线性回归方程

最小二乘法求线性回归

1、读取数据

import numpy as np
import matplotlib.pyplot as plt

# 读取信息
points = np.genfromtxt('data.csv',delimiter=',')
x = points[:,0]
y = points[:,1]
​
plt.scatter(x,y)  # 扫描所有点
plt.show() 

2、定义损失函数

# 将求拟合函数转化为求损失函数，（当损失函数的值最小时，拟合效果越好）
def cost_func(w,b,points):
    
    M = len(points)
    
    cost_sum = 0
    
    for i in range(M):
        
        x = points[i,0]
        y = points[i,1]
        
        cost_sum += (y - w*x - b) ** 2
    
    return cost_sum / M  

3、定义最小二乘法核心函数

## 1、求平均值
​
def average(data):
    
    M = len(data)
    data_sum = 0
    
    for i in range(M):
        data_sum += data[i]
    
    return data_sum / M
​
#2、求拟合函数
def fit(points):
    
    w = 0
    b = 0
    M = len(points)
    
    x_bar = average(points[:,0])
    sum_son = 0    
    sum_par_x2 = 0
    
    for i in range(M):
        
        x = points[i,0]
        y = points[i,1]
        
        sum_son += y * (x - x_bar)
        sum_par_x2 += x ** 2
        
    w = sum_son / (sum_par_x2 - M * x_bar ** 2) 
        
    
    for i in range(M):
        
        x = points[i,0]
        y = points[i,1]
        
        b += (y - w*x) / M
    
    return w,b

4、结果验证

cost = 
w,b = fit(points)
cost = cost_func(w,b,points)
print(cost)
​
x = points[:,0]
y = points[:,1]
y2 = w * x + b
​
plt.scatter(x,y)  # 扫描所有点
plt.plot(x,y2,'r')
plt.show() 

110.25738346621314

梯度下降法求线性回归

1、读取数据

import numpy as np
import matplotlib.pyplot as plt
# 读取数据
points = np.genfromtxt('data.csv',delimiter=',')
x = points[:,0]
y = points[:,1]
​
plt.scatter(x,y)  # 扫描所有点
plt.show() 

2、定义损失函数

def compute_cost(w,b,points):
    total_cost = 0
    
    M = len(points)
    
    for i in range(M):
        x = points[i,0]
        y = points[i,1]
        y_bar = w * x + b
        total_cost += (y - y_bar) ** 2
        
    return total_cost / M

3、定义模型的超参

# 定义步长
alpha = 0.0001
init_w = 0
init_b = 0
num_iter = 10

4、定义核心梯度下降算法函数

def grad_desc(points,init_w,init_b,alpha,num_iter):
    
    w = init_w
    b = init_b
    
    # 定义list，保存损失函数的值，显示下降的过程
    cost_list = []
    
    #通过不断迭代w，b的过程，获取最佳的拟合函数
    for i in range(num_iter):
        #迭代过程中损失函数的变化
        cost_list.append( compute_cost(w,b,points) )
        
        # 递归优化w,b
        w,b = step(w,b,points,alpha)
        
    return w,b,cost_list
        
def step(w,b,points,alpha):
    
    M = len(points)
    
    sum_w = 0
    sum_b = 0
    
    for i in range(M):
        
        x = points[i,0]
        y = points[i,1]
        
        sum_w += (w * x + b - y) * x
        sum_b += (w * x + b - y)
    
    update_w =w - alpha * (2 * sum_w / M )
    update_b =b - alpha * (2 * sum_b / M )
    
    return update_w,update_b

5、测试结果

w,b,cost = grad_desc(points,init_w,init_b,alpha,num_iter)
print("w-----"+str(w))
print("b-----"+str(b))
print("cost------"+str(cost))
w-----1.4774173755483797
b-----0.02963934787473238
cost------[5565.107834483211, 1484.5865574086486, 457.85425757376686, 199.50998572553874, 134.50591058200533, 118.14969342239947, 114.0341490603815, 112.99857731713661, 112.7379818756847, 112.67238435909097]
x = points[:,0]
y = points[:,1]
y2 = w * x + b
​
plt.scatter(x,y)  # 扫描所有点
plt.plot(x,y2,'r')
plt.show() 
print(compute_cost(w,b,points))
plt.plot(cost)
plt.show()

112.65585181499746

python库求线性回归

1、读取数据

import numpy as np
import matplotlib.pyplot as plt
# 导入机器学习的库
from sklearn.linear_model import LinearRegression
# 读取数据
points = np.genfromtxt('data.csv',delimiter = ',')
​
x = points[:,0]
y = points[:,1]

2、定义损失函数

cost_func
# 将求拟合函数转化为求损失函数，（当损失函数的值最小时，拟合效果越好）
def cost_func(w,b,points):
    
    M = len(points)
    
    cost_sum = 0
    
    for i in range(M):
        
        x = points[i,0]
        y = points[i,1]
        
        cost_sum += (y - w*x - b) ** 2
    
    return cost_sum / M 

3、使用系统库求解w,b

lr = LinearRegression()  #理解为创建线性回归一个对象
​
x_m = x.reshape(-1,1)   #转化成n行一列的数组,矩阵的形式
y_m = y.reshape(-1,1)
​
lr.fit(x_m,y_m) #需要传入二维的数组，矩阵的形式
​
#获取w,b
w = lr.coef_[0]          # 获取系数
b = lr.intercept_[0]       # 获取截距

4、结果显示

x = points[:,0]
y = points[:,1]
y2 = w * x + b
​
plt.scatter(x,y)  # 扫描所有点
plt.plot(x,y2,'r')
plt.show() 
print(cost_func(w,b,points))

[110.25738347]