python 实现一个简单的线性回归案例

#!/usr/bin/env python
# -*- coding: utf-8 -*-
# @File : 自实现一个线性回归.py
# @Author: 赵路仓
# @Date : 2020/4/12
# @Desc :
# @Contact : 398333404@qq.com
import os

import tensorflow as tf

def linear_regression():
 """
 自实现一个线性回归
 :return:
 """
 # 命名空间
 with tf.variable_scope("prepared_data"):
   # 准备数据
   x = tf.random_normal(shape=[100, 1], name="Feature")
   y_true = tf.matmul(x, [[0.08]]) + 0.7
   # x = tf.constant([[1.0], [2.0], [3.0]])
   # y_true = tf.constant([[0.78], [0.86], [0.94]])

with tf.variable_scope("create_model"):
   # 2.构造函数
   # 定义模型变量参数
   weights = tf.Variable(initial_value=tf.random_normal(shape=[1, 1], name="Weights"))
   bias = tf.Variable(initial_value=tf.random_normal(shape=[1, 1], name="Bias"))
   y_predit = tf.matmul(x, weights) + bias

with tf.variable_scope("loss_function"):
   # 3.构造损失函数
   error = tf.reduce_mean(tf.square(y_predit - y_true))

with tf.variable_scope("optimizer"):
   # 4.优化损失
   optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01).minimize(error)

# 收集变量
 tf.summary.scalar("error", error)
 tf.summary.histogram("weights", weights)
 tf.summary.histogram("bias", bias)

# 合并变量
 merged = tf.summary.merge_all()

# 创建saver对象
 saver = tf.train.Saver()

# 显式的初始化变量
 init = tf.global_variables_initializer()

# 开启会话
 with tf.Session() as sess:
   # 初始化变量
   sess.run(init)

# 创建事件文件
   file_writer = tf.summary.FileWriter("E:/tmp/linear", graph=sess.graph)

# print(x.eval())
   # print(y_true.eval())
   # 查看初始化变量模型参数之后的值
   print("训练前模型参数为:权重％f,偏置％f" ％ (weights.eval(), bias.eval()))

# 开始训练
   for i in range(1000):
     sess.run(optimizer)
     print("第％d次参数为：权重％f,偏置％f,损失％f" ％ (i + 1, weights.eval(), bias.eval(), error.eval()))

# 运行合并变量操作
     summary = sess.run(merged)
     # 将每次迭代后的变量写入事件
     file_writer.add_summary(summary, i)

# 保存模型
     if i == 999:
       saver.save(sess, "./tmp/model/my_linear.ckpt")

# # 加载模型
   # if os.path.exists("./tmp/model/checkpoint"):
   #   saver.restore(sess, "./tmp/model/my_linear.ckpt")

print("参数为：权重％f,偏置％f,损失％f" ％ (weights.eval(), bias.eval(), error.eval()))
   pre = [[0.5]]
   prediction = tf.matmul(pre, weights) + bias
   sess.run(prediction)
   print(prediction.eval())

return None

if __name__ == "__main__":
 linear_regression()

来源：https://www.cnblogs.com/zlc364624/p/12686695.html