weka.classifiers.lazy.kstar
Class KStarNominalAttribute

java.lang.Object
  weka.classifiers.lazy.kstar.KStarNominalAttribute

All Implemented Interfaces:

KStarConstants

public class KStarNominalAttribute

extends java.lang.Object

implements KStarConstants

A custom class which provides the environment for computing the transformation probability of a specified test instance nominal attribute to a specified train instance nominal attribute.

Version:

$Revision 1.0 $

Author:

Len Trigg (len@reeltwo.com), Abdelaziz Mahoui (am14@cs.waikato.ac.nz)

Field Summary

protected int	m_AttrIndex The index of the nominal attribute in the test and train instances
protected double	m_AverageProb Average probability of test attribute transforming into train attribute
protected int	m_BlendFactor default sphere of influence blend setting
protected int	m_BlendMethod B_SPHERE = use specified blend, B_ENTROPY = entropic blend setting
protected KStarCache	m_Cache A cache for storing attribute values and their corresponding stop parameters
protected int	m_ClassType The class attribute type
protected int[]	m_Distribution Distribution of the attribute value in the train dataset
protected int	m_MissingMode missing value treatment
protected double	m_MissingProb Probability of test attribute transforming into train attribute with missing value
protected int	m_NumAttributes The number of attributes
protected int	m_NumClasses The number of class values
protected int	m_NumInstances The number of instances in the dataset
protected int[][]	m_RandClassCols Set of colomns: each colomn representing a randomised version of the train dataset class colomn
protected double	m_SmallestProb Smallest probability of test attribute transforming into train attribute
protected double	m_Stop The stop parameter
protected Instance	m_Test The test instance
protected int	m_TotalCount Number of trai instances with no missing attribute values
protected Instance	m_Train The train instance
protected Instances	m_TrainSet The training instances used for classification.

Fields inherited from interface weka.classifiers.lazy.kstar.KStarConstants

B_ENTROPY, B_SPHERE, EPSILON, FLOOR, FLOOR1, INITIAL_STEP, LOG2, M_AVERAGE, M_DELETE, M_MAXDIFF, M_NORMAL, NUM_RAND_COLS, OFF, ON, ROOT_FINDER_ACCURACY, ROOT_FINDER_MAX_ITER

Constructor Summary

KStarNominalAttribute(Instance test, Instance train, int attrIndex, Instances trainSet, int[][] randClassCol, KStarCache cache)
Constructor

Method Summary

private void	calculateEntropy(double stop, KStarWrapper params) Calculates the entropy of the actual class prediction and the entropy for random class prediction.
private void	calculateSphereSize(int testvalue, double stop, KStarWrapper params) Calculates the size of the "sphere of influence" defined as: sphere = sum(P^2)/sum(P)^2 P(i|j) = (1-tstop)*P(i) + ((i==j)?
private void	generateAttrDistribution() Calculates the distribution, in the dataset, of the indexed nominal attribute values.
private void	init() Initializes the m_Attributes of the class.
private double	PStar(Instance test, Instance train, int col, double stop) Calculates the nominal probability function defined as: P(i|j) = (1-stop) * P(i) + ((i==j) ?
void	setOptions(int missingmode, int blendmethod, int blendfactor) Sets the options.
private double	stopProbUsingBlend() Calculates the "stop parameter" for this attribute using the blend method: the value is computed using a root finder algorithm.
private double	stopProbUsingEntropy() Calculates the "stop parameter" for this attribute using the entropy method: the value is computed using a root finder algorithm.
double	transProb() Calculates the probability of the indexed nominal attribute of the test instance transforming into the indexed nominal attribute of the training instance.

Methods inherited from class java.lang.Object

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

Field Detail

m_TrainSet

protected Instances m_TrainSet

The training instances used for classification.

m_Test

protected Instance m_Test

The test instance

m_Train

protected Instance m_Train

The train instance

m_AttrIndex

protected int m_AttrIndex

The index of the nominal attribute in the test and train instances

m_Stop

protected double m_Stop

The stop parameter

m_MissingProb

protected double m_MissingProb

Probability of test attribute transforming into train attribute with missing value

m_AverageProb

protected double m_AverageProb

Average probability of test attribute transforming into train attribute

m_SmallestProb

protected double m_SmallestProb

Smallest probability of test attribute transforming into train attribute

m_TotalCount

protected int m_TotalCount

Number of trai instances with no missing attribute values

m_Distribution

protected int[] m_Distribution

Distribution of the attribute value in the train dataset

m_RandClassCols

protected int[][] m_RandClassCols

Set of colomns: each colomn representing a randomised version of the train dataset class colomn

m_Cache

protected KStarCache m_Cache

A cache for storing attribute values and their corresponding stop parameters

m_NumInstances

protected int m_NumInstances

The number of instances in the dataset

m_NumClasses

protected int m_NumClasses

The number of class values

m_NumAttributes

protected int m_NumAttributes

The number of attributes

m_ClassType

protected int m_ClassType

The class attribute type

m_MissingMode

protected int m_MissingMode

missing value treatment

m_BlendMethod

protected int m_BlendMethod

B_SPHERE = use specified blend, B_ENTROPY = entropic blend setting

m_BlendFactor

protected int m_BlendFactor

default sphere of influence blend setting

Constructor Detail

KStarNominalAttribute

public KStarNominalAttribute(Instance test,
                             Instance train,
                             int attrIndex,
                             Instances trainSet,
                             int[][] randClassCol,
                             KStarCache cache)

Constructor

Method Detail

init

private void init()

Initializes the m_Attributes of the class.

transProb

public double transProb()

Calculates the probability of the indexed nominal attribute of the test instance transforming into the indexed nominal attribute of the training instance.

Returns:

the value of the transformation probability.

stopProbUsingEntropy

private double stopProbUsingEntropy()

Calculates the "stop parameter" for this attribute using the entropy method: the value is computed using a root finder algorithm. The method takes advantage of the calculation to compute the smallest and average transformation probabilities once the stop factor is obtained. It also sets the transformation probability to an attribute with a missing value.

Returns:

the value of the stop parameter.

calculateEntropy

private void calculateEntropy(double stop,
                              KStarWrapper params)

Calculates the entropy of the actual class prediction and the entropy for random class prediction. It also calculates the smallest and average transformation probabilities.

Parameters:

stop - the stop parameter

params - the object wrapper for the parameters: actual entropy, random entropy, average probability and smallest probability.

Returns:

the values are returned in the object "params".

stopProbUsingBlend

private double stopProbUsingBlend()

Calculates the "stop parameter" for this attribute using the blend method: the value is computed using a root finder algorithm. The method takes advantage of this calculation to compute the smallest and average transformation probabilities once the stop factor is obtained. It also sets the transformation probability to an attribute with a missing value.

Returns:

the value of the stop parameter.

calculateSphereSize

private void calculateSphereSize(int testvalue,
                                 double stop,
                                 KStarWrapper params)

Calculates the size of the "sphere of influence" defined as: sphere = sum(P^2)/sum(P)^2 P(i|j) = (1-tstop)*P(i) + ((i==j)?tstop:0). This method takes advantage of the calculation to compute the values of the "smallest" and "average" transformation probabilities when using the specified stop parameter.

Parameters:

stop - the stop parameter

params - a wrapper of the parameters to be computed: "sphere" the sphere size "avgprob" the average transformation probability "minProb" the smallest transformation probability

Returns:

the values are returned in "params" object.

PStar

private double PStar(Instance test,
                     Instance train,
                     int col,
                     double stop)

Calculates the nominal probability function defined as: P(i|j) = (1-stop) * P(i) + ((i==j) ? stop : 0) In this case, it calculates the transformation probability of the indexed test attribute to the indexed train attribute.

Parameters:

test - the test instance

train - the train instance

col - the attribute index

Returns:

the value of the tranformation probability.

generateAttrDistribution

private void generateAttrDistribution()

Calculates the distribution, in the dataset, of the indexed nominal attribute values. It also counts the actual number of training instances that contributed (those with non-missing values) to calculate the distribution.

setOptions

public void setOptions(int missingmode,
                       int blendmethod,
                       int blendfactor)

Sets the options.