tf.train.exponential_decay(learning_rate, global_step, decay_steps, decay_rate, staircase=False, name=None)

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

"""

Created on Sun May 27 11:19:46 2018

@author: Administrator

"""

#设算是函数loss=(w+1)^2令w初始值为常熟10，反向传播求最优w，求最小loss对应的w的值

#使用指数衰减的学习率，在迭代初期有比较搞的下降速度，可以在比较小的训练轮数下更有收敛度

import tensorflow as tf

LEARNING_RATE_BASE=0.1 #最初学习率

LEARNING_RATE_DECAY=0.99#学习率衰减

LEARNING_RATE_STEP=1#喂入多少伦BATCH_SIZE后更新一次学习率，一般威威总样本数/BATCH_SIZE

#运行了几轮BATCH_SIZE的计数器，初值给0，设为不被训练

global_step=tf.Variable(0,trainable=False)

#定义指数下降学习率

learning_rate=tf.train.exponential_decay(LEARNING_RATE_BASE,global_step,

                                        LEARNING_RATE_STEP,LEARNING_RATE_DECAY

                                        ,staircase=True)

#定义待优化参数，初始值10

w=tf.Variable(tf.constant(5,dtype=tf.float32))

#定义损失函数

loss=tf.square(w+1)

#定义反向传播方法

train_step=tf.train.GradientDescentOptimizer(learning_rate).minimize(

        loss,global_step=global_step)

#生成会话，训练40论

with tf.Session() as sess:

    init_op=tf.global_variables_initializer()

    sess.run(init_op)

    for i in range(40):

        sess.run(train_step)

        #更新学习速率

        learning_rate_val=sess.run(learning_rate)

        global_step_val=sess.run(global_step)

        w_val=sess.run(w)

        loss_val=sess.run(loss)

        print("在%s次训练后 global_step为%f权重为%f,学习率为%f,损失为%f"%(i,

                                                          global_step_val,

                                                          w_val,learning_rate_val,

                                                          loss_val))