Fix function evaluations in ga

I want to fix my number of maximum function evaluation as stopping criteria in ga. How can i do it?

[Geoff Hayes]

Sandeep - do you mean that you want the genetic algorithm to iterate for some maximum number of iterations/generations?

[sandeep singh chauhan]

No, I want to use genetic algorithm that will stop with maximum function evaluations

Answer 1

[Brendan Hamm]

This is controlled in the Generations property of the Genetic Algorithm options. Obtain the default options:

>> options = gaoptimset(@ga)
options = 
        PopulationType: 'doubleVector'
          PopInitRange: []
        PopulationSize: '50 when numberOfVariables <= 5, else 200'
            EliteCount: '0.05*PopulationSize'
     CrossoverFraction: 0.8000
        ParetoFraction: []
    MigrationDirection: 'forward'
     MigrationInterval: 20
     MigrationFraction: 0.2000
           Generations: '100*numberOfVariables'
             TimeLimit: Inf
          FitnessLimit: -Inf
         StallGenLimit: 50
             StallTest: 'averageChange'
        StallTimeLimit: Inf
                TolFun: 1.0000e-06
                TolCon: 1.0000e-03
     InitialPopulation: []
         InitialScores: []
    NonlinConAlgorithm: 'auglag'
        InitialPenalty: 10
         PenaltyFactor: 100
          PlotInterval: 1
           CreationFcn: @gacreationuniform
     FitnessScalingFcn: @fitscalingrank
          SelectionFcn: @selectionstochunif
          CrossoverFcn: @crossoverscattered
           MutationFcn: {@mutationgaussian  [1]  [1]}
    DistanceMeasureFcn: []
             HybridFcn: []
               Display: 'final'
              PlotFcns: []
            OutputFcns: []
            Vectorized: 'off'
           UseParallel: 0

Now you can change the Generations and pass this in to the options input of your call to ga

>> options.Generations = '200*numberOfVariables';
>> ga(...,options); % Fill in the ... with your function, constraints, etc.

[sandeep singh chauhan]

I don't want to use 'Max Generations' as my stopping criteria. I want to use maximum function evaluations

[Brendan Hamm]

The two are proportional. The constant of proportionality will be dependent on your constraints.

[UbuntuXenial]

Can you please explain how to determine the dependence? I want to run GA/MOGA only up to certain MaxFunctionEvaluations but this isn't a valid optimoption for these algorithms.

[Emily Kambalame]

Hie sandeep singh chauhan, Brendan Hamm,UbuntuXenial

I also dont want to use the max generation but rather the maximum function evaluation as my stopping condition. How should i approach my problem? Take note that my constraints are satified using Epanet simulator and am NSGA 2 to optimise.

Answer 2

[Matt J]

The following nested function strategy might be a way to cheat. Basically, it keeps a running count, funEvals, of the number of calls to the fitness function. When funEvals exceeds a specified MaxFunEvals, the FitnessFcn output is forced to -Inf, which I think must trigger the FitnessLimit stopping criterion.

function runEverything(MaxFunEvals)

    funEvals=0;
    bestFitness=inf;

    [x,fval,exitflag,output,population,scores]  = ga(@FitnessFcn,...);

      fval=bestFitness;

   function val = FitnessFcn(x)

        ...

        bestFitness=min(bestFitness,val);

        funEvals=funEvals+1; %increment counter

        if funEvals>=MaxFunEvals
            val=-inf;
        end

   end

end