-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathmodels_2.py
More file actions
191 lines (170 loc) · 8.52 KB
/
Copy pathmodels_2.py
File metadata and controls
191 lines (170 loc) · 8.52 KB
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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @Time : 2020/11/11 18:58
# @Author : Chenchen Wei
# @Description:
from .layers import *
from .load_data import *
from tqdm import tqdm
class GA_GAN_concat(object):
"""
原始路网，重建路网经SAGE聚合后融合特征
"""
def __init__(self,
sess,
all_data,
adj,
ori_adj,
sgae_hidden_list,
ge_hidden_lists,
dis_hidden_lists,
reconstruction_coefficient=100,
epoch=1,
batch_size=2,
drop_rate=0,
learning_rate=1e-3,
save_model=False,
save_model_path=None):
self.all_data = all_data
self.adj = adj
self.epoch = epoch
self.batch_size = batch_size
self.sess = sess
self.drop_rate = drop_rate
self.sgae_hidden_list = sgae_hidden_list
self.ge_hidden_lists = ge_hidden_lists
self.dis_hidden_lists = dis_hidden_lists
self.learning_rate = learning_rate
self.reconstruction_coefficient = reconstruction_coefficient
self.save_model = save_model
self.ori_adj = ori_adj
self.save_model_path = save_model_path + '/model'
def _call(self):
self.train_x, self.train_y, self.test_x, self.test_y, = self.all_data[:4]
self.train_mask, self.test_mask, self.scaler = self.all_data[4:]
self.batch_nums = int(self.train_x.shape[0] / self.batch_size + 1)
self.tr_bat_nums = self.batch_size - self.train_x.shape[0] % self.batch_size
# Make up the number of 0 for insufficient batch
self.in_dim, self.out_dim = self.train_x.shape[-1], self.train_y.shape[-1]
self.node_num = self.train_x.shape[1]
def GraphSAGE(self, inputs, reuse=False):
with tf.variable_scope('sage', reuse=reuse):
concat1 = GraphSAGE_Mean('sage1',
input_dim=self.train_x.shape[-1],
output_dim=self.sgae_hidden_list[0],
adj=self.adj)(inputs)
concat2 = GraphSAGE_Mean('sage2',
input_dim=self.sgae_hidden_list[0] * 2,
output_dim=self.sgae_hidden_list[0],
adj=self.adj)(concat1)
return concat2
def GraphSAGE_ori(self, inputs, reuse=False):
with tf.variable_scope('sage_ori', reuse=reuse):
concat1 = GraphSAGE_Mean('sage1_ori',
input_dim=self.train_x.shape[-1],
output_dim=self.sgae_hidden_list[0],
adj=self.ori_adj)(inputs)
concat2 = GraphSAGE_Mean('sage2_ori',
input_dim=self.sgae_hidden_list[0] * 2,
output_dim=self.sgae_hidden_list[0],
adj=self.ori_adj)(concat1)
return concat2
def generate(self, inputs, reuse=False):
with tf.variable_scope('generate', reuse=reuse):
h1 = Linear('ge_h1',
input_dim=inputs.shape[-1],
output_dim=self.ge_hidden_lists[0])(inputs)
h2 = Linear('ge_h2',
input_dim=self.ge_hidden_lists[0],
output_dim=self.ge_hidden_lists[1])(h1)
out = Linear('ge_out',
input_dim=self.ge_hidden_lists[1],
output_dim=self.train_x.shape[-1],
act=tf.nn.sigmoid)(h2)
return out
def discriminator(self, inputs, reuse=False):
with tf.variable_scope('discriminator', reuse=reuse):
h1 = Linear('dis_h1',
input_dim=inputs.shape[-1],
output_dim=self.dis_hidden_lists[0])(inputs)
h2 = Linear('dis_h2',
input_dim=self.dis_hidden_lists[0],
output_dim=self.dis_hidden_lists[1])(h1)
h2_logit = Linear('dis_logit',
input_dim=self.dis_hidden_lists[1],
output_dim=1,
act=lambda x: x)(h2)
h2_prob = tf.nn.sigmoid(h2_logit)
return h2_prob, h2_logit
def get_batch(self, i, batch_nums, data):
if i != batch_nums - 1:
return data[i * self.batch_size:(i + 1) * self.batch_size, :]
else:
nums = self.tr_bat_nums
temp = data[i * self.batch_size:, :]
tm_zeros = np.zeros(shape=(nums, self.node_num, self.out_dim))
concats = np.concatenate((temp, tm_zeros), axis=0)
return concats
def _build(self):
self._call()
self.x_real = tf.placeholder(tf.float32, [None, self.node_num, self.out_dim])
self.x = tf.placeholder(tf.float32, [None, self.node_num, self.in_dim])
self.concat_features_1 = self.GraphSAGE(self.x)
self.concat_features_2 = self.GraphSAGE_ori(self.x)
self.concat_features = tf.concat([self.concat_features_1, self.concat_features_2], axis=-1)
D_real, D_logit_real = self.discriminator(self.x_real, reuse=False)
G_samples = self.generate(self.concat_features, reuse=False)
D_fake, D_logit_fake = self.discriminator(G_samples, reuse=True)
d_loss_real = -tf.reduce_mean(D_logit_real)
d_loss_fake = tf.reduce_mean(D_logit_fake)
mse_loss = tf.reduce_mean(tf.square(G_samples - self.x_real))
self.d_loss = d_loss_real + d_loss_fake
self.g_loss = - d_loss_fake + self.reconstruction_coefficient * mse_loss
self.t_vars = tf.trainable_variables()
d_vars = [var for var in self.t_vars if 'dis' in var.name]
g_vars = [var for var in self.t_vars if 'ge' in var.name]
self.d_clip = [p.assign(tf.clip_by_value(p, -0.01, 0.01)) for p in d_vars]
self.d_optim = tf.train.RMSPropOptimizer(learning_rate=self.learning_rate
).minimize(self.d_loss, var_list=d_vars)
self.g_optim = tf.train.RMSPropOptimizer(learning_rate=self.learning_rate
).minimize(self.g_loss, var_list=g_vars)
self.generate_data = self.generate(self.concat_features, reuse=True)
def train(self):
self._build()
tf.global_variables_initializer().run()
start_time = time.time()
self.d_loss_sum, self.g_loss_sum = [], []
saver = tf.train.Saver(max_to_keep=1)
for epoch in tqdm(range(self.epoch)):
d_bat, g_bat = [], []
for i in range(self.batch_nums):
x_train = self.get_batch(i, self.batch_nums, self.train_x)
y_train = self.get_batch(i, self.batch_nums, self.train_y)
d_per = []
for _ in range(5):
_, D_loss, _ = self.sess.run([self.d_optim, self.d_loss, self.d_clip],
feed_dict={self.x_real: y_train,
self.x: x_train})
d_per.append(D_loss)
_, G_loss = self.sess.run([self.g_optim, self.g_loss],
feed_dict={self.x_real: y_train,
self.x: x_train})
d_bat.append(np.mean(np.asarray(d_per)))
g_bat.append(G_loss)
d_epoch, g_epoch = np.mean(np.asarray(d_bat)), np.mean(np.asarray(g_bat))
self.d_loss_sum.append(d_epoch)
self.g_loss_sum.append(g_epoch)
if epoch % 20 == 0:
use_time = (time.time() - start_time) / 60
print('Epoch:[{}], G_Loss:{:.4f}, D_Loss:{:.8f}, usu_time:{:.2f}min'
.format(epoch, g_epoch, d_epoch, use_time))
if self.save_model:
if not os.path.exists(self.save_model_path):
os.makedirs(self.save_model_path)
saver.save(self.sess, self.save_model_path + '/GA_GAN') # Save model
rec = self.sess.run(self.generate_data, feed_dict={self.x: self.test_x, })
rec = np.transpose(rec, (0, 2, 1)).reshape(-1, self.node_num)
rec = self.scaler.inverse_transform(rec)
true = np.transpose(self.test_y[:rec.shape[0]], (0, 2, 1)).reshape(-1, self.node_num)
true = self.scaler.inverse_transform(true)
return rec, true