how to write suitable stop criteria to my GA code

Dear,

I used genetic algorithm from matlab toolbox and I got a good min result using this options.

function [x,fval,exitflag,output,population,score] = optim(nvars,PopulationSize_Data,Generations_Data,InitialPopulation_Data)
%%This is an auto generated MATLAB file from Optimization Tool.

%%Start with the default options
options = gaoptimset;
%%Modify options setting
options = gaoptimset(options,'PopulationSize', PopulationSize_Data);
options = gaoptimset(options,'Generations', Generations_Data);
options = gaoptimset(options,'InitialPopulation', InitialPopulation_Data);
options = gaoptimset(options,'Display', 'off');
options = gaoptimset(options,'PlotFcns', { @gaplotbestf @gaplotbestindiv});
%options = gaoptimset(options,'OutputFcns', { @outputBox });
[x,fval,exitflag,output,population,score] = ...
ga(@fitness,nvars,[],[],[],[],[],[],[],[],options);

But because it cost a lot off time I wrote my code by my self using the same options. See the code

clc;clear all;close all;format short g;
for ki=1:1
clear bestSofar 
clear totalFitness
tic;
m=2;
hold on

populationSize=200; 
populationSize  = 4*ceil(populationSize/4);
corssOverPercentage=0.8;
noOfElites=10;
noOfGeneration=200;

%%=========================================
 noOfCrossOvers=corssOverPercentage*populationSize;
 noOFMutaion=10;
%%Initial population
  for i=1: populationSize
here  initialize population  x;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
z=0;
for l=1:noOfGeneration
y=x;

for i=1:populationSize
% %  MY fitness function for all the population
FitnessValue;
   end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 % Normlization
    xFit= FitnessValue';
    fitCumSum=1/cumsum(xFit) ;  % i change this to 1/cumsum(xFit) to get minimum
    rouletteWheel=fitCumSum/fitCumSum(end);
    y=[y,xFit];
     [j,k]=sort(y(:,end),'descend');
     y=y(k,:); % sort rows according to last column
     y=y(:,1:end-1);
     size(y);

   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
    for crindex=1:2:noOfCrossOvers-1
          ran=rand;
          for i=1:populationSize-1
               if (ran>rouletteWheel(i))
                  parent1=x(i+1,:);
                  parent1Index=i+1;

              end
          end
          ran=rand;
          for i=1:populationSize-1
              if (ran>rouletteWheel(i))
                  parent2=x(i+1,:);
                  parent2Index=i+1;

              end
          end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%    
 % Scattered crossover
% here my crossover       
    end  %noOfCrossOvers
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%mutation Gaussian
% here my mtation  
    end

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
   x=y;
    totalFitness(l)=sum(xFit);
    sortedFit=sort(xFit,'descend'); %minimisation
    %sortedFit=sort(xFit,'ascend'); %max
    bestSofar(l)=sortedFit(end);

      if l>1 & bestSofar(l)< bestSofar(l-1)
          z=0;
      else
          z=z+1;
      end
      if z==500 break
      end

end
finalSolution=x(end,:) %Best solution

end %ki
figure;
z=finalSolution;
Plot(z);
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

from this code I got a result but is not as good as the code from toolbox. I checked the selection operator it has no effect on the result. this code works perfect for 2 dimensions with number generation=100, but for m>=3 I got this different in result, I need to minimize a shape but is not as minimum as the result from matlab toolbox. Can any one tell me what is missed in it. I tried to read some code from the GA toolbox to see what is missed but I couldn't.

I checked and I found when I set the generation less than 100 I got not bad result, the problem I need the GA stop when the result reach the minimum value by itself without specifying Number generation by my self. I think the stop criteria in this code is not reliable like the GA from toolbox.

Nadia

Thanks for any help

Answer 1

[Nadiia Huskova]

Dear Nadia,

I also looking for optimal stopping criteria for the GA. I think, number of iterations and time criteria are not good ideas. The original GA (from toolbox) calculates values of standard deviation after every iteration and compares results every time.