nd.java

代码是一个分类器的实现,其中使用了部分weka的源代码。可以将项目导入eclipse运行

        m_right.unifyTree();
      }
    }

    /**
     * Returns a description of the tree rooted at this node.
     * 
     * @param text the buffer to add the node to
     * @param id the node id
     * @param level the level of the tree
     */
    protected void toString(StringBuffer text, int[] id, int level) {

      for (int i = 0; i < level; i++) {
	text.append("   | ");
      }
      text.append(id[0] + ": " + getString() + "\n");
      if (m_left != null) {
	id[0]++;
	m_left.toString(text, id, level + 1);
	id[0]++;
	m_right.toString(text, id, level + 1);
      }
    }
  }

  /** The tree of classes */
  protected NDTree m_ndtree = null;
  
  /** The hashtable containing all the classifiers */
  protected Hashtable m_classifiers = null;

  /** Is Hashtable given from END? */
  protected boolean m_hashtablegiven = false;
    
  /**
   * Constructor.
   */
  public ND() {
    
    m_Classifier = new weka.classifiers.trees.J48();
  }
  
  /**
   * String describing default classifier.
   * 
   * @return the default classifier classname
   */
  protected String defaultClassifierString() {
    
    return "weka.classifiers.trees.J48";
  }

  /**
   * Returns an instance of a TechnicalInformation object, containing 
   * detailed information about the technical background of this class,
   * e.g., paper reference or book this class is based on.
   * 
   * @return the technical information about this class
   */
  public TechnicalInformation getTechnicalInformation() {
    TechnicalInformation 	result;
    TechnicalInformation 	additional;
    
    result = new TechnicalInformation(Type.INPROCEEDINGS);
    result.setValue(Field.AUTHOR, "Lin Dong and Eibe Frank and Stefan Kramer");
    result.setValue(Field.TITLE, "Ensembles of Balanced Nested Dichotomies for Multi-class Problems");
    result.setValue(Field.BOOKTITLE, "PKDD");
    result.setValue(Field.YEAR, "2005");
    result.setValue(Field.PAGES, "84-95");
    result.setValue(Field.PUBLISHER, "Springer");

    additional = result.add(Type.INPROCEEDINGS);
    additional.setValue(Field.AUTHOR, "Eibe Frank and Stefan Kramer");
    additional.setValue(Field.TITLE, "Ensembles of nested dichotomies for multi-class problems");
    additional.setValue(Field.BOOKTITLE, "Twenty-first International Conference on Machine Learning");
    additional.setValue(Field.YEAR, "2004");
    additional.setValue(Field.PUBLISHER, "ACM");
    
    return result;
  }

  /**
   * Set hashtable from END.
   * 
   * @param table the hashtable to use
   */
  public void setHashtable(Hashtable table) {

    m_hashtablegiven = true;
    m_classifiers = table;
  }

  /**
   * Returns default capabilities of the classifier.
   *
   * @return      the capabilities of this classifier
   */
  public Capabilities getCapabilities() {
    Capabilities result = super.getCapabilities();

    // class
    result.disableAllClasses();
    result.enable(Capability.NOMINAL_CLASS);
    result.enable(Capability.MISSING_CLASS_VALUES);

    // instances
    result.setMinimumNumberInstances(1);
    
    return result;
  }

  /**
   * Builds the classifier.
   * 
   * @param data the data to train the classifier with
   * @throws Exception if anything goes wrong
   */
  public void buildClassifier(Instances data) throws Exception {

    // can classifier handle the data?
    getCapabilities().testWithFail(data);

    // remove instances with missing class
    data = new Instances(data);
    data.deleteWithMissingClass();
    
    Random random = data.getRandomNumberGenerator(m_Seed);

    if (!m_hashtablegiven) {
      m_classifiers = new Hashtable();
    }

    // Generate random class hierarchy
    int[] indices = new int[data.numClasses()];
    for (int i = 0; i < indices.length; i++) {
      indices[i] = i;
    }

    // Randomize list of class indices
    for (int i = indices.length - 1; i > 0; i--) {
      int help = indices[i];
      int index = random.nextInt(i + 1);
      indices[i] = indices[index];
      indices[index] = help;
    }

    // Insert random class index at randomly chosen node
    m_ndtree = new NDTree();
    m_ndtree.insertClassIndexAtNode(indices[0]);
    for (int i = 1; i < indices.length; i++) {
      int nodeIndex = random.nextInt(2 * i - 1);
     
      NDTree node = m_ndtree.locateNode(nodeIndex, new int[1]);
      node.insertClassIndex(indices[i]);
    }
    m_ndtree.unifyTree();
    

    // Build classifiers
    buildClassifierForNode(m_ndtree, data);
  }

  /**
   * Builds the classifier for one node.
   * 
   * @param node the node to build the classifier for
   * @param data the data to work with
   * @throws Exception if anything goes wrong
   */
  public void buildClassifierForNode(NDTree node, Instances data) throws Exception {

    // Are we at a leaf node ?
    if (node.m_left != null) {
      
      // Create classifier
      MakeIndicator filter = new MakeIndicator();
      filter.setAttributeIndex("" + (data.classIndex() + 1));
      filter.setValueIndices(node.m_right.getString());
      filter.setNumeric(false);
      filter.setInputFormat(data);
      FilteredClassifier classifier = new FilteredClassifier();
      if (data.numInstances() > 0) {
	classifier.setClassifier(Classifier.makeCopies(m_Classifier, 1)[0]);
      } else {
	classifier.setClassifier(new ZeroR());
      }
      classifier.setFilter(filter);
      
      if (!m_classifiers.containsKey(node.m_left.getString() + "|" + node.m_right.getString())) {
	classifier.buildClassifier(data);
	m_classifiers.put(node.m_left.getString() + "|" + node.m_right.getString(), classifier);
      } else {
	classifier=(FilteredClassifier)m_classifiers.get(node.m_left.getString() + "|" + 
							 node.m_right.getString());
      }
      
      // Generate successors
      if (node.m_left.m_left != null) {
        RemoveWithValues rwv = new RemoveWithValues();
        rwv.setInvertSelection(true);
        rwv.setNominalIndices(node.m_left.getString());
        rwv.setAttributeIndex("" + (data.classIndex() + 1));
        rwv.setInputFormat(data);
        Instances firstSubset = Filter.useFilter(data, rwv);
        buildClassifierForNode(node.m_left, firstSubset);
      }
      if (node.m_right.m_left != null) {
        RemoveWithValues rwv = new RemoveWithValues();
        rwv.setInvertSelection(true);
        rwv.setNominalIndices(node.m_right.getString());
        rwv.setAttributeIndex("" + (data.classIndex() + 1));
        rwv.setInputFormat(data);
        Instances secondSubset = Filter.useFilter(data, rwv);
        buildClassifierForNode(node.m_right, secondSubset);
      }
    }
  }
    
  /**
   * Predicts the class distribution for a given instance
   *
   * @param inst the (multi-class) instance to be classified
   * @return the class distribution
   * @throws Exception if computing fails
   */
  public double[] distributionForInstance(Instance inst) throws Exception {
	
    return distributionForInstance(inst, m_ndtree);
  }

  /**
   * Predicts the class distribution for a given instance
   *
   * @param inst the (multi-class) instance to be classified
   * @param node the node to do get the distribution for
   * @return the class distribution
   * @throws Exception if computing fails
   */
  protected double[] distributionForInstance(Instance inst, NDTree node) throws Exception {

    double[] newDist = new double[inst.numClasses()];
    if (node.m_left == null) {
      newDist[node.getIndices()[0]] = 1.0;
      return newDist;
    } else {
      Classifier classifier = (Classifier)m_classifiers.get(node.m_left.getString() + "|" +
							    node.m_right.getString());
      double[] leftDist = distributionForInstance(inst, node.m_left);
      double[] rightDist = distributionForInstance(inst, node.m_right);
      double[] dist = classifier.distributionForInstance(inst);

      for (int i = 0; i < inst.numClasses(); i++) {
	if (node.m_right.contains(i)) {
	  newDist[i] = dist[1] * rightDist[i];
	} else {
	  newDist[i] = dist[0] * leftDist[i];
	}
      }
      return newDist;
    }
  }

  /**
   * Outputs the classifier as a string.
   * 
   * @return a string representation of the classifier
   */
  public String toString() {
	
    if (m_classifiers == null) {
      return "ND: No model built yet.";
    }
    StringBuffer text = new StringBuffer();
    text.append("ND\n\n");
    m_ndtree.toString(text, new int[1], 0);
	
    return text.toString();
  }
	
  /**
   * @return a description of the classifier suitable for
   * displaying in the explorer/experimenter gui
   */
  public String globalInfo() {
	    
    return 
        "A meta classifier for handling multi-class datasets with 2-class "
      + "classifiers by building a random tree structure.\n\n"
      + "For more info, check\n\n"
      + getTechnicalInformation().toString();
  }
    
  /**
   * Main method for testing this class.
   *
   * @param argv the options
   */
  public static void main(String [] argv) {
    runClassifier(new ND(), argv);
  }
}