-
Notifications
You must be signed in to change notification settings - Fork 0
/
hparam.py
327 lines (256 loc) · 9.71 KB
/
hparam.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
# coding: utf-8
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
from copy import deepcopy
import os
import numpy as np
import tensorflow.compat.v1 as tf
from dataset import DTY_FLT, DTY_INT
from dataset import generator
from dataset import preprocess_image_28
from dataset import preprocess_image_32
# ======================================
# Preliminaries
# --------------------------------------
# Argparser
os.environ["CUDA_VISIBLE_DEVICES"] = '-1'
GPU_OPTIONS = tf.compat.v1.GPUOptions(allow_growth=True)
CONFIG = tf.compat.v1.ConfigProto(gpu_options=GPU_OPTIONS)
# ======================================
# Preliminaries
# --------------------------------------
# Argparser
# Saved Parameters
def default_args_params():
parser = argparse.ArgumentParser()
parser.add_argument(
'-data', '--dataset', type=str, default='mnist',
choices=['cifar10', 'mnist', 'fashion_mnist', 'fmnist'],
help='Data set')
parser.add_argument(
'-model', '--model_setting', type=str, default='dnn',
# choices=['dnn', 'cnn'], help='Builder')
choices=["dnn", "cnn", 'linear'], help='Builder')
parser.add_argument(
'-type', '--type_pruning', type=str, default='AdaNet.O',
choices=['AdaNet.O', 'SAEP.O', 'PRS.O', 'PAP.O', 'PIE.O',
'AdaNet.W', 'SAEP.W', 'PRS.W', 'PAP.W', 'PIE.W'],
help='How to prune the AdaNet')
parser.add_argument(
'-alpha', '--thinp_alpha', type=float, default=0.5,
help='The value of alpha in PIE')
parser.add_argument(
'-cv', '--cross_validation', type=int,
default=1, # or 5, 1 means no cross_validation
help='Cross validation')
parser.add_argument(
'-bi', '--binary', action='store_true',
help='binary classification (pairs)')
parser.add_argument(
'-c0', '--label_zero', type=int, default=4,
help='the first label in pairs')
parser.add_argument(
'-c1', '--label_one', type=int, default=9,
help='the second label in pairs')
parser.add_argument(
'-device', '--cuda_device', type=str, default='0',
help='cuda_visible_devices')
parser.add_argument(
'-lr', '--learning_rate', type=float, default=0.003,
help='LEARNING_RATE')
parser.add_argument(
'-ts', '--train_steps', type=int, default=5000,
help='TRAIN_STEPS')
parser.add_argument(
'-bs', '--batch_size', type=int, default=64,
help='BATCH_SIZE')
parser.add_argument(
'-rs', '--random_seed', type=str, default='None',
help='RANDOM_SEED')
parser.add_argument(
'-it', '--adanet_iterations', type=int, default=11,
help='ADANET_ITERATIONS')
parser.add_argument(
'-mix', '--adanet_learn_mixture', action='store_true',
# type=str, default='F', choices=['T', 'F'],
help='LEARN_MIXTURE_WEIGHTS')
parser.add_argument(
'-lam', '--adanet_lambda', type=float, default=0,
help='ADANET_LAMBDA')
return parser
# --------------------------------------
# Logs
def default_logs_folder(args, saved='tmpmodels'):
LOG_TLE = args.dataset
LOG_TLE += '_cv' + str(args.cross_validation)
# LOG_TLE += '_it' + str(args.adanet_iterations)
# LOG_TLE += '_lr' + str(args.learning_rate)
# LOG_TLE += '_bs' + str(args.batch_size)
# LOG_TLE += '_ts' + str(args.train_steps // 1000)
LOG_DIR = os.path.join(os.getcwd(), saved)
LOG_DIR = os.path.join(LOG_DIR, args.dataset)
TF_LOG_TLE = args.dataset
if args.binary:
feat_pair = (args.label_zero, args.label_one)
feat_temp = 'pair' + ''.join(map(str, feat_pair))
else:
feat_temp = 'multi'
LOG_DIR = os.path.join(LOG_DIR, feat_temp)
LOG_TLE += '_' + feat_temp
TF_LOG_TLE += '_' + feat_temp
LOG_TLE += '_' + args.model_setting
TF_LOG_TLE += '_' + args.model_setting
thinp_alpha = args.thinp_alpha
type_pruning = args.type_pruning
LOG_TLE += '_' + type_pruning
TF_LOG_TLE += '_' + type_pruning
if type_pruning.startswith('PIE'):
LOG_TLE += str(thinp_alpha)
TF_LOG_TLE += str(thinp_alpha)
if type_pruning.endswith('W'):
lmw = str(args.adanet_learn_mixture)
LOG_TLE += lmw[0]
TF_LOG_TLE += lmw[0]
# if args.cross_validation > 0:
# LOG_TLE += '_cv' + str(args.cross_validation)
# else:
# LOG_TLE += '_sing'
# return LOG_TLE, LOG_DIR, feat_temp
return TF_LOG_TLE, LOG_TLE, LOG_DIR
# ======================================
# Auxilliary
# --------------------------------------
# Cross validation:
# different ways to split data
def situation_cross_validation(nb_iter, y, split_type='cross_valid_v2'):
if split_type not in ['cross_valid_v3', 'cross_valid_v2',
'cross_validation', 'cross_valid']:
raise ValueError("invalid `split_type`, {}.".format(split_type))
y = np.array(y)
vY = np.unique(y)
dY = len(vY)
iY = [np.where(y == j)[0] for j in vY] # indexes
lY = [len(j) for j in iY] # length
tY = [np.copy(j) for j in iY] # temp_index
for j in tY:
np.random.shuffle(j)
sY = [int(np.floor(j / nb_iter)) for j in lY] # split length
if nb_iter == 2:
sY = [int(np.floor(j / (nb_iter + 1))) for j in lY]
elif nb_iter == 1:
sY = [int(np.floor(j / (nb_iter + 1))) for j in lY]
split_idx = []
for k in range(1, nb_iter + 1):
i_tst, i_val, i_trn = [], [], []
for i in range(dY):
k_former = sY[i] * (k - 1)
k_middle = sY[i] * k
k_latter = sY[i] * (k + 1) if k != nb_iter else sY[i]
i_tst.append(tY[i][k_former: k_middle])
if k != nb_iter:
i_val.append(tY[i][k_middle: k_latter])
i_trn.append(
np.concatenate(
[tY[i][k_latter:], tY[i][: k_former]], axis=0))
else:
i_val.append(tY[i][: k_latter])
i_trn.append(
np.concatenate(
[tY[i][k_middle:], tY[i][k_latter: k_former]], axis=0))
i_tst = np.concatenate(i_tst, axis=0).tolist()
i_val = np.concatenate(i_val, axis=0).tolist()
i_trn = np.concatenate(i_trn, axis=0).tolist()
temp_ = (deepcopy(i_trn), deepcopy(i_val), deepcopy(i_tst))
if split_type.endswith('v2'):
temp_ = (deepcopy(i_trn + i_val), deepcopy(i_tst))
split_idx.append(deepcopy(temp_))
del k_former, k_middle, k_latter, i_tst, i_val, i_trn
del k, y, vY, dY, iY, lY, tY, sY, nb_iter
# gc.collect()
return deepcopy(split_idx)
# --------------------------------------
# 1. download the data
def feed_dataset_all_in(datafeed, binary=False, c0=4, c1=9):
if datafeed.startswith('cifar10'):
NUM_CLASS = 10
NUM_SHAPE = (32, 32, 3)
TARGETS = ['airplane', 'automobile', 'bird', 'cat', 'deer',
'dog', 'frog', 'horse', 'ship', 'truck']
elif datafeed.endswith('mnist'):
NUM_CLASS = 10
NUM_SHAPE = (28, 28, 1)
TARGETS = ["T-shirt/top", "Trouser", "Pullover", "Dress", "Coat",
"Sandal", "Shirt", "Sneaker", "Bag", "Ankle boot", ]
else:
raise ValueError("No such dataset named {}!".format(datafeed))
if datafeed == 'cifar10':
(X_train, y_train), (
X_test, y_test) = tf.keras.datasets.cifar10.load_data()
elif datafeed == 'mnist':
(X_train, y_train), (
X_test, y_test) = tf.keras.datasets.mnist.load_data()
elif datafeed == 'fmnist' or datafeed == 'fashion_mnist':
(X_train, y_train), (
X_test, y_test) = tf.keras.datasets.fashion_mnist.load_data()
else:
X_train = y_train = X_test = y_test = None
if datafeed.startswith('cifar'):
y_train = y_train.reshape(-1)
y_test = y_test.reshape(-1)
if datafeed.endswith('mnist'):
X_train = X_train.reshape(-1, 28, 28, 1)
X_test = X_test.reshape(-1, 28, 28, 1)
if binary:
mask_train = (y_train == c0) | (y_train == c1)
mask_test = (y_test == c0) | (y_test == c1)
X_train = X_train[mask_train]
y_train = y_train[mask_train]
X_test = X_test[mask_test]
y_test = y_test[mask_test]
return NUM_CLASS, NUM_SHAPE, TARGETS, X_train, y_train, X_test, y_test
# --------------------------------------
# Data Set
def default_data_feedin(args):
datafeed = args.dataset
if not args.binary:
return feed_dataset_all_in(datafeed, False)
# binary = args.binary
c0, c1 = args.label_zero, args.label_one
return feed_dataset_all_in(datafeed, True, c0, c1)
# --------------------------------------
# 1. download the data
def super_input_fn(X_train, y_train, X_test, y_test,
NUM_SHAPE, RANDOM_SEED):
def input_fn(partition, training, batch_size):
# Generate an input_fn for the Estimator.
def _input_fn():
if partition == "train":
dataset = tf.data.Dataset.from_generator(generator(
X_train, y_train), (DTY_FLT, DTY_INT), (NUM_SHAPE, ()))
elif partition == "predict":
dataset = tf.data.Dataset.from_generator(generator(
X_test[:30], y_test[:30]), (DTY_FLT, DTY_INT), (NUM_SHAPE, ()))
else:
dataset = tf.data.Dataset.from_generator(generator(
X_test, y_test), (DTY_FLT, DTY_INT), (NUM_SHAPE, ()))
# We call repeat after shuffling, rather than before, to prevent
# separate epochs from blending together.
if training:
dataset = dataset.shuffle(10 * batch_size,
seed=RANDOM_SEED).repeat()
if NUM_SHAPE[0] == 32: # NUM_SHAPE == (32, 32, 3):
dataset = dataset.map(preprocess_image_32)
elif NUM_SHAPE[0] == 28: # NUM_SHAPE == (28, 28, 1):
dataset = dataset.map(preprocess_image_28)
else:
raise ValueError("invalid `NUM_SHAPE`, {}.".format(NUM_SHAPE))
dataset = dataset.shuffle(buffer_size=10000)
dataset = dataset.batch(batch_size)
dataset = dataset.repeat(10) # num_epochs)
iterator = dataset.make_one_shot_iterator()
features, labels = iterator.get_next()
return features, labels
return _input_fn
return input_fn