source: java/ecj/cecj/archive/CandidateTestArchive.java @ 224

/*
  Copyright 2009 by Marcin Szubert
  Licensed under the Academic Free License version 3.0
 */

package cecj.archive;

import java.util.List;

import cecj.eval.ArchivingCoevolutionaryEvaluator;
import cecj.interaction.LearnerTeacherInteractionScheme;
import cecj.interaction.LearnerTeacherInteractionScheme.Role;
import cecj.problems.TestBasedProblem;

import ec.EvolutionState;
import ec.Individual;
import ec.util.Parameter;

/**
 * Represents archive dedicated for learner-teacher coevolution, where two distinct roles can be
 * distinguished. The first one (leaner) represents candidate solution to the problem and it is
 * rewarded for performing well on tests which are representatives of the second role (teacher).
 * 
 * It requires <code>LearnerTeacherInteractionScheme</code> and <code>TestBasedProblem</code>
 * definition.
 * 
 * @author Marcin Szubert
 * 
 */
public abstract class CandidateTestArchive extends CoevolutionaryArchive {

        /**
         * The interaction scheme between coevolving populations.
         */
        protected LearnerTeacherInteractionScheme interactionScheme;

        /**
         * Problem which determines the interaction form between candidates and tests.
         */
        protected TestBasedProblem problem;

        @Override
        public void setup(EvolutionState state, Parameter base) {
                super.setup(state, base);

                ArchivingCoevolutionaryEvaluator e = (ArchivingCoevolutionaryEvaluator) state.evaluator;
                if (!(e.getInteractionScheme() instanceof LearnerTeacherInteractionScheme)) {
                        state.output
                                        .fatal("This archive can be used only with learner-teacher interaction scheme.\n");
                }
                interactionScheme = (LearnerTeacherInteractionScheme) e.getInteractionScheme();
                problem = e.getProblem();
        }

        @Override
        public void submit(EvolutionState state) {
                List<Integer> candidatePops = interactionScheme.getSubpopulationIndices(Role.LEARNER);
                List<Integer> testPops = interactionScheme.getSubpopulationIndices(Role.TEACHER);
                for (int candidatePop : candidatePops) {
                        List<Individual> cArchive = getArchive(state, candidatePop);
                        List<Individual> candidates = getIndividuals(state, candidatePop);
                        for (int testPop : testPops) {
                                List<Individual> tArchive = getArchive(state, testPop);
                                List<Individual> tests = getIndividuals(state, testPop);
                                submit(state, candidates, cArchive, tests, tArchive);
                        }
                }
        }

        /**
         * Submits new candidates and tests to the archives. The role of the archive is to determine
         * which individuals of both roles are useful and should be kept in the archive. It is possible
         * also to remove some existing individual because new one is strictly better.
         * 
         * @param state
         *            the current evolution state
         * @param candidates
         *            the list of newly generated candidate solutions
         * @param cArchive
         *            the list of archival candidate solutions
         * @param tests
         *            the list of newly generated tests
         * @param tArchive
         *            the list of archival tests
         */
        protected abstract void submit(EvolutionState state, List<Individual> candidates,
                        List<Individual> cArchive, List<Individual> tests, List<Individual> tArchive);
}