tf.train.MonitoredTrainingSession()的使用案例

Posted 2021-02-13 ingemar

2019-03-19 22:07:22

本文主要介绍tf.train.MonitoredTrainingSession（）：在TensorFlow版本1.2.1中有12个参数，但本例中只用到了两个参数：

checkpoint_dir：用于指定检查点保存的路径和文件名

save_checkpoint_secs：用于指定保存检查点的时间间隔，以秒为单位

 

# -*- coding:utf-8 -*-

from cifar10_pro import cifar10_input
import tensorflow as tf
import numpy as np


def weight_variable(shape):
    """According to the parameter ‘shape‘ to generate weight"""
    initial = tf.truncated_normal(shape=shape, stddev=0.1)
    return tf.Variable(initial_value=initial)


def bias_variable(shape):
    """According to the parameter ‘shpae‘ to create bias"""
    initial = tf.constant(value=0.1, shape=shape)
    return tf.Variable(initial_value=initial)


def conv2d(x, W):
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding=‘SAME‘)


def max_pool_2x2(x):
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding=‘SAME‘)


def avg_pool_6x6(x):
    return tf.nn.avg_pool(x, ksize=[1, 6, 6, 1], strides=[1, 6, 6, 1], padding=‘SAME‘)


tf.reset_default_graph()

batch_size = 128
data_dir = ‘/tmp/cifar10_data/cifar-10-batches-bin‘
# 读取训练集
images_train, labels_train = cifar10_input.inputs(eval_data=False, data_dir=data_dir, batch_size=batch_size)
# 读取测试集
images_test, labels_test = cifar10_input.inputs(eval_data=True, data_dir=data_dir, batch_size=batch_size)


x = tf.placeholder(shape=[None, 24, 24, 3], dtype=tf.float32)
y = tf.placeholder(shape=[None, 10], dtype=tf.float32)

x_image = tf.reshape(x, [-1, 24, 24, 3])

# 64个卷积核，输出64个通道
W_conv1 = weight_variable([5, 5, 3, 64])
b_conv1 = bias_variable([64])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool_1 = max_pool_2x2(h_conv1)


W_conv2 = weight_variable([5, 5, 64, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool_1, W_conv2) + b_conv2)
h_pool_2 = max_pool_2x2(h_conv2)

# 降维
W_conv3 = weight_variable([5, 5, 64, 10])
b_conv3 = bias_variable([10])
h_conv3 = tf.nn.relu(conv2d(h_pool_2, W_conv3) + b_conv3)


nt_hpool3 = avg_pool_6x6(h_conv3)
nt_hpool3_flat = tf.reshape(nt_hpool3, [-1, 10])
y_conv = tf.nn.softmax(nt_hpool3_flat)

cross_entropy = -tf.reduce_sum(y * tf.log(y_conv))

train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)

correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, ‘float‘))

# 使用MonitoredTrainingSession()之前，必须定义global_step变量
global_step = tf.train.get_or_create_global_step()
checkpoint_step = tf.assign_add(global_step, 1)

# 2秒保存一次检查点
save_filename = ‘log/checkpoints‘
sess = tf.train.MonitoredTrainingSession(checkpoint_dir=save_filename, save_checkpoint_secs=2)
# MonitoredTrainingSession() 已经初始化了全局变量，所以不需要再初始化全局变量
# tf.global_variables_initializer().run(session=sess)　　
tf.train.start_queue_runners(sess=sess)

print(‘Start Training...‘)

max_epochs = 15000

# 读取上次最后一次训练的步数，i为当前迭代步数
i = sess.run(global_step)
while i < max_epochs:
    image_batch, label_batch = sess.run([images_train, labels_train])

    label_b = np.eye(10, dtype=float)[label_batch]

    train_step.run(feed_dict={x: image_batch, y: label_b}, session=sess)

    if i % 200 == 0:
        train_accuracy = accuracy.eval(feed_dict={x: image_batch, y: label_b}, session=sess)
        print(‘step: %d, training accuracy: %g‘ % (i, train_accuracy))
    i = sess.run(checkpoint_step)

# 训练结束，测试模型
image_batch_test, label_batch_test = sess.run([images_test, labels_test])
label_b_test = np.eye(10, dtype=float)[label_batch_test]
print(‘Finished! test accuracy %g‘ % accuracy.eval(feed_dict={x: image_batch_test, y: label_b_test}, session=sess))