generatingforwardselection.java

一个很好的LIBSVM的JAVA源码。对于要研究和改进SVM算法的学者。可以参考。来自数据挖掘工具YALE工具包。

/*
 *  YALE - Yet Another Learning Environment
 *  Copyright (C) 2001-2004
 *      Simon Fischer, Ralf Klinkenberg, Ingo Mierswa, 
 *          Katharina Morik, Oliver Ritthoff
 *      Artificial Intelligence Unit
 *      Computer Science Department
 *      University of Dortmund
 *      44221 Dortmund,  Germany
 *  email: [e-mail]
 *  web:   http://yale.cs.uni-dortmund.de/
 *
 *  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., 59 Temple Place, Suite 330, Boston, MA 02111-1307
 *  USA.
 */
package edu.udo.cs.yale.operator.features;

import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.operator.Value;
import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.operator.UserError;
import edu.udo.cs.yale.operator.IOObject;
import edu.udo.cs.yale.example.Attribute;
import edu.udo.cs.yale.example.ExampleSet;
import edu.udo.cs.yale.example.AttributeWeightedExampleSet;
import edu.udo.cs.yale.generator.*;
import edu.udo.cs.yale.tools.LogService;
import edu.udo.cs.yale.tools.Tools;
import edu.udo.cs.yale.operator.performance.PerformanceVector;

import java.util.Iterator;
import java.util.List;
import java.util.LinkedList;
import java.util.ListIterator;

/** This operator is a kind of nested forward selection and thus is (in contrast to
 *  a genetic algorithm) a directed search.
 *  <ol>
 *    <li>use forward selection in order to determine the best attributes</li>
 *    <li>Create a new attribute by multiplying any of the original attributes with any
 *        of the attributes selected by the forward selection in the last turn</li>
 *    <li>loop as long as performance increases</li>
 *  </ol>
 *
 *  @yale.xmlclass GeneratingForwardSelection
 *  @author simon
 *  @version $Id: GeneratingForwardSelection.java,v 2.11 2004/09/14 08:39:05 ingomierswa Exp $
 */
public class GeneratingForwardSelection extends FeatureSelectionOperator {

    private int degree;

    /** List of AttributeReferences. */
    private Attribute[] originalAttributes;

    private ExampleSet bestIndividual;

    private List useGenerators;

    private int newAttributeStart;

    private int turn;

    public GeneratingForwardSelection() {
	addValue(new Value("turn", "The number of the current turn.") {
		public double getValue() {
		    return turn;
		}
	    });
    }


    int getDefaultDirection() {
	return FORWARD_SELECTION;
    }

    public IOObject[] apply() throws OperatorException {
	useGenerators = new LinkedList();

	if (getParameterAsBoolean("reciprocal_value")) {
	    FeatureGenerator g =  new ReciprocalValueGenerator();
	    useGenerators.add(g);
	}
	if (getParameterAsBoolean("function_characteristica")) {
	    FeatureGenerator g =  new FunctionCharacteristicaGenerator();
	    useGenerators.add(g);
	}
	if (getParameterAsBoolean("use_plus")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(0);
	    useGenerators.add(g);
	}
	if (getParameterAsBoolean("use_diff")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(1);
	    useGenerators.add(g);
	}
	if (getParameterAsBoolean("use_mult")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(2);
	    useGenerators.add(g);
	}
	if (getParameterAsBoolean("use_div")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(3);
	    useGenerators.add(g);
	}

	if (useGenerators.size()==0) {
	    LogService.logMessage("No FeatureGenerators specified for " + getName() + ".", LogService.WARNING);
	}

	return super.apply();
    }

    public Population createInitialPopulation(AttributeWeightedExampleSet es) {
	newAttributeStart = 0;
	turn = 0;
	// remember the original attributes
	originalAttributes = new Attribute[es.getNumberOfAttributes()];
	for (int i = 0; i < originalAttributes.length; i++) {
	    originalAttributes[i] = es.getAttribute(i);
	}
	return super.createInitialPopulation(es);
    }

    public boolean solutionGoodEnough(Population pop) throws OperatorException {
	if (super.solutionGoodEnough(pop)) {
	    // The forward selection is finished
	    AttributeWeightedExampleSet fsBest = (AttributeWeightedExampleSet)pop.lastBest().clone();
	    // Check whether the performance was improved by this turn 
	    if ((bestIndividual == null) ||
		(((PerformanceVector)fsBest.getUserData("performance")).compareTo(((PerformanceVector)bestIndividual.getUserData("performance"))) > 0)) {
		turn++;
		bestIndividual = (ExampleSet)fsBest.clone();

		fsBest = new AttributeWeightedExampleSet(fsBest.createCleanExampleSet());
		LogService.logMessage(getName() +": "+Tools.ordinalNumber(turn) + " turn's FS result: " + 
				      fsBest, LogService.TASK);
		// and generate all new attributes using the generators
		List generators = new LinkedList();
		ListIterator i = useGenerators.listIterator();
		// for all generator types
		while (i.hasNext()) {
		    FeatureGenerator fg = (FeatureGenerator)i.next();
		    // for all new arguments
		    for (int a = newAttributeStart; a < fsBest.getNumberOfAttributes(); a++) {
			// for all original attributes
			for (int o = 0; o < originalAttributes.length; o++) {
			    FeatureGenerator g = (FeatureGenerator)fg.newInstance();
			    g.setArguments(new Attribute[] { originalAttributes[o], fsBest.getAttribute(a) });
			    generators.add(g);
			}
		    }
		}
		LogService.logMessage(getName() + ": generating " + generators.size() + " new attributes.", 
				      LogService.OPERATOR);
		newAttributeStart = fsBest.getNumberOfAttributes();
		// generate the new attributes
		try {
		    List attributes = FeatureGenerator.generateAll(fsBest.getExampleTable(), 
								   generators);
		    Iterator j = attributes.iterator();
		    while (j.hasNext()) {
			Attribute attr = (Attribute)j.next();
			fsBest.addAttribute(attr);
			fsBest.setAttributeUsed(attr, false);
		    }
		} catch (GenerationException e) {
		    throw new UserError(this, e, 108, e.getMessage());
		}
		// clear the population, add the generated set
		pop.clear();
		pop.add(fsBest);
		return false;		
	    } else {
		// otherwise quit
		return true;
	    }
	}
	// go on with the forward selection
	return false;
    }

    public List getParameterTypes() {
	List types = super.getParameterTypes();
	types.add(new ParameterTypeBoolean("reciprocal_value", "Generate reciprocal values.", true));
	types.add(new ParameterTypeBoolean("function_characteristica", "Generate function characteristica (for C9).", false));
	types.add(new ParameterTypeBoolean("use_plus", "Generate sums.", true));
	types.add(new ParameterTypeBoolean("use_diff", "Generate differences.", true));
	types.add(new ParameterTypeBoolean("use_mult", "Generate products.", true));
	types.add(new ParameterTypeBoolean("use_div", "Generate quotients.", true));
	return types;
    }

}