算法梳理进阶线性回归(二)

上次简单介绍了线性回归原理以及公式推导，今天用python代码简单实现下单变量线性回归：

# -*- coding:utf-8 -*-
import numpy as np
from matplotlib import pylab as pl


class LRArithmetic:

    def __init__(self, x, y):
        """
        Parameters of the loading
        :param x:
        :param y:
        """
        self.x = np.array(x)
        self.y = np.array(y)

        # Check data
        if len(x) != len(y):
            raise ValueError('There is a problem with the input data!!!')

        # Initialize the parameters of the unary equation: y = wx + b
        self.w = 0
        self.b = 0
        self.is_fit = False

    def fit_arithmetic(self):
        """
         Calculate the parameters of w and b by arithmetic method
        :return:
        """
        # Calculate the mean of x and y
        x_mean = np.mean(self.x)
        y_mean = np.mean(self.y)

        # Calculate the numerator and denominator
        numerator = 0
        denominator = 0
        for i in range(len(self.x)):
            numerator += (self.x[i] - x_mean) * (self.y[i] - y_mean)
            denominator += np.square((self.x[i] - x_mean))
        self.w = numerator / denominator
        self.b = y_mean - self.w * x_mean
        self.is_fit = True

    def fit_vector(self):
        """
        Calculate the parameters of w and b by vector method
       :return:
       """
        # Calculate the mean of x and y
        x_mean = np.mean(self.x)
        y_mean = np.mean(self.y)

        # Calculate the numerator and denominator
        numerator = np.sum((self.x - x_mean) * (self.y - y_mean))
        denominator = np.sum(np.square(self.x - x_mean))

        self.w = numerator / denominator
        self.b = y_mean - self.w * x_mean
        self.is_fit = True

    def predit(self, x):
        """
        Predicted data
        :param x:
        :return:
        """
        if not self.is_fit:
            raise RuntimeError('Please call the fit method first!!!')
        return np.array(x) * self.w + self.b


if __name__ == '__main__':
    # train
    x = np.array([1, 3, 2, 1, 3])
    y = np.array([14, 24, 18, 17, 27])
    LR = LRArithmetic(x, y)
    LR.fit_vector()

    # predit
    x_test = np.array([2, 3, 5, 6, 10])
    y_test = LR.predit(x_test)
    print(y_test)

    xx = np.linspace(0, max(x_test))
    yy = LR.w * xx + LR.b
    pl.plot(xx, yy, 'k-')
    pl.scatter(x, y, cmap=pl.cm.Paired)
    pl.scatter(x_test, y_test, cmap=pl.cm.Paired)
    pl.show()

运行效果如下图：

注： 蓝色的是训练数据，橙色的是预测的结果
使用fit_vector向量计算方法进行训练对比fit_arithmetic循环方式，数据量少时看不出太大的时间差，不过向量方式计算在性能上肯定是原因比循环要快的。推荐使用numpy和pandas下的包来进行计算，毕竟做过优化。