Examples with ScoredVec ntuple.ScoredVec used on opensource projects

Example 1 with ScoredVec

use of ntuple.ScoredVec in project SimpleAsteroids by ljialin.

the class PairedDifferenceLearner method learn.

@Override
public int[] learn(List<ScoredVec> scoredVecs, NoisySolutionEvaluator evaluator) {
    int n = evaluator.searchSpace().nDims();
    fitness = new double[scoredVecs.size()];
    GeneArrayModel geneArrayModel = new GeneArrayModel(evaluator.searchSpace());
    for (int i = 0; i < scoredVecs.size(); i++) {
        for (int j = 0; j < i; j++) {
            // System.out.println(i + " <> " + j);
            ScoredVec svi = scoredVecs.get(i);
            ScoredVec svj = scoredVecs.get(j);
            if (diffUpdate) {
                geneArrayModel.updateModelDiff(svi, svj);
            } else {
                if (svi.score > svj.score) {
                    geneArrayModel.updateModel(svi, svj);
                } else {
                    geneArrayModel.updateModel(svj, svi);
                }
            }
        }
        fitness[i] = evaluator.trueFitness(geneArrayModel.argMax());
    }
    // geneArrayModel.report();
    return geneArrayModel.argMax();
}

Example 2 with ScoredVec

use of ntuple.ScoredVec in project SimpleAsteroids by ljialin.

the class MeanLearner method learn.

@Override
public int[] learn(List<ScoredVec> scoredVecs, NoisySolutionEvaluator evaluator) {
    int n = evaluator.searchSpace().nDims();
    fitness = new double[scoredVecs.size()];
    // set up the models
    GeneMeanModel[] models = new GeneMeanModel[n];
    for (int i = 0; i < n; i++) models[i] = new GeneMeanModel(evaluator.searchSpace().nValues(i));
    // now train the,
    int ix = 0;
    for (ScoredVec sv : scoredVecs) {
        // update it here
        for (int i = 0; i < n; i++) {
            models[i].updateMean(sv.p[i], sv.score);
        }
        int[] p = argmax(models);
        // System.out.println(Arrays.toString(p));
        fitness[ix++] = evaluator.trueFitness(p);
    }
    // now return the argmax array
    return argmax(models);
}

Example 3 with ScoredVec

use of ntuple.ScoredVec in project SimpleAsteroids by ljialin.

the class ProblemInstance method useVecsAroundRandomPoint.

public ProblemInstance useVecsAroundRandomPoint() {
    scoredVecs = new ArrayList<>();
    int[] px = SearchSpaceUtil.randomPoint(evaluator.searchSpace());
    DefaultMutator mutator = new DefaultMutator(evaluator.searchSpace());
    // 1.0; // 3 / n;
    mutator.pointProb = 0.0;
    // mutator
    DefaultMutator.flipAtLeastOneValueDefault = true;
    DefaultMutator.totalRandomChaosMutation = false;
    for (int i = 0; i < k; i++) {
        // int[] p = SearchSpaceUtil.randomPoint(searchSpace);
        int[] p = mutator.randMut(px);
        double score = evaluator.evaluate(p);
        scoredVecs.add(new ScoredVec(p, score));
    // System.out.println(Arrays.toString(p) + " : " + String.format("%.2f\t %.0f", score, evaluator.trueFitness(p)));
    }
    return this;
}

Example 4 with ScoredVec

use of ntuple.ScoredVec in project SimpleAsteroids by ljialin.

the class ProblemInstance method useRandomVecs.

public ProblemInstance useRandomVecs() {
    scoredVecs = new ArrayList<>();
    for (int i = 0; i < k; i++) {
        // int[] p = SearchSpaceUtil.randomPoint(searchSpace);
        int[] p = SearchSpaceUtil.randomPoint(evaluator.searchSpace());
        double score = evaluator.evaluate(p);
        scoredVecs.add(new ScoredVec(p, score));
    // System.out.println(Arrays.toString(p) + " : " + String.format("%.2f\t %.0f", score, evaluator.trueFitness(p)));
    }
    return this;
}

Example 5 with ScoredVec

use of ntuple.ScoredVec in project SimpleAsteroids by ljialin.

the class RankCorrelationEDA method runTrial.

@Override
public int[] runTrial(SolutionEvaluator evaluator, int nEvals) {
    this.evaluator = evaluator;
    // set  up some convenient references
    SearchSpace searchSpace = evaluator.searchSpace();
    int n = searchSpace.nDims();
    history = new ArrayList<>();
    geneArrayModel = new GeneArrayMeanModel(searchSpace);
    int nSteps = 0;
    Long endTime = null;
    if (timeLimit != null) {
        // endTime = timeLimit + System.currentTimeMillis();
        endTime = timeLimit + System.nanoTime() / 1000000;
    }
    while (evaluator.nEvals() < nEvals && (endTime == null || System.nanoTime() / 1000000 < endTime)) {
        int prevEvals = evaluator.nEvals();
        // each time around evaluate a single new individual: x
        // but occasionally have the possibility of sampling the best guess so far
        int[] x = geneArrayModel.generate();
        double f = fitness(evaluator, x, nSamples).mean();
        ScoredVec scoredVec = new ScoredVec(x, f);
        // now treat the history like a circular buffer and update it
        // always add the ScoredVector in
        geneArrayModel.updateModelMean(scoredVec);
        // geneArrayModel.report();
        if (history.size() < historyLength) {
            history.add(scoredVec);
        } else {
            // if we're replacing one in the history
            // then remove it from our stats
            int ix = nSteps % historyLength;
            geneArrayModel.removeVec(history.get(ix));
            history.set(ix, scoredVec);
        }
        nSteps++;
        int diffEvals = evaluator.nEvals() - prevEvals;
        for (int i = 0; i < diffEvals; i++) {
            evaluator.logger().logBestYest(geneArrayModel.argMax());
        }
    // if (verbose) {
    // int[] solution = CompactGAUtil.argmax(pVec);
    // System.out.format("%.3f\t %s\n", evaluator.evaluate(solution), Arrays.toString(solution));
    // // System.out.println(Arrays.toString(pVec));
    // for (double p : pVec) {
    // System.out.format("%.4f\t", p);
    // }
    // System.out.println();
    // System.out.println();
    // 
    // }
    }
    // now draw each x and y vec according to pVec
    // indeed, what to return
    // finally, return the argmax of each dimension
    int[] solution = geneArrayModel.argMax();
    // logger.
    evaluator.logger().keepBest(solution, evaluator.evaluate(solution));
    return solution;
}