forked from lykeven/IDGAN
/
rnn.py
295 lines (227 loc) · 11.9 KB
/
rnn.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
# -*- coding: utf-8 -*-
__author__ = 'keven'
import numpy as np
import tensorflow as tf
import networkx as nx
from tensorflow.python.ops import tensor_array_ops, control_flow_ops
import gan_seq_tree
import utils
class Generator(object):
def __init__(self, num_emb, batch_size, emb_dim, hidden_dim, sequence_length, start_token, input_length=5,
learning_rate=0.01, num_epochs=100):
self.num_emb = num_emb
self.batch_size = batch_size
self.emb_dim = emb_dim
self.hidden_dim = hidden_dim
self.sequence_length = sequence_length
self.input_length = input_length
self.start_token = tf.constant([start_token] * self.batch_size, dtype=tf.int32)
self.learning_rate = tf.Variable(float(learning_rate), trainable=False)
self.num_epochs = num_epochs
self.params = []
self.temperature = 1.0
self.grad_clip = 5.0
with tf.variable_scope('generator'):
self.embeddings = tf.Variable(self.init_matrix([self.num_emb, self.emb_dim]))
self.params.append(self.embeddings)
self.recurrent_unit = self.create_recurrent_unit(self.params) # maps h_tm1 to h_t for generator
self.output_unit = self.create_output_unit(self.params) # maps h_t to o_t (output token logits)
# placeholder definition
self.x = tf.placeholder(tf.int32, shape=[self.batch_size, self.sequence_length]) # sequence of tokens generated by generator
# processed for batch
self.processed_x = tf.transpose(tf.nn.embedding_lookup(self.embeddings, self.x), perm=[1, 0, 2]) # seq_length x batch_size x emb_dim
self.x_seq = tf.transpose(self.x, perm=[1, 0])
# Initial states
self.h0 = tf.zeros([self.batch_size, self.hidden_dim])
self.h0 = tf.stack([self.h0, self.h0])
self.generate()
self.update()
def generate(self,):
# supervised pretraining for generator
ta_emb_x = tensor_array_ops.TensorArray(
dtype=tf.float32, size=self.sequence_length)
ta_emb_x = ta_emb_x.unstack(self.processed_x)
ta_ind_x = tensor_array_ops.TensorArray(
dtype=tf.int32, size=self.sequence_length)
ta_ind_x = ta_ind_x.unstack(self.x_seq)
def _pretrain_recurrence_2(i, x_t, h_tm1, g_predictions):
h_t = self.recurrent_unit(x_t, h_tm1)
o_t = self.output_unit(h_t)
prob = tf.nn.softmax(o_t)
prob = tf.nn.l2_normalize(prob, dim=1)
g_predictions = g_predictions.write(i, prob) # batch x vocab_size
x_tp1 = ta_emb_x.read(i)
x_ind = ta_ind_x.read(i)
return i + 1, x_tp1, h_t, g_predictions
def _g_recurrence_2(i, x_t, h_tm1, gen_o, gen_x, g_predictions):
h_t = self.recurrent_unit(x_t, h_tm1) # hidden_memory_tuple
o_t = self.output_unit(h_t)
prob = tf.nn.softmax(o_t)
prob = tf.nn.l2_normalize(prob, dim=1)
g_predictions = g_predictions.write(i + self.input_length, prob) # batch x vocab_size
log_prob = tf.log(prob)
next_token = tf.cast(
tf.reshape(tf.multinomial(log_prob, 1), [self.batch_size]), tf.int32)
x_tp1 = tf.nn.embedding_lookup(self.embeddings, next_token) # batch x emb_dim
gen_o = gen_o.write(i, tf.reduce_sum(tf.multiply(tf.one_hot(next_token, self.num_emb, 1.0, 0.0),
tf.nn.softmax(o_t)), 1)) # [batch_size] , prob
gen_x = gen_x.write(i, next_token) # indices, batch_size
return i + 1, x_tp1, h_t, gen_o, gen_x, g_predictions
predictions = tensor_array_ops.TensorArray(
dtype=tf.float32, size=self.sequence_length,
dynamic_size=False, infer_shape=True)
gen_o_f2 = tensor_array_ops.TensorArray(dtype=tf.float32, size=self.sequence_length - self.input_length,
dynamic_size=False, infer_shape=True)
gen_x_f2 = tensor_array_ops.TensorArray(dtype=tf.int32, size=self.sequence_length - self.input_length,
dynamic_size=False, infer_shape=True)
_, x_f2, h_f2, predictions = control_flow_ops.while_loop(
cond=lambda i, _1, _2, _3: i < self.input_length,
body=_pretrain_recurrence_2,
loop_vars=(tf.constant(0, dtype=tf.int32),
tf.nn.embedding_lookup(self.embeddings, self.start_token),
self.h0, predictions))
_, _, _, _, gen_seq, self.predictions = control_flow_ops.while_loop(
cond=lambda i, _1, _2, _3, _4, _5: i < self.sequence_length - self.input_length,
body=_g_recurrence_2,
loop_vars=(tf.constant(0, dtype=tf.int32),
x_f2, h_f2, gen_o_f2, gen_x_f2, predictions))
self.predictions = tf.transpose(self.predictions.stack(), perm=[1, 0, 2]) # batch_size x seq_length x vocab_size
self.gen_seq = tf.transpose(gen_seq.stack(), perm=[1, 0]) # batch_size x seq_length / 2
self.gen_x = tf.concat([self.x[:, :self.input_length], self.gen_seq], axis=1)
def compute_accuracy(x, y):
intersection = tf.sets.set_intersection(x, y)
union = tf.sets.set_union(x, y)
correct_number = tf.cast(tf.sets.set_size(intersection), tf.float32)
total_number = tf.cast(tf.sets.set_size(union), tf.float32)
return tf.reduce_mean(correct_number * 1.0 / total_number)
ground_truth = self.x[:, self.sequence_length - self.input_length:]
self.accuracy = compute_accuracy(self.gen_seq, ground_truth)
self.train_loss = -tf.reduce_sum(
tf.one_hot(tf.to_int32(tf.reshape(self.x, [-1])), self.num_emb, 1.0, 0.0) * tf.log(
tf.clip_by_value(tf.reshape(self.predictions, [-1, self.num_emb]), 1e-20, 1.0)
)
) / (self.sequence_length * self.batch_size)
def update(self,):
self.loss = -tf.reduce_sum(
tf.one_hot(tf.to_int32(tf.reshape(self.x, [-1])), self.num_emb, 1.0, 0.0) * tf.log(
tf.clip_by_value(tf.reshape(self.predictions, [-1, self.num_emb]), 1e-20, 1.0)
)
) / (self.sequence_length * self.batch_size)
opt = self.g_optimizer(self.learning_rate)
self.grad, _ = tf.clip_by_global_norm(tf.gradients(self.loss, self.params), self.grad_clip)
self.updates = opt.apply_gradients(zip(self.grad, self.params))
def generate_step(self, sess, x):
feed_dict = {self.x: x}
generate_sequence, generate_prob_table = sess.run([self.gen_x, self.predictions], feed_dict=feed_dict)
return generate_sequence, generate_prob_table
def get_accuracy(self, sess, x):
feed_dict = {self.x: x}
accuracy, loss = sess.run([self.accuracy, self.train_loss], feed_dict=feed_dict)
return accuracy, loss
def update_step(self, sess, x):
feed_dict = {self.x: x}
_ = sess.run(self.updates, feed_dict)
def init_matrix(self, shape):
return tf.random_normal(shape, stddev=0.1)
def init_vector(self, shape):
return tf.zeros(shape)
def create_recurrent_unit(self, params):
# Weights and Bias for input and hidden tensor
self.Wi = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
self.Ui = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
self.bi = tf.Variable(self.init_matrix([self.hidden_dim]))
self.Wf = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
self.Uf = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
self.bf = tf.Variable(self.init_matrix([self.hidden_dim]))
self.Wog = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
self.Uog = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
self.bog = tf.Variable(self.init_matrix([self.hidden_dim]))
self.Wc = tf.Variable(self.init_matrix([self.emb_dim, self.hidden_dim]))
self.Uc = tf.Variable(self.init_matrix([self.hidden_dim, self.hidden_dim]))
self.bc = tf.Variable(self.init_matrix([self.hidden_dim]))
params.extend([
self.Wi, self.Ui, self.bi,
self.Wf, self.Uf, self.bf,
self.Wog, self.Uog, self.bog,
self.Wc, self.Uc, self.bc])
def unit(x, hidden_memory_tm1):
previous_hidden_state, c_prev = tf.unstack(hidden_memory_tm1)
# Input Gate
i = tf.sigmoid(
tf.matmul(x, self.Wi) +
tf.matmul(previous_hidden_state, self.Ui) + self.bi
)
# Forget Gate
f = tf.sigmoid(
tf.matmul(x, self.Wf) +
tf.matmul(previous_hidden_state, self.Uf) + self.bf
)
# Output Gate
o = tf.sigmoid(
tf.matmul(x, self.Wog) +
tf.matmul(previous_hidden_state, self.Uog) + self.bog
)
# New Memory Cell
c_ = tf.nn.tanh(
tf.matmul(x, self.Wc) +
tf.matmul(previous_hidden_state, self.Uc) + self.bc
)
# Final Memory cell
c = f * c_prev + i * c_
# Current Hidden state
current_hidden_state = o * tf.nn.tanh(c)
return tf.stack([current_hidden_state, c])
return unit
def create_output_unit(self, params):
self.Wo = tf.Variable(self.init_matrix([self.hidden_dim, self.num_emb]))
self.bo = tf.Variable(self.init_matrix([self.num_emb]))
params.extend([self.Wo, self.bo])
def unit(hidden_memory_tuple):
hidden_state, c_prev = tf.unstack(hidden_memory_tuple)
# hidden_state : batch x hidden_dim
logits = tf.matmul(hidden_state, self.Wo) + self.bo
# output = tf.nn.softmax(logits)
return logits
return unit
def g_optimizer(self, *args, **kwargs):
return tf.train.AdamOptimizer(*args, **kwargs)
def main():
args = utils.parse_args_new()
# Data Parameters
origin_data_file = args.data_file
graph_file = args.graph_file
generated_num = args.num_train_sample
generated_num_test = args.num_test_sample
seq_length = args.seq_length
vocab_size = args.num_node
batch_size = args.batch_size
num_epochs = args.num_epochs
# Generator Hyper-parameters
g_emb_dim = args.g_dim_emb
g_hidden_size = args.g_hidden_size
train_percent = args.train_percent
g_num_expend = args.g_num_expend
input_length = int(seq_length * train_percent)
train_batch = int(generated_num / batch_size)
test_batch = int(generated_num_test / batch_size)
# Model
START_TOKEN = 0
utils.prepare_data(origin_data_file)
graph = nx.read_edgelist(graph_file, nodetype=int, create_using=nx.DiGraph())
adjacency_matrix = np.asarray(nx.adjacency_matrix(graph).todense()).transpose()
generator = Generator(vocab_size, batch_size, g_emb_dim, g_hidden_size, seq_length, START_TOKEN, input_length)
init = tf.global_variables_initializer()
sess = tf.InteractiveSession()
sess.run(init)
print 'Start training...'
for epoch in range(num_epochs):
for it in range(train_batch):
batch = utils.train_next_batch(generator.batch_size, hard=True)
generator.update_step(sess, batch)
if epoch % 5 == 0:
accuracy, test_loss, p_n, n_n = gan_seq_tree.test_accuracy_epoch(sess, generator, generator.batch_size,
test_batch, generator.input_length,
adjacency_matrix, g_num_expend)
print 'training epoch:%d loss:%.5f jaccard:%.5f p@n:%.5f, n@n:%.5f' % (epoch, test_loss, accuracy, p_n, n_n)
if __name__ == '__main__':
main()