.#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

	}
      }
      ((MPCKMeans)m_Clusterer).buildClusterer(labeledTrainPairs, unlabeledData, labeledTrain, labeledTrain.numClasses(), labeledTrain.numInstances(),homologHash); // KLUGE: have to generalize later
    } else {
      throw new Exception ("Inappropriate clusterer: " + m_Clusterer.getClass().getName());
    } 

    //    ((SeededKMeans)m_Clusterer).printClusters();

    int numClusters = labeledTrain.numClasses();
    if (m_Clusterer instanceof SemiSupClusterer) {
      numClusters = ((SemiSupClusterer)m_Clusterer).getNumClusters();
    }

    SemiSupClustererEvaluation eval = new SemiSupClustererEvaluation(labeledTrainPairs, test, labeledTrain.numClasses(), numClusters);

    long trainTimeElapsed = System.currentTimeMillis() - trainTimeStart;
    long testTimeStart = System.currentTimeMillis();
    eval.evaluateModel(m_Clusterer, test, unlabeledTest, homologHash);
    long testTimeElapsed = System.currentTimeMillis() - testTimeStart;
    m_result = eval.toSummaryString();

    // The results stored are all per instance -- can be multiplied by the
    // number of instances to get absolute numbers
    int current = 0;    

    // Unsupervised stats: 3
    result[current++] = new Double(eval.purity());
    result[current++] = new Double(eval.entropy());
    result[current++] = new Double(eval.objectiveFunction());

    // Supervised stats: 3
    result[current++] = new Double(eval.klDivergence());
    result[current++] = new Double(eval.mutualInformation());
    result[current++] = new Double(eval.supervisedDispersion());

    // Training data stats: 2
    result[current++] = new Double(eval.numSameClassPairs());
    result[current++] = new Double(eval.numDiffClassPairs());

    // IR stats: 3
    result[current++] = new Double(eval.pairwisePrecision());
    result[current++] = new Double(eval.pairwiseRecall());
    result[current++] = new Double(eval.pairwiseFMeasure());

    // Timing stats: 2
    result[current++] = new Double(trainTimeElapsed / 1000.0);
    result[current++] = new Double(testTimeElapsed / 1000.0);

    // Clusterer defined extras: 1
    if (m_Clusterer instanceof Summarizable) {
      result[current++] = ((Summarizable)m_Clusterer).toSummaryString();
    } else {
      result[current++] = null;
    }

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

  /**
   * Gets the results for the supplied train and test datasets.
   *
   * @param labeledTrain the labeled training Instances.
   * @param unlabeledTrain the unlabeled training Instances.
   * @param test the testing Instances.
   * @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(Instances labeledTrain, Instances unlabeledTrain, Instances test, int numClasses) 
    throws Exception {
    try {
      return getResult(labeledTrain, 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 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(Instances labeledTrain, Instances unlabeledData, Instances totalTrainWithLabels, Instances test, int startingIndexOfTest) 
    throws Exception {
    
    if (labeledTrain.classAttribute().type() != Attribute.NOMINAL) {
      throw new WekaException("Class attribute is not nominal!");
    }
    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();
    int classIndex = labeledTrain.numAttributes()-1; // assuming that the last attribute is always the class

    ((SeededKMeans)m_Clusterer).buildClusterer(labeledTrain, unlabeledData, classIndex, totalTrainWithLabels, startingIndexOfTest);

    int numClusters = totalTrainWithLabels.numClasses();
    if (m_Clusterer instanceof SemiSupClusterer) {
      numClusters = ((SemiSupClusterer)m_Clusterer).getNumClusters();
    }

    SemiSupClustererEvaluation eval = new SemiSupClustererEvaluation(test, totalTrainWithLabels.numClasses(), numClusters);

    long trainTimeElapsed = System.currentTimeMillis() - trainTimeStart;
    long testTimeStart = System.currentTimeMillis();
    Instances unlabeledTest = new Instances (test);
    unlabeledTest.deleteClassAttribute();
    eval.evaluateModel(m_Clusterer, test, unlabeledTest, null);
    long testTimeElapsed = System.currentTimeMillis() - testTimeStart;
    m_result = eval.toSummaryString();

    // The results stored are all per instance -- can be multiplied by the
    // number of instances to get absolute numbers
    int current = 0;    

    // Unsupervised stats: 3
    result[current++] = new Double(eval.purity());
    result[current++] = new Double(eval.entropy());
    result[current++] = new Double(eval.objectiveFunction());

    // Supervised stats: 3
    result[current++] = new Double(eval.klDivergence());
    result[current++] = new Double(eval.mutualInformation());
    result[current++] = new Double(eval.supervisedDispersion());

    // Training data stats: 2 - there are no training pairs in this case
    result[current++] = new Double(0);
    result[current++] = new Double(0);

    // IR stats: 3
    result[current++] = new Double(eval.pairwisePrecision());
    result[current++] = new Double(eval.pairwiseRecall());
    result[current++] = new Double(eval.pairwiseFMeasure());

    // Timing stats: 2
    result[current++] = new Double(trainTimeElapsed / 1000.0);
    result[current++] = new Double(testTimeElapsed / 1000.0);

    // Clusterer defined extras: 1
    if (m_Clusterer instanceof Summarizable) {
      result[current++] = ((Summarizable)m_Clusterer).toSummaryString();
    } else {
      result[current++] = null;
    }

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


  /**
   * Gets the results for the supplied train and test datasets.
   *
   * @param labeledTrain the labeled training Instances.
   * @param unlabeledTrain the unlabeled training 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(Instances labeledTrain, Instances unlabeledTrain, Instances test, int numClasses, int startingIndexOfTest) 
    throws Exception {
    
    if (labeledTrain.classAttribute().type() != Attribute.NOMINAL) {
      throw new WekaException("Class attribute is not nominal!");
    }
    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();
    int classIndex = labeledTrain.numAttributes()-1; // assuming that the last attribute is always the class

    ((SemiSupClusterer)m_Clusterer).buildClusterer(labeledTrain, unlabeledTrain, classIndex, numClasses, startingIndexOfTest);

    int numClusters = numClasses;
    if (m_Clusterer instanceof SemiSupClusterer) {
      numClusters = ((SemiSupClusterer)m_Clusterer).getNumClusters();
    }

    SemiSupClustererEvaluation eval = new SemiSupClustererEvaluation(test, numClasses, numClusters);

    long trainTimeElapsed = System.currentTimeMillis() - trainTimeStart;
    long testTimeStart = System.currentTimeMillis();
    Instances unlabeledTest = new Instances (test);
    unlabeledTest.deleteClassAttribute();
    eval.evaluateModel(m_Clusterer, test, unlabeledTest, null);
    long testTimeElapsed = System.currentTimeMillis() - testTimeStart;
    m_result = eval.toSummaryString();

    // The results stored are all per instance -- can be multiplied by the
    // number of instances to get absolute numbers
    int current = 0;    

    // Unsupervised stats: 3
    result[current++] = new Double(eval.purity());
    result[current++] = new Double(eval.entropy());
    result[current++] = new Double(eval.objectiveFunction());

    // Supervised stats: 3
    result[current++] = new Double(eval.klDivergence());
    result[current++] = new Double(eval.mutualInformation());
    result[current++] = new Double(eval.supervisedDispersion());

    // Training data stats: 2 - there are no training pairs in this case
    result[current++] = new Double(0);
    result[current++] = new Double(0);

    // IR stats: 3
    result[current++] = new Double(eval.pairwisePrecision());
    result[current++] = new Double(eval.pairwiseRecall());
    result[current++] = new Double(eval.pairwiseFMeasure());

    // Timing stats: 2
    result[current++] = new Double(trainTimeElapsed / 1000.0);
    result[current++] = new Double(testTimeElapsed / 1000.0);

    // Clusterer defined extras: 1
    if (m_Clusterer instanceof Summarizable) {
      result[current++] = ((Summarizable)m_Clusterer).toSummaryString();
    } else {
      result[current++] = null;
    }

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

  /**
   * Returns the tip text for this property
   * @return tip text for this property suitable for
   * displaying in the explorer/experimenter gui
   */
  public String clustererTipText() {
    return "The clusterer to use.";
  }

  /**
   * Get the value of Clusterer.
   *
   * @return Value of Clusterer.
   */
  public Clusterer getClusterer() {
    
    return m_Clusterer;
  }
  
  /**
   * Sets the clusterer.
   *
   * @param newClusterer the new clusterer to use.
   */
  public void setClusterer(Clusterer newClusterer) {
    
    m_Clusterer = newClusterer;
    updateOptions();
    
    System.err.println("SemiSupClustererSplitEvaluator: In set clusterer");
  }
  
  /**
   * Get the value of ClassForIRStatistics.
   * @return Value of ClassForIRStatistics.
   */
  public int getClassForIRStatistics() {
    
    return m_IRclass;
  }
  
  /**
   * Set the value of ClassForIRStatistics.
   * @param v  Value to assign to ClassForIRStatistics.
   */
  public void setClassForIRStatistics(int v) {
    
    m_IRclass = v;
  }
  
  /**
   * Updates the options that the current clusterer is using.
   */
  protected void updateOptions() {
    
    if (m_Clusterer instanceof OptionHandler) {
      m_ClustererOptions = Utils.joinOptions(((OptionHandler)m_Clusterer)
					      .getOptions());
    } else {
      m_ClustererOptions = "";
    }
    if (m_Clusterer instanceof Serializable) {
      ObjectStreamClass obs = ObjectStreamClass.lookup(m_Clusterer
						       .getClass());
      m_ClustererVersion = "" + obs.getSerialVersionUID();
    } else {
      m_ClustererVersion = "";
    }
  }

  /**
   * Set the Clusterer to use, given it's class name. A new clusterer will be
   * instantiated.
   *
   * @param newClusterer the Clusterer class name.
   * @exception Exception if the class name is invalid.
   */
  public void setClustererName(String newClustererName) throws Exception {

    try {
      setClusterer((Clusterer)Class.forName(newClustererName)
		    .newInstance());
    } catch (Exception ex) {
      throw new Exception("Can't find Clusterer with class name: "
			  + newClustererName);
    }
  }

  /**
   * Gets the raw output from the clusterer
   * @return the raw output from the clusterer
   */
  public String getRawResultOutput() {
    StringBuffer result = new StringBuffer();

    if (m_Clusterer == null) {
      return "<null> clusterer";
    }
    result.append(toString());
    result.append("Clusterer model: \n"+m_Clusterer.toString()+'\n');

    // append the performance statistics
    if (m_result != null) {
      result.append(m_result);
    }
    return result.toString();
  }

  /**
   * Returns a text description of the split evaluator.
   *
   * @return a text description of the split evaluator.
   */
  public String toString() {

    String result = "SemiSupClustererSplitEvaluator: ";
    if (m_Clusterer == null) {
      return result + "<null> clusterer";
    }
    return result + m_Clusterer.getClass().getName() + " " 
      + m_ClustererOptions + "(version " + m_ClustererVersion + ")";
  }
} // SemiSupClustererSplitEvaluator