#22 Allow different implementations of Initialization interface

[Anonymous]

For an application I'm considering I would like to be able to provide my own implementation of the Initialization interface to create an initial population (rather than always using RandomInitialization).

I'm quite happy to make the necessary changes myself - I'm just not sure what the procedure is for submitting patches. If there isn't one, I'd likely just post my modified MOEA Framework to GitHub.

[D. Hadka]

This sounds like a great addition! I've created a Git repo with the latest source code so that you can submit the changes. You should be able to clone with:

git clone https://@git.code.sf.net/p/moeaframework/git moeaframework-git

[Brad]

Hello! Is there currently a way to insert a single individual into the randomly initialized population at initialization?

[Anonymous]

Hi Brad,

There currently isn't a way to insert a single individual using the default Executor. However, you could manually create a new algorithm with a custom initialization procedure:

import org.moeaframework.algorithm.NSGAII;
import org.moeaframework.core.Initialization;
import org.moeaframework.core.NondominatedPopulation;
import org.moeaframework.core.NondominatedSortingPopulation;
import org.moeaframework.core.Problem;
import org.moeaframework.core.Solution;
import org.moeaframework.core.Variation;
import org.moeaframework.core.comparator.ChainedComparator;
import org.moeaframework.core.comparator.CrowdingComparator;
import org.moeaframework.core.comparator.ParetoDominanceComparator;
import org.moeaframework.core.operator.RandomInitialization;
import org.moeaframework.core.operator.TournamentSelection;
import org.moeaframework.core.spi.OperatorFactory;
import org.moeaframework.problem.DTLZ.DTLZ2;
import org.moeaframework.util.TypedProperties;

public class Test {

    public static void main(String[] args) {
        final Problem problem = new DTLZ2(2);
        final int populationSize = 100;

        Initialization initialization = new Initialization() {
            public Solution[] initialize() {
                Solution[] result = new RandomInitialization(problem, populationSize).initialize();
                //result[0] = your custom solution
                return result;
            }
        };

        NondominatedSortingPopulation population = 
                new NondominatedSortingPopulation();

        TournamentSelection selection = new TournamentSelection(2, 
                new ChainedComparator(
                        new ParetoDominanceComparator(),
                        new CrowdingComparator()));

        Variation variation = OperatorFactory.getInstance().getVariation(null, 
                new TypedProperties(), problem);

        NSGAII algorithm = new NSGAII(problem, population, null, selection, variation,
                initialization);

        while (!algorithm.isTerminated() && (algorithm.getNumberOfEvaluations() < 100000)) {
            algorithm.step();
        }

        NondominatedPopulation result = algorithm.getResult();
    }

}

[Brad]

Thanks for your response! I believe that will work to use
results[0].setVariable( ) with my solution. I will try it here shortly.
Brad

On Mon, Nov 10, 2014 at 7:19 AM, noreply@in.sf.net wrote:

Hi Brad,

There currently isn't a way to insert a single individual using the
default Executor. However, you could manually create a new algorithm with a
custom initialization procedure:

import org.moeaframework.algorithm.NSGAII;import org.moeaframework.core.Initialization;import org.moeaframework.core.NondominatedPopulation;import org.moeaframework.core.NondominatedSortingPopulation;import org.moeaframework.core.Problem;import org.moeaframework.core.Solution;import org.moeaframework.core.Variation;import org.moeaframework.core.comparator.ChainedComparator;import org.moeaframework.core.comparator.CrowdingComparator;import org.moeaframework.core.comparator.ParetoDominanceComparator;import org.moeaframework.core.operator.RandomInitialization;import org.moeaframework.core.operator.TournamentSelection;import org.moeaframework.core.spi.OperatorFactory;import org.moeaframework.problem.DTLZ.DTLZ2;import org.moeaframework.util.TypedProperties;
public class Test {

public static void main(String[] args) {
    final Problem problem = new DTLZ2(2);
    final int populationSize = 100;

    Initialization initialization = new Initialization() {
        public Solution[] initialize() {
            Solution[] result = new RandomInitialization(problem, populationSize).initialize();
            //result[0] = your custom solution
            return result;
        }
    };

    NondominatedSortingPopulation population =
            new NondominatedSortingPopulation();

    TournamentSelection selection = new TournamentSelection(2,
            new ChainedComparator(
                    new ParetoDominanceComparator(),
                    new CrowdingComparator()));

    Variation variation = OperatorFactory.getInstance().getVariation(null,
            new TypedProperties(), problem);

    NSGAII algorithm = new NSGAII(problem, population, null, selection, variation,
            initialization);

    while (!algorithm.isTerminated() && (algorithm.getNumberOfEvaluations() < 100000)) {
        algorithm.step();
    }

    NondominatedPopulation result = algorithm.getResult();
}

}

---

• [feature-requests:#22]
  http://sourceforge.net/p/moeaframework/feature-requests/22 Allow
  different implementations of Initialization interface*

Status: open
Group: Any Version
Created: Thu Apr 24, 2014 11:31 PM UTC by Anonymous
Last Updated: Wed Nov 05, 2014 07:34 PM UTC
Owner: nobody

For an application I'm considering I would like to be able to provide my
own implementation of the Initialization interface to create an initial
population (rather than always using RandomInitialization).

I'm quite happy to make the necessary changes myself - I'm just not sure
what the procedure is for submitting patches. If there isn't one, I'd
likely just post my modified MOEA Framework to GitHub.

---

Sent from sourceforge.net because you indicated interest in
https://sourceforge.net/p/moeaframework/feature-requests/22/

To unsubscribe from further messages, please visit
https://sourceforge.net/auth/subscriptions/