-
Notifications
You must be signed in to change notification settings - Fork 76
/
Copy pathtrain.py
186 lines (154 loc) · 7.69 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
180
181
182
183
184
185
186
import os
import argparse
import numpy as np
import tensorflow as tf
import time
from dataloader.data_generator import load_data
from models.cliquenet import build_model
def into_batch(data, label, batch_size, shuffle):
if shuffle:
rand_indexes = np.random.permutation(data.shape[0])
data = data[rand_indexes]
label = label[rand_indexes]
batch_count=len(data)/batch_size
batches_data = np.split(data[:batch_count*batch_size], batch_count)
batches_data.append(data[batch_count*batch_size:])
batches_labels = np.split(label[:batch_count * batch_size], batch_count)
batches_labels.append(label[batch_count*batch_size:])
batch_count+=1
return batches_data, batches_labels, batch_count
def count_params():
total_params=0
for variable in tf.trainable_variables():
shape=variable.get_shape()
params=1
for dim in shape:
params=params*dim.value
total_params+=params
print("Total training params: %.2fM" % (total_params / 1e6))
if __name__=='__main__':
##
train_params={'normalize_type': 'by-channels', ## by-channels, divide-255, divide-256
'initial_lr': 0.1,
'weight_decay': 1e-4,
'batch_size': 64,
'total_epoch': 300,
'keep_prob':0.8
}
parser = argparse.ArgumentParser()
parser.add_argument('--gpu', default="0")
parser.add_argument('--dataset',
choices=['cifar-10', 'cifar-100', 'svhn'])
parser.add_argument('--k', type=int,
help='filters per layer')
parser.add_argument('--T', type=int,
help='total layers in all blocks')
parser.add_argument('--dir',
help='folder to store models')
parser.add_argument('--if_a', default=False, type=bool,
help='if use attentional transition')
parser.add_argument('--if_b', default=False, type=bool,
help='if use bottleneck architecture')
parser.add_argument('--if_c', default=False, type=bool,
help='if use compression')
args = parser.parse_args()
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
dataset=args.dataset
if dataset=='svhn':
total_epoches=40
else:
total_epoches = train_params['total_epoch']
result_dir=args.dir
batch_size = train_params['batch_size']
lr = train_params['initial_lr']
kp = train_params['keep_prob']
weight_decay = train_params['weight_decay']
if os.path.exists(result_dir)==False:
os.mkdir(result_dir)
train_data, train_label, test_data, test_label=load_data(dataset, train_params['normalize_type'])
image_size=train_data.shape[1:]
label_num=train_label.shape[-1]
graph=tf.Graph()
with graph.as_default():
input_images=tf.placeholder(tf.float32, [None, image_size[0], image_size[1], image_size[2]], name='input_images')
true_labels=tf.placeholder(tf.float32, [None, label_num], name='labels')
is_train=tf.placeholder(tf.bool, shape=[])
learning_rate=tf.placeholder(tf.float32, shape=[], name='learning_rate')
keep_prob=tf.placeholder(tf.float32, shape=[], name='keep_prob')
### build model
logits, prob=build_model(input_images, args.k, args.T, label_num, is_train, keep_prob, args.if_a, args.if_b, args.if_c)
### loss and accuracy
loss_cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=true_labels))
if_correct = tf.equal(tf.argmax(prob, 1), tf.argmax(true_labels, 1))
accuracy = tf.reduce_mean(tf.cast(if_correct, tf.float32))
l2_loss = tf.add_n([tf.nn.l2_loss(var) for var in tf.trainable_variables()])
### optimizer
optimizer = tf.train.MomentumOptimizer(learning_rate, 0.9, use_nesterov=True)
train_op = optimizer.minimize(loss_cross_entropy + l2_loss*weight_decay)
saver=tf.train.Saver()
### begin training ###
config=tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config, graph=graph) as sess:
sess.run(tf.global_variables_initializer())
count_params()
### train batch data ###
# to shuffle before each epoch
### test batch data ###
batches_data_test, batches_labels_test, batch_count_test = into_batch(test_data, test_label, batch_size, shuffle=False)
loss_train=[]
acc_train=[]
loss_test=[]
acc_test=[]
best_acc=0
for epoch in range(1, total_epoches+1):
if epoch == total_epoches/2 : lr=lr*0.1
if epoch == total_epoches*3/4 : lr=lr*0.1
batches_data, batches_labels, batch_count=into_batch(train_data, train_label, batch_size, shuffle=True)
### train ###
loss_per_bat=[]
acc_per_bat=[]
for batch_id in range(batch_count):
data_per_bat = batches_data[batch_id]
label_per_bat = batches_labels[batch_id]
result_per_bat = sess.run([train_op, loss_cross_entropy, accuracy],
feed_dict={input_images : data_per_bat,
true_labels : label_per_bat,
learning_rate : lr,
is_train : True,
keep_prob: kp})
loss_per_bat.append(result_per_bat[1])
acc_per_bat.append(result_per_bat[2])
if (batch_id+1) % 100==0:
print 'epoch:', epoch, 'batch:', batch_id+1, 'in', batch_count
print 'loss:', result_per_bat[1], 'accuracy:', result_per_bat[2]
saver.save(sess, os.path.join(result_dir, dataset+'_epoch_%d.ckpt' % epoch))
loss_train.append(np.mean(loss_per_bat))
acc_train.append(np.mean(acc_per_bat))
### test ###
loss_per_bat=[]
acc_per_bat=[]
for batch_id in range(batch_count_test):
data_per_bat = batches_data_test[batch_id]
label_per_bat = batches_labels_test[batch_id]
result_per_bat = sess.run([loss_cross_entropy, accuracy],
feed_dict={input_images : data_per_bat,
true_labels : label_per_bat,
is_train: False,
keep_prob: 1})
loss_per_bat.append(result_per_bat[0]) ## result[0]->loss
acc_per_bat.append(result_per_bat[1]) ## result[1]->acc
loss_test.append(np.mean(loss_per_bat))
acc_test.append(np.mean(acc_per_bat))
if acc_test[-1]>best_acc:
best_acc=acc_test[-1]
print time.ctime()
print 'epoch:',epoch
print 'train loss:', loss_train[-1],'acc:',acc_train[-1]
print 'test loss:', loss_test[-1], 'acc:', acc_test[-1]
print 'best test acc:', best_acc
print '\n'
np.save(os.path.join(result_dir, result_dir+'_loss_train.npy'), np.array(loss_train))
np.save(os.path.join(result_dir, result_dir+'_acc_train.npy'), np.array(acc_train))
np.save(os.path.join(result_dir, result_dir+'_loss_test.npy'), np.array(loss_test))
np.save(os.path.join(result_dir, result_dir+'_acc_test.npy'), np.array(acc_test))