package org.neuroph.contrib.model.errorestimation;

import org.neuroph.contrib.model.sampling.RandomSamplingWithRepetition;
import org.neuroph.core.data.DataSet;

import java.util.List;

/**
 * Error estimation method which uses sampling with repetition. Each sample contains N elements which forms
 * so called E632 Bootstrap
 * 
 * U odnosu na KFold nema nikakvu dodatnu logiku, samo drugi sampling i izbor datasetova
 * tako da mogu kfold da spojim sa error estimation i da ovaj dobijam kroz setere ili parametrizaciju
 * 
 * 
 * Vidi konacni kakva j erazlika izmedju ova dva
 * ako je samo smpling metod, onda se moze koristiti korsvalifacija a drugim smpling metodom 
 * 
 * moze da se izvuce ErrorEstimation method kao osnovna klasa
 * 
 * Boostraraping koristi sabsamples ne subsets
 * http://www.faqs.org/faqs/ai-faq/neural-nets/part3/section-12.html
 * 
 * mozes da koristis 632+ botstraping
 *  50 - 2000 subsamples
 * 
 * mogao bih da napravim sve to sa jednom klasom; KFold samo sa drugim sampling algoritmom
 *
 * 
 */
public class Bootstrapping/* extends KFoldCrossValidation */{

    /**
     * Default constructor for creating BootstrapEstimationMethod error estimation
     *
     * @param numberOfSamples defines number of bootstrap samples
     */
    public Bootstrapping(int numberOfSamples) {
       // super(numberOfSamples);
       // setSampling(new RandomSamplingWithRepetition(numberOfSamples));
    }

//    @Override
//    protected DataSet createTrainingSet(List samples, int selectedSampleIndex) {
//        return samples.get(selectedSampleIndex);
//    }

   // @Override
    protected DataSet createTestSet(List samples, int selectedSampleIndex) {
        return samples.get((selectedSampleIndex + 1) % samples.size());
    }


}