tensorflow线性回归

一、构造1000组高斯分布数据

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

#定义一个空的列表
vector = []
#生成1000组随机数
for i in range(1000):
    x = np.random.normal(0,0.6)
    y = 0.2*x + 0.4 + np.random.normal(0,0.08)
    vector.append([x,y])
    
x_data = [v[0] for v in vector]
y_data =[v[1] for v in vector]
plt.scatter(x_data,y_data,c='r')
plt.show()

二、训练模型

#初始化权重w和偏置b
w = tf.Variable(tf.random_uniform([1],-1,1),name='w')
b = tf.Variable(tf.random_uniform([1],-1,1),name='b')

#经过计算得出预估值
y = w * x_data + b

#定义损失函数（均方误差）
loss = tf.reduce_mean(tf.square(y-y_data),name='loss')

#采用梯度下降来更新参数
optimizer = tf.train.GradientDescentOptimizer(0.5)

#训练：训练的过程就是最小化损失函数
train = optimizer.minimize(loss,name='train')

sess = tf.Session()
sess.run(tf.global_variables_initializer())
#打印初始权重和偏置
print('w = ',sess.run(w),'b = ',sess.run(b),'loss = ',sess.run(loss))
#执行30次训练
for i in range(60):
    sess.run(train)
    #输出训练好的权重和偏置（每迭代10次显示一下）
    if i%10==0:
        print('w = ',sess.run(w),'b = ',sess.run(b),'loss = ',sess.run(loss))

####运行结果如下：
w =  [-0.24027228] b =  [-0.346241] loss =  0.64620763
w =  [-0.06341013] b =  [0.40470338] loss =  0.033992454
w =  [0.20420161] b =  [0.40080273] loss =  0.006240865
w =  [0.2064087] b =  [0.40077135] loss =  0.0062389774
w =  [0.20642689] b =  [0.40077108] loss =  0.0062389774
w =  [0.20642704] b =  [0.40077108] loss =  0.006238977
w =  [0.20642704] b =  [0.40077108] loss =  0.006238977

三、画出这组数据的拟合曲线

plt.scatter(x_data,y_data,c='r')
plt.plot(x_data,sess.run(w*x_data+b))
plt.show()