/
load_corrupted_data.py
309 lines (261 loc) · 11.7 KB
/
load_corrupted_data.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
#adapted from https://github.com/xjtushujun/meta-weight-net
from PIL import Image
import os
import os.path
import errno
import numpy as np
import sys
import pickle
import torch.utils.data as data
from torchvision.datasets.utils import download_url, check_integrity
import torch
import torch.nn.functional as F
from torch.autograd import Variable as V
import wideresnet as wrn
import torchvision.transforms as transforms
def uniform_mix_C(mixing_ratio, num_classes):
'''
returns a linear interpolation of a uniform matrix and an identity matrix
'''
return mixing_ratio * np.full((num_classes, num_classes), 1 / num_classes) + \
(1 - mixing_ratio) * np.eye(num_classes)
def flip_labels_C(corruption_prob, num_classes, seed=1):
'''
returns a matrix with (1 - corruption_prob) on the diagonals, and corruption_prob
concentrated in only one other entry for each row
'''
np.random.seed(seed)
C = np.eye(num_classes) * (1 - corruption_prob)
row_indices = np.arange(num_classes)
for i in range(num_classes):
C[i][np.random.choice(row_indices[row_indices != i])] = corruption_prob
return C
def flip_labels_C_two(corruption_prob, num_classes, seed=1):
'''
returns a matrix with (1 - corruption_prob) on the diagonals, and corruption_prob
concentrated in only one other entry for each row
'''
np.random.seed(seed)
C = np.eye(num_classes) * (1 - corruption_prob)
row_indices = np.arange(num_classes)
for i in range(num_classes):
C[i][np.random.choice(row_indices[row_indices != i], 2, replace=False)] = corruption_prob / 2
return C
class CIFAR10(data.Dataset):
base_folder = 'cifar-10-batches-py'
url = "http://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz"
filename = "cifar-10-python.tar.gz"
tgz_md5 = 'c58f30108f718f92721af3b95e74349a'
train_list = [
['data_batch_1', 'c99cafc152244af753f735de768cd75f'],
['data_batch_2', 'd4bba439e000b95fd0a9bffe97cbabec'],
['data_batch_3', '54ebc095f3ab1f0389bbae665268c751'],
['data_batch_4', '634d18415352ddfa80567beed471001a'],
['data_batch_5', '482c414d41f54cd18b22e5b47cb7c3cb'],
]
test_list = [
['test_batch', '40351d587109b95175f43aff81a1287e'],
]
def __init__(self, root='', train=True, meta=True, num_meta=1000,
corruption_prob=0, corruption_type='unif', transform=None, target_transform=None,
download=False, seed=1):
self.root = root
self.transform = transform
self.target_transform = target_transform
self.train = train # training set or test set
self.meta = meta
self.corruption_prob = corruption_prob
self.num_meta = num_meta
np.random.seed(seed)
if download:
self.download()
if not self._check_integrity():
raise RuntimeError('Dataset not found or corrupted.' +
' You can use download=True to download it')
# now load the picked numpy arrays
if self.train:
self.train_data = []
self.train_labels = []
self.train_coarse_labels = []
for fentry in self.train_list:
f = fentry[0]
file = os.path.join(root, self.base_folder, f)
fo = open(file, 'rb')
if sys.version_info[0] == 2:
entry = pickle.load(fo)
else:
entry = pickle.load(fo, encoding='latin1')
self.train_data.append(entry['data'])
if 'labels' in entry:
self.train_labels += entry['labels']
img_num_list = [int(self.num_meta/10)] * 10
num_classes = 10
else:
self.train_labels += entry['fine_labels']
self.train_coarse_labels += entry['coarse_labels']
img_num_list = [int(self.num_meta/100)] * 100
num_classes = 100
fo.close()
self.train_data = np.concatenate(self.train_data)
self.train_data = self.train_data.reshape((50000, 3, 32, 32))
self.train_data = self.train_data.transpose((0, 2, 3, 1)) # convert to HWC
data_list_val = {}
for j in range(num_classes):
data_list_val[j] = [i for i, label in enumerate(self.train_labels) if label == j]
idx_to_meta = []
idx_to_train = []
print(img_num_list)
for cls_idx, img_id_list in data_list_val.items():
np.random.shuffle(img_id_list)
img_num = img_num_list[int(cls_idx)]
idx_to_meta.extend(img_id_list[:img_num])
idx_to_train.extend(img_id_list[img_num:])
if meta is True:
self.train_data = self.train_data[idx_to_meta]
self.train_labels = list(np.array(self.train_labels)[idx_to_meta])
else:
self.train_data = self.train_data[idx_to_train]
self.train_labels = list(np.array(self.train_labels)[idx_to_train])
if corruption_type == 'hierarchical':
self.train_coarse_labels = list(np.array(self.train_coarse_labels)[idx_to_meta])
if corruption_type == 'unif':
C = uniform_mix_C(self.corruption_prob, num_classes)
print(C)
self.C = C
elif corruption_type == 'flip':
C = flip_labels_C(self.corruption_prob, num_classes)
print(C)
self.C = C
elif corruption_type == 'flip2':
C = flip_labels_C_two(self.corruption_prob, num_classes)
print(C)
self.C = C
elif corruption_type == 'hierarchical':
assert num_classes == 100, 'You must use CIFAR-100 with the hierarchical corruption.'
coarse_fine = []
for i in range(20):
coarse_fine.append(set())
for i in range(len(self.train_labels)):
coarse_fine[self.train_coarse_labels[i]].add(self.train_labels[i])
for i in range(20):
coarse_fine[i] = list(coarse_fine[i])
C = np.eye(num_classes) * (1 - corruption_prob)
for i in range(20):
tmp = np.copy(coarse_fine[i])
for j in range(len(tmp)):
tmp2 = np.delete(np.copy(tmp), j)
C[tmp[j], tmp2] += corruption_prob * 1/len(tmp2)
self.C = C
print(C)
elif corruption_type == 'clabels':
net = wrn.WideResNet(40, num_classes, 2, dropRate=0.3).cuda()
model_name = './cifar{}_labeler'.format(num_classes)
net.load_state_dict(torch.load(model_name))
net.eval()
else:
assert False, "Invalid corruption type '{}' given. Must be in {'unif', 'flip', 'hierarchical'}".format(corruption_type)
if corruption_type == 'clabels':
mean = [x / 255 for x in [125.3, 123.0, 113.9]]
std = [x / 255 for x in [63.0, 62.1, 66.7]]
test_transform = transforms.Compose(
[transforms.ToTensor(), transforms.Normalize(mean, std)])
# obtain sampling probabilities
sampling_probs = []
print('Starting labeling')
for i in range((len(self.train_labels) // 64) + 1):
current = self.train_data[i*64:(i+1)*64]
current = [Image.fromarray(current[i]) for i in range(len(current))]
current = torch.cat([test_transform(current[i]).unsqueeze(0) for i in range(len(current))], dim=0)
data = V(current).cuda()
logits = net(data)
smax = F.softmax(logits / 5) # temperature of 1
sampling_probs.append(smax.data.cpu().numpy())
sampling_probs = np.concatenate(sampling_probs, 0)
print('Finished labeling 1')
new_labeling_correct = 0
argmax_labeling_correct = 0
for i in range(len(self.train_labels)):
old_label = self.train_labels[i]
new_label = np.random.choice(num_classes, p=sampling_probs[i])
self.train_labels[i] = new_label
if old_label == new_label:
new_labeling_correct += 1
if old_label == np.argmax(sampling_probs[i]):
argmax_labeling_correct += 1
print('Finished labeling 2')
print('New labeling accuracy:', new_labeling_correct / len(self.train_labels))
print('Argmax labeling accuracy:', argmax_labeling_correct / len(self.train_labels))
else:
for i in range(len(self.train_labels)):
self.train_labels[i] = np.random.choice(num_classes, p=C[self.train_labels[i]])
self.corruption_matrix = C
else:
f = self.test_list[0][0]
file = os.path.join(root, self.base_folder, f)
fo = open(file, 'rb')
if sys.version_info[0] == 2:
entry = pickle.load(fo)
else:
entry = pickle.load(fo, encoding='latin1')
self.test_data = entry['data']
if 'labels' in entry:
self.test_labels = entry['labels']
else:
self.test_labels = entry['fine_labels']
fo.close()
self.test_data = self.test_data.reshape((10000, 3, 32, 32))
self.test_data = self.test_data.transpose((0, 2, 3, 1)) # convert to HWC
def __getitem__(self, index):
if self.train:
img, target = self.train_data[index], self.train_labels[index]
else:
img, target = self.test_data[index], self.test_labels[index]
# doing this so that it is consistent with all other datasets
# to return a PIL Image
img = Image.fromarray(img)
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target
def __len__(self):
if self.train:
if self.meta is True:
return self.num_meta
else:
return 50000 - self.num_meta
else:
return 10000
def _check_integrity(self):
root = self.root
for fentry in (self.train_list + self.test_list):
filename, md5 = fentry[0], fentry[1]
fpath = os.path.join(root, self.base_folder, filename)
if not check_integrity(fpath, md5):
return False
return True
def download(self):
import tarfile
if self._check_integrity():
print('Files already downloaded and verified')
return
root = self.root
download_url(self.url, root, self.filename, self.tgz_md5)
# extract file
cwd = os.getcwd()
tar = tarfile.open(os.path.join(root, self.filename), "r:gz")
os.chdir(root)
tar.extractall()
tar.close()
os.chdir(cwd)
class CIFAR100(CIFAR10):
base_folder = 'cifar-100-python'
url = "http://www.cs.toronto.edu/~kriz/cifar-100-python.tar.gz"
filename = "cifar-100-python.tar.gz"
tgz_md5 = 'eb9058c3a382ffc7106e4002c42a8d85'
train_list = [
['train', '16019d7e3df5f24257cddd939b257f8d'],
]
test_list = [
['test', 'f0ef6b0ae62326f3e7ffdfab6717acfc'],
]