wekametalearner.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.learner.weka;

import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.operator.IODescription;
import edu.udo.cs.yale.operator.IOObject;
import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.operator.IllegalInputException;
import edu.udo.cs.yale.operator.OperatorChain;
import edu.udo.cs.yale.operator.UserError;
import edu.udo.cs.yale.operator.learner.Learner;
import edu.udo.cs.yale.operator.learner.Model;
import edu.udo.cs.yale.operator.performance.PerformanceVector;
import edu.udo.cs.yale.example.Attribute;
import edu.udo.cs.yale.example.AttributeWeights;
import edu.udo.cs.yale.example.Example;
import edu.udo.cs.yale.example.ExampleReader;
import edu.udo.cs.yale.example.ExampleSet;
import edu.udo.cs.yale.tools.Ontology;
import edu.udo.cs.yale.tools.LogService;
import edu.udo.cs.yale.tools.WekaTools;

import weka.classifiers.Classifier;
import weka.core.Instances;

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

/** This operator can build all meta classifiers from the 
 *  <a href="http://www.cs.waikato.ac.nz/~ml/weka/">Weka</a> package.<br/>
 *  The classifier type can be selected by the parameter <var>weka_learner_name</var>. 
 *  Parameters can be passed to the Weka classifier using the Yale parameter list
 *  <var>weka_parameters</var>. The leading dash &quot;-&quot; for the keys must be omitted. 
 *  See the Weka javadoc for classifier and parameter descriptions.<br/>
 *
 *  The inner learning scheme must be emebedded as operator child. This is a more straightforward
 *  and easy to use than the usage of a normal Weka learner and specifying the inner scheme and its
 *  parameters as parameters of the meta learning scheme. See {@link WekaLearner} for details.
 *
 *  All Weka meta learning schemes which uses one inner learning scheme can be used with this operator.
 *  Some learning schemes need more than one inner operator or other parameters and must be used by using the
 *  {@link WekaLearner} with parameters. These are Grading, MultiScheme, RegressionByDiscretization, and Vote.
 *  Other meta learning schemes can only be used for certain learning types like classification or regression.
 *
 *  @version $Id: WekaMetaLearner.java,v 1.2 2004/08/27 11:57:42 ingomierswa Exp $
 */
public class WekaMetaLearner extends OperatorChain implements Learner {

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

    public static final String[] WEKA_CLASSIFIERS = WekaTools.getWekaClasses(weka.classifiers.Classifier.class, ".meta.", true);

    /** 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 exampleSet = (ExampleSet)getInput(ExampleSet.class);
	Model model = learn(exampleSet);

	String filename = getParameterAsString("model_file");
	try {
	    if (filename != null) model.writeModel(getExperiment().resolveFileName(filename));
	} catch (IOException e) {
	    LogService.logMessage("WekaMetaLearner'" + getName() + "': " +
				  "Cannot write model to file!", LogService.ERROR);
	}
	return new IOObject[] { model };
    }

    public Model learn(ExampleSet exampleSet) throws OperatorException {
	String operatorName = getParameterAsString("weka_meta_learner_name");
	String[] parameters = getWekaParameters();
	Classifier classifier = null;
	try {
	    classifier = Classifier.forName(operatorName, parameters);
	} catch (Exception e) {
	    throw new UserError(this, e, 904, new Object[] { operatorName, e});
	}

	LogService.logMessage(getName() + ": Converting to Weka instances.", LogService.MINIMUM);
	Instances instances = WekaTools.toWekaInstances(exampleSet, "TempInstances", exampleSet.getLabel(), true);
	try {
	    LogService.logMessage(getName() + ": Building Weka classifier.", LogService.MINIMUM);
	    classifier.buildClassifier(instances);
	} catch (Exception e) {
	    throw new UserError(this, e, 905, new Object[] {operatorName, e});
	}
	boolean useDist = getParameterAsBoolean("use_distribution");
	return new WekaClassifier(exampleSet.getLabel(), classifier, useDist);
    }
    
    public String[] getWekaParameters() {
	List wekaParameters = getParameterList("weka_parameters");
	String[] parameters = WekaTools.getWekaParameters(wekaParameters);
	String[] innerParameters = ((WekaLearner)getOperator(0)).getWekaParameters();
	
	int n = 0;
	String[] result = new String[parameters.length + 2 + innerParameters.length];
	for (int i = 0; i < parameters.length; i++)
	    result[n++] = parameters[i];
	result[n++] = "-W";
	result[n++] = ((WekaLearner)getOperator(0)).getWekaLearnerName();
	for (int i = 0; i < innerParameters.length; i++)
	    result[n++] = "-" + parameters[i];
	return result;
    }

    public int getNumberOfSteps() { return 1; }
    
    public int getMinNumberOfInnerOperators() { return 1; }
    public int getMaxNumberOfInnerOperators() { return 1; }

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

    public boolean canCalculateWeights() { return false; }
    public AttributeWeights getWeights(ExampleSet exampleSet) { return null; }
    public boolean canEstimatePerformance() { return false; }
    public PerformanceVector getEstimatedPerformance() { return null; }

    public Class[] checkIO(Class[] input) throws IllegalInputException {
	if (!(getOperator(0) instanceof WekaLearner))
	    throw new IllegalInputException(this, 
					    "Inner operator of a Weka meta learning operator must be another Weka learning scheme.");
	if (!IODescription.containsClass(ExampleSet.class, input))
	    throw new IllegalInputException(this, ExampleSet.class);
	return OUTPUT_CLASSES;
    }

    public List getParameterTypes() {
	List types = super.getParameterTypes();
	ParameterType type = new ParameterTypeFile("model_file", "If this parameter is set, the model is written to a file.", true);
	type.setExpert(false);
	types.add(type);
	type = new ParameterTypeStringCategory("weka_meta_learner_name", "The fully qualified classname of the weka classifier.", WEKA_CLASSIFIERS);
	type.setExpert(false);
	types.add(type);
	types.add(new ParameterTypeList("weka_parameters", "Parameters for the Weka classifier as described in the Weka manual.", new ParameterTypeString(null, null)));
	types.add(new ParameterTypeBoolean("use_distribution", "If set to true, the prediction of the model will not be the class, but the confidence for that class.", false));
	return types;
    }
}