Examples with EvalMaxM evodef.EvalMaxM used on opensource projects

Example 1 with EvalMaxM

use of evodef.EvalMaxM in project SimpleAsteroids by ljialin.

the class BanditLandscapeEATest method main.

public static void main(String[] args) {
    int nDims = 5;
    int mValues = 4;
    double noiseLevel = 1.0;
    double kExplore = 10;
    boolean useTrap = true;
    // EvalMaxM is like Noisy OneMax but generalised to M values
    // instead of binary
    EvalMaxM problem = new EvalMaxM(nDims, mValues, noiseLevel).setTrap(useTrap);
    NTupleBanditEA banditEA = new NTupleBanditEA().setKExplore(kExplore);
    // set a particlar NTuple System as the model
    // if this is not set, then it will use a default model
    NTupleSystem model = new NTupleSystem();
    // set up a non-standard tuple pattern
    model.use1Tuple = true;
    model.use2Tuple = false;
    model.useNTuple = true;
    banditEA.setModel(model);
    ElapsedTimer timer = new ElapsedTimer();
    int nEvals = 500;
    int[] solution = banditEA.runTrial(problem, nEvals);
    System.out.println("Report: ");
    new NTupleSystemReport().setModel(model).printDetailedReport();
    new NTupleSystemReport().setModel(model).printSummaryReport();
    System.out.println("Model created: ");
    System.out.println(model);
    System.out.println("Model used: ");
    System.out.println(banditEA.getModel());
    System.out.println();
    System.out.println("Solution returned: " + Arrays.toString(solution));
    System.out.println("Solution fitness:  " + problem.trueFitness(solution));
    System.out.println("k Explore: " + banditEA.kExplore);
    System.out.println(timer);
}

Example 2 with EvalMaxM

use of evodef.EvalMaxM in project SimpleAsteroids by ljialin.

the class EvoSGATest method runTrial.

public static double runTrial(EvoAlg evoAlg) {
    // ok, so the idea here is to modify ...
    SimpleGASearchSpace eval = new SimpleGASearchSpace();
    System.out.println("Search space size: " + SearchSpaceUtil.size(eval.searchSpace()));
    eval.setEvaluator(new EvalMaxM(50, 2, 0.0));
    int[] solution = evoAlg.runTrial(eval, 5000);
    // int[] solution = {0, 0, 0};
    System.out.println("Checking fitness");
    StatSummary ss = new StatSummary("Mean fitness");
    int nChecks = 50;
    for (int i = 0; i < nChecks; i++) {
        ss.add(eval.evaluate(solution));
    }
    System.out.println(ss);
    System.out.println("Solution: " + Arrays.toString(solution));
    System.out.println(eval.report(solution));
    return ss.mean();
}

Example 3 with EvalMaxM

use of evodef.EvalMaxM in project SimpleAsteroids by ljialin.

the class TestSimpleGA method main.

public static void main(String[] args) {
    NoisySolutionEvaluator evaluator = new EvalMaxM(10, 2, 1.0);
    EvoAlg evoAlg = new SimpleGA().setPopulationSize(100);
    int[] solution = evoAlg.runTrial(evaluator, 10000);
    System.out.println(Arrays.toString(solution));
}

Example 4 with EvalMaxM

use of evodef.EvalMaxM in project SimpleAsteroids by ljialin.

the class TestPlotter method main.

public static void main(String[] args) {
    int nDims = 5;
    int mValues = 4;
    double noiseLevel = 1.0;
    double kExplore = 10;
    boolean useTrap = true;
    // EvalMaxM is like Noisy OneMax but generalised to M values
    // instead of binary
    EvalMaxM problem = new EvalMaxM(nDims, mValues, noiseLevel).setTrap(useTrap);
    NTupleBanditEA banditEA = new NTupleBanditEA().setKExplore(kExplore);
    // set a particlar NTuple System as the model
    // if this is not set, then it will use a default model
    NTupleSystem model = new NTupleSystem();
    // set up a non-standard tuple pattern
    model.use1Tuple = true;
    model.use2Tuple = true;
    model.useNTuple = false;
    banditEA.setModel(model);
    ElapsedTimer timer = new ElapsedTimer();
    int nEvals = 200;
    int[] solution = banditEA.runTrial(problem, nEvals);
    System.out.println(Arrays.toString(solution));
    System.out.println(timer);
    new Plotter().setModel(model).defaultPlot().plot1Tuples();
    System.out.println(timer);
}

Example 5 with EvalMaxM

use of evodef.EvalMaxM in project SimpleAsteroids by ljialin.

the class PowerOfDifferencePairsTest method main.

// okay this is interesting: the paired idea does not work well when the
// vectors are far apart
// this might have been expected from the way that having the
// sliding history window too big causes deterioration in performance
public static void main(String[] args) {
    // create the random vectors, score them
    // and put them in a list
    // now run an experiment each way to determine the arg max
    // and then evaluate the quality of that
    ScoredVectorLearner meanLearner = new MeanLearner();
    ScoredVectorLearner diffLearner = new PairedDifferenceLearner();
    ScoredVectorLearner[] learners = new ScoredVectorLearner[] { meanLearner, diffLearner };
    int nTrials = 30;
    int n = 100, m = 2;
    double noise = 1.0;
    NoisySolutionEvaluator evaluator = new EvalMaxM(n, m, noise);
    int k = 500;
    List<StatSummary> stats = new ArrayList<>();
    for (ScoredVectorLearner learner : learners) {
        stats.add(new StatSummary(learner.getClass().getSimpleName()));
    }
    LineChart lineChart = new LineChart();
    for (int i = 0; i < nTrials; i++) {
        // ProblemInstance problem = new ProblemInstance(n, m, k, evaluator).useRandomVecs();
        ProblemInstance problem = new ProblemInstance(n, m, k, evaluator).useVecsAroundRandomPoint();
        int ix = 0;
        for (ScoredVectorLearner learner : learners) {
            System.out.println("Testing: " + learner.getClass().getSimpleName());
            int[] p = learner.learn(problem.scoredVecs, problem.evaluator);
            // System.out.println(Arrays.toString(p));
            System.out.println("True fitness is: " + evaluator.trueFitness(p));
            stats.get(ix).add(evaluator.trueFitness(p));
            System.out.println();
            // now show evolution of fitness
            // for (double x : learner.getFitness()) {
            // System.out.println(x);
            // }
            Color color = ix++ % 2 == 0 ? Color.red : Color.blue;
            LinePlot linePlot = new LinePlot().setData(learner.getFitness()).setColor(color);
            lineChart.addLine(linePlot);
            System.out.println(learner.getFitness().length);
        }
    }
    new JEasyFrame(lineChart, "Fitness v. vectors processes");
    for (StatSummary ss : stats) System.out.println(ss);
}