classifiertree.java

一个数据挖掘软件ALPHAMINERR的整个过程的JAVA版源代码

	.append(") {\n");
      text.append("      p = ")
	.append(m_localModel.distribution().maxClass(0))
	.append(";\n");
      text.append("    } ");
      for (int i = 0; i < m_sons.length; i++) {
	text.append("else if (" + m_localModel.sourceExpression(i, m_train) 
		    + ") {\n");
	if (m_sons[i].m_isLeaf) {
	  text.append("      p = " 
		      + m_localModel.distribution().maxClass(i) + ";\n");
	} else {
	  StringBuffer [] sub = m_sons[i].toSource(className);
	  text.append(sub[0]);
	  atEnd.append(sub[1]);
	}
	text.append("    } ");
	if (i == m_sons.length - 1) {
	  text.append('\n');
	}
      }

      text.append("    return p;\n  }\n");

      result[0] = new StringBuffer("    p = " + className + ".N");
      result[0].append(Integer.toHexString(m_localModel.hashCode()) +  printID)
	.append("(i);\n");
      result[1] = text.append(atEnd);
    }
    return result;
  }

  /**
   * Returns number of leaves in tree structure.
   */
  public int numLeaves() {
    
    int num = 0;
    int i;
    
    if (m_isLeaf)
      return 1;
    else
      for (i=0;i<m_sons.length;i++)
	num = num+m_sons[i].numLeaves();
        
    return num;
  }

  /**
   * Returns number of nodes in tree structure.
   */
  public int numNodes() {
    
    int no = 1;
    int i;
    
    if (!m_isLeaf)
      for (i=0;i<m_sons.length;i++)
	no = no+m_sons[i].numNodes();
    
    return no;
  }

  /**
   * Prints tree structure.
   */
  public String toString() {

		try {
			StringBuffer text = new StringBuffer(); 
			
			if (m_isLeaf) {
				text.append(": ");
				text.append(m_localModel.dumpLabel(0, m_train));
				
			} else
				dumpTree(1, text);
			text.append("\r\n\r\nNumber of Leaves  : \t" + numLeaves() + "\r\n");
			text.append("\r\nSize of the tree : \t" + numNodes() + "\r\n");

			return text.toString();
		} catch (Exception e) {
			return "Can't print classification tree.";
		}
	}

  /**
   * Returns a newly created tree.
   * 
   * @param data
   *            the training data
   * @exception Exception
   *                if something goes wrong
   */
  protected ClassifierTree getNewTree(Instances data) throws Exception {
	 
    ClassifierTree newTree = new ClassifierTree(m_toSelectModel);
    newTree.buildTree(data, false);
    
    return newTree;
  }

  /**
   * Returns a newly created tree.
   *
   * @param data the training data
   * @param test the pruning data.
   * @exception Exception if something goes wrong
   */
  protected ClassifierTree getNewTree(Instances train, Instances test) 
       throws Exception {
	 
    ClassifierTree newTree = new ClassifierTree(m_toSelectModel);
    newTree.buildTree(train, test, false);
    
    return newTree;
  }

  /**
   * Help method for printing tree structure.
   *
   * @exception Exception if something goes wrong
   */
  private void dumpTree(int depth,StringBuffer text) 
       throws Exception {
    
    int i,j; 
    
    for (i = 0; i < m_sons.length; i++) {
			text.append("\r\n");
			
			for (j = 0; j < depth; j++)
			{
				if (j==depth-1) text.append("+---");
				else text.append("|   ");
			}
			text.append(m_localModel.leftSide(m_train));
			text.append(m_localModel.rightSide(i, m_train)); 
			
			// Create statistic distribution string
			StringBuffer statStringBuffer = new StringBuffer();
			double[] stats = getM_localModel().distribution().getClassDist();
			double total=0;
			
			// Calculate total
			for(int index=0; index<stats.length; index++){   
				total += stats[index];	
			}
			statStringBuffer.append("Total:(" + NumberFormatter.decimalFormat(total,0) + ")  ");

			// Calculate distribution
			for(int index=0; index<stats.length; index++){ 
			    statStringBuffer.append(getClassLabel(index)+":(" + NumberFormatter.decimalFormat(stats[index],0) + "," + weka.core.Utils.roundDouble(100*stats[index]/total, NumberFormatter.MAX_FRACTION_DIGIT)+"%)  ");
			}

			if (m_sons[i].m_isLeaf) {
				text.append(": ");
				text.append(m_localModel.dumpLabel(i, m_train));
				
				text.append("\r\n");
				for (j = 0; j < depth; j++)
					text.append("|   ");
				text.append(statStringBuffer);
				
				// Record the leaf node content,e.g., "petallength <= 4.9: Iris-A (48.-/1.0)"
				// TWang Jan 26, 2005.	 
				m_sons[i].setM_NodeContent(new String(m_localModel.leftSide(m_train) + "" + m_localModel.rightSide(i, m_train) + ":" + m_localModel.dumpLabel(i, m_train)), m_sons[i].m_localModel.distribution().getClassDist());				
			} else{
				
				text.append("\r\n");
				for (j = 0; j < depth; j++)
					text.append("|   ");
				text.append(statStringBuffer);

				// Record the internal node content,e.g., "petallength > 4.9"
				// TWang Jan 26, 2005.			
				m_sons[i].setM_NodeContent( new String(m_localModel.leftSide(m_train) + "" + m_localModel.rightSide(i, m_train)), m_sons[i].m_localModel.distribution().getClassDist());
				
				m_sons[i].dumpTree(depth + 1, text);
			}
			
			
		}
  }

  /**
   * Help method for printing tree structure as a graph.
   *
   * @exception Exception if something goes wrong
   */
  private void graphTree(StringBuffer text) throws Exception {
    
    for (int i = 0; i < m_sons.length; i++) {
      text.append("N" + m_id  
		  + "->" + 
		  "N" + m_sons[i].m_id +
		  " [label=\"" + m_localModel.rightSide(i,m_train).trim() + 
		  "\"]\n");
      if (m_sons[i].m_isLeaf) {
	text.append("N" + m_sons[i].m_id +
		    " [label=\""+m_localModel.dumpLabel(i,m_train)+"\" "+ 
		    "shape=box style=filled ");
	if (m_train != null && m_train.numInstances() > 0) {
	  text.append("data =\n" + m_sons[i].m_train + "\n");
	  text.append(",\n");
	}
	text.append("]\n");
      } else {
	text.append("N" + m_sons[i].m_id +
		    " [label=\""+m_sons[i].m_localModel.leftSide(m_train) + 
		    "\" ");
	if (m_train != null && m_train.numInstances() > 0) {
	  text.append("data =\n" + m_sons[i].m_train + "\n");
	  text.append(",\n");
	}
	text.append("]\n");
	m_sons[i].graphTree(text);
      }
    }
  }

  /**
   * Prints the tree in prefix form
   */
  private void prefixTree(StringBuffer text) throws Exception {

    text.append("[");
    text.append(m_localModel.leftSide(m_train)+":");
    for (int i = 0; i < m_sons.length; i++) {
      if (i > 0) {
	text.append(",\n");
      }
      text.append(m_localModel.rightSide(i, m_train));
    }
    for (int i = 0; i < m_sons.length; i++) {
      if (m_sons[i].m_isLeaf) {
	text.append("[");
	text.append(m_localModel.dumpLabel(i,m_train));
	text.append("]");
      } else {
	m_sons[i].prefixTree(text);
      }
    }
    text.append("]");
  }

  /**
   * Help method for computing class probabilities of 
   * a given instance.
   *
   * @exception Exception if something goes wrong
   */
  private double getProbsLaplace(int classIndex, Instance instance, double weight) 
    throws Exception {
    
    double [] weights;
    double prob = 0;
    int treeIndex;
    int i;
    
    if (m_isLeaf) {
      return weight * localModel().classProbLaplace(classIndex, instance, -1);
    } else {
      treeIndex = localModel().whichSubset(instance);
      if (treeIndex == -1) {
	weights = localModel().weights(instance);
	for (i = 0; i < m_sons.length; i++) {
	  if (!son(i).m_isEmpty) {
	    if (!son(i).m_isLeaf) {
	      prob += son(i).getProbsLaplace(classIndex, instance, 
					     weights[i] * weight);
	    } else {
	      prob += weight * weights[i] * 
		localModel().classProbLaplace(classIndex, instance, i);
	    }
	  }
	}
	return prob;
      } else {
	if (son(treeIndex).m_isLeaf) {
	  return weight * localModel().classProbLaplace(classIndex, instance, 
							treeIndex);
	} else {
	  return son(treeIndex).getProbsLaplace(classIndex, instance, weight);
	}
      }
    }
  }

  /**
   * Help method for computing class probabilities of 
   * a given instance.
   *
   * @exception Exception if something goes wrong
   */
  private double getProbs(int classIndex, Instance instance, double weight) 
    throws Exception {

		double[] weights;
		double prob = 0;
		int treeIndex;
		int i;

		if (m_isLeaf) {
			return weight * localModel().classProb(classIndex, instance, -1);
		} else {
			treeIndex = localModel().whichSubset(instance);
			if (treeIndex == -1) {
				weights = localModel().weights(instance);
				for (i = 0; i < m_sons.length; i++) {
					if (!son(i).m_isEmpty) {
						prob += son(i).getProbs(classIndex, instance,
								weights[i] * weight);
					}
				}
				return prob;
			} else {
				if (son(treeIndex).m_isEmpty) {
					return weight
							* localModel().classProb(classIndex, instance,
									treeIndex);
				} else {
					return son(treeIndex)
							.getProbs(classIndex, instance, weight);
				}
			}
		}
	}

  /**
   * Method just exists to make program easier to read.
   */
  private ClassifierSplitModel localModel() {
    
    return (ClassifierSplitModel)m_localModel;
  }
  
  /**
   * Method just exists to make program easier to read.
   */
  private ClassifierTree son(int index) {
    
    return (ClassifierTree)m_sons[index];
  }

    /**
	 * TWang.
	 * @return Returns the m_NodeContent.
	 */
	public ClassifierSplitModel getClassifierSplitModel() {
		return (ClassifierSplitModel)m_localModel;
	}
	 /**
	 * TWang.
	 * @return Returns the m_NodeContent.
	 */
	public String getM_NodeContent() {
		return m_NodeContent;
	}
	public double[] get_NodeClassDist() {
		return m_NodeClassDist;
	}
	
	public String getClassLabel(int index) throws Exception
	{
	    return localModel().getClassLabel(index,m_train);
	}
	/**
	 * TWang.
	 * @param nodeContent The m_NodeContent to set.
	 */
	public void setM_NodeContent(String a_nodeContent, double[] a_classDist) {
		m_NodeContent = null;
		m_NodeContent = a_nodeContent;
		
		m_NodeClassDist = null;
		m_NodeClassDist = a_classDist;
	} 
	
	/**
	 * TWang.
	 * @param nodeContent The m_NodeContent to set.
	 */ 
	
	/**
	 * ONLY call this method after calling the classifyInstance(instance_A)
	 * method. It will return the probablity of classifiing  instance_A into
	 * the returned class.
	 * 
	 * TWang.
	 * @param nodeContent The m_NodeContent to set.
	 */
	public double getProbablilty() {
		return m_ClassPro;
	}
	
public ClassifierSplitModel getM_localModel() {
	return m_localModel;
}
public void setM_localModel(ClassifierSplitModel model) {
	m_localModel = model;
}
}