-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathtrain.py
179 lines (158 loc) · 8.32 KB
/
train.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
import numpy as np
import os
import pickle
from datetime import datetime
import tensorflow as tf
from time import time
from model import Model
def read_data(file_path):
with open(os.path.join(file_path, 'train_dataset.pkl'), 'rb') as f:
train = pickle.load(f)
with open(os.path.join(file_path, 'val_dataset.pkl'), 'rb') as f:
val = pickle.load(f)
[train_q, train_a, train_c, train_l, train_c_real_len, train_q_real_len] = train
[val_q, val_a, val_c, val_l, val_c_real_len, val_q_real_len] = val
return ([train_q, train_a, train_c, train_l, train_c_real_len, train_q_real_len],\
[val_q, val_a, val_c, val_l, val_c_real_len, val_q_real_len])
def batch_iter(c, q, l, a, c_real_len, q_real_len, batch_size, num_epochs, shuffle=True):
"""
Generates a batch iterator for a dataset.
"""
c = np.array(c)
q = np.array(q)
l = np.array(l)
a = np.array(a)
c_real_len = np.array(c_real_len)
q_real_len = np.array(q_real_len)
data_size = len(q)
num_batches_per_epoch = int(data_size / batch_size) + 1
for epoch in range(num_epochs):
print("In epoch >> " + str(epoch + 1))
print("num batches per epoch is: " + str(num_batches_per_epoch))
# Shuffle the data at each epoch
if shuffle:
shuffle_indices = np.random.permutation(np.arange(data_size))
c_shuffled = c[shuffle_indices]
q_shuffled = q[shuffle_indices]
l_shuffled = l[shuffle_indices]
a_shuffled = a[shuffle_indices]
c_real_len_shuffled = c_real_len[shuffle_indices]
q_real_len_shuffled = q_real_len[shuffle_indices]
else:
c_shuffled = c
q_shuffled = q
l_shuffled = l
a_shuffled = a
c_real_len_shuffled = c_real_len
q_real_len_shuffled = q_real_len
for batch_num in range(num_batches_per_epoch):
start_index = batch_num * batch_size
end_index = (batch_num + 1) * batch_size
if end_index < data_size:
c_batch, q_batch, l_batch, a_batch, c_real_len_batch, q_real_len_batch = c_shuffled[start_index:end_index], \
q_shuffled[start_index:end_index], \
l_shuffled[start_index:end_index], \
a_shuffled[start_index:end_index], \
c_real_len_shuffled[start_index:end_index], \
q_real_len_shuffled[start_index:end_index]
yield list(zip(c_batch, q_batch, l_batch, a_batch, c_real_len_batch, q_real_len_batch))
def parse_config(string):
#parsing txt file
result = dict()
args = string.split("\t")
result['c_max_len'] = int(args[0])
result['s_max_len'] = int(args[1])
result['q_max_len'] = int(args[2])
result['babi_processed'] = args[3] # path
result['batch_size'] = int(args[4])
result['s_hidden'] = int(args[5])
result['q_hidden'] = int(args[5])
result['learning_rate'] = 2e-4
result['iter_time'] = 150
result['display_step'] = 100
return result
def main():
global config
with open('config.txt', 'r') as f:
config = parse_config(f.readline())
date = datetime.fromtimestamp(time()).strftime('%Y-%m-%d_%H:%M:%S')
model_id = "RN-" + date
save_dir = "./result/" + model_id
save_summary_path = os.path.join(save_dir, 'model_summary')
save_variable_path = os.path.join(save_dir, 'model_variables')
if not os.path.exists(save_dir):
os.makedirs(save_dir)
os.makedirs(save_summary_path)
os.makedirs(save_variable_path)
(train_dataset, val_dataset) = read_data(config['babi_processed'])
with tf.Graph().as_default():
sess = tf.Session()
start_time = time()
with sess.as_default():
rn = Model(config)
# Define Training procedure
global_step = tf.Variable(0, name='global_step', trainable = False)
opt = tf.train.AdamOptimizer(config['learning_rate'])
optimizer = opt.minimize(rn.loss, global_step = global_step)
loss_train = tf.summary.scalar("loss_train", rn.loss)
accuracy_train = tf.summary.scalar("accuracy_train", rn.accuracy)
train_summary_ops = tf.summary.merge([loss_train, accuracy_train])
loss_val = tf.summary.scalar("loss_val", rn.loss)
accuracy_val = tf.summary.scalar("accuracy_val", rn.accuracy)
val_summary_ops = tf.summary.merge([loss_val, accuracy_val])
saver = tf.train.Saver(tf.global_variables(),max_to_keep=4)
sess.run(tf.global_variables_initializer())
summary_writer = tf.summary.FileWriter(save_summary_path, sess.graph)
batch_train = batch_iter(c=train_dataset[2],
q=train_dataset[0],
l=train_dataset[3],
a=train_dataset[1],
c_real_len=train_dataset[4],
q_real_len=train_dataset[5],
num_epochs=config['iter_time'],
batch_size= config['batch_size'])
for train in batch_train:
c_batch, q_batch, l_batch, a_batch, c_real_len_batch, q_real_len_batch = zip(*train)
feed_dict = {rn.context: c_batch,
rn.question: q_batch,
rn.label: l_batch,
rn.answer: a_batch,
rn.context_real_len: c_real_len_batch,
rn.question_real_len: q_real_len_batch,
rn.is_training: True}
current_step = sess.run(global_step, feed_dict=feed_dict)
optimizer.run(feed_dict=feed_dict)
train_summary = sess.run(train_summary_ops, feed_dict=feed_dict)
summary_writer.add_summary(train_summary, current_step)
if current_step % (config['display_step']) == 0:
print("step: {}".format(current_step))
print("====validation start====")
batch_val = batch_iter(c = val_dataset[2],
q = val_dataset[0],
l = val_dataset[3],
a = val_dataset[1],
c_real_len=val_dataset[4],
q_real_len=val_dataset[5],
num_epochs=1,
batch_size=config['batch_size'])
accs = []
for val in batch_val:
c_val, q_val, l_val, a_val, c_real_len_val, q_real_len_val = zip(*val)
feed_dict = {rn.context: c_val,
rn.question: q_val,
rn.label: l_val,
rn.answer: a_val,
rn.context_real_len: c_real_len_val,
rn.question_real_len: q_real_len_val,
rn.is_training: False}
acc = rn.accuracy.eval(feed_dict=feed_dict)
accs.append(acc)
val_summary = sess.run(val_summary_ops, feed_dict=feed_dict)
summary_writer.add_summary(val_summary, current_step)
print("Mean accuracy=" + str(sum(accs) / len(accs)))
saver.save(sess, save_path=save_summary_path, global_step=current_step)
print("====training====")
end_time = time()
print("Training finished in {}sec".format(end_time-start_time))
if __name__ == '__main__' :
main()