package jsat.classifiers.evaluation;

import java.util.Arrays;
import jsat.classifiers.CategoricalData;
import jsat.classifiers.CategoricalResults;

/**
 * This class implements the Balanced Accuracy metric. If the number of test
 * points has an equal total weight for each class, then Balanced Accuracy
 * returns the same result as {@link Accuracy}. If not, this class will re-scale
 * the importance of errors in each class and return an accuracy score as if
 * each class had equal total weight. That is to say, it evaluates the data as
 * if it was balanced. 
*
* See: Brodersen, K. H., Ong, C. S., Stephan, K. E., & * Buhmann, J. M. (2010). * The Balanced Accuracy and Its Posterior Distribution. In Proceedings * of the 2010 20th International Conference on Pattern Recognition (pp. * 3121–3124). Washington, DC, USA: IEEE Computer Society. * http://doi.org/10.1109/ICPR.2010.764 * @author Edward Raff */ public class BalancedAccuracy implements ClassificationScore { private int classes; double[] class_correct; double[] total_class_weight; public BalancedAccuracy() { } public BalancedAccuracy(BalancedAccuracy toClone) { this.classes = toClone.classes; if(toClone.class_correct != null) this.class_correct = Arrays.copyOf(toClone.class_correct, toClone.class_correct.length); if(toClone.total_class_weight != null) this.total_class_weight = Arrays.copyOf(toClone.total_class_weight, toClone.total_class_weight.length); } @Override public double getScore() { double score = 0; for(int i = 0; i < classes; i++) { if(total_class_weight[i] > 1e-15) score += class_correct[i]/total_class_weight[i]; else score += 1; } score /= classes; return score; } @Override public boolean lowerIsBetter() { return false; } @Override public BalancedAccuracy clone() { return new BalancedAccuracy(this); } @Override public String getName() { return "BalancedAccuracy"; } @Override public int hashCode() { return getName().hashCode(); } @Override public boolean equals(Object obj) { if(this.getClass().isAssignableFrom(obj.getClass()) && obj.getClass().isAssignableFrom(this.getClass())) { return true; } return false; } @Override public void prepare(CategoricalData toPredict) { classes = toPredict.getNumOfCategories(); total_class_weight = new double[classes]; class_correct = new double[classes]; } @Override public void addResult(CategoricalResults prediction, int trueLabel, double weight) { total_class_weight[trueLabel] += weight; if(prediction.mostLikely() == trueLabel) class_correct[trueLabel] += weight; } @Override public void addResults(ClassificationScore other) { if(other instanceof BalancedAccuracy) { BalancedAccuracy o = (BalancedAccuracy) other; for(int i = 0; i < classes; i++) { this.class_correct[i] += o.class_correct[i]; this.total_class_weight[i] += o.total_class_weight[i]; } } } }

Ads help maintain this website.