qboost.java

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

  /**
   * Get whether resampling is turned on
   *
   * @return true if resampling output is on
   */
  public boolean getUseResampling() {

    return m_UseResampling;
  }

  /**
   * Boosting method.
   *
   * @param data the training data to be used for generating the
   * boosted classifier.
   * @exception Exception if the classifier could not be built successfully
   */

  public void buildClassifier(Instances data) throws Exception {
      
      //Initialize measures
      initMeasures();
      
    if (data.checkForStringAttributes()) {
      throw new UnsupportedAttributeTypeException("Cannot handle string attributes!");
    }
    data = new Instances(data);
    data.deleteWithMissingClass();
    if (data.numInstances() == 0) {
      throw new Exception("No train instances without class missing!");
    }
    if (data.classAttribute().isNumeric()) {
      throw new UnsupportedClassTypeException("QBoost can't handle a numeric class!");
    }
    if (m_Classifier == null) {
      throw new Exception("A base classifier has not been specified!");
    }
    m_NumClasses = data.numClasses();
    m_Classifiers = Classifier.makeCopies(m_Classifier, getMaxIterations());
    if ((!m_UseResampling) && 
	(m_Classifier instanceof WeightedInstancesHandler)) {
      buildClassifierWithWeights(data);
    } else {
      buildClassifierUsingResampling(data);
    }

    //=============== BEGIN EDIT melville ===============
    m_EnsembleWts = new double [m_NumIterations];
    if(m_NumIterations==1)
	m_EnsembleWts[0] = 1.0;
    else{
	for(int i=0; i<m_NumIterations; i++)
	    m_EnsembleWts[i] =  m_Betas[i];
    } 
    computeEnsembleMeasures(data);
    //=============== END EDIT melville ===============
  }

  /**
   * Boosting method. Boosts using resampling
   *
   * @param data the training data to be used for generating the
   * boosted classifier.
   * @exception Exception if the classifier could not be built successfully
   */
  protected void buildClassifierUsingResampling(Instances data) 
    throws Exception {

    Instances trainData, sample, training;
    double epsilon, reweight, beta = 0, sumProbs;
    double oldSumOfWeights, newSumOfWeights;
    Evaluation evaluation;
    int numInstances = data.numInstances();
    Random randomInstance = new Random(m_Seed);
    double[] probabilities;
    int resamplingIterations = 0;
    int k, l;

    // Initialize data
    m_Betas = new double [m_Classifiers.length];
    m_NumIterations = 0;
    // Create a copy of the data so that when the weights are diddled
    // with it doesn't mess up the weights for anyone else
    training = new Instances(data, 0, numInstances);
    sumProbs = training.sumOfWeights();
    for (int i = 0; i < training.numInstances(); i++) {
      training.instance(i).setWeight(training.instance(i).
				      weight() / sumProbs);
    }
    
    // Do boostrap iterations
    for (m_NumIterations = 0; m_NumIterations < m_Classifiers.length; 
	 m_NumIterations++) {
      if (m_Debug) {
	System.err.println("Training classifier " + (m_NumIterations + 1));
      }

      // Select instances to train the classifier on
      if (m_WeightThreshold < 100) {
	trainData = selectWeightQuantile(training, 
					 (double)m_WeightThreshold / 100);
      } else {
	trainData = new Instances(training);
      }
      
      // Resample
      resamplingIterations = 0;
      double[] weights = new double[trainData.numInstances()];
      for (int i = 0; i < weights.length; i++) {
	weights[i] = trainData.instance(i).weight();
      }
      do {
	sample = trainData.resampleWithWeights(randomInstance, weights);

	// Build and evaluate classifier
	m_Classifiers[m_NumIterations].buildClassifier(sample);
	evaluation = new Evaluation(data);
	evaluation.evaluateModel(m_Classifiers[m_NumIterations], 
				 training);
	epsilon = evaluation.errorRate();
	resamplingIterations++;
      } while (Utils.eq(epsilon, 0) && 
	      (resamplingIterations < MAX_NUM_RESAMPLING_ITERATIONS));
      	
      // Stop if error too big or 0
      if (Utils.grOrEq(epsilon, 0.5) || Utils.eq(epsilon, 0)) {
	if (m_NumIterations == 0) {
	  m_NumIterations = 1; // If we're the first we have to to use it
	}
	break;
      }
      
      // Determine the weight to assign to this model
      m_Betas[m_NumIterations] = beta = Math.log((1 - epsilon) / epsilon);
      reweight = (1 - epsilon) / epsilon;
      if (m_Debug) {
	System.err.println("\terror rate = " + epsilon
			   +"  beta = " + m_Betas[m_NumIterations]);
      }
 
      // Update instance weights
      oldSumOfWeights = training.sumOfWeights();
      Enumeration enum = training.enumerateInstances();
      while (enum.hasMoreElements()) {
	Instance instance = (Instance) enum.nextElement();
	if (!Utils.eq(m_Classifiers[m_NumIterations].classifyInstance(instance), 
		     instance.classValue()))
	  instance.setWeight(instance.weight() * reweight);
      }
      // Renormalize weights
      newSumOfWeights = training.sumOfWeights();
      enum = training.enumerateInstances();
      while (enum.hasMoreElements()) {
	Instance instance = (Instance) enum.nextElement();
	instance.setWeight(instance.weight() * oldSumOfWeights 
			   / newSumOfWeights);
      }
    }
  }

  /**
   * Boosting method. Boosts any classifier that can handle weighted
   * instances.
   *
   * @param data the training data to be used for generating the
   * boosted classifier.
   * @exception Exception if the classifier could not be built successfully
   */
  protected void buildClassifierWithWeights(Instances data) 
    throws Exception {

    Instances trainData, training;
    double epsilon, reweight, beta = 0;
    double oldSumOfWeights, newSumOfWeights;
    Evaluation evaluation;
    int numInstances = data.numInstances();

    // Initialize data
    m_Betas = new double [m_Classifiers.length];
    m_NumIterations = 0;

    // Create a copy of the data so that when the weights are diddled
    // with it doesn't mess up the weights for anyone else
    training = new Instances(data, 0, numInstances);
    
    // Do boostrap iterations
    for (m_NumIterations = 0; m_NumIterations < m_Classifiers.length; 
	 m_NumIterations++) {
      if (m_Debug) {
	System.err.println("Training classifier " + (m_NumIterations + 1));
      }
      // Select instances to train the classifier on
      if (m_WeightThreshold < 100) {
	trainData = selectWeightQuantile(training, 
					 (double)m_WeightThreshold / 100);
      } else {
	trainData = new Instances(training, 0, numInstances);
      }

      // Build the classifier
      m_Classifiers[m_NumIterations].buildClassifier(trainData);

      // Evaluate the classifier
      evaluation = new Evaluation(data);
      evaluation.evaluateModel(m_Classifiers[m_NumIterations], training);
      epsilon = evaluation.errorRate();

      // Stop if error too small or error too big and ignore this model
      if (Utils.grOrEq(epsilon, 0.5) || Utils.eq(epsilon, 0)) {
	  if (m_NumIterations == 0) {
	  m_NumIterations = 1; // If we're the first we have to to use it
	}
	break;
      }
      // Determine the weight to assign to this model
      m_Betas[m_NumIterations] = beta = Math.log((1 - epsilon) / epsilon);
      reweight = (1 - epsilon) / epsilon;
      if (m_Debug) {
	System.err.println("\terror rate = " + epsilon
			   +"  beta = " + m_Betas[m_NumIterations]);
      }
 
      // Update instance weights
      oldSumOfWeights = training.sumOfWeights();
      Enumeration enum = training.enumerateInstances();
      while (enum.hasMoreElements()) {
	Instance instance = (Instance) enum.nextElement();
	if (!Utils.eq(m_Classifiers[m_NumIterations]
		      .classifyInstance(instance), 
		      instance.classValue()))
	  instance.setWeight(instance.weight() * reweight);
      }
      // Renormalize weights
      newSumOfWeights = training.sumOfWeights();
      enum = training.enumerateInstances();
      while (enum.hasMoreElements()) {
	Instance instance = (Instance) enum.nextElement();
	instance.setWeight(instance.weight() * oldSumOfWeights
			   / newSumOfWeights);
      }
    }
  }
  
  /**
   * Calculates the class membership probabilities for the given test instance.
   *
   * @param instance the instance to be classified
   * @return predicted class probability distribution
   * @exception Exception if instance could not be classified
   * successfully
   */
  public double [] distributionForInstance(Instance instance) 
    throws Exception {
      if (m_NumIterations == 0) {
      throw new Exception("No model built");
    }
    double [] sums = new double [instance.numClasses()]; 
    
    if (m_NumIterations == 1) {
      if (m_Classifiers[0] instanceof DistributionClassifier) {
	return ((DistributionClassifier)m_Classifiers[0]).
	distributionForInstance(instance);
      } else {
	sums[(int)m_Classifiers[0].classifyInstance(instance)] ++;
      }
    } else {
      for (int i = 0; i < m_NumIterations; i++) {
	sums[(int)m_Classifiers[i].classifyInstance(instance)] += 
	m_Betas[i];
      }
    }
    Utils.normalize(sums);
    return sums;
  }

    /** 
     * Given a set of unlabeled examples, select a specified number of examples to be labeled.
     * @param unlabeledActivePool pool of unlabeled examples
     * @param num number of examples to selcted for labeling
     * @exception Exception if selective sampling fails
     */
    public int [] selectInstances(Instances unlabeledActivePool,int num) throws Exception{
	//Make a list of pairs of indices and the corresponding measure of informativenes of examples
	//Sort this in the order of informativeness and return the list of num indices
	int poolSize = unlabeledActivePool.numInstances();
	Pair []pairs = new Pair[poolSize];
	for(int i=0; i<poolSize; i++){
	    pairs[i] = new Pair(i,calculateMargin(unlabeledActivePool.instance(i)));
	}
	//sort in ascending order
	Arrays.sort(pairs, new Comparator() {
                public int compare(Object o1, Object o2) {
		    double diff = ((Pair)o2).second - ((Pair)o1).second; 
		    return(diff < 0 ? 1 : diff > 0 ? -1 : 0);
		}
            });
	int []selected = new int[num];
	if(m_Debug) System.out.println("Sorted list:");
	for(int j=0; j<num; j++){
	    if(m_Debug) System.out.println("\t"+pairs[j].second+"\t"+pairs[j].first);
	    selected[j] = (int) pairs[j].first;
	}
	return selected;
    }
    

    //=============== BEGIN EDIT melville ===============
    /** Returns class predictions of each ensemble member */
    public double []getEnsemblePredictions(Instance instance) throws Exception{
	double preds[] = new double [m_NumIterations];
	for(int i=0; i<m_NumIterations; i++)
	    preds[i] = m_Classifiers[i].classifyInstance(instance);
	
	return preds;
    }
    
    /** 
     * Returns vote weights of ensemble members.
     *
     * @return vote weights of ensemble members
     */
    public double []getEnsembleWts(){
	return m_EnsembleWts;
    }
    
    /** Returns size of ensemble */
    public double getEnsembleSize(){
	return m_NumIterations;
    }
    //=============== END EDIT melville ===============
    
   /**
   * Returns the boosted model as Java source code.
   *
   * @return the tree as Java source code
   * @exception Exception if something goes wrong
   */
  public String toSource(String className) throws Exception {

    if (m_NumIterations == 0) {
      throw new Exception("No model built yet");
    }
    if (!(m_Classifiers[0] instanceof Sourcable)) {
      throw new Exception("Base learner " + m_Classifier.getClass().getName()
			  + " is not Sourcable");
    }

    StringBuffer text = new StringBuffer("class ");
    text.append(className).append(" {\n\n");

    text.append("  public static double classify(Object [] i) {\n");

    if (m_NumIterations == 1) {
      text.append("    return " + className + "_0.classify(i);\n");
    } else {
      text.append("    double [] sums = new double [" + m_NumClasses + "];\n");
      for (int i = 0; i < m_NumIterations; i++) {
	text.append("    sums[(int) " + className + '_' + i 
		    + ".classify(i)] += " + m_Betas[i] + ";\n");
      }
      text.append("    double maxV = sums[0];\n" +
		  "    int maxI = 0;\n"+
		  "    for (int j = 1; j < " + m_NumClasses + "; j++) {\n"+
		  "      if (sums[j] > maxV) { maxV = sums[j]; maxI = j; }\n"+
		  "    }\n    return (double) maxI;\n");
    }
    text.append("  }\n}\n");

    for (int i = 0; i < m_Classifiers.length; i++) {
	text.append(((Sourcable)m_Classifiers[i])
		    .toSource(className + '_' + i));
    }
    return text.toString();
  }

  /**
   * Returns description of the boosted classifier.
   *
   * @return description of the boosted classifier as a string
   */
  public String toString() {
    
    StringBuffer text = new StringBuffer();
    
    if (m_NumIterations == 0) {
      text.append("QBoost: No model built yet.\n");
    } else if (m_NumIterations == 1) {
      text.append("QBoost: No boosting possible, one classifier used!\n");
      text.append(m_Classifiers[0].toString() + "\n");
    } else {
      text.append("QBoost: Base classifiers and their weights: \n\n");
      for (int i = 0; i < m_NumIterations ; i++) {
	text.append(m_Classifiers[i].toString() + "\n\n");
	text.append("Weight: " + Utils.roundDouble(m_Betas[i], 2) + "\n\n");
      }
      text.append("Number of performed Iterations: " 
		  + m_NumIterations + "\n");
    }
    
    return text.toString();
  }
    
    
  /**
   * Main method for testing this class.
   *
   * @param argv the options
   */
  public static void main(String [] argv) {

    try {
      System.out.println(Evaluation.evaluateModel(new QBoost(), argv));
    } catch (Exception e) {
      System.err.println(e.getMessage());
    }
  }
}