-
Notifications
You must be signed in to change notification settings - Fork 1.1k
/
utils.py
770 lines (616 loc) · 27.3 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
# MIT License
#
# Copyright (C) IBM Corporation 2018
#
# Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated
# documentation files (the "Software"), to deal in the Software without restriction, including without limitation the
# rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit
# persons to whom the Software is furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all copies or substantial portions of the
# Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
# WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
"""
Module providing convenience functions.
"""
from __future__ import absolute_import, division, print_function, unicode_literals
import logging
import os
import numpy as np
logger = logging.getLogger(__name__)
try:
# Conditional import of `torch` to avoid segmentation fault errors this framework generates at import
import torch
except ImportError:
logger.info('Could not import PyTorch in utilities.')
# -------------------------------------------------------------------------------------------- RANDOM NUMBER GENERATORS
def master_seed(seed):
"""
Set the seed for all random number generators used in the library. This ensures experiments reproducibility and
stable testing.
:param seed: The value to be seeded in the random number generators.
:type seed: `int`
"""
import numbers
import random
if not isinstance(seed, numbers.Integral):
raise TypeError('The seed for random number generators has to be an integer.')
# Set Python seed
random.seed(seed)
# Set Numpy seed
np.random.seed(seed)
np.random.RandomState(seed)
# Now try to set seed for all specific frameworks
try:
import tensorflow as tf
logger.info('Setting random seed for TensorFlow.')
if tf.__version__[0] == '2':
tf.random.set_seed(seed)
else:
tf.set_random_seed(seed)
except ImportError:
logger.info('Could not set random seed for TensorFlow.')
try:
import mxnet as mx
logger.info('Setting random seed for MXNet.')
mx.random.seed(seed)
except ImportError:
logger.info('Could not set random seed for MXNet.')
try:
logger.info('Setting random seed for PyTorch.')
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
except ImportError:
logger.info('Could not set random seed for PyTorch.')
# ----------------------------------------------------------------------------------------------------- MATH OPERATIONS
def projection(values, eps, norm_p):
"""
Project `values` on the L_p norm ball of size `eps`.
:param values: Array of perturbations to clip.
:type values: `np.ndarray`
:param eps: Maximum norm allowed.
:type eps: `float`
:param norm_p: L_p norm to use for clipping. Only 1, 2 and `np.Inf` supported for now.
:type norm_p: `int`
:return: Values of `values` after projection.
:rtype: `np.ndarray`
"""
# Pick a small scalar to avoid division by 0
tol = 10e-8
values_tmp = values.reshape((values.shape[0], -1))
if norm_p == 2:
values_tmp = values_tmp * np.expand_dims(np.minimum(1., eps / (np.linalg.norm(values_tmp, axis=1) + tol)),
axis=1)
elif norm_p == 1:
values_tmp = values_tmp * np.expand_dims(
np.minimum(1., eps / (np.linalg.norm(values_tmp, axis=1, ord=1) + tol)), axis=1)
elif norm_p == np.inf:
values_tmp = np.sign(values_tmp) * np.minimum(abs(values_tmp), eps)
else:
raise NotImplementedError('Values of `norm_p` different from 1, 2 and `np.inf` are currently not supported.')
values = values_tmp.reshape(values.shape)
return values
def random_sphere(nb_points, nb_dims, radius, norm):
"""
Generate randomly `m x n`-dimension points with radius `radius` and centered around 0.
:param nb_points: Number of random data points
:type nb_points: `int`
:param nb_dims: Dimensionality
:type nb_dims: `int`
:param radius: Radius
:type radius: `float`
:param norm: Current support: 1, 2, np.inf
:type norm: `int`
:return: The generated random sphere
:rtype: `np.ndarray`
"""
if norm == 1:
a_tmp = np.zeros(shape=(nb_points, nb_dims + 1))
a_tmp[:, -1] = np.sqrt(np.random.uniform(0, radius ** 2, nb_points))
for i in range(nb_points):
a_tmp[i, 1:-1] = np.sort(np.random.uniform(0, a_tmp[i, -1], nb_dims - 1))
res = (a_tmp[:, 1:] - a_tmp[:, :-1]) * np.random.choice([-1, 1], (nb_points, nb_dims))
elif norm == 2:
# pylint: disable=E0611
from scipy.special import gammainc
a_tmp = np.random.randn(nb_points, nb_dims)
s_2 = np.sum(a_tmp ** 2, axis=1)
base = gammainc(nb_dims / 2.0, s_2 / 2.0) ** (1 / nb_dims) * radius / np.sqrt(s_2)
res = a_tmp * (np.tile(base, (nb_dims, 1))).T
elif norm == np.inf:
res = np.random.uniform(float(-radius), float(radius), (nb_points, nb_dims))
else:
raise NotImplementedError("Norm {} not supported".format(norm))
return res
def original_to_tanh(x_original, clip_min, clip_max, tanh_smoother=0.999999):
"""
Transform input from original to tanh space.
:param x_original: An array with the input to be transformed.
:type x_original: `np.ndarray`
:param clip_min: Minimum clipping value.
:type clip_min: `float` or `np.ndarray`
:param clip_max: Maximum clipping value.
:type clip_max: `float` or `np.ndarray`
:param tanh_smoother: Scalar for multiplying arguments of arctanh to avoid division by zero.
:type tanh_smoother: `float`
:return: An array holding the transformed input.
:rtype: `np.ndarray`
"""
x_tanh = np.clip(x_original, clip_min, clip_max)
x_tanh = (x_tanh - clip_min) / (clip_max - clip_min)
x_tanh = np.arctanh(((x_tanh * 2) - 1) * tanh_smoother)
return x_tanh
def tanh_to_original(x_tanh, clip_min, clip_max, tanh_smoother=0.999999):
"""
Transform input from tanh to original space.
:param x_tanh: An array with the input to be transformed.
:type x_tanh: `np.ndarray`
:param clip_min: Minimum clipping value.
:type clip_min: `float` or `np.ndarray`
:param clip_max: Maximum clipping value.
:type clip_max: `float` or `np.ndarray`
:param tanh_smoother: Scalar for dividing arguments of tanh to avoid division by zero.
:type tanh_smoother: `float`
:return: An array holding the transformed input.
:rtype: `np.ndarray`
"""
x_original = (np.tanh(x_tanh) / tanh_smoother + 1) / 2
return x_original * (clip_max - clip_min) + clip_min
# --------------------------------------------------------------------------------------------------- LABELS OPERATIONS
def to_categorical(labels, nb_classes=None):
"""
Convert an array of labels to binary class matrix.
:param labels: An array of integer labels of shape `(nb_samples,)`
:type labels: `np.ndarray`
:param nb_classes: The number of classes (possible labels)
:type nb_classes: `int`
:return: A binary matrix representation of `y` in the shape `(nb_samples, nb_classes)`
:rtype: `np.ndarray`
"""
labels = np.array(labels, dtype=np.int32)
if nb_classes is None:
nb_classes = np.max(labels) + 1
categorical = np.zeros((labels.shape[0], nb_classes), dtype=np.float32)
categorical[np.arange(labels.shape[0]), np.squeeze(labels)] = 1
return categorical
def check_and_transform_label_format(labels, nb_classes=None, return_one_hot=True):
"""
Check label format and transform to one-hot-encoded labels if necessary
:param labels: An array of integer labels of shape `(nb_samples,)`, `(nb_samples, 1)` or `(nb_samples, nb_classes)`
:type labels: `np.ndarray`
:param nb_classes: The number of classes
:type nb_classes: `int`
:param return_one_hot: True if returning one-hot encoded labels, False if returning index labels.
:return: Labels with shape `(nb_samples, nb_classes)` (one-hot) or `(nb_samples,)` (index)
:rtype: `np.ndarray`
"""
if labels is not None:
if len(labels.shape) == 2 and labels.shape[1] > 1:
if not return_one_hot:
labels = np.argmax(labels, axis=1)
elif len(labels.shape) == 2 and labels.shape[1] == 1:
labels = np.squeeze(labels)
if return_one_hot:
labels = to_categorical(labels, nb_classes)
elif len(labels.shape) == 1:
if return_one_hot:
labels = to_categorical(labels, nb_classes)
else:
raise ValueError('Shape of labels not recognised.'
'Please provide labels in shape (nb_samples,) or (nb_samples, nb_classes)')
return labels
def random_targets(labels, nb_classes):
"""
Given a set of correct labels, randomly choose target labels different from the original ones. These can be
one-hot encoded or integers.
:param labels: The correct labels
:type labels: `np.ndarray`
:param nb_classes: The number of classes for this model
:type nb_classes: `int`
:return: An array holding the randomly-selected target classes, one-hot encoded.
:rtype: `np.ndarray`
"""
if len(labels.shape) > 1:
labels = np.argmax(labels, axis=1)
result = np.zeros(labels.shape)
for class_ind in range(nb_classes):
other_classes = list(range(nb_classes))
other_classes.remove(class_ind)
in_cl = labels == class_ind
result[in_cl] = np.random.choice(other_classes)
return to_categorical(result, nb_classes)
def least_likely_class(x, classifier):
"""
Compute the least likely class predictions for sample `x`. This strategy for choosing attack targets was used in
(Kurakin et al., 2016).
| Paper link: https://arxiv.org/abs/1607.02533
:param x: A data sample of shape accepted by `classifier`.
:type x: `np.ndarray`
:param classifier: The classifier used for computing predictions.
:type classifier: `Classifier`
:return: Least-likely class predicted by `classifier` for sample `x` in one-hot encoding.
:rtype: `np.ndarray`
"""
return to_categorical(np.argmin(classifier.predict(x), axis=1), nb_classes=classifier.nb_classes())
def second_most_likely_class(x, classifier):
"""
Compute the second most likely class predictions for sample `x`. This strategy can be used for choosing target
labels for an attack to improve its chances to succeed.
:param x: A data sample of shape accepted by `classifier`.
:type x: `np.ndarray`
:param classifier: The classifier used for computing predictions.
:type classifier: `Classifier`
:return: Second most likely class predicted by `classifier` for sample `x` in one-hot encoding.
:rtype: `np.ndarray`
"""
return to_categorical(np.argpartition(classifier.predict(x), -2, axis=1)[:, -2], nb_classes=classifier.nb_classes())
def get_label_conf(y_vec):
"""
Returns the confidence and the label of the most probable class given a vector of class confidences
:param y_vec: (np.ndarray) vector of class confidences, nb of instances as first dimension
:return: (np.ndarray, np.ndarray) confidences and labels
"""
assert len(y_vec.shape) == 2
confs, labels = np.amax(y_vec, axis=1), np.argmax(y_vec, axis=1)
return confs, labels
def get_labels_np_array(preds):
"""
Returns the label of the most probable class given a array of class confidences.
:param preds: (np.ndarray) array of class confidences, nb of instances as first dimension
:return: (np.ndarray) labels
"""
preds_max = np.amax(preds, axis=1, keepdims=True)
y = (preds == preds_max)
return y
def compute_success(classifier, x_clean, labels, x_adv, targeted=False, batch_size=1):
"""
Compute the success rate of an attack based on clean samples, adversarial samples and targets or correct labels.
:param classifier: Classifier used for prediction.
:type classifier: :class:`.Classifier`
:param x_clean: Original clean samples.
:type x_clean: `np.ndarray`
:param labels: Correct labels of `x_clean` if the attack is untargeted, or target labels of the attack otherwise.
:type labels: `np.ndarray`
:param x_adv: Adversarial samples to be evaluated.
:type x_adv: `np.ndarray`
:param targeted: `True` if the attack is targeted. In that case, `labels` are treated as target classes instead of
correct labels of the clean samples.s
:type targeted: `bool`
:param batch_size: Batch size
:type batch_size: `int`
:return: Percentage of successful adversarial samples.
:rtype: `float`
"""
adv_preds = np.argmax(classifier.predict(x_adv, batch_size=batch_size), axis=1)
if targeted:
rate = np.sum(adv_preds == np.argmax(labels, axis=1)) / x_adv.shape[0]
else:
preds = np.argmax(classifier.predict(x_clean, batch_size=batch_size), axis=1)
rate = np.sum(adv_preds != preds) / x_adv.shape[0]
return rate
def compute_accuracy(preds, labels, abstain=True):
"""
Compute the accuracy rate and coverage rate of predictions
In the case where predictions are abstained, those samples are ignored.
:param preds: Predictions.
:type preds: `np.ndarray`
:param labels: Correct labels of `x`.
:type labels: `np.ndarray`
:param abstain: True if ignore abstained prediction, False if count them as incorrect.
:type abstain: `boolean`
:return: Tuple of accuracy rate and coverage rate
:rtype: `tuple`
"""
has_pred = np.sum(preds, axis=1)
idx_pred = np.where(has_pred)[0]
labels = np.argmax(labels[idx_pred], axis=1)
num_correct = np.sum(np.argmax(preds[idx_pred], axis=1) == labels)
coverage_rate = len(idx_pred) / preds.shape[0]
if abstain:
acc_rate = num_correct / preds[idx_pred].shape[0]
else:
acc_rate = num_correct / preds.shape[0]
return acc_rate, coverage_rate
# -------------------------------------------------------------------------------------------------- DATASET OPERATIONS
def load_cifar10(raw=False):
"""
Loads CIFAR10 dataset from config.CIFAR10_PATH or downloads it if necessary.
:param raw: `True` if no preprocessing should be applied to the data. Otherwise, data is normalized to 1.
:type raw: `bool`
:return: `(x_train, y_train), (x_test, y_test), min, max`
:rtype: `(np.ndarray, np.ndarray), (np.ndarray, np.ndarray), float, float`
"""
def load_batch(fpath):
"""
Utility function for loading CIFAR batches, as written in Keras.
:param fpath: Full path to the batch file.
:return: `(data, labels)`
"""
import sys
from six.moves import cPickle
with open(fpath, 'rb') as file_:
if sys.version_info < (3,):
content = cPickle.load(file_)
else:
content = cPickle.load(file_, encoding='bytes')
content_decoded = {}
for key, value in content.items():
content_decoded[key.decode('utf8')] = value
content = content_decoded
data = content['data']
labels = content['labels']
data = data.reshape(data.shape[0], 3, 32, 32)
return data, labels
from art import DATA_PATH
path = get_file('cifar-10-batches-py', extract=True, path=DATA_PATH,
url='http://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz')
num_train_samples = 50000
x_train = np.zeros((num_train_samples, 3, 32, 32), dtype=np.uint8)
y_train = np.zeros((num_train_samples,), dtype=np.uint8)
for i in range(1, 6):
fpath = os.path.join(path, 'data_batch_' + str(i))
data, labels = load_batch(fpath)
x_train[(i - 1) * 10000: i * 10000, :, :, :] = data
y_train[(i - 1) * 10000: i * 10000] = labels
fpath = os.path.join(path, 'test_batch')
x_test, y_test = load_batch(fpath)
y_train = np.reshape(y_train, (len(y_train), 1))
y_test = np.reshape(y_test, (len(y_test), 1))
# Set channels last
x_train = x_train.transpose(0, 2, 3, 1)
x_test = x_test.transpose(0, 2, 3, 1)
min_, max_ = 0, 255
if not raw:
min_, max_ = 0., 1.
x_train, y_train = preprocess(x_train, y_train, clip_values=(0, 255))
x_test, y_test = preprocess(x_test, y_test, clip_values=(0, 255))
return (x_train, y_train), (x_test, y_test), min_, max_
def load_mnist(raw=False):
"""
Loads MNIST dataset from `DATA_PATH` or downloads it if necessary.
:param raw: `True` if no preprocessing should be applied to the data. Otherwise, data is normalized to 1.
:type raw: `bool`
:return: `(x_train, y_train), (x_test, y_test), min, max`
:rtype: `(np.ndarray, np.ndarray), (np.ndarray, np.ndarray), float, float`
"""
from art import DATA_PATH
path = get_file('mnist.npz', path=DATA_PATH, url='https://s3.amazonaws.com/img-datasets/mnist.npz')
dict_mnist = np.load(path)
x_train = dict_mnist['x_train']
y_train = dict_mnist['y_train']
x_test = dict_mnist['x_test']
y_test = dict_mnist['y_test']
dict_mnist.close()
# Add channel axis
min_, max_ = 0, 255
if not raw:
min_, max_ = 0.0, 1.0
x_train = np.expand_dims(x_train, axis=3)
x_test = np.expand_dims(x_test, axis=3)
x_train, y_train = preprocess(x_train, y_train)
x_test, y_test = preprocess(x_test, y_test)
return (x_train, y_train), (x_test, y_test), min_, max_
def load_stl():
"""
Loads the STL-10 dataset from `DATA_PATH` or downloads it if necessary.
:return: `(x_train, y_train), (x_test, y_test), min, max`
:rtype: `(np.ndarray, np.ndarray), (np.ndarray, np.ndarray), float, float`
"""
from os.path import join
from art import DATA_PATH
min_, max_ = 0.0, 1.0
# Download and extract data if needed
path = get_file('stl10_binary', path=DATA_PATH, extract=True,
url='https://ai.stanford.edu/~acoates/stl10/stl10_binary.tar.gz')
with open(join(path, 'train_X.bin'), 'rb') as f_numpy:
x_train = np.fromfile(f_numpy, dtype=np.uint8)
x_train = np.reshape(x_train, (-1, 3, 96, 96))
with open(join(path, 'test_X.bin'), 'rb') as f_numpy:
x_test = np.fromfile(f_numpy, dtype=np.uint8)
x_test = np.reshape(x_test, (-1, 3, 96, 96))
# Set channel last
x_train = x_train.transpose(0, 2, 3, 1)
x_test = x_test.transpose(0, 2, 3, 1)
with open(join(path, 'train_y.bin'), 'rb') as f_numpy:
y_train = np.fromfile(f_numpy, dtype=np.uint8)
y_train -= 1
with open(join(path, 'test_y.bin'), 'rb') as f_numpy:
y_test = np.fromfile(f_numpy, dtype=np.uint8)
y_test -= 1
x_train, y_train = preprocess(x_train, y_train)
x_test, y_test = preprocess(x_test, y_test)
return (x_train, y_train), (x_test, y_test), min_, max_
def load_iris(raw=False, test_set=0.3):
"""
Loads the UCI Iris dataset from `DATA_PATH` or downloads it if necessary.
:param raw: `True` if no preprocessing should be applied to the data. Otherwise, data is normalized to 1.
:type raw: `bool`
:param test_set: Proportion of the data to use as validation split. The value should be between o and 1.
:type test_set: `float`
:return: Entire dataset and labels.
:rtype: `(np.ndarray, np.ndarray), (np.ndarray, np.ndarray), float, float`
"""
from art import DATA_PATH, NUMPY_DTYPE
# Download data if needed
path = get_file('iris.data', path=DATA_PATH, extract=False,
url='https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data')
data = np.loadtxt(path, delimiter=',', usecols=(0, 1, 2, 3), dtype=NUMPY_DTYPE)
labels = np.loadtxt(path, delimiter=',', usecols=4, dtype=str)
# Preprocess
if not raw:
label_map = {'Iris-setosa': 0, 'Iris-versicolor': 1, 'Iris-virginica': 2}
labels = np.array([label_map[labels[i]] for i in range(labels.size)], dtype=np.int32)
data, labels = preprocess(data, labels, nb_classes=3)
min_, max_ = np.amin(data), np.amax(data)
# Split training and test sets
split_index = int((1 - test_set) * len(data) / 3)
x_train = np.vstack((data[:split_index], data[50:50 + split_index], data[100:100 + split_index]))
y_train = np.vstack((labels[:split_index], labels[50:50 + split_index], labels[100:100 + split_index]))
if split_index >= 49:
x_test, y_test = None, None
else:
x_test = np.vstack((data[split_index:50], data[50 + split_index:100], data[100 + split_index:]))
y_test = np.vstack((labels[split_index:50], labels[50 + split_index:100], labels[100 + split_index:]))
assert len(x_train) + len(x_test) == 150
# Shuffle test set
random_indices = np.random.permutation(len(y_test))
x_test, y_test = x_test[random_indices], y_test[random_indices]
# Shuffle training set
random_indices = np.random.permutation(len(y_train))
x_train, y_train = x_train[random_indices], y_train[random_indices]
return (x_train, y_train), (x_test, y_test), min_, max_
def load_dataset(name):
"""
Loads or downloads the dataset corresponding to `name`. Options are: `mnist`, `cifar10` and `stl10`.
:param name: Name of the dataset
:type name: `str`
:return: The dataset separated in training and test sets as `(x_train, y_train), (x_test, y_test), min, max`
:rtype: `(np.ndarray, np.ndarray), (np.ndarray, np.ndarray), float, float`
:raises NotImplementedError: If the dataset is unknown.
"""
if "mnist" in name:
return load_mnist()
if "cifar10" in name:
return load_cifar10()
if "stl10" in name:
return load_stl()
if "iris" in name:
return load_iris()
raise NotImplementedError("There is no loader for dataset '{}'.".format(name))
def _extract(full_path, path):
import tarfile
import zipfile
import shutil
if full_path.endswith('tar'):
if tarfile.is_tarfile(full_path):
archive = tarfile.open(full_path, "r:")
elif full_path.endswith('tar.gz'):
if tarfile.is_tarfile(full_path):
archive = tarfile.open(full_path, "r:gz")
elif full_path.endswith('zip'):
if zipfile.is_zipfile(full_path):
archive = zipfile.ZipFile(full_path)
else:
return False
else:
return False
try:
archive.extractall(path)
except (tarfile.TarError, RuntimeError, KeyboardInterrupt):
if os.path.exists(path):
if os.path.isfile(path):
os.remove(path)
else:
shutil.rmtree(path)
raise
return True
def get_file(filename, url, path=None, extract=False):
"""
Downloads a file from a URL if it not already in the cache. The file at indicated by `url` is downloaded to the
path `path` (default is ~/.art/data). and given the name `filename`. Files in tar, tar.gz, tar.bz, and zip formats
can also be extracted. This is a simplified version of the function with the same name in Keras.
:param filename: Name of the file.
:type filename: `str`
:param url: Download URL.
:type url: `str`
:param path: Folder to store the download. If not specified, `~/.art/data` is used instead.
:type: `str`
:param extract: If true, tries to extract the archive.
:type extract: `bool`
:return: Path to the downloaded file.
:rtype: `str`
"""
if path is None:
from art import DATA_PATH
path_ = os.path.expanduser(DATA_PATH)
else:
path_ = os.path.expanduser(path)
if not os.access(path_, os.W_OK):
path_ = os.path.join('/tmp', '.art')
if not os.path.exists(path_):
os.makedirs(path_)
if extract:
extract_path = os.path.join(path_, filename)
full_path = extract_path + '.tar.gz'
else:
full_path = os.path.join(path_, filename)
# Determine if dataset needs downloading
download = not os.path.exists(full_path)
if download:
logger.info('Downloading data from %s', url)
error_msg = 'URL fetch failure on {}: {} -- {}'
try:
try:
from six.moves.urllib.error import HTTPError, URLError
from six.moves.urllib.request import urlretrieve
urlretrieve(url, full_path)
except HTTPError as exception:
raise Exception(error_msg.format(url, exception.code, exception.msg))
except URLError as exception:
raise Exception(error_msg.format(url, exception.errno, exception.reason))
except (Exception, KeyboardInterrupt):
if os.path.exists(full_path):
os.remove(full_path)
raise
if extract:
if not os.path.exists(extract_path):
_extract(full_path, path_)
return extract_path
return full_path
def make_directory(dir_path):
"""
Creates the specified tree of directories if needed.
:param dir_path: (str) directory or file path
:return: None
"""
if not os.path.exists(dir_path):
os.makedirs(dir_path)
def clip_and_round(x, clip_values, round_samples):
"""
Rounds the input to the correct level of granularity.
Useful to ensure data passed to classifier can be represented
in the correct domain, e.g., [0, 255] integers verses [0,1]
or [0, 255] floating points.
:param x: Sample input with shape as expected by the model.
:type x: `np.ndarray`
:param clip_values: Tuple of the form `(min, max)` representing the minimum and maximum values allowed
for features, or `None` if no clipping should be performed.
:type clip_values: `tuple`
:param round_samples: The resolution of the input domain to round the data to, e.g., 1.0, or 1/255. Set to 0 to
disable.
:type round_samples: `float`
"""
if round_samples == 0:
return x
if clip_values is not None:
np.clip(x, clip_values[0], clip_values[1], out=x)
x = np.around(x / round_samples) * round_samples
return x
def preprocess(x, y, nb_classes=10, clip_values=None):
"""
Scales `x` to [0, 1] and converts `y` to class categorical confidences.
:param x: Data instances.
:type x: `np.ndarray`
:param y: Labels.
:type y: `np.ndarray`
:param nb_classes: Number of classes in dataset.
:type nb_classes: `int`
:param clip_values: Original data range allowed value for features, either one respective scalar or one value per
feature.
:type clip_values: `tuple(float, float)` or `tuple(np.ndarray, np.ndarray)`
:return: Rescaled values of `x`, `y`
:rtype: `tuple`
"""
if clip_values is None:
min_, max_ = np.amin(x), np.amax(x)
else:
min_, max_ = clip_values
normalized_x = (x - min_) / (max_ - min_)
categorical_y = to_categorical(y, nb_classes)
return normalized_x, categorical_y