de.julielab.geneexpbase.classification.SVMClassifier Maven / Gradle / Ivy

Go to download
package de.julielab.geneexpbase.classification;

import cc.mallet.classify.Classification;
import cc.mallet.classify.Classifier;
import cc.mallet.types.Instance;
import cc.mallet.types.InstanceList;
import cc.mallet.types.LabelVector;
import de.julielab.geneexpbase.classification.svm.SVM;
import de.julielab.geneexpbase.classification.svm.SVMModel;
import de.julielab.geneexpbase.classification.svm.SVMTrainOptions;
import libsvm.svm_model;

/**
 * This is a MALLET compatible wrapper around LibSVM.
 */
public class SVMClassifier extends Classifier {

    private SVMModel model;

    @Override
    public Classification classify(Instance instance) {
        svm_model svmModel = model.svmModel;
        if (!model.trainOptions.probability && svmModel.nr_class != 2)
            throw new IllegalArgumentException("This method can only be used for two-class problems or when the SVM was trained with probability estimates.");
        double[] scores = SVM.predict(instance, model);
        if (scores.length == 1) {
            // no probabilities; the sole value is "class 1 vs class 2" as hyperplane distance
            double[] fullScores = new double[2];
            fullScores[0] =  scores[0];
            // for the other class, the perspective is reversed
            fullScores[1] = -1 * scores[0];
            scores = fullScores;
        }
        // Reorder the SVM scores to match the Label indices.
        // The SVM labels are just the indices of the MALLET labels.
        // Thus we just need to put the values to the position that the SVM label points to.
        double[] orderedScores = new double[scores.length];
        for (int i = 0; i < scores.length; i++) {
            int labelIndex = svmModel.label[i];
            orderedScores[labelIndex] = scores[i];
        }
        return new Classification (instance, this,
                new LabelVector(getLabelAlphabet(),
                        scores));
    }

    public void train(InstanceList instances, SVMTrainOptions options) {
        model = SVM.train(instances, options);
        instancePipe = instances.getPipe();
    }
}