weka.classifiers.trees.lmt
Class ResidualSplit

java.lang.Object
  weka.classifiers.trees.j48.ClassifierSplitModel
      weka.classifiers.trees.lmt.ResidualSplit

All Implemented Interfaces:

java.lang.Cloneable, java.io.Serializable

public class ResidualSplit

extends ClassifierSplitModel

Helper class for logistic model trees (weka.classifiers.trees.lmt.LMT) to implement the splitting criterion based on residuals of the LogitBoost algorithm.

Version:

$Revision: 1.1 $

Author:

Niels Landwehr

See Also:

Serialized Form

Field Summary

protected int	m_attIndex The index of the attribute selected for the split
protected Attribute	m_attribute The attribute selected for the split
protected Instances	m_data The set of instances
protected double[][]	m_dataWs The LogitBoost-weights for the set of instances
protected double[][]	m_dataZs The Z-values (LogitBoost response) for the set of instances
protected int	m_numClasses Number of classed
protected int	m_numInstances Number of instances in the set
protected double	m_splitPoint The split point (for numeric attributes)

Fields inherited from class weka.classifiers.trees.j48.ClassifierSplitModel

m_distribution, m_numSubsets

Constructor Summary

ResidualSplit(int attIndex)
Creates a split object

Method Summary

void	buildClassifier(Instances data) Method not in use
void	buildClassifier(Instances data, double[][] dataZs, double[][] dataWs) Builds the split.
boolean	checkModel(int minNumInstances) Checks if there are at least 2 subsets that contain >= minNumInstances.
protected double	entropy(double[][] dataZs, double[][] dataWs) Helper function to compute entropy from Z/W values.
double	entropyGain() Computes entropy gain for current split.
protected boolean	getSplitPoint() Selects split point for numeric attribute.
java.lang.String	leftSide(Instances data) Returns name of splitting attribute (left side of condition).
java.lang.String	rightSide(int index, Instances data) Prints the condition satisfied by instances in a subset.
java.lang.String	sourceExpression(int index, Instances data) Method not in use
double[]	weights(Instance instance) Method not in use
int	whichSubset(Instance instance) Returns index of subset instance is assigned to.

Methods inherited from class weka.classifiers.trees.j48.ClassifierSplitModel

checkModel, classifyInstance, classProb, classProbLaplace, clone, codingCost, distribution, dumpLabel, dumpModel, numSubsets, resetDistribution, sourceClass, split

Methods inherited from class java.lang.Object

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

Field Detail

m_attribute

protected Attribute m_attribute

The attribute selected for the split

m_attIndex

protected int m_attIndex

The index of the attribute selected for the split

m_numInstances

protected int m_numInstances

Number of instances in the set

m_numClasses

protected int m_numClasses

Number of classed

m_data

protected Instances m_data

The set of instances

m_dataZs

protected double[][] m_dataZs

The Z-values (LogitBoost response) for the set of instances

m_dataWs

protected double[][] m_dataWs

The LogitBoost-weights for the set of instances

m_splitPoint

protected double m_splitPoint

The split point (for numeric attributes)

Constructor Detail

ResidualSplit

public ResidualSplit(int attIndex)

Creates a split object

Parameters:

attIndex - the index of the attribute to split on

Method Detail

buildClassifier

public void buildClassifier(Instances data,
                            double[][] dataZs,
                            double[][] dataWs)
                     throws java.lang.Exception

Builds the split. Needs the Z/W values of LogitBoost for the set of instances.

Throws:

java.lang.Exception

getSplitPoint

protected boolean getSplitPoint()
                         throws java.lang.Exception

Selects split point for numeric attribute.

Throws:

java.lang.Exception

entropyGain

public double entropyGain()
                   throws java.lang.Exception

Computes entropy gain for current split.

Throws:

java.lang.Exception

entropy

protected double entropy(double[][] dataZs,
                         double[][] dataWs)

Helper function to compute entropy from Z/W values.

checkModel

public boolean checkModel(int minNumInstances)

Checks if there are at least 2 subsets that contain >= minNumInstances.

leftSide

public final java.lang.String leftSide(Instances data)

Returns name of splitting attribute (left side of condition).

Specified by:

leftSide in class ClassifierSplitModel

Parameters:

data - the data.

rightSide

public final java.lang.String rightSide(int index,
                                        Instances data)

Prints the condition satisfied by instances in a subset.

Specified by:

rightSide in class ClassifierSplitModel

whichSubset

public final int whichSubset(Instance instance)
                      throws java.lang.Exception

Description copied from class: ClassifierSplitModel

Returns index of subset instance is assigned to. Returns -1 if instance is assigned to more than one subset.

Specified by:

whichSubset in class ClassifierSplitModel

Throws:

java.lang.Exception - if something goes wrong

buildClassifier

public void buildClassifier(Instances data)

Method not in use

Specified by:

buildClassifier in class ClassifierSplitModel

weights

public final double[] weights(Instance instance)

Method not in use

Specified by:

weights in class ClassifierSplitModel

sourceExpression

public final java.lang.String sourceExpression(int index,
                                               Instances data)

Method not in use

Specified by:

sourceExpression in class ClassifierSplitModel