神经网络的一般结构（模板）

import tensorflow as tf

#定义训练数据batch的大小
batch_size = n

#定义学习率
learning_rate = k

#设置总共要训练的轮数
STEPS = m
#定义占位符
x = tf.placeholder(tf.float32, shape(维度1,维度2), name=‘x-input’)
y_ = tf.placeholder(tf.float32, shape(维度1,维度2), name=‘y-input’)

#定义神经网络

#损失函数和优化器
loss = …
train_step = tf.train.AdamOptimizer(learning_rate).minimize(loss)

#训练神经网络
with tf.Session() as sess:
#变量初始化
init_op = tf.global_variables_initializer()
sess.run(init_op)

#迭代更新参数
for i in range(STEPS):
    current_X, current_Y =.....
    sess.run(train＿step, feed_dict={x: current_X, y_; current_Y})