多元线性回归梯度下降代码实现

多元线性回归模型
$$\begin{gathered} w=[w_1,w_2,...,w_n] \\ {\hat{y}}^{(i)}=f_{\vec{w},b}(\vec{x})=\vec{w}\cdot \vec{x}+b \end{gathered}$$

代价函数 cost function
平均平方误差函数 squared error function
$$J(w,b)=\frac{1}{2m}\sum _{i=1}^m({\hat{y}}^{(i)}-y^{(i)}{)}^2$$

计算偏导
$$\begin{gathered} \frac{\partial }{\partial w_j}J(\vec{w},b)=\frac{\partial }{\partial w_j}\frac{1}{2m}\sum _{i=1}^m(\vec{w}\cdot {\vec{x}}^{(i)}+b-y^{(i)}{)}^2 \\ =\frac{1}{2m}\sum _{i=1}^{m}2(\vec{w}\cdot {\vec{x}}^{(i)}+b-y^{(i)})x_j^{(i)}=\frac{1}{m}\sum _{i=1}^m{x}_j^{(i)}(\vec{w}\cdot {\vec{x}}^{(i)}+b-y^{(i)}) \end{gathered}$$
$$\begin{gathered} \frac{\partial }{\partial b}J(\vec{w},b)=\frac{\partial }{\partial b}\frac{1}{2m}\sum _{i=1}^m(\vec{w}\cdot {\vec{x}}^{(i)}+b-y^{(i)}{)}^2 \\ =\frac{1}{2m}\sum _{i=1}^{m}2(\vec{w}\cdot {\vec{x}}^{(i)}+b-y^{(i)})=\frac{1}{m}\sum _{i=1}^m(\vec{w}\cdot {\vec{x}}^{(i)}+b-y^{(i)}) \end{gathered}$$

梯度下降
$$\begin{gathered} w_j=w_j-\alpha \frac{\partial }{\partial w_j}J(\vec{w},b)(j=1、\mathrm{2...}、n) \\ b=b-\alpha \frac{\partial }{\partial b}J(\vec{w},b) \end{gathered}$$

import nummpy as np

#计算代价的函数
def get_cost(x,y,w,b):
    #数据集的大小
    m=x.shape[0]
    total_cost=0
    for i in range(m):
        error=np.dot(x[i,:],w)+b-y[i]
        total_cost=total_cost+error**2
    cost=total_cost/(2*m)
    return cost

#计算梯度的函数
def get_gradient(x,y,w,b):
   #获取数据量和特征数量
   m=x.shape[0]
   n=x.shape[1]
   dj_dw=np.zeros((n,))
   dj_db=0
   for i in range(m):
       error=np.dot(x[i,:],w)+b-y[i]
       dj_db+=error
       for j in range(n):
           dj_dw[j]+=(error*x[i,j])
   dj_db=dj_db/m
   dj_dw=dj_dw/m
   return dj_dw,dj_db


#梯度下降函数
def gradient_descent(x, y, w_in, b_in, alpha, num_iters, cost_function, gradient_function): 
   '''
     x:输入向量,numpy.ndarray
     y:输出向量，numpy.ndarray
     w_in:初始w向量
     b_in:初始b
     alpha:学习率
     num_iters:迭代次数
     cost_function:代价函数
     gradient_function:计算梯度函数
   '''
   
    J_history = [] #记录训练过程中的所有代价
    b = b_in
    w = w_in
    
    for i in range(num_iters):
        # 计算偏导，更新参数w，b
        dj_dw,dj_db=gradient_function(x,y,w,b)
        w=w-dj_dw*alpha
        b=b-dj_db*alpha                           

        # 保存当前代价J和参数(w,b)->可用于后续可视化
        J_history.append( cost_function(x, y, w , b))
        # 打印其中十次训练信息
        if i% math.ceil(num_iters/10) == 0:
            print(f"Iteration {i}: Cost {J_history[-1]} ",
                  f"dj_dw: {dj_dw}, dj_db: {dj_db}  ",
                  f"w: {w}, b:{b}")
    print{f'final w:{w},b:{b}'}
    
    #输出目标值与预测值对比
    y_hat=np.dot(x,w)+b 
    for i in range(x.shape[0])：
         print(f'target value:{y[i]},predicted value:{y_hat[i]},error:{y[i]-y_hat[i]}')
    
    return w, b, J_history