org.wdssii.decisiontree
Class QuinlanC45AxialDecisionTreeCreator

java.lang.Object
  org.wdssii.decisiontree.QuinlanC45AxialDecisionTreeCreator

All Implemented Interfaces:

DecisionTreeCreator

public class QuinlanC45AxialDecisionTreeCreator
extends Object
implements DecisionTreeCreator

C45 learning algorithm to create an axial decision tree. J. R. Quinlan. Improved use of continuous attributes in c4.5. Journal of Artificial Intelligence Research, 4:77-90, 1996 Usage:

   float[][] data = new float[numTraining][numAttr];
   int[] categories = new int[numTraining];
   // populate arrays
   ...
   QuinlanC45AxialDecisionTreeCreator classifier = new QuinlanC45AxialDecisionTreeCreator(0.1); // pruning fraction
   DecisionTree tree = classifier.learn(data, categories);
 

Author:

lakshman

Nested Class Summary

static class

QuinlanC45AxialDecisionTreeCreator.TreeCreationException

Field Summary

private FitnessFunction	fitness By default, InformationGain is used.
private int	maxDepth how deep can this tree go? The deeper the tree the less general it is.
private int	numCategories how many classes are there?
private int	populationToConsiderSplitting How many members can be in a population before it is split?
private float	pruningFraction fraction of the training data set to keep aside so that the learned tree is not overfit.

Constructor Summary

QuinlanC45AxialDecisionTreeCreator()

QuinlanC45AxialDecisionTreeCreator(float pruningFraction)

Method Summary

private AxialTreeNode	buildTree(float[][] inputData, int[] targetClass, int[] toConsider, int depth) Helper method that creates a sub-tree and returns a node
private int	getMostLikelyCategory(float[][] inputData, int[] targetClass, int[] toConsider)
int	getNumCategories() Corresponds to previously learnt data set
AxialDecisionTree	learn(float[][] inputData, int[] targetClass)
private AxialTreeNode	pruneTree(AxialTreeNode node, float[][] inputData, int[] targetClass, int[] toConsider) removes nodes that do not perform well on the validation dataset.
private int[][]	split(float[][] inputData, int[] toConsider, int bestAttribute, float thresh)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

populationToConsiderSplitting

private int populationToConsiderSplitting

How many members can be in a population before it is split?

pruningFraction

private float pruningFraction

fraction of the training data set to keep aside so that the learned tree is not overfit. A value of 0.1f may be a pretty good choice. The pruning points will be the last few instances of the training data set. Pass in a randomized sample if this simply won't do.

maxDepth

private int maxDepth

how deep can this tree go? The deeper the tree the less general it is.

numCategories

private int numCategories

how many classes are there?

fitness

private FitnessFunction fitness

By default, InformationGain is used.

Constructor Detail

QuinlanC45AxialDecisionTreeCreator

public QuinlanC45AxialDecisionTreeCreator(float pruningFraction)

QuinlanC45AxialDecisionTreeCreator

public QuinlanC45AxialDecisionTreeCreator()

Method Detail

learn

public AxialDecisionTree learn(float[][] inputData,
                               int[] targetClass)
                        throws QuinlanC45AxialDecisionTreeCreator.TreeCreationException,
                               IllegalArgumentException

Specified by:

learn in interface DecisionTreeCreator

Parameters:

inputData - an array where each row corresponds to a single instance (to be classified) and the columns hold the attributes of that instance

targetClass - an array where each row corresponds to a single instance, specifically the actual classification of that instance. The class needs to be a number 0,1,2,...,N-1 where N is the number of classes. Some of these classes may have no examples.

Returns:

decisiontree

Throws:

QuinlanC45AxialDecisionTreeCreator.TreeCreationException

IllegalArgumentException

pruneTree

private AxialTreeNode pruneTree(AxialTreeNode node,
                                float[][] inputData,
                                int[] targetClass,
                                int[] toConsider)

removes nodes that do not perform well on the validation dataset.

buildTree

private AxialTreeNode buildTree(float[][] inputData,
                                int[] targetClass,
                                int[] toConsider,
                                int depth)

Helper method that creates a sub-tree and returns a node

Returns:

split

private int[][] split(float[][] inputData,
                      int[] toConsider,
                      int bestAttribute,
                      float thresh)

getMostLikelyCategory

private int getMostLikelyCategory(float[][] inputData,
                                  int[] targetClass,
                                  int[] toConsider)

getNumCategories

public int getNumCategories()

Corresponds to previously learnt data set