directedgga.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.ga;

import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.operator.features.*;
import edu.udo.cs.yale.tools.LogService;
import edu.udo.cs.yale.example.Attribute;
import edu.udo.cs.yale.example.ExampleSet;
import edu.udo.cs.yale.example.Tools;

import java.util.List;

/** DirectedGGA is an acronym for a Generating Genetic Algorithm which uses probability directed
 *  search heuristics to select attributes for generation or removing. 
 *  Its approach to generating new attributes differs from the original one and is the same as the one
 *  of {@link YAGGA}. <br/>
 *
 *  The (generating) mutation can do one of the following things with 
 *  different probabilities:
 *  <ul>
 *    <li>Probability {@yale.math p/4}: Add a newly generated attribute to the feature vector</li>
 *    <li>Probability {@yale.math p/4}: Add a randomly chosen original attribute to the feature vector</li>
 *    <li>Probability {@yale.math p/2}: Remove a randomly chosen attribute from the feature vector</li>
 *  </ul>
 *  Thus it is guaranteed that the length of the feature vector can both
 *  grow and shrink. On average it will keep its original length, unless
 *  longer or shorter individuals prove to have a better fitness.<br/>
 *  
 *  In addition to these mutation heuristics probablilities based on the weights of the attributes are
 *  calculated. It is more likely for attributes with a great weight to be selected for generating new
 *  attributes. On the other hand the probability for removing an attribute from the example set will 
 *  decrease for attributes with great weights. This decreases the amount of needed generations drastically. <br/>
 * 
 *  Another enhancement in comparison to the original GGA is the addition of several generators like
 *  the ones for trigonometric or exponential functions. In this way a sinple linear working learning scheme which 
 *  can deliver weights can be used as inner operator. If this learner can also estimate its performance it is not
 *  longer necessary to use a inner cross-validation which also decreases learning time. Such a learner is 
 *  for example the {@link edu.udo.cs.yale.operator.learner.kernel.JMySVMLearner} which delivers the xi-alpha
 *  performance estimation at least for classification tasks. <br/>.
 *  
 *  Summarized the advantages of this feature construction algorithm are smaller runtimes and smaller 
 *  attribute sets as result. These attribute sets increase performance and can be used to explain the
 *  models of more complex learning schemes like SVMs. The additional generators allow the construction 
 *  of features which are not possible by the known kernel functions. <br/>
 *
 *  Since this operator does not contain algorithms to extract features from value series, it is restricted
 *  to example sets with only single attributes. For (automatic) feature extraction from values series the 
 *  value series plugin for Yale written by Ingo Mierswa should be used. It is available at 
 *  <tt>http://yale.cs.uni-dortmund.de</tt>.
 *
 *  @version $Id: DirectedGGA.java,v 2.9 2004/08/27 11:57:36 ingomierswa Exp $
 */
public class DirectedGGA extends YAGGA2 {

    /** Returns the {@link DirectedGeneratingMutation}. */
    protected PopulationOperator getMutationPopulationOperator() throws OperatorException {
	List generators = getGenerators();
	if (generators.size()==0)
	    LogService.logMessage("No FeatureGenerators specified for " + getName() + ".", LogService.WARNING);
	
 	ExampleSet eSet = (ExampleSet)getInput(ExampleSet.class, false);
	Attribute[] attributes = new Attribute[eSet.getNumberOfAttributes()];
	for (int i = 0; i < eSet.getNumberOfAttributes(); i++) {
 	    attributes[i] = eSet.getAttribute(i);
 	}

 	return new DirectedGeneratingMutation(attributes, 
					      getParameterAsDouble("p_mutation"), 
					      generators,
					      getParameterAsInt("max_generated"), 
					      getParameterAsInt("max_original"),
					      getParameterAsInt("max_construction_depth"),
					      getParameterAsString("unused_functions").split(" "));
    }    

    public List getParameterTypes() {
	List types = super.getParameterTypes();
	types.add(new ParameterTypeInt("max_generated", 
				       "The maximum number of generated attributes per generation.", 
				       1, Integer.MAX_VALUE, 2));
	types.add(new ParameterTypeInt("max_original", 
				       "The maximum number of original attributes added per generation.", 
				       1, Integer.MAX_VALUE, 2));
	types.add(new ParameterTypeInt("max_construction_depth", 
				       "The maximum depth for the argument attributes used for attribute construction (-a: allow all depths).", 
				       -1, Integer.MAX_VALUE, -1));
	types.add(new ParameterTypeString("unused_functions", 
					  "Space separated list of functions which are not allowed in arguments for attribute construction."));
	return types;
    }
}