使用TensorFlow实现简单线性回归模型

本文使用TensorFlow实现最简单的线性回归模型，供大家参考，具体内容如下

线性拟合y=2.7x+0.6，代码如下:

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

n = 201 # x点数
X = np.linspace(-1, 1, n)[:,np.newaxis] # 等差数列构建X，[:,np.newaxis]这个是shape，这一行构建了一个n维列向量（[1,n]的矩阵）
noise = np.random.normal(0, 0.5, X.shape) # 噪声值，与X同型
Y = X*2.7 + 0.6 + noise # Y

xs = tf.placeholder(tf.float32, [None, 1]) # 下面两行是占位符tf.placeholder(dtype, shape)
ys = tf.placeholder(tf.float32, [None, 1])

w = tf.Variable(1.1) # 这两行是weight变量，bias变量，括号中是初始值
b = tf.Variable(0.2)

ypredict = tf.add(w*xs,b) # 根据 w, b 产生的预测值

loss = tf.reduce_sum(tf.pow(ys-ypredict,2.0))/n # 损失函数，tf.reduce_sum()按某一维度元素求和，默认为按列

optimizer = tf.train.GradientDescentOptimizer(0.01).minimize(loss) # 梯度下降优化器，0.01学习率，最小化losss

init = tf.global_variables_initializer() # 初始化所有变量

with tf.Session() as sess:
sess.run(init) # 运行初始化
for i in range (1000): # 迭代1000次
 sess.run(optimizer, feed_dict = {xs:X,ys:Y}) # 运行优化器，梯度下降用到loss，计算loss需要xs, ys所以后面需要feed_dict
 if i％50==0: # 每隔50次迭代输出w,b,loss
    # 下面sess.run(w),sess.run(b)里面没有feed_dict是因为打印w,b不需要xs,ys，而打印loss需要
    print ("w:",sess.run(w),"\t b:", sess.run(b), "\t loss:", sess.run(loss,feed_dict={xs:X,ys:Y}))

plt.plot(X,X*sess.run(w)+sess.run(b)) # 运行迭代之后绘制拟合曲线，这需要在sess里面运行是因为要用到w,b
plt.scatter(X,Y) # 绘制被拟合数据（散点）
plt.show() # 绘制图像

结果：

w: 1.1106868  b: 0.2086223 loss: 1.2682248
w: 1.5626049  b: 0.4772562 loss: 0.7024503
w: 1.8849733  b: 0.57508457 loss: 0.47280872
w: 2.1149294  b: 0.61071056 loss: 0.36368176
w: 2.278966  b: 0.6236845 loss: 0.30917725
w: 2.3959787  b: 0.6284093 loss: 0.2815788
w: 2.4794474  b: 0.6301298 loss: 0.26755357
w: 2.5389886  b: 0.63075644 loss: 0.26041925
w: 2.5814607  b: 0.6309848 loss: 0.2567894
w: 2.611758  b: 0.6310678 loss: 0.25494233
w: 2.6333694  b: 0.6310981 loss: 0.25400248
w: 2.6487865  b: 0.631109  loss: 0.2535242
w: 2.659784  b: 0.63111293 loss: 0.25328085
w: 2.6676288  b: 0.6311139 loss: 0.25315702
w: 2.6732242  b: 0.6311139 loss: 0.25309405
w: 2.6772156  b: 0.6311139 loss: 0.25306198
w: 2.6800632  b: 0.6311139 loss: 0.25304565
w: 2.6820953  b: 0.6311139 loss: 0.25303733
w: 2.6835444  b: 0.6311139 loss: 0.25303313
w: 2.684578  b: 0.6311139 loss: 0.25303096

来源：https://blog.csdn.net/weixin_38275649/article/details/80233579