forked from Philip-Bachman/Sequential-Generation
-
Notifications
You must be signed in to change notification settings - Fork 0
/
load_data.py
357 lines (315 loc) · 13.1 KB
/
load_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
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
import numpy as np
import numpy.random as npr
import cPickle
import gzip
import os
import sys
import theano
import theano.tensor as T
def row_shuffle(X):
"""
Return a copy of X with shuffled rows.
"""
shuf_idx = np.arange(X.shape[0])
npr.shuffle(shuf_idx)
X_shuf = X[shuf_idx]
return X_shuf
def _shared_dataset(data_xy):
"""
Function that loads the dataset into shared variables
The reason we store our dataset in shared variables is to allow
Theano to copy it into the GPU memory (when code is run on GPU).
Since copying data into the GPU is slow, copying a minibatch everytime
is needed (the default behaviour if the data is not in a shared
variable) would lead to a large decrease in performance.
"""
data_x, data_y = data_xy
shared_x = theano.shared(np.asarray(data_x,
dtype=theano.config.floatX))
shared_y = theano.shared(np.asarray(data_y,
dtype=theano.config.floatX))
# When storing data on the GPU it has to be stored as floats
# therefore we will store the labels as ``floatX`` as well
# (``shared_y`` does exactly that).
return shared_x, shared_y
def load_binarized_mnist(data_path='./'):
#binarized_mnist_test.amat binarized_mnist_train.amat binarized_mnist_valid.amat
print 'loading binary MNIST, sampled version'
train_x = np.loadtxt(data_path + 'binarized_mnist_train.amat').astype('float32')
valid_x = np.loadtxt(data_path + 'binarized_mnist_valid.amat').astype('float32')
test_x = np.loadtxt(data_path + 'binarized_mnist_test.amat').astype('float32')
# shuffle dataset
train_x = row_shuffle(train_x)
valid_x = row_shuffle(valid_x)
test_x = row_shuffle(test_x)
return train_x, valid_x, test_x
def load_mnist(path, zero_mean=True):
mnist = np.load(path)
train_set_x = mnist['train_data']
train_set_y = mnist['train_labels'] + 1
test_set_x = mnist['test_data']
test_set_y = mnist['test_labels'] + 1
if zero_mean:
obs_mean = np.mean(train_set_x, axis=0, keepdims=True)
train_set_x = train_set_x - obs_mean
test_set_x = test_set_x - obs_mean
train_set_x, train_set_y = _shared_dataset((train_set_x, train_set_y))
test_set_x, test_set_y = _shared_dataset((test_set_x, test_set_y))
valid_set_x, valid_set_y = test_set_x, test_set_y
rval = [(train_set_x, train_set_y), (valid_set_x, valid_set_y),
(test_set_x, test_set_y)]
return rval
def load_udm_ss(dataset, sup_count, rng, zero_mean=True):
"""
Load semi-supervised version of the standard UdM MNIST data.
For this, the training data is split into labeled and unlabeled portions.
The number of labeled examples is 'sup_count', and an equal number of
labeled examples will be selected for each class. The remaining (50000 -
sup_count) examples are provided as unlabeled training data. The validate
and test sets are left unchanged.
Note: labels for the normal digit classes will range from 1-10, i.e. +1
compared to their standard value, as 'un-classed' examples take label 0.
"""
udm_data = load_udm(dataset, as_shared=False, zero_mean=zero_mean)
Xtr = udm_data[0][0]
Ytr = udm_data[0][1][:,np.newaxis]
all_count = Xtr.shape[0]
pc_count = int(np.ceil(sup_count / 10.0))
sup_count = int(10 * pc_count)
unsup_count = all_count - sup_count
Xtr_su = []
Ytr_su = []
Xtr_un = []
Ytr_un = []
# Sample supervised and unsupervised subsets of each class' observations
for c_label in np.unique(Ytr):
c_idx = [i for i in range(all_count) if (Ytr[i] == c_label)]
rng.shuffle(c_idx)
Xtr_su.append(Xtr[c_idx[0:pc_count],:])
Ytr_su.append(Ytr[c_idx[0:pc_count],:])
Xtr_un.append(Xtr[c_idx[pc_count:],:])
Ytr_un.append(Ytr[c_idx[pc_count:],:])
# Stack per-class supervised/unsupervised splits into matrices
Xtr_su = np.vstack(Xtr_su)
Ytr_su = np.vstack(Ytr_su)
Xtr_un = np.vstack(Xtr_un)
Ytr_un = np.vstack(Ytr_un)
# Also keep "unsupervised" copies of the "supervised" data
Xtr_un = Xtr_un #np.vstack([Xtr_un, Xtr_su])
Ytr_un = 0 * Ytr_un #np.vstack([Ytr_un, Ytr_su])
# Shuffle the rows so that observations are not grouped by class
shuf_idx = rng.permutation(Xtr_su.shape[0])
Xtr_su = Xtr_su[shuf_idx,:]
Ytr_su = Ytr_su[shuf_idx].ravel() + 1
shuf_idx = rng.permutation(Xtr_un.shape[0])
Xtr_un = Xtr_un[shuf_idx,:]
Ytr_un = Ytr_un[shuf_idx].ravel()
# Put matrices into GPU shared variables, for great justice
Xtr_su, Ytr_su = _shared_dataset((Xtr_su, Ytr_su))
Xtr_un, Ytr_un = _shared_dataset((Xtr_un, Ytr_un))
Xva, Yva = _shared_dataset((udm_data[1][0], (udm_data[1][1] + 1)))
Xte, Yte = _shared_dataset((udm_data[2][0], (udm_data[2][1] + 1)))
rval = [(Xtr_su, Ytr_su), (Xtr_un, Ytr_un), (Xva, Yva), (Xte, Yte)]
return rval
def load_udm(dataset, as_shared=True, zero_mean=True):
"""Loads the UdM train/validate/test split of MNIST."""
#############
# LOAD DATA #
#############
# Download the MNIST dataset if it is not present
data_dir, data_file = os.path.split(dataset)
if (not os.path.isfile(dataset)) and data_file == 'mnist.pkl.gz':
import urllib
origin = 'http://www.iro.umontreal.ca/~lisa/deep/data/mnist/mnist.pkl.gz'
print 'Downloading data from %s' % origin
urllib.urlretrieve(origin, dataset)
print '... loading data'
# Load the dataset
f = gzip.open(dataset, 'rb')
train_set, valid_set, test_set = cPickle.load(f)
f.close()
#train_set, valid_set, test_set format: tuple(input, target)
#input is an np.ndarray of 2 dimensions (a matrix)
#witch row's correspond to an example. target is a
#np.ndarray of 1 dimensions (vector)) that have the same length as
#the number of rows in the input. It should give the target
#target to the example with the same index in the input.
train_set = [v for v in train_set]
valid_set = [v for v in valid_set]
test_set = [v for v in test_set]
train_set[0] = np.asarray(train_set[0]).astype(np.float32)
valid_set[0] = np.asarray(valid_set[0]).astype(np.float32)
test_set[0] = np.asarray(test_set[0]).astype(np.float32)
if zero_mean:
obs_mean = np.mean(train_set[0], axis=0, keepdims=True)
train_set[0] = train_set[0] - obs_mean
valid_set[0] = valid_set[0] - obs_mean
test_set[0] = test_set[0] - obs_mean
if as_shared:
test_set_x, test_set_y = _shared_dataset((test_set[0],test_set[1]+1))
valid_set_x, valid_set_y = _shared_dataset((valid_set[0],valid_set[1]+1))
train_set_x, train_set_y = _shared_dataset((train_set[0],train_set[1]+1))
else:
test_set_x, test_set_y = (test_set[0], test_set[1]+1)
valid_set_x, valid_set_y = (valid_set[0], valid_set[1]+1)
train_set_x, train_set_y = (train_set[0], train_set[1]+1)
rval = [(train_set_x, train_set_y), (valid_set_x, valid_set_y),
(test_set_x, test_set_y)]
return rval
def load_svhn(tr_file, te_file, ex_file=None, ex_count=None):
"""
Loads the full SVHN train/test sets and an additional number of randomly
selected examples from the "extra set".
"""
# load the training set as a numpy arrays
pickle_file = open(tr_file)
data_dict = cPickle.load(pickle_file)
Xtr = data_dict['X'].astype(theano.config.floatX)
Ytr = data_dict['y'].astype(np.int32) + 1
Xtr_vec = np.zeros((Xtr.shape[3], 32*32*3)).astype(theano.config.floatX)
for i in range(Xtr.shape[3]):
c_pix = 32*32
for c in range(3):
Xtr_vec[i,c*c_pix:((c+1)*c_pix)] = \
Xtr[:,:,c,i].reshape((32*32,))
Xtr = Xtr_vec
pickle_file.close()
# load the test set as numpy arrays
pickle_file = open(te_file)
data_dict = cPickle.load(pickle_file)
Xte = data_dict['X'].astype(theano.config.floatX)
Yte = data_dict['y'].astype(np.int32) + 1
Xte_vec = np.zeros((Xte.shape[3], 32*32*3)).astype(theano.config.floatX)
for i in range(Xte.shape[3]):
c_pix = 32*32
for c in range(3):
Xte_vec[i,c*c_pix:((c+1)*c_pix)] = \
Xte[:,:,c,i].reshape((32*32,))
Xte = Xte_vec
pickle_file.close()
if ex_file is None:
Xex = None
else:
# load the extra digit examples and only keep a random subset
pickle_file = open(ex_file)
data_dict = cPickle.load(pickle_file)
ex_full_size = data_dict['X'].shape[3]
idx = npr.randint(low=0, high=ex_full_size, size=(ex_count))
Xex = data_dict['X'].take(idx, axis=3).astype(theano.config.floatX)
Xex_vec = np.zeros((Xex.shape[3], 32*32*3)).astype(theano.config.floatX)
for i in range(Xex.shape[3]):
c_pix = 32*32
for c in range(3):
Xex_vec[i,c*c_pix:((c+1)*c_pix)] = \
Xex[:,:,c,i].reshape((32*32,))
Xex = Xex_vec
pickle_file.close()
# package data up for easy returnage
data_dict = {'Xtr': Xtr, 'Ytr': Ytr, \
'Xte': Xte, 'Yte': Yte, \
'Xex': Xex}
return data_dict
def load_svhn_gray(tr_file, te_file, ex_file=None, ex_count=None):
"""
Load pickle files with grayscale versions of the SVHN data.
"""
# load the training set as a numpy arrays
pickle_file = open(tr_file)
data_dict = cPickle.load(pickle_file)
Xtr = data_dict['X']
print("Xtr.shape: {0:s}".format(str(Xtr.shape)))
Ytr = data_dict['y'].astype(np.int32)
Xtr_vec = np.zeros((Xtr.shape[2], 32*32), dtype=theano.config.floatX)
for i in range(Xtr.shape[2]):
c_pix = 32*32
Xtr_vec[i,:] = Xtr[:,:,i].reshape((32*32,)).astype(theano.config.floatX)
del Xtr
Xtr = Xtr_vec
pickle_file.close()
# load the test set as numpy arrays
pickle_file = open(te_file)
data_dict = cPickle.load(pickle_file)
pickle_file.close()
Xte = data_dict['X']
print("Xte.shape: {0:s}".format(str(Xte.shape)))
Yte = data_dict['y'].astype(np.int32)
Xte_vec = np.zeros((Xte.shape[2], 32*32), dtype=theano.config.floatX)
for i in range(Xte.shape[2]):
c_pix = 32*32
Xte_vec[i,:] = Xte[:,:,i].reshape((32*32,)).astype(theano.config.floatX)
del Xte
Xte = Xte_vec
# process extra data as desired
if ex_file is None:
Xex = None
else:
if ex_count is None:
ex_count = 100000000
# load the extra digit examples and only keep a subset
pickle_file = open(ex_file)
data_dict = cPickle.load(pickle_file)
pickle_file.close()
Xex = data_dict['X']
print("Xex.shape: {0:s}".format(str(Xex.shape)))
max_idx = min(ex_count, Xex.shape[2])
Xex_vec = np.zeros((max_idx, 32*32)).astype(theano.config.floatX)
for i in range(max_idx):
c_pix = 32*32
Xex_vec[i,:] = Xex[:,:,i].reshape((32*32,)).astype(theano.config.floatX)
del Xex
Xex = Xex_vec
# package data up for easy returnage
data_dict = {'Xtr': Xtr, 'Ytr': Ytr, \
'Xte': Xte, 'Yte': Yte, \
'Xex': Xex}
return data_dict
def load_svhn_all_gray_zca(all_file):
"""
Load a pickle file with whitened grayscale versions of the SVHN data.
"""
# load the training set as a numpy arrays
pickle_file = open(all_file)
data_dict = cPickle.load(pickle_file)
return data_dict
def load_tfd(tfd_pkl_name='', which_set='', fold=0):
"""
Load TFD dataset, stored as pickled dict rather than a .mat file.
"""
assert(((fold >= 0) and (fold < 5)) or (fold == 'all'))
# setup a map for grabbing indices to access the requested images
set_type_map = {'unlabeled': 0,
'train': 1,
'valid': 2,
'test': 3}
assert(which_set in set_type_map)
data = cPickle.load(open(tfd_pkl_name))
# get indices of images in the requested set
set_indices = np.zeros((data['folds'][:,0].shape[0],))
for i in range(data['folds'].shape[1]):
if ((i == fold) or (fold == 'all')):
set_indices = set_indices + \
(data['folds'][:,i] == set_type_map[which_set])
set_indices = set_indices > 0.1
assert(set_indices.sum() > 0)
# get the requested images and cast to theano.config.floatX
data_x = data['images'][set_indices].astype(theano.config.floatX)
# scale data into range [0...1]
data_x = data_x / 255.
# reshape to 1-d images
data_x = data_x.reshape((data_x.shape[0], data_x.shape[1]*data_x.shape[2]))
# some of the original unlabeled faces are all zero?
good_idx = np.sum(data_x, axis=1) > 0.1
data_x = data_x[good_idx]
# get labels if they were requested
if which_set != 'unlabeled':
data_y = data['labs_ex'][set_indices] - 1
data_y_identity = data['labs_id'][set_indices]
y_labels = 7
else:
data_y = None
data_y_identity = None
y_labels = None
# check label info
mask = data['labs_ex'][set_indices] > -1
print("lab_ex > -1: {0:d}".format(np.sum(mask)))
return [data_x, data_y, data_y_identity, y_labels]