source: trunk/CrossPare/src/de/ugoe/cs/cpdp/dataselection/SeparatabilitySelection.java @ 29

package de.ugoe.cs.cpdp.dataselection;

import java.util.Arrays;
import java.util.Random;

import org.apache.commons.collections4.list.SetUniqueList;

import weka.classifiers.Evaluation;
import weka.classifiers.functions.Logistic;
import weka.core.DenseInstance;
import weka.core.Instance;
import weka.core.Instances;

/**
 * A setwise data selection strategy based on the separatability of the training data from the test data after Z. He, F. Peters, T. Menzies, Y. Yang: Learning from Open-Source Projects: An Empirical Study on Defect Prediction.
 * <br><br>
 * This is calculated through the error of a logistic regression classifier that tries to separate the sets. 
 * @author Steffen Herbold
 */
public class SeparatabilitySelection implements ISetWiseDataselectionStrategy {

        /**
         * size of the random sample that is drawn from both test data and training data 
         */
        private final int sampleSize = 500;
        
        /**
         * number of repetitions of the sample drawing
         */
        private final int maxRep = 10;
        
        /**
         * number of neighbors that are selected
         */
        private int neighbors = 10;
        
        /**
         * Sets the number of neighbors that are selected.
         */
        @Override
        public void setParameter(String parameters) {
                if( !"".equals(parameters) ) {
                        neighbors = Integer.parseInt(parameters);
                }
        }

        /**
         * @see de.ugoe.cs.cpdp.dataselection.SetWiseDataselectionStrategy#apply(weka.core.Instances, org.apache.commons.collections4.list.SetUniqueList)
         */
        @Override
        public void apply(Instances testdata, SetUniqueList<Instances> traindataSet) {
                final Random rand = new Random(1);
                
                // calculate distances between testdata and traindata
                final double[] distances = new double[traindataSet.size()]; 
                
                int i=0;
                for( Instances traindata : traindataSet ) {
                        double distance = 0.0;
                        for( int rep=0; rep<maxRep ; rep++ ) {
                                // sample instances
                                Instances sample = new Instances(testdata);
                                for( int j=0; j<sampleSize; j++ ) {
                                        Instance inst = new DenseInstance(testdata.instance(rand.nextInt(testdata.numInstances())));
                                        inst.setDataset(sample);
                                        inst.setClassValue(1.0);
                                        sample.add(inst);
                                        inst = new DenseInstance(traindata.instance(rand.nextInt(traindata.numInstances())));
                                        inst.setDataset(sample);
                                        inst.setClassValue(0.0);
                                        sample.add(inst);
                                }
                                
                                // calculate separation
                                Evaluation eval;
                                try {
                                        eval = new Evaluation(sample);
                                        eval.crossValidateModel(new Logistic(), sample, 5, rand);
                                } catch (Exception e) {
                                        throw new RuntimeException("cross-validation during calculation of separatability failed", e);
                                }
                                distance += eval.pctCorrect()/100.0;
                        }
                        distances[i++] = 2*((distance/maxRep)-0.5);
                }
                
                // select closest neighbors
                final double[] distancesCopy = Arrays.copyOf(distances, distances.length);
                Arrays.sort(distancesCopy);
                final double cutoffDistance = distancesCopy[neighbors];
                
                for( i=traindataSet.size()-1; i>=0 ; i-- ) {
                        if( distances[i]>cutoffDistance ) {
                                traindataSet.remove(i);
                        }
                }
        }
}