python一元线性回归：梯度下降法、损失函数

# -*- coding: utf-8 -*-


import numpy as np
import matplotlib.pyplot as plt

# %matplotlib inline

# 1.读入数据
data = np.genfromtxt('一元线性回归.csv', delimiter=',')
x_data = data[:, 0]  # 1维
y_data = data[:, 1]  # 1维
plt.scatter(x_data, y_data)

# 2.用梯度下降法调整参数
k = 0  # 先随便设置一个k
b = 0  # 先随便设置一个b
learing_rate = 0.0001  # 学习率
epoches = 30  # 循环30次
m = len(x_data)  # 一共100条数据

# 不断的调用k和b
for i in range(epoches):
    # for j in range(m):
    k_grad = (k * x_data + b - y_data) * x_data
    b_grad = k * x_data + b - y_data

    k_grad = (-1 / m) * sum(k_grad)
    b_grad = (-1 / m) * sum(b_grad)

    k = k + learing_rate * k_grad
    b = b + learing_rate * b_grad

print('最终结果：k={}, b={}'.format(k, b))
plt.scatter(x_data, y_data)
plt.plot(x_data, x_data * k + b, 'r')

# 改进1

k = 0  # 先随便设置一个k
b = 0  # 先随便设置一个b
learing_rate = 0.0001  # 学习率
epoches = 30  # 循环30次
m = len(x_data)  # 一共100条数据


def loss_function(k, b):
    total_error = (k * x_data + b - y_data) ** 2
    total_error = sum(total_error)
    return total_error / (2 * m)


# 不断的调用k和b
for i in range(1, epoches + 1):

    k_grad = (k * x_data + b - y_data) * x_data
    b_grad = k * x_data + b - y_data

    k_grad = (-1 / m) * sum(k_grad)
    b_grad = (-1 / m) * sum(b_grad)

    k = k + learing_rate * k_grad
    b = b + learing_rate * b_grad

    # 每循环5次，输出此时k, b和损失函数
    if i % 3 == 0:
        print('epoch{}:k={},b={},loss function={}'.format(i, k, b, loss_function(k, b)))
        plt.scatter(x_data, y_data)
        plt.plot(x_data, x_data * k + b, 'r')
        plt.show()

print('最终结果：k={}, b={}'.format(k, b))
plt.scatter(x_data, y_data)
plt.plot(x_data, x_data * k + b, 'r')
plt.show()