package tr.gov.ulakbim.jDenetX.classifiers;
/*
* OzaBoostAdwin.java
* Copyright (C) 2010 University of Waikato, Hamilton, New Zealand
* @author Albert Bifet (abifet@cs.waikato.ac.nz)
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
*/
import tr.gov.ulakbim.jDenetX.core.DoubleVector;
import tr.gov.ulakbim.jDenetX.core.Measurement;
import tr.gov.ulakbim.jDenetX.core.MiscUtils;
import tr.gov.ulakbim.jDenetX.core.SizeOf;
import tr.gov.ulakbim.jDenetX.options.ClassOption;
import tr.gov.ulakbim.jDenetX.options.FlagOption;
import tr.gov.ulakbim.jDenetX.options.FloatOption;
import tr.gov.ulakbim.jDenetX.options.IntOption;
import weka.core.Instance;
public class OzaBoostAdwin extends AbstractClassifier {
private static final long serialVersionUID = 1L;
public ClassOption baseLearnerOption = new ClassOption("baseLearner", 'l',
"Classifier to train.", Classifier.class, "HoeffdingTree");
public IntOption ensembleSizeOption = new IntOption("ensembleSize", 's',
"The number of models to boost.", 10, 1, Integer.MAX_VALUE);
public FlagOption pureBoostOption = new FlagOption("pureBoost", 'p',
"Boost with weights only; no poisson.");
public FloatOption deltaAdwinOption = new FloatOption("deltaAdwin", 'a',
"Delta of Adwin change detection", 0.002, 0.0, 1.0);
public FlagOption outputCodesOption = new FlagOption("outputCodes", 'o',
"Use Output Codes to use binary classifiers.");
public FlagOption sammeOption = new FlagOption("same", 'e',
"Use Samme Algorithm.");
protected Classifier[] ensemble;
protected double[] scms;
protected double[] swms;
protected ADWIN[] ADError;
protected int numberOfChangesDetected;
protected int[][] matrixCodes;
protected boolean initMatrixCodes = false;
protected double logKm1 = 0.0;
protected int Km1 = 1;
protected boolean initKm1 = false;
@Override
public int measureByteSize() {
int size = (int) SizeOf.sizeOf(this);
for (Classifier classifier : this.ensemble) {
size += classifier.measureByteSize();
}
for (ADWIN adwin : this.ADError) {
size += adwin.measureByteSize();
}
return size;
}
@Override
public void resetLearningImpl() {
this.ensemble = new Classifier[this.ensembleSizeOption.getValue()];
Classifier baseLearner = (Classifier) getPreparedClassOption(this.baseLearnerOption);
baseLearner.resetLearning();
for (int i = 0; i < this.ensemble.length; i++) {
this.ensemble[i] = baseLearner.copy();
}
this.scms = new double[this.ensemble.length];
this.swms = new double[this.ensemble.length];
this.ADError = new ADWIN[this.ensemble.length];
for (int i = 0; i < this.ensemble.length; i++) {
this.ADError[i] = new ADWIN((double) this.deltaAdwinOption.getValue());
}
this.numberOfChangesDetected = 0;
if (this.outputCodesOption.isSet()) {
this.initMatrixCodes = true;
}
if (this.sammeOption.isSet()) {
this.initKm1 = true;
}
}
@Override
public void trainOnInstanceImpl(Instance inst) {
int numClasses = inst.numClasses();
// Set log (k-1) and (k-1) for SAMME Method
if (this.sammeOption.isSet()) {
this.Km1 = numClasses - 1;
this.logKm1 = Math.log(this.Km1);
this.initKm1 = false;
}
//Output Codes
if (this.initMatrixCodes) {
this.matrixCodes = new int[this.ensemble.length][inst.numClasses()];
for (int i = 0; i < this.ensemble.length; i++) {
int numberOnes;
int numberZeros;
do { // until we have the same number of zeros and ones
numberOnes = 0;
numberZeros = 0;
for (int j = 0; j < numClasses; j++) {
int result = 0;
if (j == 1 && numClasses == 2) {
result = 1 - this.matrixCodes[i][0];
} else {
result = (this.classifierRandom.nextBoolean() ? 1 : 0);
}
this.matrixCodes[i][j] = result;
if (result == 1)
numberOnes++;
else
numberZeros++;
}
} while ((numberOnes - numberZeros) * (numberOnes - numberZeros) > (this.ensemble.length % 2));
}
this.initMatrixCodes = false;
}
boolean Change = false;
double lambda_d = 1.0;
Instance weightedInst = (Instance) inst.copy();
for (int i = 0; i < this.ensemble.length; i++) {
double k = this.pureBoostOption.isSet() ? lambda_d : MiscUtils
.poisson(lambda_d * this.Km1, this.classifierRandom);
if (k > 0.0) {
if (this.outputCodesOption.isSet()) {
weightedInst.setClassValue((double) this.matrixCodes[i][(int) inst.classValue()]);
}
weightedInst.setWeight(inst.weight() * k);
this.ensemble[i].trainOnInstance(weightedInst);
}
boolean correctlyClassifies = this.ensemble[i].correctlyClassifies(weightedInst);
if (correctlyClassifies) {
this.scms[i] += lambda_d;
lambda_d *= this.trainingWeightSeenByModel / (2 * this.scms[i]);
} else {
this.swms[i] += lambda_d;
lambda_d *= this.trainingWeightSeenByModel / (2 * this.swms[i]);
}
double ErrEstim = this.ADError[i].getEstimation();
if (this.ADError[i].setInput(correctlyClassifies ? 0 : 1))
if (this.ADError[i].getEstimation() > ErrEstim) Change = true;
}
if (Change) {
numberOfChangesDetected++;
double max = 0.0;
int imax = -1;
for (int i = 0; i < this.ensemble.length; i++) {
if (max < this.ADError[i].getEstimation()) {
max = this.ADError[i].getEstimation();
imax = i;
}
}
if (imax != -1) {
this.ensemble[imax].resetLearning();
//this.ensemble[imax].trainOnInstance(inst);
this.ADError[imax] = new ADWIN((double) this.deltaAdwinOption.getValue());
this.scms[imax] = 0;
this.swms[imax] = 0;
}
}
}
protected double getEnsembleMemberWeight(int i) {
double em = this.swms[i] / (this.scms[i] + this.swms[i]);
if ((em == 0.0) || (em > 0.5)) {
return this.logKm1;
}
return Math.log((1.0 - em) / em) + this.logKm1;
}
public double[] getVotesForInstance(Instance inst) {
if (this.outputCodesOption.isSet()) {
return getVotesForInstanceBinary(inst);
}
DoubleVector combinedVote = new DoubleVector();
for (int i = 0; i < this.ensemble.length; i++) {
double memberWeight = getEnsembleMemberWeight(i);
if (memberWeight > 0.0) {
DoubleVector vote = new DoubleVector(this.ensemble[i]
.getVotesForInstance(inst));
if (vote.sumOfValues() > 0.0) {
vote.normalize();
vote.scaleValues(memberWeight);
combinedVote.addValues(vote);
}
} else {
break;
}
}
return combinedVote.getArrayRef();
}
public double[] getVotesForInstanceBinary(Instance inst) {
double combinedVote[] = new double[(int) inst.numClasses()];
Instance weightedInst = (Instance) inst.copy();
if (this.initMatrixCodes) {
for (int i = 0; i < this.ensemble.length; i++) {
//Replace class by OC
weightedInst.setClassValue((double) this.matrixCodes[i][(int) inst.classValue()]);
double vote[];
vote = this.ensemble[i]
.getVotesForInstance(weightedInst);
//Binary Case
int voteClass = 0;
if (vote.length == 2) {
voteClass = (vote[1] > vote[0] ? 1 : 0);
}
//Update votes
for (int j = 0; j < inst.numClasses(); j++) {
if (this.matrixCodes[i][j] == voteClass) {
combinedVote[j] += getEnsembleMemberWeight(i);
}
}
}
}
return combinedVote;
}
public boolean isRandomizable() {
return true;
}
@Override
public void getModelDescription(StringBuilder out, int indent) {
// TODO Auto-generated method stub
}
@Override
protected Measurement[] getModelMeasurementsImpl() {
return new Measurement[]{new Measurement("ensemble size",
this.ensemble != null ? this.ensemble.length : 0),
new Measurement("change detections", this.numberOfChangesDetected)
};
}
@Override
public Classifier[] getSubClassifiers() {
return this.ensemble.clone();
}
}