len(unlabeled_test_inputs)), log)
# gt
correct_array, _ = test(unlabeled_test_inputs, model2test,
num_classes=num_classes)
misclass_array = (correct_array==0).astype(int)
def uncertainty_test(model, test_set):
test_loader = torch.utils.data.DataLoader(test_set, batch_size=256, shuffle=False, num_workers=16)
model.train()
model.dropout = True
T = 100
unct_list_v1 = []
unct_list_v2 = []
for data, target in test_loader:
output_list = []
data, target = data.to(device), target.to(device)
for i in range(T):
output = model(data)[0].detach()
output_list.append(torch.unsqueeze(F.softmax(output, dim=1), 0))
# print (torch.cat(output_list, 0).var(0).mean(1).shape)
output_mean = torch.cat(output_list, 0).mean(0).cpu().numpy()
# v1: variance
output_var_v1 = torch.cat(output_list, 0).var(0).mean(1).data.cpu().numpy()
# v2: mean and entropy
output_var_v2 = - np.sum(output_mean * np.log(output_mean), axis=1)
unct_list_v1.extend(output_var_v1)
unct_list_v2.extend(output_var_v2)
model.dropout = False
return unct_list_v1, unct_list_v2
unct_list_v1, unct_list_v2 = uncertainty_test(model2test, unlabeled_test_inputs)
print_log('uncertainty values: {}'.format(unct_list_v2), log)
# use v2
ranked_indexes = np.argsort(np.array(unct_list_v2))[::-1]
print_log(np.array(unct_list_v2).shape, log)
# plot histogram: relation between uncertainty and correctness
bins = np.linspace(min(unct_list_v2), max(unct_list_v2), 100)
plt.hist(np.array(unct_list_v2), bins, alpha=0.3, label='total')
plt.hist(np.array(unct_list_v2)[correct_array.astype(bool)], bins, log=True, alpha=0.5, color='b', label='correct')
plt.hist(np.array(unct_list_v2)[misclass_array.astype(bool)], bins, log=True, alpha=0.5, color='r', label='misclassified')
plt.legend()
plt.savefig(os.path.join(args.save_path, 'uncertainty_label.pdf'))
for p_budget in p_budget_lst:
print_log("\n ###### budget percent is {} ######".format(p_budget), log)
model2test_temp = copy.deepcopy(model2test)
budget = int(p_budget*len(unlabeled_indices)/100.0)
selected = ranked_indexes[:budget]
pos_count = misclass_array[selected].sum()
# store result
p_fault_detected = 100.0*pos_count/misclass_array.sum()
ideal_fault_detected = 100.0*budget/misclass_array.sum()
if ideal_fault_detected > 100.0:
ideal_fault_detected = 100.0
random_p_fault_detected = (100.0*budget/misclass_array.shape[0])
print_log("Percentage of fault detected: %s "%(p_fault_detected), log)
print_log("Percentage of fault detected (random): %s "%(random_p_fault_detected), log)
# write results and logging
out_file = os.path.join(args.save_path, 'dropout_uncertainty_result.csv')
print_log ("writing output to csv file: {}".format(out_file), log)
budget_lst.append(p_budget/100.0)
pfd_lst.append(p_fault_detected/100.0)
ideal_pfd_lst.append(ideal_fault_detected/100.0)
apfd = 0
ideal_apfd = 0
if p_budget == 100:
# print (budget_lst, pfd_lst, ideal_pfd_lst)
apfd = get_APFD(copy.copy(budget_lst), pfd_lst)
ideal_apfd = get_APFD(budget_lst, ideal_pfd_lst)
with open(out_file, 'a+') as f:
writer = csv.writer(f, delimiter=',')
writer.writerow([args.model2test_arch,
args.model2test_path,
budget,
p_budget,
args.sel_method,
'FaultDetected',
p_fault_detected,
ideal_fault_detected,
random_p_fault_detected,
'APFD',
apfd,
ideal_apfd,
'TRC',
p_fault_detected/ideal_fault_detected
])
print_log('success!', log)
log.close()
return
#####################################################
# ------ Test Center: Get the result ground truth -----
# Test model2test use all test cases
print_log('Get the ground truth', log)
correct_array, logits = test(test_set, model2test, num_classes=num_classes)
misclass_array = (correct_array==0).astype(int)
# ---- Test Center: Get latent vectors of test dataset------
fe_config = ['unsup', 'self']
feature_extractor = fe_config[args.feature_extractor_id]
if feature_extractor == 'unsup':
print_log('Use unsupervised BYOL model as feature extractor', log)
if args.dataset == 'cifar10':
unsup_img_size = 224
test_transform = torchvision.transforms.Compose([
torchvision.transforms.Resize(size=unsup_img_size),
torchvision.transforms.ToTensor()])
train_set_t = torchvision.datasets.CIFAR10(
root=data_dir,
train=True,
transform=test_transform,
download=False )
test_set_t = torchvision.datasets.CIFAR10(root=data_dir,
train=False,
transform=test_transform,
download=False)
mix_test_set_t = torch.utils.data.ConcatDataset([train_set_t, test_set_t])
test_set_t = torch.utils.data.Subset(mix_test_set_t, rest_indices)
feature_extractor_net = models.resnet18(pretrained=False).to(device)
feature_extractor_net.load_state_dict(torch.load('./ckpt_byol/cifar10_fc_224.pt', map_location=device))
if args.dataset == 'stl10':
unsup_img_size = 96
test_transform = torchvision.transforms.Compose([
torchvision.transforms.Resize(size=unsup_img_size),
torchvision.transforms.ToTensor()])
train_set_t = torchvision.datasets.STL10(
root=data_dir,
split='train',
transform=test_transform,
download=False )
test_set_t = torchvision.datasets.STL10(root=data_dir,
split='test',
transform=test_transform,
download=False)
mix_test_set_t = torch.utils.data.ConcatDataset([train_set_t, test_set_t])
test_set_t = torch.utils.data.Subset(mix_test_set_t, rest_indices)
feature_extractor_net = models.resnet18(pretrained=False, num_classes=num_classes).to(device)
feature_extractor_net.load_state_dict(torch.load('./ckpt_byol/stl10_fe_96.pt', map_location=device))
elif args.dataset == 'svhn':
unsup_img_size = 224
test_transform = torchvision.transforms.Compose([
torchvision.transforms.Resize(size=unsup_img_size),
torchvision.transforms.ToTensor()])
train_set_t = torchvision.datasets.SVHN(root=data_dir,
split='train',
transform=test_transform,
download=False)
test_set_t = torchvision.datasets.SVHN(root=data_dir,
split='test',
transform=test_transform,
download=False)
mix_test_set_t = torch.utils.data.ConcatDataset([train_set_t, test_set_t])
test_set_t = torch.utils.data.Subset(mix_test_set_t, rest_indices)
feature_extractor_net = models.resnet18(pretrained=False, num_classes=num_classes).to(device)
feature_extractor_net.load_state_dict(torch.load('./ckpt_byol/svhn_fe_224.pt', map_location=device))
# remove the fc layer
feature_extractor_net = nn.Sequential(*list(feature_extractor_net.children())[:-1])
latents = get_features(test_set_t, feature_extractor_net)
elif feature_extractor == 'self':
# bug 512*8*8 instead of 512?
print_log("Use model2test as feature extractor", log)
feature_extractor_net = model2test
feature_extractor_net = nn.Sequential(*list(feature_extractor_net.children())[:-1])
latents = get_features(test_set, feature_extractor_net)
latents = latents.reshape(len(test_set), -1)
else:
raise ValueError("Unknown feature extractor configuration")
#latents = sklearn.preprocessing.normalize(latents, norm='l2')
# print_log("latents shape {}".format(latents.shape), log)
#####################################################
# # ---- Test Center: Visulize test latent vectors------
# tSNE
full_loader = torch.utils.data.DataLoader(test_set, batch_size=len(test_set), shuffle=False)
labels = next(iter(full_loader))[1]
print ("labels: ", labels[:100])
class_inds = labels >= 8
print ("boolean class indexes: ", class_inds)
print ('length of selected visualize instances: ', class_inds.sum())
if args.latent_space_plot:
print_log('visulize latent embeddings via tsne ...', log)
# only visualize latents from class 0 and class 1
codes_embedded = TSNE(n_components=2).fit_transform(latents[class_inds])
plot_2d_scatter(codes_embedded, misclass_array[class_inds], save_path=args.save_path, fig_name='feature_extractor_embedding')
# return
#####################################################
# # ---- Test Center: Select cases for testing ------
print_log("\nTest Center Starts Working...", log)
neg_case_indexes = []
pos_case_indexes = []
print_log("Test the labeled samples", log)
labled_neg_indices = list(labeled_indices[np.nonzero(correct_array[labeled_indices])[0]])
labled_pos_indices = list(labeled_indices[np.nonzero(misclass_array[labeled_indices])[0]])
neg_case_indexes += labled_neg_indices
pos_case_indexes += labled_pos_indices
selected = set()
# Iterative
print_log(' ** classification using GNN algorithm: {}'.format(args.graph_nn), log)
print_log ("# of unlabeled test inputs: {}".format(len(unlabeled_indices)), log)
# get classification result rank
prob = softmax(logits, axis=1)
confidence = np.sum(np.square(prob), axis=1)
prob = torch.from_numpy(prob)
# Step 2: rank samples
# method1: apply label propagation algorithm
if not args.graph_nn:
if args.bf_mixed or args.learn_mixed: # mix
_, output_distribution = propogate(latents, pos_case_indexes, neg_case_indexes)
print_log("Mixed method enabled: combine gini and correlation based method", log)
mix_rank_indicator = output_distribution[:, 1] * (1 - confidence) # the bigger, the more likely to be positive case
ranked_indexes = np.argsort(mix_rank_indicator)[::-1] # cases having high positive probability are put in the front
else: # correlation alone
_, output_distribution = propogate(latents, pos_case_indexes, neg_case_indexes)
ranked_indexes = np.argsort(output_distribution[:, 1])[::-1] # cases having high positive probability are put in the front
return
# method2: apply GNN classification algorithm
old_time = time.time()