multiclasslearner.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.learner;

import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.operator.OperatorChain;
import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.operator.IOObject;
import edu.udo.cs.yale.operator.IOContainer;
import edu.udo.cs.yale.operator.IllegalInputException;
import edu.udo.cs.yale.operator.performance.RunVector;          // RK/2003/06/06: error estimation added
import edu.udo.cs.yale.operator.performance.PerformanceVector;  // RK/2003/06/06: error estimation added
import edu.udo.cs.yale.operator.learner.Learner;                // RK/2003/06/06: error estimation added
import edu.udo.cs.yale.example.ExampleSet;
import edu.udo.cs.yale.example.Attribute;
import edu.udo.cs.yale.tools.TempFileService;
import edu.udo.cs.yale.tools.LogService;

import java.io.File;
import java.io.IOException;
import java.util.List;

/** This operator chain must contain exactly one inner operator which must (currently) be a SVMLightLearner
 *  for classification. In order to classify example sets with more than two classifications, the inner
 *  learner is applied once for each class. Thus, it generates a MultiModel consisting of numberOfClasses 
 *  submodels. When the MultiModel is applied, all submodels are applied and the class belonging to the
 *  model with the highest confidence for the "positive" class is chosen.
 *
 *  @yale.xmlclass MultiClassLearner
 *  @see edu.udo.cs.yale.operator.learner.MultiModel
 *  @see edu.udo.cs.yale.operator.learner.SVMLightLearner
 *  @author Simon Fischer, Ingo Mierswa, Ralf Klinkenberg
 *  @version $Id: MultiClassLearner.java,v 2.6 2003/09/03 09:57:02 fischer Exp $
 */
public class MultiClassLearner extends OperatorChain {
    // History:
    // - RK/2003/06/06: error estimation added;
    //
    // To do:
    // - generalize class to allow other learners than just SVMLightLearner as inner learner;

    private static final Class[]  INPUT_CLASSES  = { ExampleSet.class };    
    private static final Class[]  OUTPUT_CLASSES = { MultiModel.class };

    private int     numberOfClasses;

    /** Iterates over all classes <i>i</i>. The
     *  inner Learner is notified about the current class by setPositiveLabelIndex(i). Then
     *  the learning algorithm is applied. A MultiModel is created from the generated models.
     */
    public IOObject[] apply() throws OperatorException {
	ExampleSet eSet = (ExampleSet)getInput(ExampleSet.class);
	numberOfClasses = eSet.getLabel().getNumberOfClasses();

	String filename = getParameterAsString("model_file");

	Model[] models = new Model[numberOfClasses];
	SVMLightLearner learner = (SVMLightLearner)getOperator(0);

	// collect performance estimations of the enclosed learners for averaging  // RK/2003/06/06: performance estimation added
	RunVector  innerPerformanceEstimations  = new RunVector();                 // RK/2003/06/06: performance estimation added

	for (int i = 0; i < numberOfClasses; i++) {
	    learner.setPositiveLabelIndex(i + Attribute.FIRST_CLASS_INDEX);
	    IOContainer learnResult = learner.apply(getInput().append(new IOObject[] { eSet }));
	    models[i] = (Model)learnResult.getInput(Model.class);
	    innerPerformanceEstimations.add(learner.getEstimatedPerformance());  // RK/2003/06/06: performance estimation added
	}

	MultiModel multiModel = new MultiModel(models);

	try {
	    if (filename != null) multiModel.writeModel(getExperiment().resolveFileName(filename));
	} catch (IOException e) {
	    LogService.logMessage("MultiClassLearner'" + getName() + "': " +
				  "Cannot write MultiModel to disc!", LogService.ERROR);
	}
	return new IOObject[] { multiModel, innerPerformanceEstimations.average() };
    }

    public Class[] getInputClasses() { return INPUT_CLASSES; }

    public Class[] getOutputClasses() { return OUTPUT_CLASSES; }

    /** Ok if the only inner operator returns a learner.
     */
    public Class[] checkIO(Class[] input) throws IllegalInputException {
	SVMLightLearner learner;
	try {
	    learner = (SVMLightLearner)getOperator(0);
	} catch (ClassCastException e) {
	    throw new IllegalInputException("MultiClassLearner '" + getName() + "': " + 
					    "Inner operator is not a SVMLightLearner.", this);
	}

	learner.checkIO(input);
	return new Class[] { MultiModel.class };
    }

    /** Returns the maximum number of innner operators. */
    public int getMaxNumberOfInnerOperators() { return 1; }
    /** Returns the minimum number of innner operators. */
    public int getMinNumberOfInnerOperators() { return 1; }

    public List getParameterTypes() {
	List types = super.getParameterTypes();
	types.add(new ParameterTypeFile("model_file", "If this parameter is set, the model is written to a file.", true));
	return types;
    }

    public int getNumberOfSteps() {
	return getNumberOfChildrensSteps() * numberOfClasses + 1;
    }
}