-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_network_synthetic.py
176 lines (137 loc) · 7.43 KB
/
train_network_synthetic.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
import json
import shutil
import tensorflow as tf
import numpy as np
import network_ttn_synthetic as network
import os.path
import scipy.io
from configuration import configure
from configuration.general import placeholder_inputs, load_data
from configuration import general
from test_network_synthetic import test_ttn
def train_ttn(dict_cus):
datadir = os.path.join(configure.parameters_dict["database_dir"], 'ende',
dict_cus["dataset_name"],
str(dict_cus["distance_int"]),
'npy')
cutdatadir = os.path.join(datadir, str(configure.parameters_dict["ksplit"]))
val_acc_dict = {"0": {}, "1": {}, "2": {}}
for k in range(configure.parameters_dict["ksplit"]):
# load data
x_train, y_train, x_test, y_test, y_true, nb_classes, y_true_train, enc = load_data(
dict_cus["dataset_name"],
cutdatadir,
dict_cus["seq_len"],
k)
train_data = x_train
train_label = y_train # 需要为 onehot 编码格式
test_data = x_test
test_label = y_test # 需要为 onehot 编码格式
num_classes = nb_classes
configure.parameters_dict["num_classes"] = num_classes
num_train = configure.parameters_dict["num_train"] = len(train_data)
configure.parameters_dict = general.Merge(configure.parameters_dict, dict_cus) # TODO: 不确定对不对
batch_size = configure.parameters_dict["batch_size"]
learning_rate_1 = configure.parameters_dict["learning_rate_1"]
numBatches = int(num_train / batch_size)
maxIters = configure.parameters_dict["maxIters"]
num_classes = configure.parameters_dict["num_classes"] # TODO: 类别数量, 根据需要设置
seq_len = configure.parameters_dict["seq_len"]
with tf.Graph().as_default():
x_placeholder, y_placeholder, learning_rate_placeholder = placeholder_inputs(batch_size, num_classes, configure.parameters_dict["seq_len"])
output, sequence_unwarped, gamma, sequence1 = network.mapping(x_placeholder, batch_size, configure.parameters_dict["seq_len"])
loss = network.loss(output, y_placeholder)
var_list_ttn = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='ttn')
var_list_classifier = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, scope='classifier')
train_op_classifier = network.training(loss, learning_rate_placeholder, var_list_classifier)
train_op_ttn = network.training(loss, learning_rate_placeholder / 10.0, var_list_ttn)
train_op = tf.group(train_op_classifier, train_op_ttn)
# train_op = network.training(loss, learning_rate_placeholder)
init = tf.initialize_all_variables()
saver = tf.train.Saver(max_to_keep=500)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
sess.run(init)
batchIdx = 0
c_loss = 9999 # 初始化 loss
train_loss = []
for step in range(maxIters * numBatches + 1):
batchIdx = batchIdx % numBatches
if batchIdx == 0:
randIdx = np.random.permutation(train_data.shape[0])
trainIdx = randIdx[int(batchIdx * batch_size):int((batchIdx + 1) * batch_size)]
if len(trainIdx) != batch_size:
print("trainIdx: {}".format(len(trainIdx)))
continue
_, loss_value = sess.run([train_op, loss],
feed_dict={x_placeholder: train_data[trainIdx, :],
y_placeholder: train_label[trainIdx, :],
learning_rate_placeholder: learning_rate_1})
# print("y_placeholder: {}".format(sess.run(y_placeholder)))
train_loss.append(loss_value)
if step % 1 == 0: # #TODO：1000
# print('---------------------------------')
# print(step)
# print(loss_value)
print("k: {}\tstep: {}/{}, loss_value: {}".format(k, step, maxIters * numBatches, loss_value))
batchIdx = batchIdx + 1
if step in np.linspace(0, maxIters * numBatches, 21):
# weights_dir = os.path.join("weights")
weights_dir = general.getWeightsDir(configure.parameters_dict, k)
general.create_folder(weights_dir, remake=True)
weights_name = os.path.join(weights_dir, "%s_%s_gaussians_github_ttn_%s" % (str(k), str(step), str(loss_value)))
saver.save(sess, weights_name)
print(general.colorstr("model saved at: {}".format(weights_name)))
val_acc = test_ttn(dict_cus, k, weights_name="%s_%s_gaussians_github_ttn_%s" % (
str(k), str(step), str(loss_value)))
val_acc_dict[str(k)].update(
{
"%s" % (str(step)): val_acc
}
)
if loss_value <= c_loss:
weights_name = os.path.join(weights_dir, "%s_best_model" % (str(k)))
saver.save(sess, weights_name)
print(general.colorstr("model saved at: {}".format(weights_name)))
val_acc = test_ttn(dict_cus, k, weights_name="%s_best_model" % (str(k)))
val_acc_dict[str(k)].update(
{
"best": val_acc
}
)
weights_name = os.path.join(weights_dir, "%s_last_model" % (str(k)))
saver.save(sess, weights_name)
print(general.colorstr("model saved at: {}".format(weights_name)))
val_acc = test_ttn(dict_cus, k, weights_name="%s_last_model" % (str(k)))
val_acc_dict[str(k)].update(
{
"last": val_acc
}
)
val_acc_json_name = os.path.join(weights_dir, "%s_val_acc.json" % (str(k)))
with open(val_acc_json_name, "w", encoding="utf-8") as fw:
json.dump(val_acc_dict, fw)
print(general.colorstr("val_acc_json saved at: {}".format(val_acc_json_name)))
log_name = os.path.join(weights_dir, "%s_train_loss.txt" % (str(k)))
np.savetxt(log_name, np.array(train_loss, dtype=np.float), delimiter=', ')
print(general.colorstr("log saved at: {}".format(log_name)))
# print(val_acc_dict)
print("\n\nk:\tbest\tlast\n0:\t{}\t{}\n1:\t{}\t{}\n2:\t{}\t{}\n".format(
val_acc_dict["0"]["best"], val_acc_dict["0"]["last"],
val_acc_dict["1"]["best"], val_acc_dict["1"]["last"],
val_acc_dict["2"]["best"], val_acc_dict["2"]["last"]))
if __name__ == '__main__':
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
# 默认为0：输出所有log信息
# 设置为1：进一步屏蔽INFO信息
# 设置为2：进一步屏蔽WARNING信息
# 设置为3：进一步屏蔽ERROR信息
dict_cus = {
"batch_size": 1,
"maxIters": 2000, # 100000
"seq_len": 2000,
"distance_int": 999,
"dataset_name": "kyoto7",
}
train_ttn(dict_cus)