机器学习之线性回归python实现

最新推荐文章于 2024-08-05 22:24:28 发布

Sayram_0

最新推荐文章于 2024-08-05 22:24:28 发布

阅读量217

点赞数

分类专栏：机器学习文章标签： python 机器学习

本文链接：https://blog.csdn.net/weixin_45476502/article/details/108628745

版权

机器学习专栏收录该内容

10 篇文章 0 订阅

订阅专栏

线性回归

粗略的记录了一下今天上课将的内容……

1 损失函数

$J(a,b)=\frac{1}{2n}\displaystyle \sum_{i=0}^n(y_i- \hat y_i)^2$

2 优化方法

$\frac{\partial J}{\partial a} =\frac{\partial\frac{1}{2n}\displaystyle \sum_{i=0}^n(y_i- \hat{y_i})^2}{\partial a}=\frac{1}{n}\displaystyle \sum_{i=0}^n(y_i-ax_i-b)\frac{\partial(y_i-ax_i-b)(-x_i)}{\partial a}$
$=\frac{1}{n}\displaystyle \sum_{i=0}^n(y_i-ax_i-b)(-x_i) =\frac{1}{n} \displaystyle \sum_{i=0}^nx(\hat{y_i}-y_i)$

$\frac{\partial J}{\partial b} =\frac{\partial\frac{1}{2n}\displaystyle \sum_{i=0}^n(y_i- \hat{y_i})^2}{\partial b} =\frac{1}{n}\displaystyle \sum_{i=0}^n(y_i-ax_i-b)\frac{\partial(y_i-ax_i-b)(-x_i)}{\partial b}$
$=\frac{1}{n}\displaystyle \sum_{i=0}^n(y_i-ax_i-b)(-1) =\frac{1}{n} \displaystyle \sum_{i=0}^n(\hat{y_i}-y_i)$

更新 $a$ 和 $b$ 的值（梯度下降）：
$a-\alpha \frac{\partial J}{\partial a}$
$b-\alpha \frac{\partial J}{\partial b}$

3 代码实现

'''
Description: 
Author: Weijian Ma
Date: 2020-09-16 18:47:40
LastEditTime: 2020-09-16 19:23:12
LastEditors: Weijian Ma
'''
import numpy as np
import matplotlib.pyplot as plt

plt.rcParams['font.sans-serif'] = ['SimHei']
plt.rcParams['axes.unicode_minus'] = False

## 数据及参数的初始化
x = [13854,12213,11009,10655,9503] 
x = np.reshape(x,newshape=(5,1)) / 10000.0
y =  [21332, 20162, 19138, 18621, 18016] 
y = np.reshape(y,newshape=(5,1)) / 10000.0
a = 1 
b = 1
alpha = 1e-1
n = len(x)

## 模型
def myModel(x):
    return a*x + b

## 损失函数
def costFunction(x, y, a, b):
    return 0.5/n*(np.square(a*x+b-y)).sum()

## 优化
def opt(x, y, a, b):
    yi = myModel(x)
    da = (1/n) * ((yi-y)*x).sum()
    db = (1/n) * ((yi-y).sum())
    a = a-alpha*da
    b = a-alpha*db
    return a, b

## 训练模型
fig = plt.figure(figsize=(8,4))
sub01 = plt.subplot(121)
sub02 = plt.subplot(122)
costList = []

for i in range(50):
    print('训练次数：{}'.format(i+1))
    cost = costFunction(x, y, a, b)
    costList.append(cost)
    a, b = opt(x, y, a, b)
    sub01.cla()
    sub02.cla()
    sub01.plot(x, a*x+b)
    sub01.scatter(x, y)
    sub01.set_xlabel('x')
    sub01.set_ylabel('y')
    sub01.set_title('a={0}, b={1}'.format(a, b))
    sub02.set_xlabel('训练次数')
    sub02.set_ylabel('损失函数值')
    sub02.set_title('当前损失函数值：{}'.format(cost))
    sub02.plot(costList)
    plt.pause(0.001)
plt.show()