-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathCNN_train.py
159 lines (136 loc) · 6.41 KB
/
CNN_train.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
import datetime
import os
import numpy as np
import tensorflow as tf
from openpyxl import load_workbook
from tensorflow import keras
import time
from ssgan_dataset_tf2 import test_Y, test_X, train_X, train_Y, valid_Y, valid_X
from CNN_model_tf2 import CNN
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")
devices = tf.config.experimental.list_physical_devices('GPU')
tf.config.experimental.set_memory_growth(devices[0], True)
def prepare_extended_label(label):
# add extra label for fake data
extended_label = tf.concat([tf.zeros([tf.shape(label)[0], 1]), label], axis=1)
return extended_label
def loss_fn(CNN_model, batch_x, batch_label, is_training):
D_real_features, D_real_logits, D_real_prob = CNN_model(batch_x, is_training)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=batch_label, logits=D_real_logits))
return loss
def accuracy(Cnn, batch_x, extended_label, is_training):
D_real_features, D_real_logits, D_real_prob = Cnn(batch_x, is_training)
prediction_value = tf.argmax(D_real_prob[:, 1:], 1)
prediction = tf.equal(tf.argmax(D_real_prob[:, 1:], 1),
tf.argmax(extended_label[:, 1:], 1))
acc = tf.reduce_mean(tf.cast(prediction, tf.float32))
return acc, prediction_value
def Draw(hist, name, epoch, show=False, save=False, is_loss=True):
plt.figure()
Time = time.strftime("%Y-%m-%d %H-%M-%S", time.localtime())
if is_loss:
plt.plot(hist, 'b', label='CNN')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()
if save:
if not os.path.exists('Loss'):
os.mkdir('Loss')
plt.savefig("Loss/loss_lr[{}]epoch[{}]time[{}].png".format(name, epoch, Time))
else:
plt.plot(hist, 'b', label='acc')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.legend()
if save:
if not os.path.exists('plot'):
os.mkdir('plot')
plt.savefig("plot/acc_lr[{}]epoch[{}]time[{}].png".format(name, epoch, Time))
if show:
plt.show()
else:
plt.close()
def main(learning_rate, epochs):
batch_size = 64
is_training = True
Train_acc = []
test_acc = []
train_hist = []
file = ['', '', ''] # 模型文件名称，删除之前保存的文件名称
CNN_model = CNN()
CNN_model.build(input_shape=(None, 32, 32, 1))
optimizer = keras.optimizers.Adam(learning_rate=learning_rate, beta_1=0.5)
CNN_model.summary()
Time = datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
path = 'D:/python/ssgan_tf2.0/log_dir'
log_dir = os.path.join(path, Time)
if not os.path.exists(log_dir):
os.mkdir(log_dir)
writer = tf.summary.create_file_writer(log_dir)
no_of_batches = int(train_X.shape[0] / batch_size) + 1
for epoch in range(epochs):
train_accuracies, train_losses = [], []
for i in range(no_of_batches - 1):
# 准备训练数据
batch_x = train_X[i * batch_size:batch_size + i * batch_size, ]
batch_label = train_Y[i * batch_size:batch_size + i * batch_size, ]
batch_reshaped = batch_x.reshape([-1, 32, 32, 1])
extended_label = prepare_extended_label(batch_label)
# 准备验证数据
valid_data = valid_X
valid_label = valid_Y
valid_data_reshaped = valid_data.reshape([-1, 32, 32, 1])
valid_extended_label = prepare_extended_label(valid_label)
# 判别器前向计算
with tf.GradientTape() as tape:
loss = loss_fn(CNN_model, batch_reshaped, extended_label, is_training)
grads = tape.gradient(loss, CNN_model.trainable_variables)
optimizer.apply_gradients(zip(grads, CNN_model.trainable_variables))
train_accuracy, _ = accuracy(CNN_model, valid_data_reshaped, valid_extended_label, None)
train_accuracies.append(train_accuracy)
print('Epoch [{}]/[{}]'.format(epoch, epochs),
'Batch evaluated [{}]/[{}]'.format(str(i + 1), no_of_batches - 1))
train_losses.append(loss)
tr_Loss = np.mean(train_losses)
tr_acc = np.mean(train_accuracies)
train_hist.append(tr_Loss)
Train_acc.append(tr_acc)
print('After epoch: ' + str(epoch + 1) + ' loss: ' + str(tr_Loss) + ' Accuracy: ' + str(tr_acc))
with writer.as_default():
tf.summary.scalar("train/tr_DL", tr_Loss, epoch)
tf.summary.scalar("train/tr_acc", tr_acc, epoch)
test_data = test_X
test_label = test_Y
test_data_reshaped = test_data.reshape([-1, 32, 32, 1])
test_extended_label = prepare_extended_label(test_label)
test_features, test_logits, test_prob = CNN_model(test_data_reshaped, False)
test_accuracy, _ = accuracy(CNN_model, test_data_reshaped, test_extended_label, False)
print('测试集：' + str(test_accuracy.numpy()))
epoch_accuracy = test_accuracy
with writer.as_default():
tf.summary.scalar("test/test_accuracy", epoch_accuracy.numpy(), epoch)
print(np.array(test_acc))
print('目前准确率最大为' + str(tf.reduce_max(test_acc)))
print(str(epoch_accuracy.numpy()), str(tf.reduce_max(test_acc).numpy()))
if epoch_accuracy.numpy() >= tf.reduce_max(test_acc).numpy():
Time = time.strftime("%Y-%m-%d %H-%M-%S", time.localtime()) # 本次 循环开始时间，放到文件命名
for i in file: # 删除上一个模型文件，保存新的模型
if os.path.exists(i):
os.remove(i)
else:
print('no such file:%s' % i)
print('*************************模型保存***************************************')
CNN_model.save_weights('CNN_model/model_time[{}]'.format(Time))
file = ['CNN_model/model_time[{}].index'.format(Time),
'CNN_model/model_time[{}].data-00000-of-00002'.format(Time),
'CNN_model/model_time[{}].data-00001-of-00002'.format(Time)]
test_acc.append(test_accuracy)
return train_hist, Train_acc
if __name__ == '__main__':
Learning_rate = 0.0005
Epochs = 100
train_loss, train_acc = main(Learning_rate, Epochs)
Draw(train_loss, Learning_rate, Epochs, show=True, save=True)
Draw(train_acc, Learning_rate, Epochs, show=True, save=True, is_loss=False)