aode.java

MacroWeka扩展了著名数据挖掘工具weka

/*
 *    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.
 */

/*
 *    AODE.java
 *    Copyright (C) 2003
 *    Algorithm developed by: Geoff Webb
 *    Code written by: Janice Boughton & Zhihai Wang
 */

package weka.classifiers.bayes;

import java.io.*;
import java.util.*;
import weka.core.*;
import weka.classifiers.Evaluation;
import weka.classifiers.Classifier;
import weka.classifiers.UpdateableClassifier;

/**
 * AODE achieves highly accurate classification by averaging over all
 * of a small space of alternative naive-Bayes-like models that have
 * weaker (and hence less detrimental) independence assumptions than
 * naive Bayes. The resulting algorithm is computationally efficient while
 * delivering highly accurate classification on many learning tasks.<br>
 * For more information, see<p>
 * G. Webb, J. Boughton & Z. Wang (2004). <i>Not So Naive Bayes.</i>
 * To be published in Machine Learning.<br>
 * G. Webb, J. Boughton & Z. Wang (2002). <i>Averaged One-Dependence
 * Estimators: Preliminary Results.</i> AI2002 Data Mining Workshop, Canberra.
 *
 * Valid options are:<p>
 *
 * -D <br>
 * Debugging information is printed if this flag is specified.<p>
 * 
 * -F <br>
 * Specify the frequency limit for parent attributes.<p>
 *
 * @author Janice Boughton ([e-mail]) & Zhihai Wang ([e-mail])
 * @version $Revision: 1.1 $
 *  this version resolves errors in the handling of missing attribute values.
 */
public class AODE extends Classifier
    implements OptionHandler, WeightedInstancesHandler, UpdateableClassifier {
    
  /**
   * 3D array (m_NumClasses * m_TotalAttValues * m_TotalAttValues)
   * of attribute counts, i.e. the number of times an attribute value occurs
   * in conjunction with another attribute value and a class value.  
   */
  private double [][][] m_CondiCounts;
    
  /** The number of times each class value occurs in the dataset */
  private double [] m_ClassCounts;
    
  /** The sums of attribute-class counts  
   *    -- if there are no missing values for att, then m_SumForCounts[classVal][att] 
   *       will be the same as m_ClassCounts[classVal] 
   */
  private double [][] m_SumForCounts;

  /** The number of classes */
  private int m_NumClasses;
    
  /** The number of attributes in dataset, including class */
  private int m_NumAttributes;
    
  /** The number of instances in the dataset */
  private int m_NumInstances;
    
  /** The index of the class attribute */
  private int m_ClassIndex;
    
  /** The dataset */
  private Instances m_Instances;
    
  /**
   * The total number of values (including an extra for each attribute's 
   * missing value, which are included in m_CondiCounts) for all attributes 
   * (not including class).  Eg. for three atts each with two possible values,
   * m_TotalAttValues would be 9 (6 values + 3 missing).
   * This variable is used when allocating space for m_CondiCounts matrix.
   */
  private int m_TotalAttValues;
    
  /** The starting index (in the m_CondiCounts matrix) of the values for each attribute */
  private int [] m_StartAttIndex;
    
  /** The number of values for each attribute */
  private int [] m_NumAttValues;
    
  /** The frequency of each attribute value for the dataset */
  private double [] m_Frequencies;

  /** The number of valid class values observed in dataset 
   *  -- with no missing classes, this number is the same as m_NumInstances.
   */
  private double m_SumInstances;

  /** An att's frequency must be this value or more to be a superParent */
  private int m_Limit;

  /** If true, outputs debugging info */
  private boolean m_Debug = false;

 
  /**
   * Returns a string describing this classifier
   * @return a description of the classifier suitable for
   * displaying in the explorer/experimenter gui
   */
  public String globalInfo() {
    return "AODE achieves highly accurate classification by averaging over all "
      +"of a small space of alternative naive-Bayes-like models that have "
      +"weaker (and hence less detrimental) independence assumptions than "
      +"naive Bayes. The resulting algorithm is computationally efficient while "
      +"delivering highly accurate classification on many learning tasks.\n\n"
      +"For more information, see\n\n"
      +"G. Webb, J. Boughton & Z. Wang (2004). Not So Naive Bayes. "
      +"To be published in Machine Learning. "
      +"G. Webb, J. Boughton & Z. Wang (2002). <i>Averaged One-Dependence "
      +"Estimators: Preliminary Results. AI2002 Data Mining Workshop, Canberra.";
  }
 
  /**
   * Generates the classifier.
   *
   * @param instances set of instances serving as training data
   * @exception Exception if the classifier has not been generated
   * successfully
   */
  public void buildClassifier(Instances instances) throws Exception {

    // reset variable for this fold
    m_SumInstances = 0;
    
    m_NumClasses = instances.numClasses();
    if(instances.classAttribute().isNumeric()) {
       throw new Exception("AODE: Class is numeric!");
    }
    if(m_NumClasses < 2) {
       throw new Exception ("Dataset has no class attribute");
    }
    if(instances.checkForStringAttributes()) {
       throw new Exception("AODE: String attributes are not allowed.");
    }

    m_ClassIndex = instances.classIndex();
    m_NumAttributes = instances.numAttributes();
    for(int att = 0; att < m_NumAttributes; att++) {
       Attribute attribute = (Attribute)instances.attribute(att);
       if(!attribute.isNominal()) {
          throw new Exception("Attributes must be nominal.  " +
                              "Discretize dataset with FilteredClassifer.");
       }
    }

    // copy the instances
    m_Instances = instances;
    m_NumInstances = m_Instances.numInstances();
 
    // allocate space for attribute reference arrays
    m_StartAttIndex = new int[m_NumAttributes];
    m_NumAttValues = new int[m_NumAttributes];
 
    m_TotalAttValues = 0;
    for(int i = 0; i < m_NumAttributes; i++) {
       if(i != m_ClassIndex) {
          m_StartAttIndex[i] = m_TotalAttValues;
          m_NumAttValues[i] = m_Instances.attribute(i).numValues();
          m_TotalAttValues += m_NumAttValues[i] + 1;  // + 1 so room for missing value count
       } else {
          // m_StartAttIndex[i] = -1;  // class isn't included
	  m_NumAttValues[i] = m_NumClasses;
       }
    }

    // allocate space for counts and frequencies
    m_CondiCounts = new double[m_NumClasses][m_TotalAttValues][m_TotalAttValues];
    m_ClassCounts = new double[m_NumClasses];
    m_SumForCounts = new double[m_NumClasses][m_NumAttributes];
    m_Frequencies = new double[m_TotalAttValues];

    // calculate the counts
    for(int k = 0; k < m_NumInstances; k++) {
       addToCounts((Instance)m_Instances.instance(k));
    }

    // free up some space
    m_Instances = new Instances(m_Instances, 0);
  }
 

  /**
   * Updates the classifier with the given instance.
   *
   * @param instance the new training instance to include in the model 
   * @exception Exception if the instance could not be incorporated in
   * the model.
   */
    public void updateClassifier(Instance instance) {
	this.addToCounts(instance);
    }

    /** Puts an instance's values into m_CondiCounts, m_ClassCounts and 
     * m_SumInstances.
     *
     * @param instance the instance whose values are to be put into the counts variables
     *
     */
  private void addToCounts(Instance instance) {
 
    double [] countsPointer;
 
    if(instance.classIsMissing())
       return;   // ignore instances with missing class

    int classVal = (int)instance.classValue();
    int weight = (int)instance.weight();
 
    m_ClassCounts[classVal] += weight;
    m_SumInstances += weight;
   
    // store instance's att val indexes in an array, b/c accessing it 
    // in loop(s) is more efficient
    int [] attIndex = new int[m_NumAttributes];
    for(int i = 0; i < m_NumAttributes; i++) {
       if(i == m_ClassIndex)
          attIndex[i] = -1;  // we don't use the class attribute in counts
       else {
          if(instance.isMissing(i))
             attIndex[i] = m_StartAttIndex[i] + m_NumAttValues[i];
          else
             attIndex[i] = m_StartAttIndex[i] + (int)instance.value(i);
       }
    }

    for(int Att1 = 0; Att1 < m_NumAttributes; Att1++) {
       if(attIndex[Att1] == -1)
          continue;   // avoid pointless looping as Att1 is currently the class attribute

       m_Frequencies[attIndex[Att1]] += weight;
       
       // if this is a missing value, we don't want to increase sumforcounts
       if(!instance.isMissing(Att1))
          m_SumForCounts[classVal][Att1] += weight;

       // save time by referencing this now, rather than do it repeatedly in the loop
       countsPointer = m_CondiCounts[classVal][attIndex[Att1]];

       for(int Att2 = 0; Att2 < m_NumAttributes; Att2++) {
          if(attIndex[Att2] != -1) {