-
Notifications
You must be signed in to change notification settings - Fork 45
/
model.py
187 lines (153 loc) · 10.5 KB
/
model.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
# -*- coding:utf-8 -*-
'''
the idea of this script came from LUNA2016 champion paper.
This model conmposed of three network,namely Archi-1(size of 10x10x6),Archi-2(size of 30x30x10),Archi-3(size of 40x40x26)
'''
import tensorflow as tf
from data_prepare import get_train_batch,get_all_filename,get_test_batch
import random
import time
#import tensorflow.python.debug as tf_debug
class model(object):
def __init__(self,learning_rate,keep_prob,batch_size,epoch):
print(" network begin...")
self.learning_rate = learning_rate
self.keep_prob = keep_prob
self.batch_size = batch_size
self.epoch = epoch
self.cubic_shape = [[6, 20, 20], [10, 30, 30], [26, 40, 40]]
def archi_1(self,input,keep_prob):
with tf.name_scope("Archi-1"):
# input size is batch_sizex20x20x6
# 5x5x3 is the kernel size of conv1,1 is the input depth,64 is the number output channel
w_conv1 = tf.Variable(tf.random_normal([3,5,5,1,64],stddev=0.001),dtype=tf.float32,name='w_conv1')
b_conv1 = tf.Variable(tf.constant(0.01,shape=[64]),dtype=tf.float32,name='b_conv1')
out_conv1 = tf.nn.relu(tf.add(tf.nn.conv3d(input,w_conv1,strides=[1,1,1,1,1],padding='VALID'),b_conv1))
out_conv1 = tf.nn.dropout(out_conv1,keep_prob)
# max pooling ,pooling layer has no effect on the data size
hidden_conv1 = tf.nn.max_pool3d(out_conv1,strides=[1,1,1,1,1],ksize=[1,1,1,1,1],padding='SAME')
# after conv1 ,the output size is batch_size x x20x20x64 ([batch_size,in_deep,width,height,output_deep])
w_conv2 = tf.Variable(tf.random_normal([3,5, 5, 64,64], stddev=0.001), dtype=tf.float32,name='w_conv2')
b_conv2 = tf.Variable(tf.constant(0.01, shape=[64]), dtype=tf.float32, name='b_conv2')
out_conv2 = tf.nn.relu(tf.add(tf.nn.conv3d(hidden_conv1, w_conv2, strides=[1, 1, 1,1, 1], padding='VALID'), b_conv2))
out_conv2 = tf.nn.dropout(out_conv2, keep_prob)
# after conv2 ,the output size is batch_sizex2x12x12x64([batch_size,in_deep,width,height,output_deep])
w_conv3 = tf.Variable(tf.random_normal([1,5, 5, 64,64], stddev=0.001), dtype=tf.float32,
name='w_conv3')
b_conv3 = tf.Variable(tf.constant(0.01, shape=[64]), dtype=tf.float32, name='b_conv3')
out_conv3 = tf.nn.relu(
tf.add(tf.nn.conv3d(out_conv2, w_conv3, strides=[1, 1, 1, 1,1], padding='VALID'),b_conv3))
out_conv3 = tf.nn.dropout(out_conv3, keep_prob)
out_conv3_shape = tf.shape(out_conv3)
tf.summary.scalar('out_conv3_shape', out_conv3_shape[0])
# after conv2 ,the output size is batch_sizex2x8x8x64([batch_size,in_deep,width,height,output_deep])
# all feature map flatten to one dimension vector,this vector will be much long
out_conv3 = tf.reshape(out_conv3,[-1,64*8*8*2])
w_fc1 = tf.Variable(tf.random_normal([64*8*8*2,150],stddev=0.001),name='w_fc1')
out_fc1 = tf.nn.relu(tf.add(tf.matmul(out_conv3,w_fc1),tf.constant(0.001,shape=[150])))
out_fc1 = tf.nn.dropout(out_fc1,keep_prob)
out_fc1_shape = tf.shape(out_fc1)
tf.summary.scalar('out_fc1_shape', out_fc1_shape[0])
w_fc2 = tf.Variable(tf.random_normal([150, 2], stddev=0.001), name='w_fc2')
out_fc2 = tf.nn.relu(tf.add(tf.matmul(out_fc1, w_fc2), tf.constant(0.001, shape=[2])))
out_fc2 = tf.nn.dropout(out_fc2, keep_prob)
w_sm = tf.Variable(tf.random_normal([2, 2], stddev=0.001), name='w_sm')
b_sm = tf.constant(0.001, shape=[2])
out_sm = tf.add(tf.matmul(out_fc2, w_sm), b_sm)
return out_sm
def archi_2(self, input, keep_prob):
with tf.name_scope("Archi-2"):
# input size is batch_sizex20x20x6
# 5x5x3 is the kernel size of conv1,1 is the input depth,64 is the number output channel
w_conv1 = tf.Variable(tf.random_normal([3, 5, 5, 1, 64], stddev=0.001), dtype=tf.float32, name='w_conv1')
b_conv1 = tf.Variable(tf.constant(0.01, shape=[64]), dtype=tf.float32, name='b_conv1')
out_conv1 = tf.nn.relu(
tf.add(tf.nn.conv3d(input, w_conv1, strides=[1, 1, 1, 1, 1], padding='VALID'), b_conv1))
out_conv1 = tf.nn.dropout(out_conv1, keep_prob)
# max pooling ,pooling layer has no effect on the data size
hidden_conv1 = tf.nn.max_pool3d(out_conv1, strides=[1, 2, 2, 1, 1], ksize=[1, 1, 1, 1, 1], padding='SAME')
# after conv1 ,the output size is batch_sizex4x16x16x64([batch_size,in_deep,width,height,output_deep])
w_conv2 = tf.Variable(tf.random_normal([3, 5, 5, 64, 64], stddev=0.001), dtype=tf.float32, name='w_conv2')
b_conv2 = tf.Variable(tf.constant(0.01, shape=[64]), dtype=tf.float32, name='b_conv2')
out_conv2 = tf.nn.relu(
tf.add(tf.nn.conv3d(hidden_conv1, w_conv2, strides=[1, 1, 1, 1, 1], padding='VALID'), b_conv2))
out_conv2 = tf.nn.dropout(out_conv2, keep_prob)
# after conv2 ,the output size is batch_sizex2x12x12x64([batch_size,in_deep,width,height,output_deep])
w_conv3 = tf.Variable(tf.random_normal([3, 5, 5, 64, 64], stddev=0.001), dtype=tf.float32,
name='w_conv3')
b_conv3 = tf.Variable(tf.constant(0.01, shape=[64]), dtype=tf.float32, name='b_conv3')
out_conv3 = tf.nn.relu(
tf.add(tf.nn.conv3d(out_conv2, w_conv3, strides=[1, 1, 1, 1, 1], padding='VALID'), b_conv3))
out_conv3 = tf.nn.dropout(out_conv3, keep_prob)
out_conv3_shape = tf.shape(out_conv3)
tf.summary.scalar('out_conv3_shape', out_conv3_shape[0])
# after conv2 ,the output size is batch_sizex2x8x8x64([batch_size,in_deep,width,height,output_deep])
# all feature map flatten to one dimension vector,this vector will be much long
out_conv3 = tf.reshape(out_conv3, [-1, 64 * 8 * 8 * 2])
w_fc1 = tf.Variable(tf.random_normal([64 * 8 * 8 * 2, 250], stddev=0.001), name='w_fc1')
out_fc1 = tf.nn.relu(tf.add(tf.matmul(out_conv3, w_fc1), tf.constant(0.001, shape=[250])))
out_fc1 = tf.nn.dropout(out_fc1, keep_prob)
out_fc1_shape = tf.shape(out_fc1)
tf.summary.scalar('out_fc1_shape', out_fc1_shape[0])
w_fc2 = tf.Variable(tf.random_normal([150, 2], stddev=0.001), name='w_fc2')
out_fc2 = tf.nn.relu(tf.add(tf.matmul(out_fc1, w_fc2), tf.constant(0.001, shape=[2])))
out_fc2 = tf.nn.dropout(out_fc2, keep_prob)
w_sm = tf.Variable(tf.random_normal([2, 2], stddev=0.001), name='w_sm')
b_sm = tf.constant(0.001, shape=[2])
out_sm = tf.add(tf.matmul(out_fc2, w_sm), b_sm)
return out_sm
def inference(self,npy_path,test_path,model_index,train_flag=True):
# some statistic index
highest_acc = 0.0
highest_iterator = 1
all_filenames = get_all_filename(npy_path,self.cubic_shape[model_index][1])
print("file size is :\t",len(all_filenames))
# how many time should one epoch should loop to feed all data
times = int(len(all_filenames) / self.batch_size)
if (len(all_filenames) % self.batch_size) != 0:
times = times + 1
# keep_prob used for dropout
keep_prob = tf.placeholder(tf.float32)
# take placeholder as input
x = tf.placeholder(tf.float32, [None, self.cubic_shape[model_index][0], self.cubic_shape[model_index][1], self.cubic_shape[model_index][2]])
x_image = tf.reshape(x, [-1, self.cubic_shape[model_index][0], self.cubic_shape[model_index][1], self.cubic_shape[model_index][2], 1])
net_out = self.archi_1(x_image,keep_prob)
saver = tf.train.Saver() # default to save all variable,save mode or restore from path
if train_flag:
# softmax layer
real_label = tf.placeholder(tf.float32, [None, 2])
cross_entropy = tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(logits=net_out, labels=real_label))
#cross_entropy = -tf.reduce_sum(real_label * tf.log(net_out))
net_loss = tf.reduce_mean(cross_entropy)
train_step = tf.train.MomentumOptimizer(self.learning_rate, 0.9).minimize(net_loss)
correct_prediction = tf.equal(tf.argmax(net_out, 1), tf.argmax(real_label, 1))
accruacy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
merged = tf.summary.merge_all()
with tf.Session() as sess:
#sess = tf_debug.LocalCLIDebugWrapperSession(sess)
#sess.add_tensor_filter("has_inf_or_nan", tf_debug.has_inf_or_nan)
sess.run(tf.global_variables_initializer())
train_writer = tf.summary.FileWriter('./tensorboard/', sess.graph)
# loop epoches
for i in range(self.epoch):
epoch_start =time.time()
# the data will be shuffled by every epoch
random.shuffle(all_filenames)
for t in range(times):
batch_files = all_filenames[t*self.batch_size:(t+1)*self.batch_size]
batch_data, batch_label = get_train_batch(batch_files)
feed_dict = {x: batch_data, real_label: batch_label,
keep_prob: self.keep_prob}
_,summary = sess.run([train_step, merged],feed_dict =feed_dict)
train_writer.add_summary(summary, i)
saver.save(sess, './ckpt/archi-1', global_step=i + 1)
epoch_end = time.time()
test_batch,test_label = get_test_batch(test_path)
print("test batch data:\t",test_batch)
print("test batch label:\t", test_label)
test_dict = {x: test_batch, real_label: test_label, keep_prob:self.keep_prob}
acc_test,loss = sess.run([accruacy,net_loss],feed_dict=test_dict)
print('accuracy is %f' % acc_test)
print("loss is ", loss)
print(" epoch %d time consumed %f seconds"%(i,(epoch_end-epoch_start)))
print("training finshed..highest accuracy is %f,the iterator is %d " % (highest_acc, highest_iterator))