.#semisupclusterersplitevaluator.java.1.11

wekaUT是 university texas austin 开发的基于weka的半指导学习(semi supervised learning)的分类器

/*
 *    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.
 */ 
/*
 *    SemiSupClustererSplitEvaluator.java
 *    Copyright (C) 2002 Sugato Basu
 *
 */


package weka.experiment;

import java.io.*;
import java.util.*;
import weka.core.*;
import weka.clusterers.*;

/**
 * A SplitEvaluator that produces results for a semi-supervised clustering scheme
 * on a nominal class attribute.
 *
 * -W clustername <br>
 * Specify the full class name of the clusterer to evaluate. <p>
 *
 * -C class index <br>
 * The index of the class for which statistics are to
 * be output. (default 1) <p>
 *
 * @author Sugato Basu
 */

public class SemiSupClustererSplitEvaluator implements SplitEvaluator, 
  OptionHandler {
  
  /** The semi-supervised clusterer used for evaluation */
  protected Clusterer m_Clusterer = new MPCKMeans();

  /** Holds the statistics for the most recent application of the clusterer */
  protected String m_result = null;

  /** The clusterer options (if any) */
  protected String m_ClustererOptions = "";

  /** The clusterer version */
  protected String m_ClustererVersion = "";

  /** The length of a key */
  private static final int KEY_SIZE = 3;

  /** The length of a result */
  private static final int RESULT_SIZE = 14;

  /** Class index for information retrieval statistics (default 0) */
  private int m_IRclass = 0;

  /**
   * No args constructor.
   */
  public SemiSupClustererSplitEvaluator() {

    updateOptions();
  }

  /** Does nothing, since cluster evaluation does not allow additional measures */
  public void setAdditionalMeasures(String [] additionalMeasures){}
  
  /**
   * Returns a string describing this split evaluator
   * @return a description of the split evaluator suitable for
   * displaying in the explorer/experimenter gui
   */
  public String globalInfo() {
    return " A SplitEvaluator that produces results for a semi-supervised "
      + "clustering scheme on a nominal class attribute.";
  }

  /**
   * Returns an enumeration describing the available options..
   *
   * @return an enumeration of all the available options.
   */
  public Enumeration listOptions() {

    Vector newVector = new Vector(2);

    newVector.addElement(new Option(
	     "\tThe full class name of the clusterer.\n"
	      +"\teg: weka.clusterers.SimpleKMeans", 
	     "W", 1, 
	     "-W <class name>"));
    newVector.addElement(new Option(
	     "\tThe index of the class for which IR statistics\n" +
	     "\tare to be output. (default 1)",
	     "C", 1, 
	     "-C <index>"));

    if ((m_Clusterer != null) &&
	(m_Clusterer instanceof OptionHandler)) {
      newVector.addElement(new Option(
	     "",
	     "", 0, "\nOptions specific to clusterer "
	     + m_Clusterer.getClass().getName() + ":"));
      Enumeration enum = ((OptionHandler)m_Clusterer).listOptions();
      while (enum.hasMoreElements()) {
	newVector.addElement(enum.nextElement());
      }
    }
    return newVector.elements();
  }

  /**
   * Parses a given list of options. Valid options are:<p>
   *
   * -W classname <br>
   * Specify the full class name of the clusterer to evaluate. <p>
   *
   * -C class index <br>
   * The index of the class for which IR statistics are to
   * be output. (default 1) <p>
   *
   * All option after -- will be passed to the clusterer.
   *
   * @param options the list of options as an array of strings
   * @exception Exception if an option is not supported
   */
  public void setOptions(String[] options) throws Exception {
    
    String cName = Utils.getOption('W', options);
    if (cName.length() == 0) {
      throw new Exception("A clusterer must be specified with"
			  + " the -W option.");
    }
    // Do it first without options, so if an exception is thrown during
    // the option setting, listOptions will contain options for the actual
    // Clusterer.
    setClusterer(Clusterer.forName(cName, null));
    if (getClusterer() instanceof OptionHandler) {
      ((OptionHandler) getClusterer())
	.setOptions(Utils.partitionOptions(options));
      updateOptions();
    }

    String indexName = Utils.getOption('C', options);
    if (indexName.length() != 0) {
      m_IRclass = (new Integer(indexName)).intValue() - 1;
    } else {
      m_IRclass = 0;
    }
  }

  /**
   * Gets the current settings of the Clusterer.
   *
   * @return an array of strings suitable for passing to setOptions
   */
  public String [] getOptions() {

    String [] clustererOptions = new String [0];
    if ((m_Clusterer != null) && 
	(m_Clusterer instanceof OptionHandler)) {
      clustererOptions = ((OptionHandler)m_Clusterer).getOptions();
    }
    
    String [] options = new String [clustererOptions.length + 5];
    int current = 0;

    if (getClusterer() != null) {
      options[current++] = "-W";
      options[current++] = getClusterer().getClass().getName();
    }
    options[current++] = "-C"; 
    options[current++] = "" + (m_IRclass + 1);
    options[current++] = "--";

    System.arraycopy(clustererOptions, 0, options, current, 
		     clustererOptions.length);
    current += clustererOptions.length;
    while (current < options.length) {
      options[current++] = "";
    }
    return options;
  }


  /**
   * Gets the data types of each of the key columns produced for a single run.
   * The number of key fields must be constant
   * for a given SplitEvaluator.
   *
   * @return an array containing objects of the type of each key column. The 
   * objects should be Strings, or Doubles.
   */
  public Object [] getKeyTypes() {

    Object [] keyTypes = new Object[KEY_SIZE];
    keyTypes[0] = "";
    keyTypes[1] = "";
    keyTypes[2] = "";
    return keyTypes;
  }

  /**
   * Gets the names of each of the key columns produced for a single run.
   * The number of key fields must be constant
   * for a given SplitEvaluator.
   *
   * @return an array containing the name of each key column
   */
  public String [] getKeyNames() {

    String [] keyNames = new String[KEY_SIZE];
    keyNames[0] = "Scheme";
    keyNames[1] = "Scheme_options";
    keyNames[2] = "Scheme_version_ID";
    return keyNames;
  }

  /**
   * Gets the key describing the current SplitEvaluator. For example
   * This may contain the name of the clusterer used for clusterer
   * predictive evaluation. The number of key fields must be constant
   * for a given SplitEvaluator.
   *
   * @return an array of objects containing the key.
   */
  public Object [] getKey(){

    Object [] key = new Object[KEY_SIZE];
    key[0] = m_Clusterer.getClass().getName();
    key[1] = m_ClustererOptions;
    key[2] = m_ClustererVersion;
    return key;
  }

  /**
   * Gets the data types of each of the result columns produced for a 
   * single run. The number of result fields must be constant
   * for a given SplitEvaluator.
   *
   * @return an array containing objects of the type of each result column. 
   * The objects should be Strings, or Doubles.
   */
  public Object [] getResultTypes() {
    int overall_length = RESULT_SIZE;
    Object [] resultTypes = new Object[overall_length];
    Double doub = new Double(0);
    int current = 0;
    
    // Unsupervised stats: 3
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    // Supervised stats: 3
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    // Training data stats: 2  
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    // IR stats: 3
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    // Timing stats: 2
    resultTypes[current++] = doub;
    resultTypes[current++] = doub;

    // Clusterer defined extras: 1
    resultTypes[current++] = "";

    if (current != overall_length) {
      throw new Error("ResultTypes didn't fit RESULT_SIZE");
    }
    return resultTypes;
  }

  /**
   * Gets the names of each of the result columns produced for a single run.
   * The number of result fields must be constant for a given SplitEvaluator.
   *
   * @return an array containing the name of each result column
   */
  public String [] getResultNames() {
    int overall_length = RESULT_SIZE;
    String [] resultNames = new String[overall_length];
    int current = 0;

    // Unsupervised stats: 3
    resultNames[current++] = "Purity";
    resultNames[current++] = "Entropy";
    resultNames[current++] = "Objective_function";

    // Supervised stats: 3
    resultNames[current++] = "KL_divergence";
    resultNames[current++] = "Mutual_information";
    resultNames[current++] = "Supervised_dispersion";

    // Training data stats: 2  
    resultNames[current++] = "SameClassPairs";
    resultNames[current++] = "DiffClassPairs";

    // IR stats: 3
    resultNames[current++] = "Pairwise_ir_precision";
    resultNames[current++] = "Pairwise_ir_recall";
    resultNames[current++] = "Pairwise_f_measure";

    // Timing stats: 2
    resultNames[current++] = "Time_training";
    resultNames[current++] = "Time_testing";

    // Clusterer defined extras: 1
    resultNames[current++] = "Summary";

    if (current != overall_length) {
      throw new Error("ResultNames didn't fit RESULT_SIZE");
    }
    return resultNames;
  }

  /** Dummy function, exists just for compatibility with SplitEvaluator interface
   */
  public Object [] getResult(Instances unlabeledTrain, Instances test) {
    try {
      return getResult(null, unlabeledTrain, test, test.numClasses(), -1); // labeled set is null
    }
    catch (Exception e) {
      e.printStackTrace();
    }
    return null;
  }

  /**
   * Gets the results for the supplied train and test datasets.
   *
   * @param labeledTrainPairs the constraint pairs having labels on them
   * @param labeledTrain the labeled training Instances.
   * @param unlabeledData the unlabeled training (+ test for transductive) Instances.
   * @param test the testing Instances.
   * @param startingIndexOfTest from where test data starts in unlabeledData, useful if clustering is transductive
   * @return the results stored in an array. The objects stored in
   * the array may be Strings, Doubles, or null (for the missing value).
   * @exception Exception if a problem occurs while getting the results
   */
  public Object [] getResult(ArrayList labeledTrainPairs, Instances labeledTrain, Instances unlabeledData, Instances test, Instances unlabeledTest, HashMap homologHash) throws Exception{
    if (m_Clusterer == null) {
      throw new WekaException("No clusterer has been specified");
    }
    if (!(m_Clusterer instanceof SemiSupClusterer)) {
      throw new WekaException("Clusterer should implement SemiSupClusterer interface!!\n"); // KLUGE (we could not make m_Clusterer of type SemiSupClusterer, since SemiSupClusterer is an interface and not an abstract class ... so we have to make the check here)
    }

    int overall_length = RESULT_SIZE;
    Object [] result = new Object[overall_length];
    long trainTimeStart = System.currentTimeMillis();

    if (m_Clusterer instanceof PCKMeans) {
      ((PCKMeans)m_Clusterer).buildClusterer(labeledTrainPairs, unlabeledData, labeledTrain, labeledTrain.numInstances()); // KLUGE: have to generalize later
    } else if (m_Clusterer instanceof MPCKMeans) {
      // Couldn't bypass it any more!!! Adding test to labeledTrain ...
      if (test.instance(0) instanceof SparseInstance) {
	for (int i=0; i<test.numInstances(); i++) {
	  labeledTrain.add(new SparseInstance(test.instance(i)));
	}
      } else {
	for (int i=0; i<test.numInstances(); i++) {
	  labeledTrain.add(new Instance(test.instance(i)));