directedgeneratinggeneticalgorithm.java

著名的开源仿真软件yale

/*
 *  YALE - Yet Another Learning Environment
 *  Copyright (C) 2002, 2003
 *      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.ga;

import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.operator.FatalException;
import edu.udo.cs.yale.generator.*;
import edu.udo.cs.yale.tools.LogService;
import edu.udo.cs.yale.operator.features.*;

import java.util.ArrayList;
import java.util.List;
import java.util.ListIterator;

/** By using a generating genetic algorithm which generates new
 *  attributes and do not only select them it can happen that many
 *  irrelevant attributes are generated. In addition, these individuals
 *  can be randomly generated and deleted several times.
 *  <br/>
 *  It might be a good idea to make the generating genetic algorithm a
 *  sort of smarter by using an information gain criterion to decide, which
 *  attribute should be selected or should be used to generate another
 *  one. With the new regression information gain criterion this is
 *  possible for regression problems too.
 *  <br/>
 *  The attributes get their information gain values. Then the more
 *  informative attributes will be preferably selected and used for
 *  generating new attributes. Therefore it is postulated that it is
 *  better to generate new attributes from the informative ones.
 *
 *  @yale.xmlclass DirectedGeneratingGeneticAlgorithm
 *  @author ingo
 *  @version $Id: DirectedGeneratingGeneticAlgorithm.java,v 2.3 2003/04/04 11:59:28 fischer Exp $
 */
public class DirectedGeneratingGeneticAlgorithm extends GeneratingGeneticAlgorithm {

    /** Ruft <tt>initApply()</tt> der Oberklasse auf und setzt noch entsprechende Parameter. Zus&auml;tzlich wird noch 
     *  InformationGain in die preEvaluation - Liste gepackt.
     */
    public void initApply() throws OperatorException {
	super.initApply();

	if (getNumberOfOperators() != 3) {
	    throw new FatalException("DirectedGeneratingGeneticAlgorithm needs three operators: " + 
				     "a normal GGA experiment chain, a learner and a model applier to find the information gain!");
	}

	//addPostEvaluationPopulationOperator(new RemoveUnusedAttributes());
    }

    PopulationOperator[] getPreProcessingPopulationOperators() {

	// ratio gain
	boolean ratioGain = getParameterAsBoolean("use_ratio_gain");

	// epsilon
	double epsilon = getParameterAsDouble("epsilon");

	// usePredictedLabel
	boolean usePredictedLabel = getParameterAsBoolean("use_predicted_label");

	return new PopulationOperator[] {  
	    new InformationGain(getOperator(1), getOperator(2), epsilon, 
				usePredictedLabel, ratioGain) 
		};
    }

    /** Liefert den Mutations <tt>PopulationOperator</tt>, kann von Unterklassen &uuml;berschireben werden.
     */
    PopulationOperator getMutationPopulationOperator() {
	double pMutation   = getParameterAsDouble("p_mutation");
	double leftBound   = getParameterAsDouble("lower_mutation_bound");
	double rightBound  = getParameterAsDouble("upper_mutation_bound");
	return new DirectedMutation(pMutation, leftBound, rightBound);
    }


    /** Liefert den generierenden <tt>PopulationOperator</tt>, kann von Unterklassen &uuml;berschrieben werden.
     */
    PopulationOperator getGeneratingPopulationOperator() {
	int    noOfNewAttributes = getParameterAsInt("max_number_of_new_attributes"); 
	double pGenerate         = getParameterAsDouble("p_generate");
	
	// erzeugt die Generatoren
	ArrayList generators = new ArrayList();
	
	if (getParameterAsBoolean("reciprocal_value")) {
	    FeatureGenerator g =  new ReciprocalValueGenerator(true);
	    generators.add(g);
	}
	if (getParameterAsBoolean("function_characteristica")) {
	    FeatureGenerator g =  new FunctionCharacteristicaGenerator();
	    generators.add(g);
	}
	if (getParameterAsBoolean("use_plus")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(0, true);
	    generators.add(g);
	}
	if (getParameterAsBoolean("use_diff")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(1, true);
	    generators.add(g);
	}
	if (getParameterAsBoolean("use_mult")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(2, true);
	    generators.add(g);
	}
	if (getParameterAsBoolean("use_div")) {
	    FeatureGenerator g =  new BasicArithmeticOperationGenerator(3, true);
	    generators.add(g);
	}

	if (generators.size()==0) {
	    LogService.logMessage("No FeatureGenerators specified for " + getName() + ".", LogService.WARNING);
	} 
	
	double leftBound   = getParameterAsDouble("lower_generation_bound");
	double rightBound  = getParameterAsDouble("upper_generation_bound");

	// fuegt das Generieren in die PreEval - Liste ein.
	return new DirectedAttributeGenerator(pGenerate, noOfNewAttributes, generators, leftBound, rightBound);
    } 

    public List getParameterTypes() {
	List types = super.getParameterTypes();
	types.add(new ParameterTypeDouble("lower_generation_bound", "Lower bound for the generation probability.", 0, 1, 0.1));
	types.add(new ParameterTypeDouble("upper_generation_bound", "Upper bound for the generation probability.", 0, 1, 0.9));
	types.add(new ParameterTypeBoolean("use_ratio_gain", "If set to true the ratio gain criterion is used.", true));
	types.add(new ParameterTypeDouble("epsilon", "Variation range of attribute values for the attribute information gain.", 0, 1, 0.1));
	types.add(new ParameterTypeBoolean("use_predicted_label", "If set to true, the predicted label is used for the attribute information gain.", true));
	types.add(new ParameterTypeDouble("lower_mutation_bound", "Lower bound for the mutation probability.", 0, 1, 0.1));
	types.add(new ParameterTypeDouble("upper_mutation_bound", "Upper bound for the mutation probability.", 0, 1, 0.9));

	types.add(new ParameterTypeBoolean("reciprocal_value", "Generate reciprocal values.", true));
	types.add(new ParameterTypeBoolean("function_charactersitica", "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;
    }
}