-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathload_corrupted_data.py
231 lines (195 loc) · 8.25 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
from PIL import Image
import os
import os.path
import numpy as np
import sys
import pickle
import random
import torch.utils.data as data
from torchvision.datasets.utils import download_url, check_integrity
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
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)
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 == '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
else:
assert False, "Invalid corruption type '{}' given. Must be in {'unif', 'flip'}".format(corruption_type)
self.clabels = np.zeros_like(self.train_labels)
for i in range(len(self.clabels)):
self.clabels[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):
true_target = None
if self.train and not self.meta:
img, target, true_target = self.train_data[index], self.clabels[index], self.train_labels[index]
elif 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)
if true_target is not None:
true_target = self.target_transform(true_target)
if true_target is not None:
return img, target, true_target, index
else:
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
return len(self.train_data) if self.train else len(self.test_data)
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'],
]