/*********************************************************************** 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/ **********************************************************************/ /** * <p> * @author Written by Manuel Chica Serrano (University of Jaen) 01/09/2005 * @author Modified by Jose Joaquin Aguilera Garcia (University of Jaen) 19/12/2008 * @author Modified by Cristobal Jose Carmona del Jesus (University of Jaen) 19/12/2008 * @author Modified by Jose Joaquin Aguilera Garcia (University of Jaen) 03/02/2009 * @version 1.0 * @since JDK1.5 * </p> */ package keel.Algorithms.Preprocess.Feature_Selection.nonevolutionary_algorithms.RELIEF; import java.util.*; import org.core.*; import keel.Dataset.*; import keel.Algorithms.Preprocess.Feature_Selection.Datos; public class ReliefDiff { /** * <p> * Main class of relief method using difference between nearest neighbours as evalution measure. * Relief Method are based in a selection of relevant features. A feature will be relevant if (ideal case): * I is a random instance, H and J the nearest instances to I, and H's class is equal to I's class and different to J's class. * Then, a feature X will be relevant if its instance values are equal to I and H ones, and very different to I and J * </p> */ /** Datos class with all information about datasets and feature selection methods */ private Datos data; /** needed parameters for backward method */ private Parametros params; /** a boolean array with selected features */ private boolean features[]; /** interior class using for reading all parameters */ private class Parametros{ /** algorithm name */ String nameAlgorithm; /** number of nearest neighbours for KNN Classifier */ int paramKNN; /** pathname of training dataset */ String trainFileNameInput; /** pathname of test dataset */ String testFileNameInput; /** pathname of test dataset only with selected features */ String testFileNameOutput; /** pathname of training dataset only with selected features */ String trainFileNameOutput; /** pathname of an extra file with additional information about the algorithm results */ String extraFileNameOutput; /** seed for pseudo-aleatorian generator */ long seed; /** relevance threshold to decide the relevant features */ double relevanceThreshold; /** indicates the number of randomly instances sampled to calculate the relevant features */ int numInstancesSampled; /** * <p> * Constructor of the Parametros Class * </p> * @param nombreFileParametros is the pathname of input parameter file */ Parametros (String nombreFileParametros){ try{ int i; String fichero, linea, tok; StringTokenizer lineasFile, tokens; /* read the parameter file using Files class */ fichero = Files.readFile(nombreFileParametros); fichero += "\n"; /* remove all \r characters. it is neccesary for a correst use in Windows and UNIX */ fichero = fichero.replace('\r', ' '); /* extract the differents tokens of the file */ lineasFile = new StringTokenizer(fichero, "\n"); i=0; while(lineasFile.hasMoreTokens()) { linea = lineasFile.nextToken(); i++; tokens = new StringTokenizer(linea, " ,\t"); if(tokens.hasMoreTokens()){ tok = tokens.nextToken(); if(tok.equalsIgnoreCase("algorithm")) nameAlgorithm = getParamString(tokens); else if(tok.equalsIgnoreCase("inputdata")) getInputFiles(tokens); else if(tok.equalsIgnoreCase("outputdata")) getOutputFiles(tokens); else if(tok.equalsIgnoreCase("paramKNN")) paramKNN = getParamInt(tokens); else if(tok.equalsIgnoreCase("seed")) seed = getParamLong(tokens); else if(tok.equalsIgnoreCase("relevanceThreshold")) relevanceThreshold = getParamFloat(tokens); else if(tok.equalsIgnoreCase("numInstancesSampled")) numInstancesSampled = getParamInt(tokens); else throw new java.io.IOException("Syntax error on line " + i + ": [" + tok + "]\n"); } } } catch(java.io.FileNotFoundException e){ System.err.println(e + "Parameter file"); }catch(java.io.IOException e){ System.err.println(e + "Aborting program"); System.exit(-1); } /** show the read parameter in the standard output */ String contents = "-- Parameters echo --- \n"; contents += "Algorithm name: " + nameAlgorithm +"\n"; contents += "Input Train File: " + trainFileNameInput +"\n"; contents += "Input Test File: " + testFileNameInput +"\n"; contents += "Output Train File: " + trainFileNameOutput +"\n"; contents += "Output Test File: " + testFileNameOutput +"\n"; contents += "Parameter k of KNN Algorithm: " + paramKNN + "\n"; contents += "Relevance Threshold: " + relevanceThreshold + "\n"; contents += "Number of Instances sampled: " + numInstancesSampled + "\n"; contents += "Seed: " + seed + "\n"; System.out.println(contents); } /** obtain an integer value from the parameter file @param s is the StringTokenizer */ private int getParamInt(StringTokenizer s){ String val = s.nextToken(); val = s.nextToken(); return Integer.parseInt(val); } /** obtain a float value from the parameter file @param s is the StringTokenizer */ private float getParamFloat(StringTokenizer s){ String val = s.nextToken(); val = s.nextToken(); return Float.parseFloat(val); } /** obtain a long value from the parameter file @param s is the StringTokenizer */ private long getParamLong(StringTokenizer s){ String val = s.nextToken(); val = s.nextToken(); return Long.parseLong(val); } /** obtain a string value from the parameter file @param s is the StringTokenizer */ private String getParamString(StringTokenizer s){ String contenido = ""; String val = s.nextToken(); while(s.hasMoreTokens()) contenido += s.nextToken() + " "; return contenido.trim(); } /** obtain the names of the input files from the parameter file @param s is the StringTokenizer */ private void getInputFiles(StringTokenizer s){ String val = s.nextToken(); trainFileNameInput = s.nextToken().replace('"', ' ').trim(); testFileNameInput = s.nextToken().replace('"', ' ').trim(); } /** obtain the names of the output files from the parameter file @param s is the StringTokenizer */ private void getOutputFiles(StringTokenizer s){ String val = s.nextToken(); trainFileNameOutput = s.nextToken().replace('"', ' ').trim(); testFileNameOutput = s.nextToken().replace('"', ' ').trim(); extraFileNameOutput = s.nextToken().replace('"', ' ').trim(); } } /** * <p> * Creates a new instance of ReliefDiff * </p> * @param ficParametros is the name of the param file */ public ReliefDiff(String ficParametros) { /* loads the parameter file */ params = new Parametros(ficParametros); /* set the pseudo-aleatorian generator and loads the training and test dataset */ Randomize.setSeed(params.seed); data = new Datos (params.trainFileNameInput, params.testFileNameInput, params.paramKNN); } private void relief(){ /** * <p> * main method of ReliefDiff. Selects a random instance, calculates its hit nearest neighbour (with the same class) and its miss nearest neighbour (different class) * Later, calculates the differences between the random instance and the others. Finally, gets the more relevants features (that satisfy the relevance threshold) * </p> */ boolean featuresVector[]; double w[]; int posI, posH, posJ; /* w contains the differents feature weights */ w = new double[data.returnNumFeatures()]; for(int i=0; i<params.numInstancesSampled; i++){ /* selects a random instance */ posI = Randomize.Randint(0, data.returnNumInstances()); /* returns the position of the hit nearest neighbour */ posH = data.findNearestHit(posI); /* returns the position of the miss nearest neighbour */ posJ = data.findNearestMiss(posI); /* calculates features's weights, and adds to an array */ for(int j=0; j<w.length; j++) w[j] = w[j] - Math.pow(data.diff(j, posI, posH), 2)/params.numInstancesSampled + Math.pow(data.diff(j, posI, posJ), 2)/params.numInstancesSampled; } /* selects the features which satisfy the relevance threshold */ features = new boolean[data.returnNumFeatures()]; for(int i=0; i<w.length; i++) if(w[i] > params.relevanceThreshold) features[i] = true; /* checks if the number of selected features is more than 0. If not, an exception ocurred */ boolean vacio = true; for(int i=0; i<features.length && vacio; i++) if(features[i] == true) vacio = false; if(vacio){ System.err.println("ERROR: It couldn't be possible to find any solution."); System.err.println("Please to reduce the relevance threshold parameter"); System.exit(0); } } /** * <p> * Method interface for Relief Algorithm * </p> */ public void ejecutar(){ String resultado; int i, numFeatures; Date d; d = new Date(); resultado = "RESULTS generated at " + String.valueOf((Date)d) + " \n--------------------------------------------------\n"; resultado += "Algorithm Name: " + params.nameAlgorithm + "\n"; /* call of Relief algorithm */ relief(); /* una vez ejecutado el algoritmo componemos los datos en un String */ resultado += "\nPARTITION Filename: "+ params.trainFileNameInput +"\n---------------\n\n"; resultado += "Features selected: \n"; for(i=numFeatures=0; i<features.length; i++) if(features[i] == true){ resultado += Attributes.getInputAttribute(i).getName() + " - "; numFeatures++; } resultado += "\n\n" + String.valueOf(numFeatures) + " features of " + Attributes.getInputNumAttributes() + "\n\n" ; resultado += "Error in test (using train for prediction): " + String.valueOf(data.validacionCruzada(features)) + "\n"; resultado += "Error in test (using test for prediction): " + String.valueOf(data.LVOTest(features)) + "\n"; resultado += "---------------\n"; System.out.println("Experiment completed successfully"); /* creates the new training and test datasets only with the selected features */ Files.writeFile(params.extraFileNameOutput, resultado); data.generarFicherosSalida(params.trainFileNameOutput, params.testFileNameOutput, features); } }