tensorflow

#1、创建图,启动图

import tensorflow as tf

#创建两个常量op:m1,m2

m1=tf.constant([[1,2,3],[2,3,4]])

m2=tf.constant([[3,5],[2,3],[4,5]])

print(m1,m2)

#创建矩阵乘法op

p=tf.matmul(m1,m2)

print(p)

---

#创建会话，启动默认图

sess=tf.Session()

#调用运行图的方法:run

result=sess.run(p)

print(result)

#关闭会话

sess.close()

---

#会话启用的简易操作,自动关闭会话

with tf.Session() as sess:

    result=sess.run(p)

    print(result)

---

#2、变量

#定义变量

x=tf.Variable([7,8])

#定义常量

a=tf.constant([3,5])

#加法运算op

sub=tf.subtract(x,a)

#减法运算op

add=tf.add(x,a)

#全局变量初始化

init=tf.global_variables_initializer()

#启动会话

with tf.Session() as sess:

    sess.run(init)

    print(sess.run(sub))

    print(sess.run(add))

---

#创建一个循环变量

state=tf.Variable(0,name='counter')

new_state=tf.add(state,1)

upstate=tf.assign(state,new_state)

init=tf.global_variables_initializer()

with tf.Session() as sess:

    sess.run(init)

    print(sess.run(state))

    for _ in range(5):

        sess.run(upstate)

        print(sess.run(state),end=',')

---

#3、fetch and feed

#fetch同时执行多个节点OP

input1=tf.constant(2)

input2=tf.constant(5)

input3=tf.constant(7)

add=tf.add(input1,input2)

mul=tf.multiply(add,input3)

with tf.Session() as sess:

    result=sess.run([add,mul])

    print(result)

---

#Feed

#创建占位符

input1=tf.placeholder(tf.float32)

input2=tf.placeholder(tf.float32)

mul=tf.multiply(input1,input2)

with tf.Session() as sess:

    print(sess.run(mul,feed_dict={input1:[7.],input2:[8.]}))#以数据字典的形式输入数据

---

#简单实例

import numpy as np

#使用numpy生成100个随机数样本

x_data=np.random.rand(100)

y_data=x_data*0.1+0.2

#构建个线性模型

b=tf.Variable(0.)

k=tf.Variable(0.)

y=k*x_data+b

#二次代价函数

loss=tf.reduce_mean(tf.square(y_data-y))

#梯度下降训练优化

optimizer=tf.train.GradientDescentOptimizer(0.2)

#代价函数最小化

train=optimizer.minimize(loss)

init=tf.global_variables_initializer()

with tf.Session() as sess:

    sess.run(init)

    for step in range(201):

        sess.run(train)

        if step%20 == 0:

            print(step,sess.run([k,b]))

---

#非线性回归模型

import tensorflow as tf

import numpy as np

import matplotlib.pyplot as plt

#使用numpy生成200个随机点

x_data=np.linspace(-0.5,0.5,200)[:,np.newaxis]#增加新的维度

noise=np.random.normal(0,0.02,x_data.shape)#生成噪点数据

y_data=np.square(x_data) + noise

#定义两个占位符

x=tf.placeholder(tf.float32,[None,1])#32位浮点数，[任意形状，一列]

y=tf.placeholder(tf.float32,[None,1])#32位浮点数，[任意形状，一列]

#定义神经网络中间层

Weights_l1=tf.Variable(tf.random_normal([1,10]))#权重值

biases_l1=tf.Variable(tf.zeros([1,10]))#偏置值

Wx_plus_b_l1=tf.matmul(x,Weights_l1)+biases_l1#信号总和

l1=tf.nn.tanh(Wx_plus_b_l1)#中间层输出

#输出层

Weights_l2=tf.Variable(tf.random_normal([10,1]))#权重值

biases_l2=tf.Variable(tf.zeros([1,1]))#偏置值

Wx_plus_b_l2=tf.matmul(l1,Weights_l2)+biases_l2#信号总和

prediction=tf.nn.tanh(Wx_plus_b_l2)#中间层输出层

#二次代价函数

loss=tf.reduce_mean(tf.square(y-prediction))

#梯度下降法

train_step=tf.train.GradientDescentOptimizer(0.1).minimize(loss)

#定义会话

with tf.Session() as sess:

        sess.run(tf.global_variables_initializer())

        for _ in range(2000):

            sess.run(train_step,feed_dict={x:x_data,y:y_data})

#获得预测值

        prediction_value=sess.run(prediction,feed_dict={x:x_data})

#画图

        plt.figure()

        plt.scatter(x_data,y_data)

        plt.plot(x_data,prediction_value,'r-',lw=5)

        plt.show()