梯度下降法解决线性回归问题

class linear_model:
    def __init__(self, w, b ,lr):
        self.w = w
        self.b = b
        self.lr = lr
        
        self.ypred_minus_ytrue = 0
        self.x = 0
        
        self.grad_w = 0
        self.grad_b = 0
        
    def forward(self, x):
        self.x = x
        return self.w * x + self.b
    
    def backward(self,diff,diff_mul_x):
        # update the gradient
        
        # d cost / d w = 2 * x * (y_pred - y_true)   [1]
        # d cost / d b = 2 * (y_pred - y_true)       [2]
        
        self.grad_w = diff_mul_x  # [1]
        self.grad_b = diff           # [2]
        
        self.w = self.w - self.lr * self.grad_w
        self.b = self.b - self.lr * self.grad_b


def loss_function(y_pred,y_true,x):
    diff = np.mean(y_pred - y_true)
    diff_mul_x = np.mean(x * (y_pred - y_true))
    return np.mean((y_pred - y_true)**2) , diff , diff_mul_x


# main 
if __name__=="__main__":
    
    import numpy as np
    
    # random x ,y data
    x = np.array([1,2,3,4,7])
    y = np.array([1,3,2,5,9])
    
    # inverse x array
    x = 1/x
    
    # # square y
    # y = y**2
    
    
    loss_ls = []
    w_ls = []
    b_ls = []
    
    clf = linear_model(w=1,b=0,lr=0.01)
    epoch = 1000
    for i in range(epoch):
        y_pred = clf.forward(x)
        diff_square , diff , diff_mul_x = loss_function(y_pred, y, x)
        clf.backward(diff,diff_mul_x)
        print(diff_square,diff,clf.w,clf.b)
        
        loss_ls.append(diff_square)
        w_ls.append(clf.w)
        b_ls.append(clf.b)
    
    # plot
    import matplotlib.pyplot as plt
    # plot in a subplot toghter with loss
    plt.subplot(2,1,1)
    plt.plot(range(epoch),loss_ls)
    plt.title('loss')
    plt.subplot(2,1,2)
    plt.plot(range(epoch),w_ls)
    plt.plot(range(epoch),b_ls)
    # legend
    plt.legend(['w','b'])
    plt.title('w,b')
    plt.show()
    
    # plot data and line
    plt.plot(x,y,'o')
    plt.plot(x,clf.w*x+clf.b)
    plt.show()
    plt.title(label='w*x+b')

数据1

数据2