Examples with MultiLabel edu.neu.ccs.pyramid.dataset.MultiLabel used on opensource projects

Example 41 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class GeneralF1Predictor method predictWithPMatrix.

public MultiLabel predictWithPMatrix(double[][] pMatrix, double zeroProbability) {
    int numLabels = pMatrix.length;
    int min = Math.min(maxSize, numLabels);
    MultiLabel best = new MultiLabel();
    double bestScore = zeroProbability;
    for (int k = 1; k <= min; k++) {
        double[] deltaVector = getDeltaVector(pMatrix, k);
        Pair<MultiLabel, Double> innerBest = bestWithLengthK(deltaVector, k);
        if (innerBest.getSecond() > bestScore) {
            bestScore = innerBest.getSecond();
            best = innerBest.getFirst();
        }
    }
    return best;
}

Example 42 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class GeneralF1Predictor method showSupportPrediction.

public static Analysis showSupportPrediction(List<MultiLabel> combinations, double[] probs, MultiLabel truth, MultiLabel prediction, int numClasses) {
    int truthIndex = 0;
    for (int i = 0; i < combinations.size(); i++) {
        if (combinations.get(i).equals(truth)) {
            truthIndex = i;
            break;
        }
    }
    double[] trueJoint = new double[combinations.size()];
    trueJoint[truthIndex] = 1;
    double kl = KLDivergence.kl(trueJoint, probs);
    List<Pair<MultiLabel, Double>> list = new ArrayList<>();
    for (int i = 0; i < combinations.size(); i++) {
        list.add(new Pair<>(combinations.get(i), probs[i]));
    }
    Comparator<Pair<MultiLabel, Double>> comparator = Comparator.comparing(a -> a.getSecond());
    List<Pair<MultiLabel, Double>> sorted = list.stream().sorted(comparator.reversed()).filter(pair -> pair.getSecond() > 0.01).collect(Collectors.toList());
    double expectedF1Prediction = expectedF1(combinations, probs, prediction, numClasses);
    double expectedF1Truth = expectedF1(combinations, probs, truth, numClasses);
    double actualF1 = new InstanceAverage(numClasses, truth, prediction).getF1();
    StringBuilder jointString = new StringBuilder();
    for (int i = 0; i < sorted.size(); i++) {
        jointString.append(sorted.get(i).getFirst()).append(":").append(sorted.get(i).getSecond()).append(", ");
    }
    Analysis analysis = new Analysis();
    analysis.expectedF1Prediction = expectedF1Prediction;
    analysis.expectedF1Truth = expectedF1Truth;
    analysis.actualF1 = actualF1;
    analysis.kl = kl;
    analysis.prediction = prediction;
    analysis.truth = truth;
    analysis.joint = jointString.toString();
    return analysis;
}

Example 43 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class GeneralF1Predictor method getPMatrix.

private double[][] getPMatrix(int numClasses, List<MultiLabel> multiLabels, List<Double> probabilities) {
    int min = Math.min(maxSize, numClasses);
    double[][] pMatrix = new double[numClasses][min];
    for (int j = 0; j < multiLabels.size(); j++) {
        MultiLabel multiLabel = multiLabels.get(j);
        double prob = probabilities.get(j);
        int s = multiLabel.getMatchedLabels().size();
        if (s <= maxSize) {
            for (int i : multiLabel.getMatchedLabels()) {
                double old = pMatrix[i][s - 1];
                pMatrix[i][s - 1] = old + prob;
            }
        }
    }
    return pMatrix;
}

Example 44 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class GeneralF1Predictor method predict.

/**
     *
     * @param numClasses
     * @param samples sampled multi-labels; can have duplicates; their empirical probabilities will be estimated
     * @return
     */
public MultiLabel predict(int numClasses, List<MultiLabel> samples) {
    Multiset<MultiLabel> multiset = ConcurrentHashMultiset.create();
    for (MultiLabel multiLabel : samples) {
        multiset.add(multiLabel);
    }
    int sampleSize = samples.size();
    List<MultiLabel> uniqueOnes = new ArrayList<>();
    List<Double> probs = new ArrayList<>();
    for (Multiset.Entry<MultiLabel> entry : multiset.entrySet()) {
        uniqueOnes.add(entry.getElement());
        probs.add((double) entry.getCount() / sampleSize);
    }
    return predict(numClasses, uniqueOnes, probs);
}

Example 45 with MultiLabel

use of edu.neu.ccs.pyramid.dataset.MultiLabel in project pyramid by cheng-li.

the class SubsetAccPredictor method predict.

@Override
public MultiLabel predict(Vector vector) {
    if (imlGradientBoosting.getAssignments() == null) {
        throw new RuntimeException("CRF is used but legal assignments is not specified!");
    }
    double maxScore = Double.NEGATIVE_INFINITY;
    MultiLabel prediction = null;
    double[] classScores = imlGradientBoosting.predictClassScores(vector);
    for (MultiLabel assignment : imlGradientBoosting.getAssignments()) {
        double score = imlGradientBoosting.calAssignmentScore(assignment, classScores);
        if (score > maxScore) {
            maxScore = score;
            prediction = assignment;
        }
    }
    return prediction;
}