source: branches/MetisMQI/src/test/java/weka/classifiers/AbstractClassifierTest.java

/*
 *    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.
 */

/*
 * Copyright (C) 2002-2006 University of Waikato 
 */

package weka.classifiers;

import weka.classifiers.evaluation.EvaluationUtils;
import weka.core.Attribute;
import weka.core.CheckGOE;
import weka.core.CheckOptionHandler;
import weka.core.FastVector;
import weka.core.Instances;
import weka.core.OptionHandler;
import weka.core.CheckScheme.PostProcessor;
import weka.test.Regression;

import junit.framework.TestCase;

/**
 * Abstract Test class for Classifiers. Internally it uses the class
 * <code>CheckClassifier</code> to determine success or failure of the
 * tests. It follows basically the <code>testsPerClassType</code> method.
 *
 * @author <a href="mailto:len@reeltwo.com">Len Trigg</a>
 * @author FracPete (fracpete at waikato dot ac dot nz)
 * @version $Revision: 1.23 $
 *
 * @see CheckClassifier
 * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
 * @see PostProcessor
 */
public abstract class AbstractClassifierTest 
  extends TestCase {
  
  /** a class for postprocessing the test-data: all values of numeric attributs
   * are replaced with their absolute value */
  public static class AbsPostProcessor 
    extends PostProcessor {
    
    /**
     * initializes the PostProcessor
     */
    public AbsPostProcessor() {
      super();
    }
    
    /**
     * Provides a hook for derived classes to further modify the data. Currently,
     * the data is just passed through.
     * 
     * @param data      the data to process
     * @return          the processed data
     */
    public Instances process(Instances data) {
      Instances result;
      int               i;
      int               n;
      
      result = super.process(data);
      
      for (i = 0; i < result.numAttributes(); i++) {
        if (i == result.classIndex())
        continue;
        if (!result.attribute(i).isNumeric())
        continue;
        
        for (n = 0; n < result.numInstances(); n++)
        result.instance(n).setValue(i, Math.abs(result.instance(n).value(i)));
      }
      
      return result;
    }
  }
  
  /** The classifier to be tested */
  protected Classifier m_Classifier;

  /** For testing the classifier */
  protected CheckClassifier m_Tester;
  
  /** whether classifier is updateable */
  protected boolean m_updateableClassifier;

  /** whether classifier handles weighted instances */
  protected boolean m_weightedInstancesHandler;

  /** whether classifier handles multi-instance data */
  protected boolean m_multiInstanceHandler;

  /** the number of classes to test with testNClasses() 
   * @see #testNClasses() */
  protected int m_NClasses;

  /** whether to run CheckClassifier in DEBUG mode */
  protected boolean DEBUG = false;

  /** the attribute type with the lowest value */
  protected final static int FIRST_CLASSTYPE = Attribute.NUMERIC;

  /** the attribute type with the highest value */
  protected final static int LAST_CLASSTYPE = Attribute.RELATIONAL;
  
  /** wether classifier can predict nominal attributes (array index is attribute type of class) */
  protected boolean[] m_NominalPredictors;
  
  /** wether classifier can predict numeric attributes (array index is attribute type of class) */
  protected boolean[] m_NumericPredictors;
  
  /** wether classifier can predict string attributes (array index is attribute type of class) */
  protected boolean[] m_StringPredictors;
  
  /** wether classifier can predict date attributes (array index is attribute type of class) */
  protected boolean[] m_DatePredictors;
  
  /** wether classifier can predict relational attributes (array index is attribute type of class) */
  protected boolean[] m_RelationalPredictors;
  
  /** whether classifier handles missing values */
  protected boolean[] m_handleMissingPredictors;

  /** whether classifier handles class with only missing values */
  protected boolean[] m_handleMissingClass;

  /** whether classifier handles class as first attribute */
  protected boolean[] m_handleClassAsFirstAttribute;

  /** whether classifier handles class as second attribute */
  protected boolean[] m_handleClassAsSecondAttribute;
  
  /** the results of the regression tests */
  protected FastVector[] m_RegressionResults;
  
  /** the OptionHandler tester */
  protected CheckOptionHandler m_OptionTester;
  
  /** for testing GOE stuff */
  protected CheckGOE m_GOETester;
  
  /**
   * Constructs the <code>AbstractClassifierTest</code>. Called by subclasses.
   *
   * @param name the name of the test class
   */
  public AbstractClassifierTest(String name) { 
    super(name); 
  }

  /**
   * returns a custom PostProcessor for the CheckClassifier datasets, currently
   * only null.
   * 
   * @return            a custom PostProcessor, if necessary
   * @see PostProcessor
   */
  protected PostProcessor getPostProcessor() {
    return null;
  }
  
  /**
   * configures the CheckClassifier instance used throughout the tests
   * 
   * @return    the fully configured CheckClassifier instance used for testing
   */
  protected CheckClassifier getTester() {
    CheckClassifier     result;
    
    result = new CheckClassifier();
    result.setSilent(true);
    result.setClassifier(m_Classifier);
    result.setNumInstances(20);
    result.setDebug(DEBUG);
    result.setPostProcessor(getPostProcessor());
    
    return result;
  }
  
  /**
   * Configures the CheckOptionHandler uses for testing the optionhandling.
   * Sets the classifier return from the getClassifier() method.
   * 
   * @return    the fully configured CheckOptionHandler
   * @see       #getClassifier()
   */
  protected CheckOptionHandler getOptionTester() {
    CheckOptionHandler          result;
    
    result = new CheckOptionHandler();
    result.setOptionHandler((OptionHandler) getClassifier());
    result.setUserOptions(new String[0]);
    result.setSilent(true);
    
    return result;
  }
  
  /**
   * Configures the CheckGOE used for testing GOE stuff.
   * Sets the Classifier returned from the getClassifier() method.
   * 
   * @return    the fully configured CheckGOE
   * @see       #getClassifier()
   */
  protected CheckGOE getGOETester() {
    CheckGOE            result;
    
    result = new CheckGOE();
    result.setObject(getClassifier());
    result.setSilent(true);
    
    return result;
  }
  
  /**
   * Called by JUnit before each test method. This implementation creates
   * the default classifier to test and loads a test set of Instances.
   *
   * @exception Exception if an error occurs reading the example instances.
   */
  protected void setUp() throws Exception {
    m_Classifier   = getClassifier();
    m_Tester       = getTester();
    m_OptionTester = getOptionTester();
    m_GOETester    = getGOETester();

    m_updateableClassifier         = m_Tester.updateableClassifier()[0];
    m_weightedInstancesHandler     = m_Tester.weightedInstancesHandler()[0];
    m_multiInstanceHandler         = m_Tester.multiInstanceHandler()[0];
    m_NominalPredictors            = new boolean[LAST_CLASSTYPE + 1];
    m_NumericPredictors            = new boolean[LAST_CLASSTYPE + 1];
    m_StringPredictors             = new boolean[LAST_CLASSTYPE + 1];
    m_DatePredictors               = new boolean[LAST_CLASSTYPE + 1];
    m_RelationalPredictors         = new boolean[LAST_CLASSTYPE + 1];
    m_handleMissingPredictors      = new boolean[LAST_CLASSTYPE + 1];
    m_handleMissingClass           = new boolean[LAST_CLASSTYPE + 1];
    m_handleClassAsFirstAttribute  = new boolean[LAST_CLASSTYPE + 1];
    m_handleClassAsSecondAttribute = new boolean[LAST_CLASSTYPE + 1];
    m_RegressionResults            = new FastVector[LAST_CLASSTYPE + 1];
    m_NClasses                     = 4;

    // initialize attributes
    checkAttributes(true,  false, false, false, false, false);
    checkAttributes(false, true,  false, false, false, false);
    checkAttributes(false, false, true,  false, false, false);
    checkAttributes(false, false, false, true,  false, false);
    checkAttributes(false, false, false, false, true,  false);
    
    // initialize missing values handling
    for (int i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the scheme support this type of class at all?
      if (!canPredict(i))
        continue;
      
      // 20% missing
      m_handleMissingPredictors[i] = checkMissingPredictors(i, 20, false);
      m_handleMissingClass[i]      = checkMissingClass(i, 20, false);
    }
  }

  /** Called by JUnit after each test method */
  protected void tearDown() {
    m_Classifier   = null;
    m_Tester       = null;
    m_OptionTester = null;
    m_GOETester    = null;

    m_updateableClassifier         = false;
    m_weightedInstancesHandler     = false;
    m_NominalPredictors            = new boolean[LAST_CLASSTYPE + 1];
    m_NumericPredictors            = new boolean[LAST_CLASSTYPE + 1];
    m_StringPredictors             = new boolean[LAST_CLASSTYPE + 1];
    m_DatePredictors               = new boolean[LAST_CLASSTYPE + 1];
    m_RelationalPredictors         = new boolean[LAST_CLASSTYPE + 1];
    m_handleMissingPredictors      = new boolean[LAST_CLASSTYPE + 1];
    m_handleMissingClass           = new boolean[LAST_CLASSTYPE + 1];
    m_handleClassAsFirstAttribute  = new boolean[LAST_CLASSTYPE + 1];
    m_handleClassAsSecondAttribute = new boolean[LAST_CLASSTYPE + 1];
    m_RegressionResults            = new FastVector[LAST_CLASSTYPE + 1];
    m_NClasses                     = 4;
  }

  /**
   * Used to create an instance of a specific classifier.
   *
   * @return a suitably configured <code>Classifier</code> value
   */
  public abstract Classifier getClassifier();

  /**
   * checks whether at least one attribute type can be handled with the
   * given class type
   *
   * @param type      the class type to check for
   * @return          true if at least one attribute type can be predicted with
   *                  the given class
   */
  protected boolean canPredict(int type) {
    return    m_NominalPredictors[type]
           || m_NumericPredictors[type]
           || m_StringPredictors[type]
           || m_DatePredictors[type]
           || m_RelationalPredictors[type];
  }

  /** 
   * returns a string for the class type
   * 
   * @param type        the class type
   * @return            the class type as string
   */
  protected String getClassTypeString(int type) {
    return CheckClassifier.attributeTypeToString(type);
  }

  /**
   * tests whether the classifier can handle certain attributes and if not,
   * if the exception is OK
   *
   * @param nom         to check for nominal attributes
   * @param num         to check for numeric attributes
   * @param str         to check for string attributes
   * @param dat         to check for date attributes
   * @param rel         to check for relational attributes
   * @param allowFail   whether a junit fail can be executed
   * @see CheckClassifier#canPredict(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  protected void checkAttributes(boolean nom, boolean num, boolean str, 
                                 boolean dat, boolean rel,
                                 boolean allowFail) {
    boolean[]     result;
    String        att;
    int           i;

    // determine text for type of attributes
    att = "";
    if (nom)
      att = "nominal";
    else if (num)
      att = "numeric";
    else if (str)
      att = "string";
    else if (dat)
      att = "date";
    else if (rel)
      att = "relational";
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      result = m_Tester.canPredict(nom, num, str, dat, rel, m_multiInstanceHandler, i);
      if (nom)
        m_NominalPredictors[i] = result[0];
      else if (num)
        m_NumericPredictors[i] = result[0];
      else if (str)
        m_StringPredictors[i] = result[0];
      else if (dat)
        m_DatePredictors[i] = result[0];
      else if (rel)
        m_RelationalPredictors[i] = result[0];

      if (!result[0] && !result[1] && allowFail)
        fail("Error handling " + att + " attributes (" + getClassTypeString(i) 
            + " class)!");
    }
  }

  /**
   * tests whether the toString method of the classifier works even though the
   * classifier hasn't been built yet.
   */
  public void testToString() {
    boolean[]     result;

    result = m_Tester.testToString();

    if (!result[0])
      fail("Error in toString() method!");
  }

  /**
   * tests whether the scheme declares a serialVersionUID.
   */
  public void testSerialVersionUID() {
    boolean[]     result;

    result = m_Tester.declaresSerialVersionUID();

    if (!result[0])
      fail("Doesn't declare serialVersionUID!");
  }
  
  /**
   * tests whether the classifier can handle different types of attributes and
   * if not, if the exception is OK
   *
   * @see #checkAttributes(boolean, boolean, boolean, boolean, boolean, boolean, boolean)
   */
  public void testAttributes() {
    // nominal
    checkAttributes(true,  false, false, false, false, true);
    // numeric
    checkAttributes(false, true,  false, false, false, true);
    // string
    checkAttributes(false, false, true,  false, false, true);
    // date
    checkAttributes(false, false, false, true,  false, true);
    // relational
    if (!m_multiInstanceHandler)
      checkAttributes(false, false, false, false, true,  true);
  }

  /**
   * tests whether the classifier handles instance weights correctly
   *
   * @see CheckClassifier#instanceWeights(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testInstanceWeights() {
    boolean[]     result;
    int           i;
    
    if (m_weightedInstancesHandler) {
      for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
        // does the classifier support this type of class at all?
        if (!canPredict(i))
          continue;
        
        result = m_Tester.instanceWeights(
            m_NominalPredictors[i], 
            m_NumericPredictors[i], 
            m_StringPredictors[i], 
            m_DatePredictors[i], 
            m_RelationalPredictors[i], 
            m_multiInstanceHandler, 
            i);

        if (!result[0])
          System.err.println("Error handling instance weights (" + getClassTypeString(i) 
              + " class)!");
      }
    }
  }

  /**
   * tests whether classifier handles data containing only a class attribute
   *
   * @see CheckClassifier#canHandleOnlyClass(boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testOnlyClass() {
    boolean[]   result;
    int         i;

    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      result = m_Tester.canHandleOnlyClass(
          m_NominalPredictors[i],
          m_NumericPredictors[i],
          m_StringPredictors[i],
          m_DatePredictors[i],
          m_RelationalPredictors[i],
          i);

      if (!result[0] && !result[1])
              fail("Error handling data containing only the class!");
    }
  }

  /**
   * tests whether classifier handles N classes
   *
   * @see CheckClassifier#canHandleNClasses(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   * @see #m_NClasses
   */
  public void testNClasses() {
    boolean[]     result;

    if (!canPredict(Attribute.NOMINAL))
      return;

    result = m_Tester.canHandleNClasses(
        m_NominalPredictors[Attribute.NOMINAL],
        m_NumericPredictors[Attribute.NOMINAL],
        m_StringPredictors[Attribute.NOMINAL],
        m_DatePredictors[Attribute.NOMINAL],
        m_RelationalPredictors[Attribute.NOMINAL],
        m_multiInstanceHandler,
        m_NClasses);

    if (!result[0] && !result[1])
      fail("Error handling " + m_NClasses + " classes!");
  }

  /**
   * checks whether the classifier can handle the class attribute at a given
   * position (0-based index, -1 means last).
   *
   * @param type        the class type
   * @param position    the position of the class attribute (0-based, -1 means last)
   * @return            true if the classifier can handle it
   */
  protected boolean checkClassAsNthAttribute(int type, int position) {
    boolean[]     result;
    String        indexStr;
    
    result = m_Tester.canHandleClassAsNthAttribute(
        m_NominalPredictors[type], 
        m_NumericPredictors[type], 
        m_StringPredictors[type], 
        m_DatePredictors[type], 
        m_RelationalPredictors[type], 
        m_multiInstanceHandler, 
        type,
        position);

    if (position == -1)
      indexStr = "last";
    else
      indexStr = (position + 1) + ".";
    
    if (!result[0] && !result[1])
      fail("Error handling class as " + indexStr + " attribute (" 
          + getClassTypeString(type) + " class)!");
    
    return result[0];
  }

  /**
   * Tests whether the classifier can handle class attributes as Nth
   * attribute. In case of multi-instance classifiers it performs no tests,
   * since the multi-instance data has a fixed format (bagID,bag,class).
   *
   * @see CheckClassifier#canHandleClassAsNthAttribute(boolean, boolean, boolean, boolean, boolean, boolean, int, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testClassAsNthAttribute() {
    int           i;
    
    // multi-Instance data has fixed format!
    if (m_multiInstanceHandler)
      return;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      // first attribute
      m_handleClassAsFirstAttribute[i] = checkClassAsNthAttribute(i, 0);

      // second attribute
      m_handleClassAsSecondAttribute[i] = checkClassAsNthAttribute(i, 1);
    }
  }

  /**
   * tests whether the classifier can handle zero training instances
   *
   * @see CheckClassifier#canHandleZeroTraining(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testZeroTraining() {
    boolean[]     result;
    int           i;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      result = m_Tester.canHandleZeroTraining(
          m_NominalPredictors[i], 
          m_NumericPredictors[i], 
          m_StringPredictors[i], 
          m_DatePredictors[i], 
          m_RelationalPredictors[i], 
          m_multiInstanceHandler, 
          i);

      if (!result[0] && !result[1])
        fail("Error handling zero training instances (" + getClassTypeString(i) 
            + " class)!");
    }
  }

  /**
   * checks whether the classifier can handle the given percentage of
   * missing predictors
   *
   * @param type        the class type
   * @param percent     the percentage of missing predictors
   * @param allowFail   if true a fail statement may be executed
   * @return            true if the classifier can handle it
   */
  protected boolean checkMissingPredictors(int type, int percent, boolean allowFail) {
    boolean[]     result;
    
    result = m_Tester.canHandleMissing(
        m_NominalPredictors[type], 
        m_NumericPredictors[type], 
        m_StringPredictors[type], 
        m_DatePredictors[type], 
        m_RelationalPredictors[type], 
        m_multiInstanceHandler, 
        type,
        true,
        false,
        percent);

    if (allowFail) {
      if (!result[0] && !result[1])
        fail("Error handling " + percent + "% missing predictors (" 
            + getClassTypeString(type) + " class)!");
    }
    
    return result[0];
  }

  /**
   * tests whether the classifier can handle missing predictors (20% and 100%)
   *
   * @see CheckClassifier#canHandleMissing(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testMissingPredictors() {
    int           i;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      // 20% missing
      checkMissingPredictors(i, 20, true);

      // 100% missing
      if (m_handleMissingPredictors[i])
        checkMissingPredictors(i, 100, true);
    }
  }

  /**
   * checks whether the classifier can handle the given percentage of
   * missing class labels
   *
   * @param type        the class type
   * @param percent     the percentage of missing class labels
   * @param allowFail   if true a fail statement may be executed
   * @return            true if the classifier can handle it
   */
  protected boolean checkMissingClass(int type, int percent, boolean allowFail) {
    boolean[]     result;
    
    result = m_Tester.canHandleMissing(
        m_NominalPredictors[type], 
        m_NumericPredictors[type], 
        m_StringPredictors[type], 
        m_DatePredictors[type], 
        m_RelationalPredictors[type], 
        m_multiInstanceHandler, 
        type,
        false,
        true,
        percent);

    if (allowFail) {
      if (!result[0] && !result[1])
        fail("Error handling " + percent + "% missing class labels (" 
            + getClassTypeString(type) + " class)!");
    }
    
    return result[0];
  }

  /**
   * tests whether the classifier can handle missing class values (20% and
   * 100%)
   *
   * @see CheckClassifier#canHandleMissing(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testMissingClass() {
    int           i;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      // 20% missing
      checkMissingClass(i, 20, true);

      // 100% missing
      if (m_handleMissingClass[i])
        checkMissingClass(i, 100, true);
    }
  }

  /**
   * tests whether the classifier correctly initializes in the
   * buildClassifier method
   *
   * @see CheckClassifier#correctBuildInitialisation(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testBuildInitialization() {
    boolean[]     result;
    int           i;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      result = m_Tester.correctBuildInitialisation(
          m_NominalPredictors[i], 
          m_NumericPredictors[i], 
          m_StringPredictors[i], 
          m_DatePredictors[i], 
          m_RelationalPredictors[i], 
          m_multiInstanceHandler, 
          i);

      if (!result[0] && !result[1])
        fail("Incorrect build initialization (" + getClassTypeString(i) 
            + " class)!");
    }
  }

  /**
   * tests whether the classifier alters the training set during training.
   *
   * @see CheckClassifier#datasetIntegrity(boolean, boolean, boolean, boolean, boolean, boolean, int, boolean, boolean)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testDatasetIntegrity() {
    boolean[]     result;
    int           i;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      result = m_Tester.datasetIntegrity(
          m_NominalPredictors[i], 
          m_NumericPredictors[i], 
          m_StringPredictors[i], 
          m_DatePredictors[i], 
          m_RelationalPredictors[i], 
          m_multiInstanceHandler, 
          i,
          m_handleMissingPredictors[i],
          m_handleMissingClass[i]);

      if (!result[0] && !result[1])
        fail("Training set is altered during training (" 
            + getClassTypeString(i) + " class)!");
    }
  }

  /**
   * tests whether the classifier erroneously uses the class value of test
   * instances (if provided)
   *
   * @see CheckClassifier#doesntUseTestClassVal(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testUseOfTestClassValue() {
    boolean[]     result;
    int           i;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
      
      result = m_Tester.doesntUseTestClassVal(
          m_NominalPredictors[i], 
          m_NumericPredictors[i], 
          m_StringPredictors[i], 
          m_DatePredictors[i], 
          m_RelationalPredictors[i], 
          m_multiInstanceHandler, 
          i);

      if (!result[0])
        fail("Uses test class values (" + getClassTypeString(i) + " class)!");
    }
  }

  /**
   * tests whether the classifier produces the same model when trained
   * incrementally as when batch trained.
   *
   * @see CheckClassifier#updatingEquality(boolean, boolean, boolean, boolean, boolean, boolean, int)
   * @see CheckClassifier#testsPerClassType(int, boolean, boolean, boolean)
   */
  public void testUpdatingEquality() {
    boolean[]     result;
    int           i;
    
    if (m_updateableClassifier) {
      for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
        // does the classifier support this type of class at all?
        if (!canPredict(i))
          continue;
        
        result = m_Tester.updatingEquality(
            m_NominalPredictors[i], 
            m_NumericPredictors[i], 
            m_StringPredictors[i], 
            m_DatePredictors[i], 
            m_RelationalPredictors[i], 
            m_multiInstanceHandler, 
            i);

        if (!result[0])
          System.err.println("Incremental training does not produce same result as "
              + "batch training (" + getClassTypeString(i) + " class)!");
      }
    }
  }

  /**
   * Builds a model using the current classifier using the first
   * half of the current data for training, and generates a bunch of
   * predictions using the remaining half of the data for testing.
   *
   * @param data        the instances to test the classifier on
   * @return a <code>FastVector</code> containing the predictions.
   */
  protected FastVector useClassifier(Instances data) throws Exception {
    Classifier dc = null;
    int tot = data.numInstances();
    int mid = tot / 2;
    Instances train = null;
    Instances test = null;
    EvaluationUtils evaluation = new EvaluationUtils();
    
    try {
      train = new Instances(data, 0, mid);
      test = new Instances(data, mid, tot - mid);
      dc = m_Classifier;
    } 
    catch (Exception e) {
      e.printStackTrace();
      fail("Problem setting up to use classifier: " + e);
    }

    do {
      try {
        return evaluation.getTrainTestPredictions(dc, train, test);
      } 
      catch (IllegalArgumentException e) {
        String msg = e.getMessage();
        if (msg.indexOf("Not enough instances") != -1) {
          System.err.println("\nInflating training data.");
          Instances trainNew = new Instances(train);
          for (int i = 0; i < train.numInstances(); i++) {
            trainNew.add(train.instance(i));
          }
          train = trainNew;
        } 
        else {
          throw e;
        }
      }
    } while (true);
  }

  /**
   * Returns a string containing all the predictions.
   *
   * @param predictions a <code>FastVector</code> containing the predictions
   * @return a <code>String</code> representing the vector of predictions.
   */
  public static String predictionsToString(FastVector predictions) {
    StringBuffer sb = new StringBuffer();
    sb.append(predictions.size()).append(" predictions\n");
    for (int i = 0; i < predictions.size(); i++) {
      sb.append(predictions.elementAt(i)).append('\n');
    }
    return sb.toString();
  }
  
  /**
   * Provides a hook for derived classes to further modify the data. Currently,
   * the data is just passed through.
   * 
   * @param data        the data to process
   * @return            the processed data
   */
  protected Instances process(Instances data) {
    return data;
  }

  /**
   * Runs a regression test -- this checks that the output of the tested
   * object matches that in a reference version. When this test is
   * run without any pre-existing reference output, the reference version
   * is created.
   */
  public void testRegression() throws Exception {
    int         i;
    boolean     succeeded;
    Regression  reg;
    Instances   train;
    
    // don't bother if not working correctly
    if (m_Tester.hasClasspathProblems())
      return;
    
    reg = new Regression(this.getClass());
    succeeded = false;
    train = null;
    
    for (i = FIRST_CLASSTYPE; i <= LAST_CLASSTYPE; i++) {
      // does the classifier support this type of class at all?
      if (!canPredict(i))
        continue;
        
      train = m_Tester.makeTestDataset(
          42, m_Tester.getNumInstances(), 
          m_NominalPredictors[i] ? m_Tester.getNumNominal() : 0,
          m_NumericPredictors[i] ? m_Tester.getNumNumeric() : 0, 
          m_StringPredictors[i] ? m_Tester.getNumString() : 0,
          m_DatePredictors[i] ? m_Tester.getNumDate() : 0,
          m_RelationalPredictors[i] ? m_Tester.getNumRelational() : 0,
          2, 
          i,
          m_multiInstanceHandler);
  
      try {
        m_RegressionResults[i] = useClassifier(train);
        succeeded = true;
        reg.println(predictionsToString(m_RegressionResults[i]));
      }
      catch (Exception e) {
        String msg = e.getMessage().toLowerCase();
        if (msg.indexOf("not in classpath") > -1)
          return;

        m_RegressionResults[i] = null;
      }
    }
    
    if (!succeeded) {
      fail("Problem during regression testing: no successful predictions for any class type");
    }

    try {
      String diff = reg.diff();
      if (diff == null) {
        System.err.println("Warning: No reference available, creating."); 
      } else if (!diff.equals("")) {
        fail("Regression test failed. Difference:\n" + diff);
      }
    } 
    catch (java.io.IOException ex) {
      fail("Problem during regression testing.\n" + ex);
    }
  }
  
  /**
   * tests the listing of the options
   */
  public void testListOptions() {
    if (!m_OptionTester.checkListOptions())
      fail("Options cannot be listed via listOptions.");
  }
  
  /**
   * tests the setting of the options
   */
  public void testSetOptions() {
    if (!m_OptionTester.checkSetOptions())
      fail("setOptions method failed.");
  }
  
  /**
   * tests whether the default settings are processed correctly
   */
  public void testDefaultOptions() {
    if (!m_OptionTester.checkDefaultOptions())
      fail("Default options were not processed correctly.");
  }
  
  /**
   * tests whether there are any remaining options
   */
  public void testRemainingOptions() {
    if (!m_OptionTester.checkRemainingOptions())
      fail("There were 'left-over' options.");
  }
  
  /**
   * tests the whether the user-supplied options stay the same after setting.
   * getting, and re-setting again.
   * 
   * @see       #getOptionTester()
   */
  public void testCanonicalUserOptions() {
    if (!m_OptionTester.checkCanonicalUserOptions())
      fail("setOptions method failed");
  }
  
  /**
   * tests the resetting of the options to the default ones
   */
  public void testResettingOptions() {
    if (!m_OptionTester.checkSetOptions())
      fail("Resetting of options failed");
  }
  
  /**
   * tests for a globalInfo method
   */
  public void testGlobalInfo() {
    if (!m_GOETester.checkGlobalInfo())
      fail("No globalInfo method");
  }
  
  /**
   * tests the tool tips
   */
  public void testToolTips() {
    if (!m_GOETester.checkToolTips())
      fail("Tool tips inconsistent");
  }
}