java与C语言实现一个简单线性回归模型

简介

线性回归模型是机器学习中非常基础的模型，重点就是在通过最小二乘法计算参数w和b的值。用python实现线性回归模型很简单，直接调包就好了，出来的结果也高大上。matlab实现线性回归也很方便，因为matlab做矩阵运算很方便。我出血java和C，简单做了一个二维的线性回归模型，供大家参考。多维的也好做。最后测是模型性能是用的留出法。

java代码

public class studentdemi {
    public static void main(String[] args) {
        //训练数据集
        double[] x= { 7, 8, 9.5, 10.5, 14, 16, 18, 22, 15 };
        //训练数据标签
        float[] y= { 14, 14, 14, 17, 17, 17, 17, 17, 17 };
        // 根据最小二乘参数估计公式可以得出w和b
        int size_x = x.length;//数据集的长度
        double avg_x=0;
        double sum_x = 0;
        for (int i = 0; i < size_x - 1; i++)
        {
            sum_x = sum_x + x[i];

        }
        avg_x = sum_x / size_x;
        double fz = 0;
        double fm = 0;
        for (int i = 0; i <= size_x - 1; i++)
        {
            fz =fz+ y[i] * (x[i] - avg_x);
            fm =fm+x[i] * x[i] - sum_x * sum_x / size_x;
        }
        double w = fz / fm;
        double b = 0;

        for (int i = 0; i <= size_x - 1; i++)
        {
            b = b + y[i] - w*x[i];

        }
        b = b / size_x;
        //测试集数据
        double[] x1 = { 9, 10, 7, 22, 14 };
        //测试集标签
        double[] y1 = { 14, 14, 14, 17, 17 };
        double mse = 0;
        double avg = 0;
        for (int i = 0; i <= x1.length - 1; i++)
        {
            avg = avg + x1[i];
        }
        avg = avg / x1.length;
        for (int i = 0; i <= x1.length - 1; i++)
        {
            mse = mse + y1[i] - avg;
        }
        mse = mse / x1.length;
        System.out.println("均方误差："+mse);
    }
}

java其实和C很像，这里java 其实可以定义一些函数 public static double（返回值为double类型）这样可以让代码模块化，更加方便美观。

C语言代码

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <math.h>
#include <time.h>

int main() 
{
	//训练数据集
	float x[] = { 7, 8, 9.5, 10.5, 14, 16, 18, 22, 15 };
	//训练数据标签
	float y[] = { 14, 14, 14, 17, 17, 17, 17, 17, 17 };
	// 根据最小二乘参数估计公式可以得出w和b
	int size_x = sizeof(x) / sizeof(x[0]);//数据集的长度
	float avg_x=0;
	float sum_x = 0;

	for (int i = 0; i < size_x - 1; i++)
	{
		sum_x = sum_x + x[i];

	}
	avg_x = sum_x / size_x;
	float fz = 0;
	float fm = 0;
	for (int i = 0; i <= size_x - 1; i++)
	{
		fz =fz+ y[i] * (x[i] - avg_x);
		fm =fm+x[i] * x[i] - sum_x * sum_x / size_x;
	}
	float w = fz / fm;
	float b = 0;

	for (int i = 0; i <= size_x - 1; i++)
	{
		b = b + y[i] - w*x[i];

	}
	b = b / size_x;
	//测试集数据
	float x1[] = { 9, 10, 7, 22, 14 };
	//测试集标签
	float y1[] = { 14, 14, 14, 17, 17 };
	float mse = 0;
	float avg = 0;
	for (int i = 0; i <= sizeof(x1) / sizeof(x1[0]) - 1; i++)
	{
		avg = avg + x1[i];
	}
	avg = avg / (sizeof(x1) / sizeof(x1[0]));
	for (int i = 0; i <= sizeof(x1) / sizeof(x1[0]) - 1; i++)
	{
		mse = mse + y1[i] - avg;
	}
	mse = mse / (sizeof(x1) / sizeof(x1[0]));
	printf("均方误差:%f\n", mse);

	return 0;
}

这里模型性能指标用的mse最后结果为2.8，其实还挺差劲的，主要原因可能因为数据少，而且数据都是瞎编的，哈哈哈。（欸，不过这样的模型有什么意义呢，下次严谨一点）