Manytasking optimization MATP

Manytasking Jmetal代码反向解析1_MATP测试函数集

觉得有用的话,欢迎一起讨论相互学习~Follow Me

• 这是我在写Manytask optimization时的笔记，代码地址可以下载

• 打开文件夹如下目录，选择benchmark中的任意一个测试函数集，此时我们选择MATP1

  MATP1

package momfo.problems.benchmarks;

import java.io.IOException;

import momfo.core.Problem;
import momfo.core.ProblemSet;
import momfo.problems.base.*;

public class MATP1 {

    public static ProblemSet getProblem() throws IOException {

        int taskNumber=50;

        ProblemSet problemSet = new ProblemSet(taskNumber);

        for(int i=1;i<=taskNumber;i++)
            problemSet.add(getT(i).get(0));

        return problemSet;

    }

    public static ProblemSet getT(int taskID) throws IOException {
        ProblemSet problemSet = new ProblemSet(1);

        MMDTLZ prob = new MMDTLZ(2, 50, 1, -100,100);
        prob.setGType("sphere");

        double[][] matrix = IO.readMatrixFromFile("MData/M1/M1_"+taskID+".txt");

        double shiftValues[] = IO.readShiftValuesFromFile("SVData/S1/S1_"+taskID+".txt");

        prob.setRotationMatrix(matrix);
        prob.setShiftValues(shiftValues);       

        ((Problem)prob).setName("MATP1-"+taskID);

        problemSet.add(prob);

        return problemSet;
    }

}

• int taskNumber=50; 表示MATP1问题设置的任务个数是50，初始化problemSet作为一个容器以承载每一个被初始化后的实际MATP1问题。

• for(int i=1;i<=taskNumber;i++) problemSet.add(getT(i).get(0));
  意味着此时需要关注的是getT函数，从

  MMDTLZ prob = new MMDTLZ(2, 50, 1, -100,100);
  prob.setGType("sphere");

  看出MATP_1是一个双目标的问题，并且决策变量个数为50，决策变量最小值为-100，最大值为100，而G函数的类型为sphere,而只要是双目标的函数使用的H函数都是circle
  ```java
  // MMDTLZ
  if (numberOfObjectives == 2)
  hType_ = "circle";
  else
  hType_ = "sphere";

![在这里插入图片描述](https://img-blog.csdnimg.cn/20191216155508697.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTM1NTU3MTk=,size_16,color_FFFFFF,t_70)

* **通过**

double[][] matrix = IO.readMatrixFromFile("MData/M1/M1_"+taskID+".txt");

double shiftValues[] = IO.readShiftValuesFromFile("SVData/S1/S1_"+taskID+".txt");
``* 可知matrix[][]其中保存有旋转函数，根据taskID具有不同的旋转角度，shiftValues则保存了偏移向量，通过不同的旋转函数和偏移向量可以保证MATP问题中的不同任务即使具有相同的帕累托前沿，但是对应的决策变量距离和位置却完全不同。 ![在这里插入图片描述](https://img-blog.csdnimg.cn/20191216145252251.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTM1NTU3MTk=,size_16,color_FFFFFF,t_70) ## 注意在MATP2中仅仅设置G函数的形状为sphere和alpha`值1。(后期通关观察得知，alpha值是用于计算cos函数的)

---

MATP2

public class MATP2 {

    public static ProblemSet getProblem() throws IOException {

        int taskNumber=50;

        ProblemSet problemSet = new ProblemSet(taskNumber);

        for(int i=1;i<=taskNumber;i++)
            problemSet.add(getT(i).get(0));

        return problemSet;

    }

    public static ProblemSet getT(int taskID) throws IOException {
        ProblemSet problemSet = new ProblemSet(1);

        MMZDT prob = new MMZDT(50, 1,  -100,100);
        prob.setGType("mean");
        prob.setHType("concave");

        double[][] matrix = IO.readMatrixFromFile("MData/M2/M2_"+taskID+".txt");

        double shiftValues[] = IO.readShiftValuesFromFile("SVData/S2/S2_"+taskID+".txt");

        prob.setRotationMatrix(matrix);
        prob.setShiftValues(shiftValues);       

        ((Problem)prob).setName("MATP2-"+taskID);

        problemSet.add(prob);

        return problemSet;
    }

}

和MATP1问题不同的是MATP2问题设置G函数为mean,设置H函数为concave,最终形状为concave并且是根据ZDT问题改编的，而不是DTLZ问题，表示有50个变量，并且K值为1，根据ZDT问题的默认函数来看，ZDT问题构造的都是双目标问题。

public MMZDT(int numberOfVariables, int k, double lg, double ug) {
  numberOfObjectives_ = 2;
  numberOfVariables_ = numberOfVariables;
  k_ = k;

  gType_ = "sphere";
  f1Type_ = "linear";
  hType_ = "convex";

---

MATP3

package momfo.problems.benchmarks;

import java.io.IOException;

import momfo.core.Problem;
import momfo.core.ProblemSet;
import momfo.problems.base.*;

public class MATP3 {

    public static ProblemSet getProblem() throws IOException {

        int taskNumber=50;

        ProblemSet problemSet = new ProblemSet(taskNumber);

        for(int i=1;i<=taskNumber;i++)
            problemSet.add(getT(i).get(0));

        return problemSet;

    }

    public static ProblemSet getT(int taskID) throws IOException {
        ProblemSet problemSet = new ProblemSet(1);

        MMZDT prob = new MMZDT(10,  1, -5,5);
        prob.setGType("rosenbrock");
        prob.setHType("concave");

        double[][] matrix = IO.readMatrixFromFile("MData/M3/M3_"+taskID+".txt");

        double shiftValues[] = IO.readShiftValuesFromFile("SVData/S3/S3_"+taskID+".txt");

        prob.setRotationMatrix(matrix);
        prob.setShiftValues(shiftValues);       

        ((Problem)prob).setName("MATP3-"+taskID);

        problemSet.add(prob);

        return problemSet;
    }

}

和MATP1以及MATP2问题不同的是MATP3问题设置G函数为rosenbrock,设置H函数为concave,最终形状为concave并且是根据ZDT问题改编的，而不是DTLZ问题，表示有50个这样的任务，并且K值为1，根据ZDT问题的默认函数来看，ZDT问题构造的都是双目标问题,与MATP2问题不同的是使用10个变量，并且变量范围为[-5,5]而不是[-50,50]

---

MATP4

import java.io.IOException;

import momfo.core.Problem;
import momfo.core.ProblemSet;
import momfo.problems.base.*;

public class MATP4 {

    public static ProblemSet getProblem() throws IOException {

        int taskNumber=50;

        ProblemSet problemSet = new ProblemSet(taskNumber);

        for(int i=1;i<=taskNumber;i++)
            problemSet.add(getT(i).get(0));

        return problemSet;

    }

    public static ProblemSet getT(int taskID) throws IOException {
        ProblemSet problemSet = new ProblemSet(1);

        MMDTLZ prob = new MMDTLZ(2, 50, 1, -2,2);
        prob.setGType("rastrigin");

        double[][] matrix = IO.readMatrixFromFile("MData/M4/M4_"+taskID+".txt");

        double shiftValues[] = IO.readShiftValuesFromFile("SVData/S4/S4_"+taskID+".txt");

        prob.setRotationMatrix(matrix);
        prob.setShiftValues(shiftValues);       

        ((Problem)prob).setName("MATP4-"+taskID);

        problemSet.add(prob);

        return problemSet;
    }

}

可以看出MATP4使用的是DTLZ函数其中具有两个目标函数，具有50个决策变量，alpha值为1，上界为2下界为-2.使用的G函数为rastrigin,使用的H函数为sphere

//MMDTLZ
if (numberOfObjectives == 2)
            hType_ = "circle";
        else
            hType_ = "sphere";

---

MATP5

package momfo.problems.benchmarks;

import java.io.IOException;

import momfo.core.Problem;
import momfo.core.ProblemSet;
import momfo.problems.base.*;

public class MATP5 {

    public static ProblemSet getProblem() throws IOException {

        int taskNumber=50;

        ProblemSet problemSet = new ProblemSet(taskNumber);

        for(int i=1;i<=taskNumber;i++)
            problemSet.add(getT(i).get(0));

        return problemSet;

    }

    public static ProblemSet getT(int taskID) throws IOException {
        ProblemSet problemSet = new ProblemSet(1);

        MMZDT prob = new MMZDT(50, 1,  -1,1);
        prob.setGType("ackley");
        prob.setHType("convex");

        double[][] matrix = IO.readMatrixFromFile("MData/M5/M5_"+taskID+".txt");

        double shiftValues[] = IO.readShiftValuesFromFile("SVData/S5/S5_"+taskID+".txt");

        prob.setRotationMatrix(matrix);
        prob.setShiftValues(shiftValues);       

        ((Problem)prob).setName("MATP5-"+taskID);

        problemSet.add(prob);

        return problemSet;
    }

}

决策变量个数为50，k为1，变量最小范围为-1，变量最大范围为1，G函数为ackley,H函数为convex

---

MATP6

package momfo.problems.benchmarks;

import java.io.IOException;

import momfo.core.Problem;
import momfo.core.ProblemSet;
import momfo.problems.base.*;

public class MATP6 {

    public static ProblemSet getProblem() throws IOException {

        int taskNumber=50;

        ProblemSet problemSet = new ProblemSet(taskNumber);

        for(int i=1;i<=taskNumber;i++)
            problemSet.add(getT(i).get(0));

        return problemSet;

    }

    public static ProblemSet getT(int taskID) throws IOException {
        ProblemSet problemSet = new ProblemSet(1);

        MMDTLZ prob = new MMDTLZ(2, 50, 1, -50,50);
        prob.setGType("griewank");

        double[][] matrix = IO.readMatrixFromFile("MData/M6/M6_"+taskID+".txt");

        double shiftValues[] = IO.readShiftValuesFromFile("SVData/S6/S6_"+taskID+".txt");

        prob.setRotationMatrix(matrix);
        prob.setShiftValues(shiftValues);       

        ((Problem)prob).setName("MATP6-"+taskID);

        problemSet.add(prob);

        return problemSet;
    }

}

目标函数2个，50个决策变量，alpha值设置为1，下边界为-50，上边界为50，G函数为griewank,H函数为circle

MATP算例总结

float[] rgb1 = {0, 0, 0};//黑色-circle
        float[] rgb2 = {(float) 0, (float) 0, (float) 139};//蓝黑-concave
        float[] rgb3 = {(float) 255, (float) 0, (float) 0};//红色-concave
        float[] rgb4 = {(float) 0, (float) 255, (float) 0};//绿色-circle
        float[] rgb5 = {(float) 255, (float) 255, (float) 0};//黄色-convex
        float[] rgb6 = {(float) 139, (float) 69, (float) 19};//巧克力色-circle

        float alpha = (float) 1;
        Scatter scatter1 = tools.Setpoint(TruePFT1_Matrix, rgb1, alpha, 20);
        Scatter scatter2 = tools.Setpoint(TruePFT2_Matrix, rgb2, alpha, 15);
        Scatter scatter3 = tools.Setpoint(TruePFT3_Matrix, rgb3, alpha, 8);
        Scatter scatter4 = tools.Setpoint(TruePFT4_Matrix, rgb4, alpha, 10);
        Scatter scatter5 = tools.Setpoint(TruePFT5_Matrix, rgb5, alpha, 6);
        Scatter scatter6 = tools.Setpoint(TruePFT6_Matrix, rgb6, alpha, 5);