[深度学习]使用Tensorflow实现一个简单异或逻辑的学习

import tensorflow as tf
import numpy as np

# zhonglihao 2019.04

# 训练集
x_train = [[0.0,0.0],[0.0,1.0],[1.0,0.0],[1.0,1.0]]
y_train = [[1],[0],[0],[1]]

# 网络参数 输入向量为2 隐藏层为4个cell 输出1/0单个节点
# zhonglihao: 使用2个隐藏cell也能学 但很容易卡死 所以这里用得多点 训练的时候容易收敛
Input_Node      = 2
Layer1_Node     = 4 # 4颗比较稳
Output_Node     = 1

# 输入输出动态占位符
x   = tf.placeholder(tf.float32, [None, Input_Node])
y_  = tf.placeholder(tf.float32, [None, Output_Node])

# 定义网络参数变量
w1 = tf.Variable(tf.random_normal([Input_Node, Layer1_Node]))
w2 = tf.Variable(tf.random_normal([Layer1_Node, Output_Node]))

b1 = tf.Variable(tf.zeros([Layer1_Node]))
b2 = tf.Variable(tf.zeros([Output_Node]))

# 进行神经网络相乘并使用激活函数sigmoid
Layer1_Output   = tf.sigmoid(tf.matmul(x,w1)+b1)
pred            = tf.sigmoid(tf.matmul(Layer1_Output,w2)+b2)

# 计算损失，这里使用方差
cost = tf.reduce_mean(tf.square(y_ - pred))

# 学习率
learning_rate = 0.01

# 优化器
train_step = tf.train.AdamOptimizer(learning_rate).minimize(cost) 

# 全局变量初始化
init = tf.global_variables_initializer()

with tf.Session() as sess:
    sess.run(init)
    # 自定义训练轮数
    max_step = 10000
    for i in range(max_step + 1):
        sess.run(train_step, feed_dict = {x: x_train, y_: y_train})
        loss = sess.run(cost, feed_dict = {x: x_train, y_: y_train})
        if (i % 100 == 0):
            print(sess.run(pred, feed_dict = {x: x_train}))
    print('end')