milr.java

代码是一个分类器的实现,其中使用了部分weka的源代码。可以将项目导入eclipse运行

/*
 *    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., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

/*
 * MILR.java
 * Copyright (C) 2005 University of Waikato, Hamilton, New Zealand
 *
 */

package weka.classifiers.mi;

import weka.classifiers.Classifier;
import weka.core.Capabilities;
import weka.core.Instance;
import weka.core.Instances;
import weka.core.MultiInstanceCapabilitiesHandler;
import weka.core.Optimization;
import weka.core.Option;
import weka.core.OptionHandler;
import weka.core.SelectedTag;
import weka.core.Tag;
import weka.core.Utils;
import weka.core.Capabilities.Capability;

import java.util.Enumeration;
import java.util.Vector;

/**
 <!-- globalinfo-start -->
 * Uses either standard or collective multi-instance assumption, but within linear regression. For the collective assumption, it offers arithmetic or geometric mean for the posteriors.
 * <p/>
 <!-- globalinfo-end -->
 *
 <!-- options-start -->
 * Valid options are: <p/>
 * 
 * <pre> -D
 *  Turn on debugging output.</pre>
 * 
 * <pre> -R &lt;ridge&gt;
 *  Set the ridge in the log-likelihood.</pre>
 * 
 * <pre> -A [0|1|2]
 *  Defines the type of algorithm:
 *   0. standard MI assumption
 *   1. collective MI assumption, arithmetic mean for posteriors
 *   2. collective MI assumption, geometric mean for posteriors</pre>
 * 
 <!-- options-end -->
 *
 * @author Eibe Frank ([e-mail])
 * @author Xin Xu ([e-mail])
 * @version $Revision: 1.3 $ 
 */
public class MILR
  extends Classifier 
  implements OptionHandler, MultiInstanceCapabilitiesHandler {

  /** for serialization */
  static final long serialVersionUID = 1996101190172373826L;
  
  protected double[] m_Par;

  /** The number of the class labels */
  protected int m_NumClasses;

  /** The ridge parameter. */
  protected double m_Ridge = 1e-6;

  /** Class labels for each bag */
  protected int[] m_Classes;

  /** MI data */ 
  protected double[][][] m_Data;

  /** All attribute names */
  protected Instances m_Attributes;

  protected double[] xMean = null, xSD = null;

  /** the type of processing */
  protected int m_AlgorithmType = ALGORITHMTYPE_DEFAULT;

  /** standard MI assumption */
  public static final int ALGORITHMTYPE_DEFAULT = 0;
  /** collective MI assumption, arithmetic mean for posteriors */
  public static final int ALGORITHMTYPE_ARITHMETIC = 1;
  /** collective MI assumption, geometric mean for posteriors */
  public static final int ALGORITHMTYPE_GEOMETRIC = 2;
  /** the types of algorithms */
  public static final Tag [] TAGS_ALGORITHMTYPE = {
    new Tag(ALGORITHMTYPE_DEFAULT, "standard MI assumption"),
    new Tag(ALGORITHMTYPE_ARITHMETIC, "collective MI assumption, arithmetic mean for posteriors"),
    new Tag(ALGORITHMTYPE_GEOMETRIC, "collective MI assumption, geometric mean for posteriors"),
  };

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String globalInfo() {
    return 
        "Uses either standard or collective multi-instance assumption, but "
      + "within linear regression. For the collective assumption, it offers "
      + "arithmetic or geometric mean for the posteriors.";
  }

  /**
   * Returns an enumeration describing the available options
   *
   * @return an enumeration of all the available options
   */
  public Enumeration listOptions() {
    Vector result = new Vector();
    
    result.addElement(new Option(
          "\tTurn on debugging output.",
          "D", 0, "-D"));
    
    result.addElement(new Option(
        "\tSet the ridge in the log-likelihood.",
        "R", 1, "-R <ridge>"));

    result.addElement(new Option(
        "\tDefines the type of algorithm:\n"
        + "\t 0. standard MI assumption\n"
        + "\t 1. collective MI assumption, arithmetic mean for posteriors\n"
        + "\t 2. collective MI assumption, geometric mean for posteriors",
        "A", 1, "-A [0|1|2]"));

    return result.elements();
  }

  /**
   * Parses a given list of options. 
   *
   * @param options the list of options as an array of strings
   * @throws Exception if an option is not supported
   */
  public void setOptions(String[] options) throws Exception {
    String      tmpStr;

    setDebug(Utils.getFlag('D', options));

    tmpStr = Utils.getOption('R', options);
    if (tmpStr.length() != 0) 
      setRidge(Double.parseDouble(tmpStr));
    else 
      setRidge(1.0e-6);

    tmpStr = Utils.getOption('A', options);
    if (tmpStr.length() != 0) {
      setAlgorithmType(new SelectedTag(Integer.parseInt(tmpStr), TAGS_ALGORITHMTYPE));
    } else {
      setAlgorithmType(new SelectedTag(ALGORITHMTYPE_DEFAULT, TAGS_ALGORITHMTYPE));
    }     
  }

  /**
   * Gets the current settings of the classifier.
   *
   * @return an array of strings suitable for passing to setOptions
   */
  public String[] getOptions() {
    Vector        result;
    
    result = new Vector();

    if (getDebug())
      result.add("-D");
    
    result.add("-R");
    result.add("" + getRidge());
    
    result.add("-A");
    result.add("" + m_AlgorithmType);

    return (String[]) result.toArray(new String[result.size()]);
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String ridgeTipText() {
    return "The ridge in the log-likelihood.";
  }

  /**
   * Sets the ridge in the log-likelihood.
   *
   * @param ridge the ridge
   */
  public void setRidge(double ridge) {
    m_Ridge = ridge;
  }

  /**
   * Gets the ridge in the log-likelihood.
   *
   * @return the ridge
   */
  public double getRidge() {
    return m_Ridge;
  }

  /**
   * Returns the tip text for this property
   *
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String algorithmTypeTipText() {
    return "The mean type for the posteriors.";
  }

  /**
   * Gets the type of algorithm.
   *
   * @return the algorithm type
   */
  public SelectedTag getAlgorithmType() {
    return new SelectedTag(m_AlgorithmType, TAGS_ALGORITHMTYPE);
  }

  /**
   * Sets the algorithm type.
   *
   * @param newType the new algorithm type
   */
  public void setAlgorithmType(SelectedTag newType) {
    if (newType.getTags() == TAGS_ALGORITHMTYPE) {
      m_AlgorithmType = newType.getSelectedTag().getID();
    }
  }

  private class OptEng 
    extends Optimization {
    
    /** the type to use 
     * @see MILR#TAGS_ALGORITHMTYPE */
    private int m_Type;
    
    /**
     * initializes the object
     * 
     * @param type      the type top use
     * @see MILR#TAGS_ALGORITHMTYPE
     */
    public OptEng(int type) {
      super();
      
      m_Type = type;
    }
    
    /** 
     * Evaluate objective function
     * @param x the current values of variables
     * @return the value of the objective function 
     */
    protected double objectiveFunction(double[] x){
      double nll = 0; // -LogLikelihood
      
      switch (m_Type) {
        case ALGORITHMTYPE_DEFAULT:
          for(int i=0; i<m_Classes.length; i++){ // ith bag
            int nI = m_Data[i][0].length; // numInstances in ith bag
            double bag = 0.0, // NLL of each bag 
                   prod = 0.0;   // Log-prob. 

            for(int j=0; j<nI; j++){
              double exp=0.0;
              for(int k=m_Data[i].length-1; k>=0; k--)
                exp += m_Data[i][k][j]*x[k+1];
              exp += x[0];
              exp = Math.exp(exp);

              if(m_Classes[i]==1)
                prod -= Math.log(1.0+exp);
              else
                bag += Math.log(1.0+exp);
            }

            if(m_Classes[i]==1)
              bag = -Math.log(1.0-Math.exp(prod));

            nll += bag;
          }   
          break;
        
        case ALGORITHMTYPE_ARITHMETIC:
          for(int i=0; i<m_Classes.length; i++){ // ith bag
            int nI = m_Data[i][0].length; // numInstances in ith bag
            double bag = 0;  // NLL of each bag

            for(int j=0; j<nI; j++){
              double exp=0.0;
              for(int k=m_Data[i].length-1; k>=0; k--)
                exp += m_Data[i][k][j]*x[k+1];
              exp += x[0];
              exp = Math.exp(exp);

              if(m_Classes[i] == 1)
                bag += 1.0-1.0/(1.0+exp); // To avoid exp infinite
              else
                bag += 1.0/(1.0+exp);                  
            }   
            bag /= (double)nI;

            nll -= Math.log(bag);
          }   
          break;
          
        case ALGORITHMTYPE_GEOMETRIC:
          for(int i=0; i<m_Classes.length; i++){ // ith bag
            int nI = m_Data[i][0].length; // numInstances in ith bag
            double bag = 0;   // Log-prob. 

            for(int j=0; j<nI; j++){
              double exp=0.0;
              for(int k=m_Data[i].length-1; k>=0; k--)
                exp += m_Data[i][k][j]*x[k+1];
              exp += x[0];

              if(m_Classes[i]==1)
                bag -= exp/(double)nI;
              else
                bag += exp/(double)nI;
            }

            nll += Math.log(1.0+Math.exp(bag));
          }   
          break;
      }

      // ridge: note that intercepts NOT included
      for(int r=1; r<x.length; r++)
        nll += m_Ridge*x[r]*x[r];

      return nll;
    }

    /** 
     * Evaluate Jacobian vector
     * @param x the current values of variables
     * @return the gradient vector 
     */
    protected double[] evaluateGradient(double[] x){
      double[] grad = new double[x.length];
      
      switch (m_Type) {
        case ALGORITHMTYPE_DEFAULT:
          for(int i=0; i<m_Classes.length; i++){ // ith bag
            int nI = m_Data[i][0].length; // numInstances in ith bag

            double denom = 0.0; // denominator, in log-scale       
            double[] bag = new double[grad.length]; //gradient update with ith bag

            for(int j=0; j<nI; j++){
              // Compute exp(b0+b1*Xi1j+...)/[1+exp(b0+b1*Xi1j+...)]
              double exp=0.0;
              for(int k=m_Data[i].length-1; k>=0; k--)
                exp += m_Data[i][k][j]*x[k+1];
              exp += x[0];
              exp = Math.exp(exp)/(1.0+Math.exp(exp));

              if(m_Classes[i]==1)
                // Bug fix: it used to be denom += Math.log(1.0+exp);
                // Fixed 21 Jan 2005 (Eibe)
                denom -= Math.log(1.0-exp);

              // Instance-wise update of dNLL/dBk
              for(int p=0; p<x.length; p++){  // pth variable
                double m = 1.0;
                if(p>0) m=m_Data[i][p-1][j];
                bag[p] += m*exp;
              }     
            }

            denom = Math.exp(denom);

            // Bag-wise update of dNLL/dBk
            for(int q=0; q<grad.length; q++){
              if(m_Classes[i]==1)
                grad[q] -= bag[q]/(denom-1.0);
              else
                grad[q] += bag[q];
            }   
          }
          break;
        
        case ALGORITHMTYPE_ARITHMETIC:
          for(int i=0; i<m_Classes.length; i++){ // ith bag