-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathWGAN_GP_Char.py
160 lines (128 loc) · 6.55 KB
/
WGAN_GP_Char.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
__author__= 'Weili Nie'
import os
import tensorflow as tf
from tqdm import trange
from lang_helpers import load_dataset
from config import get_config
from model_lang import *
from utils import *
class WGAN_GP_Char(object):
def __init__(self, config):
self.d_net = config.d_net
self.g_net = config.g_net
self.dataset = config.dataset_lang
self.data_path = config.data_path
self.max_train_data = config.max_train_data
self.beta1 = config.beta1
self.beta2 = config.beta2
self.optimizer = config.optimizer
self.batch_size = config.batch_size
self.kernel_size = config.kernel_size
self.z_dim = config.z_dim
self.conv_hidden_num = config.conv_hidden_num
self.g_lr = config.g_lr
self.d_lr = config.d_lr
self.seq_len = config.seq_len
self.model_dir = config.model_dir
self.start_step = 0
self.log_step = config.log_step
self.max_step = config.max_step
self.save_step = config.save_step
self.lmd = config.lmd
self.critic_iters = config.critic_iters
self.build_model()
self.saver = tf.train.Saver()
self.summary_writer = tf.summary.FileWriter(self.model_dir)
sv = tf.train.Supervisor(logdir=self.model_dir,
is_chief=True,
saver=self.saver,
summary_op=None,
summary_writer=self.summary_writer,
save_model_secs=300,
ready_for_local_init_op=None)
gpu_options = tf.GPUOptions(allow_growth=True)
sess_config = tf.ConfigProto(allow_soft_placement=True, gpu_options=gpu_options)
self.sess = sv.prepare_or_wait_for_session(config=sess_config)
def build_model(self):
self.lines, self.charmap, self.inv_charmap = load_dataset(
seq_len=self.seq_len,
max_train_data=self.max_train_data,
data_path=self.data_path
)
vocab_size = len(self.charmap)
self.z = tf.random_normal(shape=[self.batch_size, self.z_dim])
self.real_data = tf.placeholder(tf.int32, shape=[self.batch_size, self.seq_len])
real_data_one_hot = tf.one_hot(self.real_data, vocab_size)
fake_data_softmax, g_vars = generator(self.g_net, self.z, self.conv_hidden_num, self.seq_len,
self.kernel_size, vocab_size)
self.fake_data = tf.argmax(fake_data_softmax, fake_data_softmax.get_shape().ndims-1)
d_out_real, d_vars = discriminator(self.d_net, real_data_one_hot, self.conv_hidden_num,
self.seq_len, self.kernel_size, vocab_size, reuse=False)
d_out_fake, _ = discriminator(self.d_net, fake_data_softmax, self.conv_hidden_num,
self.seq_len, self.kernel_size, vocab_size)
self.d_loss = tf.reduce_mean(d_out_fake) - tf.reduce_mean(d_out_real)
self.g_loss = -tf.reduce_mean(d_out_fake)
# WGAN lipschitz-penalty
epsilon = tf.random_uniform(shape=[self.batch_size, 1, 1], minval=0., maxval=1.)
data_diff = fake_data_softmax - real_data_one_hot
interp_data = real_data_one_hot + epsilon * data_diff
disc_interp, _ = discriminator(self.d_net, interp_data, self.conv_hidden_num,
self.seq_len, self.kernel_size, vocab_size)
grad_interp = tf.gradients(disc_interp, [interp_data])[0]
print('The shape of grad_interp: {}'.format(grad_interp.get_shape().as_list()))
self.slope = tf.norm(grad_interp)
gradient_penalty = tf.reduce_mean((self.slope - 1.) ** 2)
self.d_loss += self.lmd * gradient_penalty
if self.optimizer == 'adam':
optim = tf.train.AdamOptimizer
else:
raise Exception("[!] Caution! Paper didn't use {} optim other than Adam".format(self.optimizer))
self.d_optim = optim(self.d_lr, self.beta1, self.beta2).minimize(self.d_loss, var_list=d_vars)
self.g_optim = optim(self.g_lr, self.beta1, self.beta2).minimize(self.g_loss, var_list=g_vars)
self.summary_op = tf.summary.merge([
tf.summary.scalar("loss/d_loss", self.d_loss),
tf.summary.scalar("loss/g_loss", self.g_loss),
tf.summary.scalar("grad/norm_gradient", self.slope)
])
def train(self):
z_fixed = np.random.normal(size=[self.batch_size*10, self.z_dim]) # samples of 10 times batch size
gen = inf_train_gen(self.lines, self.batch_size, self.charmap)
for step in trange(self.max_step):
# Train generator
_data = gen.next()
summary_str, _ = self.sess.run([self.summary_op, self.g_optim], feed_dict={self.real_data: _data})
self.summary_writer.add_summary(summary_str, global_step=step)
self.summary_writer.flush()
# Train critic
for i in range(self.critic_iters):
_data = gen.next()
self.sess.run(self.d_optim, feed_dict={self.real_data: _data})
if step % 100 == 99:
_data = gen.next()
g_loss, d_loss, slope = self.sess.run([self.g_loss, self.d_loss, self.slope],
feed_dict={self.real_data: _data})
print("[{}/{}] Loss_D: {:.6f} Loss_G: {:.6f} Slope: {:.6f}".
format(step+1, self.max_step, d_loss, g_loss, slope))
self.generate_samples(z_fixed, idx=step+1)
def generate_samples(self, z_fixed, idx):
gen_samples = []
for num in range(10):
z_fixed_bs = z_fixed[self.batch_size*num: self.batch_size*(num+1)]
samples = self.sess.run(self.fake_data, feed_dict={self.z: z_fixed_bs})
decoded_samples = []
for i in range(len(samples)):
decoded = []
for j in range(len(samples[i])):
decoded.append(self.inv_charmap[samples[i][j]])
decoded_samples.append(tuple(decoded))
gen_samples.extend(decoded_samples)
with open(os.path.join(self.model_dir, 'samples_{}.txt'.format(idx)), 'w') as f:
for s in gen_samples:
s = "".join(s)
f.write(s + "\n")
if __name__ == '__main__':
config, unparsed = get_config()
os.environ['CUDA_VISIBLE_DEVICES'] = config.gpus
prepare_dirs(config, config.dataset_lang)
wgan_gp_char = WGAN_GP_Char(config)
wgan_gp_char.train()