java实现多元线性回归，与matlab中计算一致

java实现多元线性回归，与matlab中计算一致

最近公司要做多元线性回归，计算的是某一列与多列的关系，入参是一个一维数组和一个二维数组。

首先，maven引入的依赖，此为apach提供的公共计算包，包含了各种各样的计算模型

<!-- https://mvnrepository.com/artifact/org.apache.commons/commons-math3 -->
    <dependency>
        <groupId>org.apache.commons</groupId>
        <artifactId>commons-math3</artifactId>
        <version>3.6.1</version>
    </dependency>

代码片段：

package success;

import org.apache.commons.math3.stat.descriptive.moment.StandardDeviation;
import org.apache.commons.math3.stat.regression.OLSMultipleLinearRegression;

/**
 * 线性回归
 */
public class ClineNone {
    //测试
    public static void main(String[] args) {
        double [] y ={114, 49, 84, 79, 87, 74, 77, 82, 80, 88, 123, 82, 98, 65, 61, 78, 51, 121, 78, 50, 75, 65, 113, 122, 78, 119, 45, 89, 102, 75};
        //标准化y
        y = dataStandardization(y);
        double[][] x ={{38,13,27,25,18,29,30,20,23,32,38,28,34,19,20,25,16,36,25,17,24,18,30,35,22,34,12,26,29,21},
                {37,15,22,21,29,24,26,27,17,28,34,25,26,21,18,21,16,30,15,14,22,18,32,40,25,34,15,26,32,27},
                {12,13,21,20,20,12,8,17,19,12,25,15,19,11,11,18,13,25,17,12,15,16,24,21,15,25,7,20,21,12},
                {31,29,44,23,21,34,37,36,26,26,18,34,22,30,34,29,50,14,26,36,27,33,23,25,31,18,35,20,21,29},
                {31,29,19,24,26,18,27,24,25,29,27,27,26,32,31,28,35,27,17,25,17,31,22,26,20,22,30,23,23,24}};
        x = dataStandardizationDouble(x);
        OLSMultipleLinearRegression regression = new OLSMultipleLinearRegression();
        regression.newSampleData(y,x);
        double rSquared = regression.calculateRSquared();
        //R方
        System.out.println("R方"+rSquared);
        double[] doubles = regression.estimateRegressionParameters();
        for (double d : doubles) {
            System.out.println("打印: " + d);
        }
        double f = getF(x, y, doubles);
        System.out.println("F:"+f);
    }

    /**
     * 数组进行数据标准化
     *
     */
    public static double[] dataStandardization(double array[]){
        StandardDeviation deviation =new StandardDeviation();
        double sum = 0;
        for(double i : array){
            sum += i;
        }
        //均值
        double avg = sum / array.length;
        //标准差
        double evaluate = deviation.evaluate(array);
        //进行标准化
        for(int i=0;i<array.length;i++){
            array[i]=(array[i] - avg)/evaluate;
        }
        return array;
    }
    /**
     * 标准化多维数组
     *
     */
    public static double[][] dataStandardizationDouble(double arrays[][]){
        double [][] result = new double[arrays[0].length][arrays.length];
        for(int i=0;i<arrays.length;i++){
            double[] doubles = dataStandardization(arrays[i]);
            for(int k=0;k<result.length;k++){
                result[k][i]=doubles[k];
            }
        }
        return result;
    }
    
    /**
     * 线性回归方程拿到F值
     * @param x
     * @param y
     * @param back
     * @return
     */
    public static double getF(double[][] x, double[] y,double[] back){
        if(x.length!=y.length){
            System.out.println("数组不相等");
        }
        double sumY = 0;
        for(double d : y){
            sumY += d;
        }
        double avgY = sumY/y.length;
        //回归的平方和
        double SSR = 0;
        //残差的平方和
        double SSE = 0;
        //y的估值
        double yTemp = 0;
        for(int k=0;k<x.length;k++){
            double temp = 0;
            for(int j =0; j<x[k].length; j++){
                if(j==0){
                    temp += back[j];
                }
                temp +=  x[k][j] * back[j + 1];
            }
             yTemp = temp;
            //回归平方
            temp = Math.pow(temp - avgY,2);
            SSR += temp;
            //残差平方
            SSE += Math.pow(y[k]-yTemp,2);

        }
        //p值，自变量的个数
        int p = x[0].length;
        //n值，为观测总值的个数
        int n = y.length;
        //求F的计算公式 f=(SSR/p)/(SSE/(n-p-1))
        double f = (SSR / p) / (SSE / (n - p - 1));
        return f;
    }


}