/*********************************************************************** This file is part of KEEL-software, the Data Mining tool for regression, classification, clustering, pattern mining and so on. Copyright (C) 2004-2010 F. Herrera (herrera@decsai.ugr.es) L. S�nchez (luciano@uniovi.es) J. Alcal�-Fdez (jalcala@decsai.ugr.es) S. Garc�a (sglopez@ujaen.es) A. Fern�ndez (alberto.fernandez@ujaen.es) J. Luengo (julianlm@decsai.ugr.es) 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 3 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, see http://www.gnu.org/licenses/ **********************************************************************/ /*********************************************************************** This file is part of the Fuzzy Instance Based Learning package, a Java package implementing Fuzzy Nearest Neighbor Classifiers as complementary material for the paper: Fuzzy Nearest Neighbor Algorithms: Taxonomy, Experimental analysis and Prospects Copyright (C) 2012 J. Derrac (jderrac@decsai.ugr.es) S. Garc�a (sglopez@ujaen.es) F. Herrera (herrera@decsai.ugr.es) 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 3 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, see http://www.gnu.org/licenses/ **********************************************************************/ /** * * File: PFKNN.java * * The PFKNN algorithm. * * @author Written by Joaqu�n Derrac (University of Granada) 13/11/2011 * @version 1.0 * @since JDK1.5 * */ package keel.Algorithms.Fuzzy_Instance_Based_Learning.PFKNN; import java.text.DecimalFormat; import java.text.DecimalFormatSymbols; import java.util.Arrays; import java.util.StringTokenizer; import org.core.Files; import org.core.Randomize; import keel.Algorithms.Fuzzy_Instance_Based_Learning.FuzzyIBLAlgorithm; import keel.Algorithms.Fuzzy_Instance_Based_Learning.ReportTool; import keel.Algorithms.Fuzzy_Instance_Based_Learning.Timer; import keel.Algorithms.Fuzzy_Instance_Based_Learning.Util; public class PFKNN extends FuzzyIBLAlgorithm { private int K; private int selected[]; private double membership[][]; private double referenceMembership [][]; private double testMembership [][]; /** * Reads the parameters of the algorithm. * * @param script Configuration script * */ @Override protected void readParameters(String script) { String file; String line; StringTokenizer fileLines, tokens; file = Files.readFile (script); fileLines = new StringTokenizer (file,"\n\r"); //Discard in/out files definition fileLines.nextToken(); fileLines.nextToken(); fileLines.nextToken(); //Getting the seed line = fileLines.nextToken(); tokens = new StringTokenizer (line, "="); tokens.nextToken(); seed = Long.parseLong(tokens.nextToken().substring(1)); //Getting the K parameter line = fileLines.nextToken(); tokens = new StringTokenizer (line, "="); tokens.nextToken(); K = Integer.parseInt(tokens.nextToken().substring(1)); } //end-method /** * Main builder. Initializes the methods' structures * * @param script Configuration script */ public PFKNN(String script){ readDataFiles(script); //Naming the algorithm name="PFKNN"; selected = new int [trainData.length]; membership = new double [trainData.length][nClasses]; referenceMembership = new double [referenceData.length][nClasses]; testMembership = new double [testData.length][nClasses]; //Initialization of Reporting tool ReportTool.setOutputFile(outFile[2]); } //end-method private void selectBoundaryPoints(){ Arrays.fill(selected, 0); //for each training instances find its K-nearest enemies //and select them double minDist[]; int nearestN[]; double dist; boolean stop; nearestN = new int[K]; minDist = new double[K]; for(int index = 0; index<trainData.length;index++){ for (int i=0; i<K; i++) { nearestN[i] = 0; minDist[i] = Double.MAX_VALUE; } //KNN Method starts here for (int i=0; i<trainData.length; i++) { if(trainOutput[i]!=trainOutput[index]){ dist = Util.euclideanDistance(trainData[i],trainData[index]); //see if it's nearer than our previous selected neighbors stop=false; for(int j=0;j<K && !stop;j++){ if (dist < minDist[j]) { for (int l = K - 1; l >= j+1; l--) { minDist[l] = minDist[l - 1]; nearestN[l] = nearestN[l - 1]; } minDist[j] = dist; nearestN[j] = i; stop=true; } } } } for(int j=0;j<K;j++){ selected[nearestN[j]]=1; } } } private void shuffleVector(int vector []){ int pos,tmp; for (int i=0; i<vector.length; i++) { pos = Randomize.Randint (0, vector.length); tmp = vector[i]; vector[i] = vector[pos]; vector[pos] = tmp; } } private void accomodateClusters(){ evaluateMembership(); //set a random order for the instances int order [] = new int [trainData.length]; for (int i=0; i<trainData.length; i++) { order[i]=i; } //test if all instances are correctly classified int index; double minDist[]; int nearestN[]; double dist; boolean stop; double classMembership[]; classMembership=new double[nClasses]; for (int i=0; i<trainData.length; i++) { index=order[i]; Arrays.fill(classMembership, 0.0); //find its K nearest neighbors nearestN = new int[K]; minDist = new double[K]; for (int i2=0; i2<K; i2++) { nearestN[i2] = -1; minDist[i2] = Double.MAX_VALUE; } //KNN Method starts here for (int i2=0; i2<trainData.length; i2++) { dist = Util.euclideanDistance(trainData[i2],trainData[index]); if (i2 != index){ //leave-one-out //see if it's nearer than our previous selected neighbors stop=false; for(int j=0;j<K && !stop;j++){ if (dist < minDist[j]) { for (int l = K - 1; l >= j+1; l--) { minDist[l] = minDist[l - 1]; nearestN[l] = nearestN[l - 1]; } minDist[j] = dist; nearestN[j] = i2; stop=true; } } } } //compute its class membership double norm[]; double sum; double MAX_NORM = 100000000; norm = new double [K]; sum = 0.0; for(int n = 0;n<K;n++){ if(nearestN[n]!=-1){ if(minDist[n]==0.0){ norm[n]=MAX_NORM; } norm[n] = 1.0/ Math.pow(minDist[n],(2.0/(2.0-1.0))); norm[n]=Math.min(norm[n],MAX_NORM); sum+=norm[n]; } } for(int n = 0;n<K;n++){ if(nearestN[n]!=-1){ for(int c=0;c<nClasses;c++){ classMembership[c]+= membership[nearestN[n]][c]*(norm[n]/sum); } } } double max= Double.MIN_VALUE; int pred=-1; for(int c=0;c<nClasses;c++){ if(max<classMembership[c]){ max=classMembership[c]; pred=c; } } if(pred!=trainOutput[index]){ selected[index]=1; evaluateMembership(); } } } private void evaluateMembership(){ for(int instance=0;instance<trainData.length;instance++){ if(selected[instance]==1){ double minDist[]; int nearestN[]; int selectedClasses[]; double dist; boolean stop; nearestN = new int[K]; minDist = new double[K]; for (int i=0; i<K; i++) { nearestN[i] = 0; minDist[i] = Double.MAX_VALUE; } //KNN Method starts here for (int i=0; i<trainData.length; i++) { dist = Util.euclideanDistance(trainData[i],trainData[instance]); if (selected[i]==1 && i != instance){ //leave-one-out //see if it's nearer than our previous selected neighbors stop=false; for(int j=0;j<K && !stop;j++){ if (dist < minDist[j]) { for (int l = K - 1; l >= j+1; l--) { minDist[l] = minDist[l - 1]; nearestN[l] = nearestN[l - 1]; } minDist[j] = dist; nearestN[j] = i; stop=true; } } } } //we have check all the instances... see what is the most present class selectedClasses= new int[nClasses]; Arrays.fill(selectedClasses, 0); for (int i=0; i<K; i++) { selectedClasses[trainOutput[nearestN[i]]]++; } Arrays.fill(membership[instance], 0.0); double term; for (int i=0; i<nClasses; i++) { term = ((double)selectedClasses[i]/(double)K); if(trainOutput[instance]==i){ membership[instance][i]=0.51+0.49*term; }else{ membership[instance][i]=0.49*term; } } } else{ for (int i=0; i<nClasses; i++) { if(trainOutput[instance]==i){ membership[instance][i]=1.0; }else{ membership[instance][i]=0; } } } } } private void editTrainingSet(){ int remove [] = new int [trainData.length]; Arrays.fill(remove, 1); int winner; double minDist,dist; for(int instance=0;instance<trainData.length;instance++){ winner=-1; minDist=Double.MAX_VALUE; for(int i=0;i<trainData.length;i++){ if(selected[i]==1 && trainOutput[i]==trainOutput[instance]){ dist=Util.euclideanDistance(trainData[instance], trainData[i]); if(minDist>dist){ minDist=dist; winner=i; } } } if(winner!=-1){ remove[winner]=0; } else{ remove[instance]=0; } } for(int instance=0;instance<trainData.length;instance++){ if(remove[instance]==1){ selected[instance]=0; } } evaluateMembership(); } /** * Classifies the training set (leave-one-out) */ public void classifyTrain(){ //Start of training time Timer.resetTime(); classifyTrainSet(); //End of training time Timer.setTrainingTime(); //Showing results System.out.println(name+" "+ relation + " Training " + Timer.getTrainingTime() + "s"); } //end-method /** * Classifies the test set */ public void classifyTest(){ //Start of training time Timer.resetTime(); classifyTestSet(); //End of test time Timer.setTestTime(); //Showing results System.out.println(name+" "+ relation + " Test " + Timer.getTestTime() + "s"); } //end-method /** * Classifies the training set */ public void classifyTrainSet(){ for(int i=0;i<trainData.length;i++){ trainPrediction[i]=classifyTrain(i,trainData[i]); } } //end-method /** * Classifies the test set */ public void classifyTestSet(){ for(int i=0;i<testData.length;i++){ testPrediction[i]=classifyTest(i,testData[i]); } } //end-method /** * Evaluates a instance to predict its class membership * * @param index Index of the instance in the test set * @param example Instance evaluated * */ private int classifyTrain(int index, double example[]) { double minDist[]; int nearestN[]; double dist; boolean stop; nearestN = new int[K]; minDist = new double[K]; for (int i=0; i<K; i++) { nearestN[i] = 0; minDist[i] = Double.MAX_VALUE; } //KNN Method starts here for (int i=0; i<trainData.length; i++) { if(selected[i]==1 && index!=i){ dist = Util.euclideanDistance(trainData[i],example); //see if it's nearer than our previous selected neighbors stop=false; for(int j=0;j<K && !stop;j++){ if (dist < minDist[j]) { for (int l = K - 1; l >= j+1; l--) { minDist[l] = minDist[l - 1]; nearestN[l] = nearestN[l - 1]; } minDist[j] = dist; nearestN[j] = i; stop=true; } } } } //compute membership double norm[]; double sum; norm = new double [K]; sum = 0.0; double MAX_NORM = 100000000; for(int i = 0;i<K;i++){ if(minDist[i]==0.0){ norm[i]=MAX_NORM; } norm[i] = 1.0/ Math.pow(minDist[i],(2.0/(2.0-1.0))); norm[i]=Math.min(norm[i],MAX_NORM); sum+=norm[i]; } for(int i = 0;i<K;i++){ for(int c=0;c<nClasses;c++){ referenceMembership [index][c]+= membership[nearestN[i]][c]*(norm[i]/sum); } } return computeClass(referenceMembership [index]); } /** * Evaluates a instance to predict its class membership * * @param index Index of the instance in the test set * @param example Instance evaluated * */ private int classifyTest(int index, double example[]) { double minDist[]; int nearestN[]; double dist; boolean stop; nearestN = new int[K]; minDist = new double[K]; for (int i=0; i<K; i++) { nearestN[i] = 0; minDist[i] = Double.MAX_VALUE; } //KNN Method starts here for (int i=0; i<trainData.length; i++) { if(selected[i]==1){ dist = Util.euclideanDistance(trainData[i],example); //see if it's nearer than our previous selected neighbors stop=false; for(int j=0;j<K && !stop;j++){ if (dist < minDist[j]) { for (int l = K - 1; l >= j+1; l--) { minDist[l] = minDist[l - 1]; nearestN[l] = nearestN[l - 1]; } minDist[j] = dist; nearestN[j] = i; stop=true; } } } } //compute membership double norm[]; double sum; norm = new double [K]; sum = 0.0; double MAX_NORM = 100000000; for(int i = 0;i<K;i++){ if(minDist[i]==0.0){ norm[i]=MAX_NORM; } norm[i] = 1.0/ Math.pow(minDist[i],(2.0/(2.0-1.0))); norm[i]=Math.min(norm[i],MAX_NORM); sum+=norm[i]; } for(int i = 0;i<K;i++){ for(int c=0;c<nClasses;c++){ testMembership [index][c]+= membership[nearestN[i]][c]*(norm[i]/sum); } } return computeClass(testMembership [index]); } /** * Computes the class of a instance given its membership array * @param pertenence Membership array * * @return Class assigned (crisp) */ private int computeClass(double pertenence[]){ double max = Double.MIN_VALUE; int output=-1; for(int i=0; i< pertenence.length;i++){ if(max<pertenence[i]){ max=pertenence[i]; output=i; } } return output; } //end-method /** * Generates the model of the algorithm */ public void generateModel(){ //Start of model time Timer.resetTime(); selectBoundaryPoints(); accomodateClusters(); editTrainingSet(); //End of model time Timer.setModelTime(); //Showing results System.out.println(name+" "+ relation + " Model " + Timer.getModelTime() + "s"); } /** * Reports the results obtained */ public void printReport(){ writeOutput(outFile[0], trainOutput, trainPrediction); writeOutput(outFile[1], testOutput, testPrediction); ReportTool.setResults(trainOutput,trainPrediction,testOutput,testPrediction,nClasses); ReportTool.printReport(); } //end-method } //end-class