svmlightmodel.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.kernel;

import edu.udo.cs.yale.operator.UserError;
import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.operator.learner.FileModel;
import edu.udo.cs.yale.operator.learner.Model;
import edu.udo.cs.yale.example.Attribute;
import edu.udo.cs.yale.example.ExampleReader;
import edu.udo.cs.yale.example.ExampleSet;
import edu.udo.cs.yale.example.Example;
import edu.udo.cs.yale.operator.Operator;
import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.tools.LogService;
import edu.udo.cs.yale.tools.TempFileService;
import edu.udo.cs.yale.tools.Tools;
import edu.udo.cs.yale.tools.Ontology;
import edu.udo.cs.yale.tools.ParameterService;

import java.io.*;

/** A <code>SVMLightModel</code> stores a classification model produced by a 
 *  <code>SVMLightLearner</code> and can apply this model to an <code>ExampleSet</code>
 *  in order to label the examples of this example set with their predicted labels.
 *  <p>
 *    For applying a stored SVM<sup><i>light</i></sup> model to an example set, a
 *    <code>SVMLightModel</code> invokes the external
 *    <a TARGET="_top" HREF="http://svmlight.joachims.org/">SVM<i><sup>light</sup></i></a>
 *    program by <a href="http://www.joachims.org/">Thorsten Joachims</a>
 *    [Joachims/1999], an implementation of Vladimir Vapnik's Support Vector Machine (SVM) 
 *    learning method from statistical learning theory [Vapnik/1998].
 *    The SVM<sup><i>light</i></sup> parameters are the same that were previously used by the 
 *    <code>SVMLightLearner</code> to learn the model and that are stored in the model. 
 *    See the description of the <code>SVMLightLearner</code> for further information.
 *  </p>
 *
 *  <p><i>Bibliography:</i><br>
 *    <b>[Joachims/1999]</b>
 *      <a href="http://www.joachims.org/">Thorsten Joachims</a>:
 *        <I>Making large-Scale SVM Learning Practical</I>. Chapter 11 in: 
 *        Advances in Kernel Methods - Support Vector Learning, 
 *        B. Sch&ouml;lkopf and C. Burges and A. Smola (editors),
 *        MIT Press, 1999. <BR>
 *        <a TARGET="_top" HREF="http://www.joachims.org/publications/joachims_99a.ps.gz">[Postscript (gz)]</a>
 *        <a TARGET="_top" HREF="http://www.joachims.org/publications/joachims_99a.pdf">[PDF]</a>
 *        <a TARGET="_top" HREF="http://svmlight.joachims.org/">[SVM<i><sup>light</sup></i>]</a>
 *      <br>
 *    <b>[Vapnik/1998]</b> Vladimir N. Vapnik.
 *      <i>Statistical Learning Theory</i>. Wiley, Chichester, UK, 1998.
 *      <br>
 *  </p>
 *
 *  @author  Ingo Mierswa, Ralf Klinkenberg
 *  @version $Id: SVMLightModel.java,v 1.3 2004/08/27 11:57:40 ingomierswa Exp $
 *  @see     edu.udo.cs.yale.operator.learner.kernel.SVMLightLearner
 *  @see     edu.udo.cs.yale.example.ExampleSet
 */
public class SVMLightModel extends FileModel {

    private boolean classificationTask;
    private int predictionType = PREDICT_CLASSIFICATION;

    public SVMLightModel(Attribute label) throws IOException {
	super(label);
    }

    public SVMLightModel(Attribute label, File modelFile, boolean classificationTask) throws IOException {
	super(label, modelFile);
	this.classificationTask = classificationTask;
    }

    
    public void writeYaleData(ObjectOutputStream out) throws IOException {
	out.writeBoolean(classificationTask);
    }

    public void readYaleData(ObjectInputStream in) throws IOException {
	classificationTask = in.readBoolean();
    }



    /** applies the SVM<i><sup>light</sup></i> model to a test set and sets the predicted labels for the examples of this test set. */
    public void  apply(ExampleSet exampleSet) throws OperatorException {
	LogService.logMessage("SVMLightModel '"+getName()+"':  SVM^light starts predicting labels ("
			      +exampleSet.getSize()+" examples).", LogService.TASK);

	if (!Ontology.ATTRIBUTE_VALUE_TYPE.isA(exampleSet.getLabel().getValueType(), 
					       Ontology.NOMINAL)) {
	    LogService.logMessage("SVMLightModel '"+getName()+"':  Example set does not have nominal labels!", 
				  LogService.ERROR);
	}

	// Prepare output and input files (Yale output = SVM^light input = test examples,
	// SVM^light output = Yale input = predicted labels for these test examples):
	File    testFile   = writeExamples(exampleSet);
	String  command    = ParameterService.getProperty("yale.svmlight.applycommand");
	File    outputFile = TempFileService.createTempFile(getName() + "_labels_");

	// Start process:
	Process process   = null;
	File    modelFile = null;

	try {
	    modelFile = writeModelToTmp();
	} catch (IOException e) {
	    TempFileService.deleteTempFile(modelFile);
	    TempFileService.deleteTempFile(outputFile);
	    throw new UserError(null, 303, new Object[] { modelFile, e.getMessage() });
	}
	String completeCommand = command + " " + testFile + " " + modelFile + " " + outputFile;

	LogService.logMessage("SVMLightModel '"+getName()+"':  SVM^light called with command '" 
			      + completeCommand + "'", LogService.TASK);
	try {
	    process = Runtime.getRuntime().exec(completeCommand);
	} catch (IOException e) {
	    TempFileService.deleteTempFile(modelFile);
	    TempFileService.deleteTempFile(outputFile);
	    throw new UserError(null, e, 310, new Object[] {completeCommand, e.getMessage()});
	}


	// Read SVM^light output:
	BufferedReader in = new BufferedReader(new InputStreamReader(process.getInputStream()));
	String output = null;
	try {
	    output = Tools.readOutput(in);
	    in.close();
	} catch (IOException e) {
	    TempFileService.deleteTempFile(modelFile); 
	    TempFileService.deleteTempFile(outputFile);
	    throw new UserError(null, e, 310, new Object[] {completeCommand, e.getMessage()});
	}

	try {
	    Tools.waitForProcess(null, process, "svmlight");
	} catch (OperatorException e) {
	    LogService.logMessage("Output of svm_classify:\n"+output, LogService.ERROR);
	    TempFileService.deleteTempFile(modelFile);
	    TempFileService.deleteTempFile(outputFile);
	    throw e;
	}

	LogService.logMessage("SVMlightModel '"+getName()+"' has succesfully "+
			      "predicted labels for the example set.", 
			      LogService.TASK);	

	try {
	    setPredictedLabels(exampleSet, outputFile);
	} catch (IOException e) {
	    throw new UserError(null, 302, new Object[] { outputFile, e.getMessage() });
	} finally {
	    TempFileService.deleteTempFile(modelFile);
	    TempFileService.deleteTempFile(outputFile);
	}
    }


    /** Creates the file with examples for the external SVM<i><sup>light</sup></i> process. */
    private File writeExamples(ExampleSet exampleSet) throws OperatorException {
	File         testFile = TempFileService.createTempFile(getName() + "_test_data_");
	PrintWriter  out      = null;
	try {
	    out = new PrintWriter(new FileWriter(testFile));
	} catch (IOException e) {
	    throw new UserError(null, e, 303, new Object[] { testFile, e.getMessage() });
	}
	ExampleReader r = exampleSet.getExampleReader();

	int numberOfAttributes = exampleSet.getNumberOfAttributes();
	while (r.hasNext()) {
	    Example e = r.next();
	    out.println(e.getAttributesAsSparseString());
	}

	out.close();
	return testFile;
    }

    /** iterates all examples and the predicted label file and sets the predicted label of each example.
     */
    private void  setPredictedLabels(ExampleSet exampleSet, File labels) throws IOException {
	BufferedReader reader = null;
	reader = new BufferedReader(new FileReader(labels));;

	ExampleReader r =  exampleSet.getExampleReader();
      
	String line = "";
	while (((line=reader.readLine())!=null) && (r.hasNext())) {
	    double readLabel = Double.NaN;
	    try {
		readLabel = Double.parseDouble(line);
	    } catch (NumberFormatException e) {
		    LogService.logMessage("SVMLightModel '"+getName()+"':  "+
					  "Read 'NaN' (Not a Number) from SVM^light", 
					  LogService.WARNING);
	    }
	    
	    double label;
	    switch (predictionType) {
	    case PREDICT_CLASSIFICATION: 
		if (classificationTask) {
		    label = (readLabel >= 0) ? Attribute.FIRST_CLASS_INDEX : Attribute.FIRST_CLASS_INDEX + 1;
		} else {
		    label = readLabel;
		}
		break;
	    case PREDICT_CONFIDENCE: 
		label = readLabel;
		break;
	    default:
		throw new IllegalArgumentException("SVMLightModel '"+getName()+"':  Illegal prediction type.");
	    }
	    r.next().setPredictedLabel(label);
	}
    }

    public boolean equals(Object o) {
	if (!super.equals(o)) return false;
	SVMLightModel other = (SVMLightModel)o;
	return this.classificationTask == other.classificationTask;
    }

    public void setPredictionType(int type) {
  	this.predictionType = type;
    }

}