机器学习一元线性回归的几种实现

给定一组样本点：

x = [0.95, 3, 4, 5.07, 6.03, 8.21, 8.85, 12.02, 15]
y = [5.1, 8.7, 11.5, 13, 15.3, 18, 21, 26.87, 32.5]

然后求y=wx+b的w和b值。

-----------------------------------------------------------------------------------------------------

方法1：调python的sklearn库，看不到具体逻辑，都已经被封装好了。

from sklearn import linear_model
import time

start = time.time()

reg=linear_model.LinearRegression(fit_intercept=True,copy_X=False)
x=[[0.95], [3], [4], [5.07], [6.03], [8.21], [8.85], [12.02], [15]]
y=[5.1, 8.7, 11.5, 13, 15.3, 18, 21, 26.87, 32.5]
reg.fit(x,y)
w=reg.coef_
b=reg.intercept_
print(str(w)+","+str(b))

end = time.time()
print(end - start, '秒')

 w和b值如下：

-----------------------------------------------------------------------------------------------------

方法2：批量梯度下降法，python实现

import matplotlib.pyplot as plt
import matplotlib
from math import pow
import time

start = time.time()

x = [0.95, 3, 4, 5.07, 6.03, 8.21, 8.85, 12.02, 15]
y = [5.1, 8.7, 11.5, 13, 15.3, 18, 21, 26.87, 32.5]

# 参数定义
w = 1.5
b = 1.5
learningRate = 0.01
length = len(x)

# 批量梯度下降法
for num in range(10000):
    derivative_b = 0  #导数
    derivative_w = 0  #导数
    # 求导
    for i in range(length):
        derivative_w += - (w * x[i] + b - y[i]) * x[i] / length
        derivative_b += - (w * x[i] + b - y[i]) / length
    # delta w , delta b = 学习率 * 偏导数
    w = w + learningRate * derivative_w
    b = b + learningRate * derivative_b

    if derivative_w <= pow(10,-13) and derivative_b <= pow(10,-13):
        break

print("y={}*x+{}".format(w, b))

end = time.time()
print(end - start, '秒')

matplotlib.rcParams['font.sans-serif'] = ['SimHei']
plt.plot(x, y, 'bo', label='样本数据', color='black')
plt.plot(x, [b + w * x for x in x], label='线性回归直线', color='red')
plt.xlabel('x')
plt.ylabel('y')
plt.legend()
plt.show()
plt.xlabel('上方为w，下方为b')
plt.show()
plt.xlabel('上方为w，下方为b')
plt.show()

运行结果如下：

 ----------------