-
Notifications
You must be signed in to change notification settings - Fork 38
/
DenoisingAE.py
187 lines (164 loc) · 8.17 KB
/
DenoisingAE.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
import tensorflow as tf
import numpy as np
import math
import matplotlib.pyplot as plt
def corrupt(x):
r = tf.add(x, tf.cast(tf.random_uniform(shape=tf.shape(x),minval=0,maxval=0.1,dtype=tf.float32), tf.float32))
# r = tf.multiply(x,tf.cast(tf.random_uniform(shape=tf.shape(x), minval=0.5, maxval=1.5, dtype=tf.float32), tf.float32))
return r
def kl_divergence(p, p_hat):
# return tf.reduce_mean(p * tf.log(tf.abs(p)) - p * tf.log(tf.abs(p_hat)) + (1 - p) * tf.log(tf.abs(1 - p)) - (1 - p) * tf.log(tf.abs(1 - p_hat)))
return tf.reduce_mean(p * tf.log(tf.abs(p)) - p * tf.log(tf.abs(p_hat)) + (1 - p) * tf.log(1 - p) - (1 - p) * tf.log(1 - p_hat))
def autoencoder(dimensions=[784, 512, 256, 64]):
x = tf.placeholder(tf.float32, [None, dimensions[0]], name='x')
corrupt_prob = tf.placeholder(tf.float32, [1])
current_input = corrupt(x) * corrupt_prob + x * (1 - corrupt_prob)
noise_input = current_input
# Build the encoder
print("========= encoder begin ==========")
encoder = []
for layer_i, n_output in enumerate(dimensions[1:]):
n_input = int(current_input.get_shape()[1])
print("encoder : layer_i - n_output - n_input",layer_i,n_output,n_input)
W = tf.Variable(tf.random_uniform([n_input, n_output],-1.0 / math.sqrt(n_input),1.0 / math.sqrt(n_input)))
b = tf.Variable(tf.zeros([n_output]))
encoder.append(W)
output = tf.nn.tanh(tf.matmul(current_input, W) + b)
current_input = output
print("========= encoder finish =========")
# latent representation
encoder_out = current_input
print(encoder_out.shape)
encoder.reverse()
# Build the decoder using the same weights
print("========= decoder begin ==========")
for layer_i, n_output in enumerate(dimensions[:-1][::-1]):
print("decoder : layer_i - n_output", layer_i, n_output)
W = tf.transpose(encoder[layer_i]) # transpose of the weights
b = tf.Variable(tf.zeros([n_output]))
output = tf.nn.tanh(tf.matmul(current_input, W) + b)
current_input = output
print("========= decoder finish =========")
# now have the reconstruction through the network
reconstruction = current_input
# kl = tf.reduce_mean(-tf.nn.softmax_cross_entropy_with_logits(logits=z, labels=z/0.01))
p_hat = tf.reduce_mean(encoder_out,0)
p = np.repeat([-0.05], 200).astype(np.float32)
dummy = np.repeat([1], 200).astype(np.float32)
kl = kl_divergence(p_hat,p)
cost = tf.reduce_mean(tf.square(reconstruction - x)) + 0.01*kl
# cost = 0.5 * tf.reduce_sum(tf.square(y - x))
return {
'x': x,
'encoder_out': encoder_out,
'reconstruction': reconstruction,
'corrupt_prob': corrupt_prob,
'cost': cost,
'noise_input' : noise_input,
'kl' : kl
}
def train_DOA():
from get_csv_data import HandleData
import csv
################ TEST DATA ################
data = HandleData(total_data=880, data_per_angle=110, num_angles=8)
antenna_data, label_data = data.get_synthatic_data(test_data=False)
antenna_data_mean = np.mean(antenna_data, axis=0)
###########################################
################ learning parameters ######
learning_rate = 0.001
batch_size = 20
n_epochs = 1000
###########################################
################ AutoEncoder ##############
ae = autoencoder(dimensions=[4, 200])
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(ae['cost'])
###########################################
################ Training #################
sess = tf.Session()
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver()
########### restore ###########
# saver_restore = tf.train.import_meta_graph('./DAE_save/DenoisingAE_save_noise_add.meta')
# saver_restore = tf.train.import_meta_graph('DenoisingAE_save_noise_multiply.meta')
# saver_restore.restore(sess, tf.train.latest_checkpoint('./DAE_save/'))
###############################
train=0
for epoch_i in range(n_epochs):
for batch_i in range(data.total_data//batch_size):
batch_xs, _ = data.next_batch(batch_size)
train = np.array([img - antenna_data_mean for img in batch_xs])
# print(train.shape)
sess.run(optimizer, feed_dict={ae['x']: train, ae['corrupt_prob']: [1.0]})
print(epoch_i,sess.run([ae['cost'],ae['kl']], feed_dict={ae['x']: train, ae['corrupt_prob']: [1.0]}))
##### debug kl ######
# tmp=sess.run(ae['encoder_out'], feed_dict={ae['x']: train, ae['corrupt_prob']: [1.0]})
# p_hat = tf.reduce_mean(tmp, 0)
# p = np.repeat([-0.05], 200).astype(np.float32)
# dummy = np.repeat([1], 200).astype(np.float32)
# p_hat = p_hat+dummy
# p = p+dummy
# kl_tmp = p * tf.log(tf.abs(p)) - p * tf.log(tf.abs(p_hat)) + (1 - p) * tf.log(p-1) - (1 - p) * tf.log(p_hat-1)
# print(sess.run( p_hat ))
# ######################
###########################################
saver.save(sess, './DAE_save/DenoisingAE_save_noise_add')
############### Test Data ################
data_test = HandleData(total_data=80, data_per_angle=10, num_angles=8)
antenna_data_test, label_data_test = data_test.get_synthatic_data(test_data=True)
antenna_data_test_mean = np.mean(antenna_data_test, axis=0)
###########################################
################ Testing trained data #####
test_xs_norm = np.array([img - antenna_data_test_mean for img in antenna_data])
a,b,output_y = sess.run([ae['cost'],ae['noise_input'],ae['reconstruction']], feed_dict={ae['x']: test_xs_norm, ae['corrupt_prob']: [1.0]})
print("Testing trained data avarage cost : ", a)
###########################################
################ Testing ##################
test_xs, _ = data_test.next_batch(80)
test_xs_norm = np.array([img - antenna_data_test_mean for img in test_xs])
a,b,output_y = sess.run([ae['cost'],ae['noise_input'],ae['reconstruction']], feed_dict={ae['x']: test_xs_norm, ae['corrupt_prob']: [1.0]})
print("avarage cost : ", a)
for i in range(len(output_y)):
comp = output_y[i]
orgi = test_xs[i]
noise = b[i]
comp += antenna_data_test_mean
noise += antenna_data_test_mean
plt.subplot(8, 10, i + 1)
plt.plot(comp,color='blue',label='rcon')
plt.plot(orgi,color='green',label='orgi')
plt.plot(noise,color='red',label='noise')
plt.xticks(())
plt.yticks(())
plt.subplots_adjust(0.08, 0.02, 0.92, 0.85, 0.08, 0.23)
plt.legend(loc='upper left')
plt.show()
print("difference between noise and origial :")
# print(b-test_xs_norm)
#############################################
################ Test Data ################
data_test_noise = HandleData(total_data=120, data_per_angle=120, num_angles=8)
antenna_data_test, label_data_test = data_test_noise.get_synthatic_data(test_data=-1)
antenna_data_test_mean = np.mean(antenna_data_test, axis=0)
###########################################
################ Testing ##################
test_xs, _ = data_test_noise.next_batch(120)
test_xs_norm = np.array([img - antenna_data_test_mean for img in test_xs])
a,b,output_y = sess.run([ae['cost'],ae['noise_input'],ae['reconstruction']], feed_dict={ae['x']: test_xs_norm, ae['corrupt_prob']: [1.0]})
print("avarage cost : ", a)
for i in range(len(output_y)):
comp = output_y[i]
orgi = test_xs[i]
noise = b[i]
comp += antenna_data_test_mean
noise += antenna_data_test_mean
plt.subplot(10, 12, i + 1)
plt.plot(comp,color='blue',label='rcon')
plt.plot(orgi,color='green',label='orgi')
plt.plot(noise,color='red',label='noise')
plt.xticks(())
plt.yticks(())
plt.subplots_adjust(0.08, 0.02, 0.92, 0.85, 0.08, 0.23)
plt.show()
#############################################
train_DOA()