-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_cnn.py
259 lines (237 loc) · 13.3 KB
/
train_cnn.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
from text_cnn import TextCNN
import tensorflow as tf
import os
import time
import datetime
import numpy as np
class TrainCNN(object):
def __init__(self, whether_word2vec, whether_tf, vocab_size, embedding_dim, sequence_length, vocab_processor, num_classes, tf_dict_size,
tf_emb_size=10, filter_sizes=[1, 2, 3], num_filters=50):
"""
:param whether_word2vec:
:param whether_tf:
:param vocab_size: 词典的长度
:param embedding_dim:
:param sequence_length: samples的最大长度
:param vocab_processor: 词典,需要保存到本地.在use model的时候需要这个词典
:param num_classes:
:param tf_dict_size:
:param tf_emb_size:
:param filter_sizes:
:param num_filters:
:return:
"""
g = tf.Graph()
with g.as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=False)
self.sess = tf.Session(config=session_conf)
with self.sess.as_default():
self.cnn = TextCNN(
whether_word2vec=whether_word2vec,
whether_tf=whether_tf,
vocab_size=vocab_size,
embedding_dim=embedding_dim,
sequence_length=sequence_length,
num_classes=num_classes,
tf_dict_size=tf_dict_size,
tf_emb_size=tf_emb_size,
filter_sizes=filter_sizes,
num_filters=num_filters,
l2_reg_lambda=0.0
)
self.optimizer = tf.train.AdamOptimizer(1e-3)
self.grads_and_vars = self.optimizer.compute_gradients(self.cnn.loss)
self.global_step = tf.Variable(0, name="global_step", trainable=False)
self.train_op = self.optimizer.apply_gradients(self.grads_and_vars, global_step=self.global_step)
# Keep track of gradient values and sparsity (optional)
self.grad_summaries = []
for g, v in self.grads_and_vars:
if g is not None:
grad_hist_summary = tf.histogram_summary("{}/grad/hist".format(v.name), g)
sparsity_summary = tf.scalar_summary("{}/grad/sparsity".format(v.name), tf.nn.zero_fraction(g))
self.grad_summaries.append(grad_hist_summary)
self.grad_summaries.append(sparsity_summary)
self.grad_summaries_merged = tf.merge_summary(self.grad_summaries)
# Output directory for models and summaries
self.timestamp = str(int(time.time()))
self.out_dir = os.path.abspath(os.path.join(os.path.curdir, "runs", self.timestamp))
# print("Writing to {}\n".format(self.out_dir))
# Summaries for loss and accuracy
self.loss_summary = tf.scalar_summary("loss", self.cnn.loss)
self.acc_summary = tf.scalar_summary("accuracy", self.cnn.accuracy)
# Train Summaries
self.train_summary_op = tf.merge_summary([self.loss_summary, self.acc_summary])
self.train_summary_dir = os.path.join(self.out_dir, "summaries", "train")
self.train_summary_writer = tf.train.SummaryWriter(self.train_summary_dir, self.sess.graph)
# Dev summaries
self.dev_summary_op = tf.merge_summary([self.loss_summary, self.acc_summary])
self.dev_summary_dir = os.path.join(self.out_dir, "summaries", "dev")
self.dev_summary_writer = tf.train.SummaryWriter(self.dev_summary_dir, self.sess.graph)
# checkpoint directory. Tensorflow assumes this directory already exits so we need to create it.
self.checkpoint_dir = os.path.abspath(os.path.join(self.out_dir, "checkpoints"))
self.checkpoint_prefix = os.path.join(self.checkpoint_dir, "model")
if not os.path.exists(self.checkpoint_dir):
os.makedirs(self.checkpoint_dir)
# 创建Saver来管理模型中的所有变量, add ops to save and restore all the variables.
self.saver = tf.train.Saver(tf.all_variables())
# Write vocabulary, save the vocabulary to disk.
vocab_processor.save(os.path.join(self.out_dir, "vocab"))
self.sess.run(tf.initialize_all_variables())
time_str = datetime.datetime.now().isoformat()
self.fp = open("result_"+str(time_str), "w")
def train_step(self, whether_word2vec, w_batch, y_batch, embedding):
print('train whether_word2vec:', whether_word2vec)
if whether_word2vec:
feed_dict = {
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
self.cnn.word_placeholder: embedding
}
_, step, summaries, loss, accuracy, predictions, embedding_init = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions, self.cnn.embedding_init], feed_dict=feed_dict)
else:
print("train not word2vec!", whether_word2vec)
feed_dict = {
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
}
_, step, summaries, loss, accuracy, predictions, = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions], feed_dict=feed_dict)
time_str = datetime.datetime.now().isoformat()
print("train# {}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
self.train_summary_writer.add_summary(summaries, step)
def train_step_tf(self, whether_word2vec, w_batch, t_tf_batch, a_tf_batch, j_tf_batch, y_batch, embedding):
print('train whether_word2vec:', whether_word2vec)
if whether_word2vec:
feed_dict = {
self.cnn.t_tf: t_tf_batch,
self.cnn.a_tf: a_tf_batch,
self.cnn.j_tf: j_tf_batch,
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
self.cnn.word_placeholder: embedding
}
_, step, summaries, loss, accuracy, predictions, embedding_init = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions, self.cnn.embedding_init], feed_dict=feed_dict)
else:
feed_dict = {
self.cnn.t_tf: t_tf_batch,
self.cnn.a_tf: a_tf_batch,
self.cnn.j_tf: j_tf_batch,
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
}
_, step, summaries, loss, accuracy, predictions, = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions], feed_dict=feed_dict)
time_str = datetime.datetime.now().isoformat()
print("train# {}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
self.train_summary_writer.add_summary(summaries, step)
def test_step(self, whether_word2vec, w_batch, y_batch, embedding, writer=None):
print("test step:")
if whether_word2vec:
feed_dict = {
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
self.cnn.word_placeholder: embedding
}
_, step, summaries, loss, accuracy, predictions, embedding_init = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions, self.cnn.embedding_init], feed_dict=feed_dict)
else:
print('test not word2vec!', whether_word2vec)
feed_dict = {
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
}
_, step, summaries, loss, accuracy, predictions, = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions], feed_dict=feed_dict)
time_str = datetime.datetime.now().isoformat()
print("test# {}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
self.fp.write("test# {}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
self.fp.write('\n')
if writer:
writer.add_summary(summaries, step)
def test_step_tf(self, whether_word2vec, w_batch, t_tf_batch, a_tf_batch, j_tf_batch, y_batch, embedding, writer=None):
if whether_word2vec:
feed_dict = {
self.cnn.t_tf: t_tf_batch,
self.cnn.a_tf: a_tf_batch,
self.cnn.j_tf: j_tf_batch,
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
self.cnn.word_placeholder: embedding
}
_, step, summaries, loss, accuracy, predictions, embedding_init = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions, self.cnn.embedding_init], feed_dict=feed_dict)
else:
feed_dict = {
self.cnn.t_tf: t_tf_batch,
self.cnn.a_tf: a_tf_batch,
self.cnn.j_tf: j_tf_batch,
self.cnn.input_x: w_batch,
self.cnn.input_y: y_batch,
self.cnn.dropout_keep_prob: 0.5, # set 0.5 at train step
}
_, step, summaries, loss, accuracy, predictions, = \
self.sess.run([self.train_op, self.global_step, self.train_summary_op, self.cnn.loss,
self.cnn.accuracy, self.cnn.predictions], feed_dict=feed_dict)
time_str = datetime.datetime.now().isoformat()
print("test# {}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
self.fp.write("test# {}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
self.fp.write('\n')
if writer:
writer.add_summary(summaries, step)
def cnn_train(self, whether_word2vec, whether_tf, embedding, w_tr, w_te, t_tf_tr, t_tf_te, a_tf_tr, a_tf_te, j_tf_tr, j_tf_te, y_tr, y_te,
test_every=200, checkpoint_every=200, batch_size=64, num_epoch=10, shuffle=True):
# Generate batches per_epoch and execute train step
data_size = len(w_tr)
num_batch_per_epoch = int(len(w_tr)/batch_size) + 1
print("train...")
print('whether_word2vec:', whether_word2vec)
for epoch in range(num_epoch):
if shuffle:
shuffle_indices = np.random.permutation(np.arange(data_size))
s_w_tr = w_tr[shuffle_indices]
s_y_tr = y_tr[shuffle_indices]
s_t_tf_tr = t_tf_tr[shuffle_indices]
s_a_tf_tr = a_tf_tr[shuffle_indices]
s_j_tf_tr = j_tf_tr[shuffle_indices]
else:
s_w_tr = w_tr
s_y_tr = y_tr
s_t_tf_tr = t_tf_tr
s_a_tf_tr = a_tf_tr
s_j_tf_tr = j_tf_tr
for batch_num in range(num_batch_per_epoch):
start = batch_num * batch_size
end = min((batch_num + 1) * batch_size, data_size)
if whether_tf:
self.train_step_tf(whether_word2vec, s_w_tr[start:end], s_t_tf_tr[start:end], s_a_tf_tr[start:end], s_j_tf_tr[start:end], s_y_tr[start:end], embedding)
else:
self.train_step(whether_word2vec, s_w_tr[start:end], s_y_tr[start:end], embedding)
current_step = tf.train.global_step(self.sess, self.global_step)
if current_step % test_every == 0:
print("\nEvaluation:")
if whether_tf:
self.test_step_tf(whether_word2vec, w_te, t_tf_te, a_tf_te, j_tf_te, y_te, embedding, writer=self.dev_summary_writer)
else:
self.test_step(whether_word2vec, w_te, y_te, embedding, writer=self.dev_summary_writer)
print("")
if current_step % checkpoint_every == 0:
path = self.saver.save(self.sess, self.checkpoint_prefix, global_step=current_step)
print("Saved model checkpoint to {}\n".format(path))