-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathrun_deep_vae.py
executable file
·207 lines (169 loc) · 6.11 KB
/
run_deep_vae.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
import numpy as np
from models import BIVA
import sys
from custom import DeepVAEEvaluator
from data import load_mnist_binarized, load_cifar
from utils import AdamaxOptimizer
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.allow_growth = False
def get_deep_vae_mnist(name):
filters = 64
no_layers = 2
enc = []
z = []
enc_z1 = [[filters, (5, 5), (1, 1)]] * no_layers
enc_z1 += [[filters, (5, 5), (2, 2)]]
z_1 = 48
enc += [enc_z1]
z += [z_1]
enc_z2 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z2 += [[filters, (3, 3), (1, 1)]]
z_2 = 40
enc += [enc_z2]
z += [z_2]
enc_z3 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z3 += [[filters, (3, 3), (1, 1)]]
z_3 = 32
enc += [enc_z3]
z += [z_3]
enc_z4 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z4 += [[filters, (3, 3), (1, 1)]]
z_4 = 24
enc += [enc_z4]
z += [z_4]
enc_z5 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z5 += [[filters, (3, 3), (1, 1)]]
z_5 = 16
enc += [enc_z5]
z += [z_5]
enc_z6 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z6 += [[filters, (3, 3), (2, 2)]]
z_6 = 8
enc += [enc_z6]
z += [z_6]
return BIVA(tf.Session(), [28, 28, 1], enc, z, tf.nn.elu, "bernoulli", model_name=name, dropout_inference=.5,
dropout_generative=.5, eps=1e-8, is_log_var=True, minimum_kl=-2.)
def get_deep_vae_cifar(name):
filters = 96
no_layers = 2
enc = []
z = []
enc_z1 = [[filters, (5, 5), (1, 1)]] * no_layers
enc_z1 += [[filters, (5, 5), (2, 2)]]
z_1 = [38, (16, 16), (1, 1)]
enc += [enc_z1]
z += [z_1]
enc_z2 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z2 += [[filters, (3, 3), (1, 1)]]
z_2 = [36, (16, 16), (1, 1)]
enc += [enc_z2]
z += [z_2]
enc_z3 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z3 += [[filters, (3, 3), (1, 1)]]
z_3 = [34, (16, 16), (1, 1)]
enc += [enc_z3]
z += [z_3]
enc_z4 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z4 += [[filters, (3, 3), (1, 1)]]
z_4 = [32, (16, 16), (1, 1)]
enc += [enc_z4]
z += [z_4]
enc_z5 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z5 += [[filters, (3, 3), (1, 1)]]
z_5 = [30, (16, 16), (1, 1)]
enc += [enc_z5]
z += [z_5]
enc_z6 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z6 += [[filters, (3, 3), (1, 1)]]
z_6 = [28, (16, 16), (1, 1)]
enc += [enc_z6]
z += [z_6]
enc_z7 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z7 += [[filters, (3, 3), (1, 1)]]
z_7 = [26, (16, 16), (1, 1)]
enc += [enc_z7]
z += [z_7]
enc_z8 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z8 += [[filters, (3, 3), (1, 1)]]
z_8 = [24, (16, 16), (1, 1)]
enc += [enc_z8]
z += [z_8]
enc_z9 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z9 += [[filters, (3, 3), (1, 1)]]
z_9 = [22, (16, 16), (1, 1)]
enc += [enc_z9]
z += [z_9]
enc_z10 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z10 += [[filters, (3, 3), (1, 1)]]
z_10 = [20, (16, 16), (1, 1)]
enc += [enc_z10]
z += [z_10]
enc_z11 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z11 += [[filters, (3, 3), (2, 2)]]
z_11 = [18, (8, 8), (1, 1)]
enc += [enc_z11]
z += [z_11]
enc_z12 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z12 += [[filters, (3, 3), (1, 1)]]
z_12 = [16, (8, 8), (1, 1)]
enc += [enc_z12]
z += [z_12]
enc_z13 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z13 += [[filters, (3, 3), (1, 1)]]
z_13 = [14, (8, 8), (1, 1)]
enc += [enc_z13]
z += [z_13]
enc_z14 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z14 += [[filters, (3, 3), (1, 1)]]
z_14 = [12, (8, 8), (1, 1)]
enc += [enc_z14]
z += [z_14]
enc_z15 = [[filters, (3, 3), (1, 1)]] * no_layers
enc_z15 += [[filters, (3, 3), (2, 2)]]
z_15 = [10, (4, 4), (1, 1)]
enc += [enc_z15]
z += [z_15]
return BIVA(tf.Session(), [32, 32, 3], enc, z, tf.nn.elu, "discretized", model_name=name, dropout_inference=.2,
eps=1e-8, is_log_var=True, minimum_kl=-2.)
if __name__ == "__main__":
dataset_arg = sys.argv[1].lower()
deep_vae = None
if "mnist" in dataset_arg:
print("Training a MNIST binarized model...")
deep_vae = get_deep_vae_mnist("mnist_binarized")
train_x, valid_x, test_x = load_mnist_binarized()
train_batch_size = 32
valid_batch_size = 50
w, h = 28, 28
train_x = np.append(train_x, valid_x, axis=0)
preprocess_batch = lambda x: x
updates_in_epoch = None
temp = [1.]
train_x = np.reshape(train_x, (-1, h, w, 1))
test_x = np.reshape(test_x, (-1, h, w, 1))
eval = DeepVAEEvaluator(test_x, n_images=10, eval_every=1, preprocess_batch=preprocess_batch)
fn_eval = [eval.deep_vae_generate_evaluator]
elif "cifar" in dataset_arg:
print("Training a CIFAR10 model...")
deep_vae = get_deep_vae_cifar("cifar")
train_x, test_x = load_cifar(levels=256)
train_x = train_x.reshape((-1, 32, 32, 3))
test_x = test_x.reshape((-1, 32, 32, 3))
train_batch_size = 48
valid_batch_size = 50
w, h = 32, 32
preprocess_batch = lambda x: x
updates_in_epoch = None
temp = [1.]
eval = DeepVAEEvaluator(test_x, iw_samples=1, n_images=5, eval_every=1)
fn_eval = [eval.deep_vae_generate_evaluator]
assert deep_vae is not None, "Please enter the name of the experiment as an argument."
deep_vae.train_multi_gpu(train_x, test_x, n_epochs=10000, temperatures=temp, eq=1, iw=1,
train_batch_size=train_batch_size,
valid_batch_size=valid_batch_size, preprocess_batch=preprocess_batch,
optimizer=AdamaxOptimizer,
optimizer_args=(0.9, 0.999,), init_learning_rate=0.002, final_learning_rate=0.0001,
fn_decay=lambda x: x * 0.999999,
gradient_clipping=lambda x: x, fn_epoch=fn_eval,
updates_in_epoch=updates_in_epoch, debug_every=100)