libsvmlearner.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.learner.AbstractLearner;
import edu.udo.cs.yale.operator.learner.Model;
import edu.udo.cs.yale.example.ExampleSet;
import edu.udo.cs.yale.example.ExampleReader;
import edu.udo.cs.yale.example.Example;
import edu.udo.cs.yale.example.Attribute;
import edu.udo.cs.yale.operator.OperatorException;
import edu.udo.cs.yale.operator.UserError;
import edu.udo.cs.yale.operator.parameter.*;
import edu.udo.cs.yale.tools.LogService;

import libsvm.svm;
import libsvm.svm_problem;
import libsvm.svm_parameter;
import libsvm.svm_node;
import libsvm.svm_model;

import java.util.List;
import java.util.LinkedList;


/** Applies the <a href="http://www.csie.ntu.edu.tw/~cjlin/libsvm">libsvm</a> learner
 *  by Chih-Chung Chang and Chih-Jen Lin. The SVM is a powerful method for both
 *  classification and regression.
 *
 *  @yale.index SVM
 *  @version $Id: LibSVMLearner.java,v 1.2 2004/08/27 11:57:40 ingomierswa Exp $
 */
public class LibSVMLearner extends AbstractLearner {

    public static final String[] SVM_TYPES    = { "C-SVC", "nu-SVC", "one-class", "epsilon-SVR", "nu-SVR" };
    public static final String[] KERNEL_TYPES = { "linear", "poly", "rbf", "sigmoid" };

    public static svm_node[] makeNodes(Example e) {
	List nodes = new LinkedList();
	for (int a = 0; a < e.getNumberOfAttributes(); a++) {
	    double value = e.getValue(a);
	    Attribute attr = e.getAttribute(a);
	    if (!attr.isDefault(value)) {
		svm_node node = new svm_node();
		node.index = a;
		node.value = value;
		nodes.add(node);
	    }
	}
	svm_node[] nodeArray = new svm_node[nodes.size()];
	nodes.toArray(nodeArray);
	return nodeArray;
    }

    public static svm_problem getProblem(ExampleSet exampleSet) {
	LogService.logMessage("Creating LibSVM problem.", LogService.MINIMUM);
	int nodeCount = 0;
	svm_problem problem = new svm_problem();
	problem.l = exampleSet.getSize();
	problem.y = new double[exampleSet.getSize()];
	problem.x = new svm_node[exampleSet.getSize()][];
	ExampleReader i = exampleSet.getExampleReader();
	int j = 0;
	while (i.hasNext()) {
	    Example e = i.next();
	    problem.x[j] = makeNodes(e);
	    problem.y[j] = e.getLabel();
	    nodeCount += problem.x[j].length;
	    j++;
	}
	LogService.logMessage("Created "+nodeCount+" nodes for "+j+" examples.", LogService.MINIMUM);
	return problem;
    }

    private svm_parameter getParameters(ExampleSet exampleSet) {
	svm_parameter params = new svm_parameter();

	params.svm_type    = getParameterAsInt("svm_type");
	params.kernel_type = getParameterAsInt("kernel_type");
	params.degree      = getParameterAsDouble("degree");
	params.gamma       = getParameterAsDouble("gamma");
	if (params.gamma == 0)
	    params.gamma = 1.0/exampleSet.getSize();
	params.coef0       = getParameterAsDouble("coef0");
	params.nu          = getParameterAsDouble("nu");
	params.cache_size  = getParameterAsInt("cache_size");
	params.C           = getParameterAsDouble("C");
	params.eps         = getParameterAsDouble("epsilon");
	params.p           = getParameterAsDouble("p");
	params.shrinking   = getParameterAsBoolean("shrinking") ? 1 : 0;

	return params;
    }

    public Model learn(ExampleSet exampleSet) throws OperatorException {
	svm_parameter params = getParameters(exampleSet);
	svm_problem problem  = getProblem(exampleSet);
	String errorMsg = svm.svm_check_parameter(problem, params);
	if (errorMsg != null)
	    throw new UserError(this, 905, new Object[] {"libsvm", errorMsg});
	LogService.logMessage("Training LibSVM.", LogService.MINIMUM);
	return new LibSVMModel(exampleSet.getLabel(), svm.svm_train(problem, params));
    }

    public List getParameterTypes() {
	List types = super.getParameterTypes();
	ParameterType type = new ParameterTypeCategory("svm_type", "SVM for classification (C-SVC, nu-SVC), regression (epsilon-SVR, nu-SVR) and distribution estimation (one-class)", SVM_TYPES, 0);
	type.setExpert(false);
	types.add(type);
	type = new ParameterTypeCategory("kernel_type", "The type of the kernel functions", KERNEL_TYPES, 0);
	type.setExpert(false);
	types.add(type);
	type = new ParameterTypeDouble("degree", "The degree for a polynomial kernel function.", 1, Double.POSITIVE_INFINITY, 3);
	type.setExpert(false);
	types.add(type);
	type = new ParameterTypeDouble("gamma", "The parameter gamma for polynomial, rbf, and sigmoid kernel functions. 0 means 1/#attributes.", 0, 1, 0);
	type.setExpert(false);
	types.add(type);
	type = new ParameterTypeDouble("C", "The cost parameter C for c_svc, epsilon_svr, and nu_svr.", 
				       Double.NEGATIVE_INFINITY, Double.POSITIVE_INFINITY, 100);
	type.setExpert(false);
	types.add(type);
	types.add(new ParameterTypeDouble("coef0", "The parameter coef0 for polynomial and sigmoid kernel functions.", Double.NEGATIVE_INFINITY, Double.POSITIVE_INFINITY, 0));
	types.add(new ParameterTypeDouble("nu", "The parameter nu for nu_svc, one_class, and nu_svr.", Double.NEGATIVE_INFINITY, Double.POSITIVE_INFINITY, 0.5));
	types.add(new ParameterTypeInt("cache_size", "Cache size in Megabyte.", 0, Integer.MAX_VALUE, 40));

	types.add(new ParameterTypeDouble("epsilon", "Tolerance of termination criterion.", Double.NEGATIVE_INFINITY, Double.POSITIVE_INFINITY, 0.001));
	types.add(new ParameterTypeDouble("p", "Tolerance of loss function of epsilon-SVR.", Double.NEGATIVE_INFINITY, Double.POSITIVE_INFINITY, 0.1));
	types.add(new ParameterTypeBoolean("shrinking", "Whether to use the shrinking heuristics.", true));
	return types;
    }
}