基于numpy的线性回归模型

实现思路

1. 导入numpy包
2. 构建线性回归损失函数模型
3. 计算损失函数的梯度，并按梯度下降方法（最小化损失函数）更新参数
4. 设置初始化参数，并按3中的方法更新参数

代码

# 导入numpy包
import numpy as np


# 采用线性回归模型
# 计算损失函数
def compute_loss(w, b, points):
    loss = 0
    for i in range(0, len(points)):
        x = points[i, 0]
        y = points[i, 1]
        loss += ((w*x + b) - y) ** 2
    return loss/len(points)

# 计算梯度
def compute_gradient(w_current, b_current, points, learningrate):
    w_gradient = 0
    b_gradient = 0
    N = float(len(points))
    for i in range(0, len(points)):
        x = points[i, 0]
        y = points[i, 1]
        w_gradient = x * ((w_current*x + b_current) - y) * 2 / N
        b_gradient = ((w_current*x + b_current) - y) * 2 / N
    w_new = w_current - (learningrate * w_gradient)
    b_new = b_current - (learningrate * b_gradient)
    return [w_new, b_new]

def gradient_decent(points, w_init, b_init, learningrate, num_train):
    w = w_init
    b = b_init
    for i in range(num_train):
        b, w = compute_gradient(w, b, np.array(points), learningrate)
    return[w, b]

def main():
    points = np.genfromtxt("data.csv", delimiter=",")
    learningrate = 0.0001
    num_train = 1000
    w_initial = 0
    b_initial = 0
    print("starting... w={0} b={1} loss={2}" .format(w_initial, b_initial, compute_loss(w_initial, b_initial, points)))
    print("loading...")
    [w, b] = gradient_decent(points, w_initial, b_initial, learningrate, num_train)
    print("after {0} times, w = {1}, b = {2}, loss = {3}".format(num_train, w, b, compute_loss(w, b, points)))

main()

参考链接

• 线性回归原理