forked from Leo-Chu/Deep-learning-for-LDPC-decoding
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Quantized_ML_Decoding_Optimization.py
173 lines (156 loc) · 10.2 KB
/
Quantized_ML_Decoding_Optimization.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
# -*- coding: utf-8 -*-
"""
Created on Mon Jun 24 21:28:08 2019
@author: user
"""
import numpy as np
import tensorflow as tf
import datetime
import DataIO
import os
import Quantized_ML_Decoder
class BP_Training:
def __init__(self, train_config_in, top_config_in, code_in):
#config
self.train_config = train_config_in
self.top_config = top_config_in
self.code = code_in
#para_alpha
self.alpha_name = {}
self.alpha = {}
self.best_alpha = {}
self.assign_best_alpha = {}
#para_beta
self.beta_name = {}
self.beta = {}
self.best_beta = {}
self.assign_best_beta = {}
def build_network(self, bp_decoder, built_for_training=False):
#x_in
x_in = tf.placeholder(tf.float32, [None, self.train_config.training_minibatch_size])
xe_0 = tf.placeholder(tf.float32, [self.train_config.feature_length, self.train_config.training_minibatch_size])
#para_name
self.alpha_name = format("alpha")
self.beta_name = format("beta")
#alpha&beta
if built_for_training:
self.alpha= tf.get_variable(name=self.alpha_name, shape=[1], dtype=tf.float32, initializer=tf.contrib.layers.xavier_initializer())
self.best_alpha = tf.Variable(tf.ones(1, tf.float32), dtype=tf.float32)
self.assign_best_alpha = self.best_alpha.assign(self.alpha)
self.beta = tf.get_variable(name=self.beta_name, shape=[1], dtype=tf.float32, initializer=tf.contrib.layers.xavier_initializer())
self.best_beta = tf.Variable(tf.ones(1, tf.float32), dtype=tf.float32)
self.assign_best_beta = self.best_beta.assign(self.beta)
else:
# self.alpha = tf.round(tf.Variable(5*tf.ones([1]), dtype=tf.float32, name=self.alpha_name)*10)/10
self.alpha = tf.Variable(1*tf.ones([1]), dtype=tf.float32, name=self.alpha_name)
self.best_alpha = tf.Variable(tf.ones(1, tf.float32), dtype=tf.float32)
self.assign_best_alpha = self.best_alpha.assign(self.alpha)
self.beta = tf.Variable(1.0*tf.ones([1]), dtype=tf.float32, name=self.beta_name)
# self.beta = tf.Variable(0*tf.ones([1]), dtype=tf.float32, name=self.beta_name, trainable=False)
self.best_beta = tf.Variable(tf.ones(1, tf.float32), dtype=tf.float32)
self.assign_best_beta = self.best_beta.assign(self.beta)
#NN
y_out = bp_decoder.one_nn_iteration(x_in, xe_0, self.alpha, self.beta)
return x_in, xe_0, y_out
def softsign(self, x_in):
y_out = x_in/(tf.abs(x_in) + 0.01)
return y_out
def save_network_temporarily(self, sess_in):
sess_in.run(self.assign_best_alpha)
sess_in.run(self.assign_best_beta)
def test_network_online(self, dataio, decoder_test, iteration, x_in, xe_0, y_label, orig_loss, loss_after_training, calc_org_loss, sess_in):
# this function is used to test the network loss online when training network
remain_samples = self.train_config.test_sample_num
load_batch_size = self.train_config.test_minibatch_size
ave_orig_loss = 0.0
ave_loss_after_train = 0.0
while remain_samples > 0:
if remain_samples < self.train_config.test_minibatch_size:
load_batch_size = remain_samples
batch_xs, batch_ys = dataio.load_batch_for_test(load_batch_size) # features, labels
bp_out_xs, bp_out_xe0 = decoder_test.quantized_decode_before_nn(batch_xs, iteration, self.train_config.alpha, self.train_config.beta)
if calc_org_loss:
orig_loss_value, loss_after_training_value = sess_in.run([orig_loss, loss_after_training], feed_dict={x_in: bp_out_xs, xe_0: bp_out_xe0, y_label: batch_ys})
ave_orig_loss += orig_loss_value * load_batch_size
else:
loss_after_training_value = sess_in.run(loss_after_training, feed_dict={x_in: bp_out_xs, xe_0: bp_out_xe0, y_label: batch_ys})
remain_samples -= load_batch_size
ave_loss_after_train += loss_after_training_value * load_batch_size
if calc_org_loss:
ave_orig_loss /= np.double(self.train_config.test_sample_num)
ave_loss_after_train /= np.double(self.train_config.test_sample_num)
if calc_org_loss:
print("Orig loss: %f, Test loss: %f" % (ave_orig_loss, ave_loss_after_train))
return ave_orig_loss, ave_loss_after_train
def cal_training_loss(self, y_out, y_label):
y_out = tf.to_double(y_out)
y_label = tf.to_double(y_label)
y_out1 = tf.div(1-self.softsign(tf.matmul(self.top_config.D,y_out)), 2)
training_loss = tf.reduce_mean(tf.square(y_out1-tf.transpose(y_label)))
return training_loss
def train_network(self, model_id, test_iter, test_SNR):
start = datetime.datetime.now()
bp_decoder = Quantized_ML_Decoder.BP_NetDecoder(self.code.H_matrix, self.train_config.training_minibatch_size)
x_in, xe_0, y_out = self.build_network(bp_decoder, False)
y_label = tf.placeholder(tf.float32, [self.train_config.training_minibatch_size, self.train_config.label_length])
training_loss = self.cal_training_loss(y_out, y_label)
test_loss = training_loss
orig_loss_for_test = self.cal_training_loss(xe_0, y_label)
# SGD_Adam
train_step = tf.train.AdamOptimizer().minimize(training_loss)
# init operation
init = tf.global_variables_initializer()
# create a session
sess = tf.Session()
for SNR in test_SNR:
print("Training for SNR = %.1f" % SNR)
training_feature_file = format('%s%s/feature_%d_%d_%d_%.1f.dat' % (self.train_config.training_data_folder, self.top_config.channel, self.train_config.feature_length, self.train_config.label_length, self.train_config.training_minibatch_size,SNR))
training_label_file = format('%s%s/label_%d_%d_%d_%.1f.dat' % (self.train_config.training_data_folder, self.top_config.channel, self.train_config.feature_length, self.train_config.label_length, self.train_config.training_minibatch_size,SNR))
test_feature_file = format('%s%s/feature_%d_%d_%d_%.1f.dat' % (self.train_config.test_data_folder, self.top_config.channel, self.train_config.feature_length, self.train_config.label_length, self.train_config.test_minibatch_size,SNR))
test_label_file = format('%s%s/label_%d_%d_%d_%.1f.dat' % (self.train_config.test_data_folder, self.top_config.channel, self.train_config.feature_length, self.train_config.label_length, self.train_config.test_minibatch_size,SNR))
dataio_train = DataIO.TrainingDataIO(training_feature_file, training_label_file, self.train_config.training_sample_num, self.train_config.feature_length, self.train_config.label_length)
dataio_test = DataIO.TestDataIO(test_feature_file, test_label_file, self.train_config.test_sample_num, self.train_config.feature_length, self.train_config.label_length)
for iteration in range(0, test_iter):
start1 = datetime.datetime.now()
sess.run(init)
# calculate the loss before training and assign it to min_loss
ave_orig_loss, min_loss = self.test_network_online(dataio_test, bp_decoder, iteration, x_in, xe_0, y_label, orig_loss_for_test, test_loss, True, sess)
self.save_network_temporarily(sess)
# Train
count = 0
epoch = 0
print('Iteration\tBest alpha\tBest beta\tCurrent loss\tCurrent alpha\tCurrent beta')
alpha_set = []
beta_set = []
while epoch < self.train_config.epoch_num:
epoch += 1
batch_xs, batch_ys = dataio_train.load_next_mini_batch(self.train_config.training_minibatch_size)
llr_into_nn_net, xe0_into_nn_net = bp_decoder.quantized_decode_before_nn(batch_xs, iteration, self.train_config.alpha, self.train_config.beta)
sess.run([train_step], feed_dict={x_in: llr_into_nn_net, xe_0: xe0_into_nn_net, y_label: batch_ys})
a,b = sess.run([self.alpha, self.beta])
alpha_set.append(a)
beta_set.append(b)
if epoch % 100 == 0 or epoch == self.train_config.epoch_num:
_, ave_loss_after_train = self.test_network_online(dataio_test, bp_decoder, iteration, x_in, xe_0, y_label, orig_loss_for_test, test_loss, False, sess)
if ave_loss_after_train < min_loss:
print('%d\t\t%f\t%f\t%f\t%f\t%f' % (epoch, sess.run(self.best_alpha), sess.run(self.best_beta), ave_loss_after_train, sess.run(self.alpha), sess.run(self.beta)))
min_loss = ave_loss_after_train
self.save_network_temporarily(sess)
count = 0
else:
print('%d\t\t%f\t%f\t%f\t%f\t%f' % (epoch, sess.run(self.best_alpha), sess.run(self.best_beta), ave_loss_after_train, sess.run(self.alpha), sess.run(self.beta)))
count += 1
if count >= 8: # no patience
break
best_alpha = sess.run(self.best_alpha)
best_beta = sess.run(self.best_beta)
self.train_config.alpha[iteration] = best_alpha
self.train_config.beta[iteration] = best_beta
para_file = format('%sSNR%.1f_Iter%d.txt' % (self.top_config.results_folder, SNR, iteration+1))
np.savetxt(para_file,np.vstack((self.train_config.alpha, self.train_config.beta)))
end1 = datetime.datetime.now()
print('Used time for %dth training: %ds'% (iteration+1, (end1-start1).seconds))
print('\n')
sess.close()
end = datetime.datetime.now()
print('Used time for training: %ds'% (end-start).seconds)