moa.classifiers.core.attributeclassobservers

Interface AttributeClassObserver

All Superinterfaces:

MOAObject, OptionHandler, Serializable

All Known Subinterfaces:

DiscreteAttributeClassObserver, NumericAttributeClassObserver

All Known Implementing Classes:

BinaryTreeNumericAttributeClassObserver, BinaryTreeNumericAttributeClassObserverRegression, FIMTDDNumericAttributeClassLimitObserver, FIMTDDNumericAttributeClassObserver, GaussianNumericAttributeClassObserver, GreenwaldKhannaNumericAttributeClassObserver, NominalAttributeClassObserver, NullAttributeClassObserver, VFMLNumericAttributeClassObserver

public interface AttributeClassObserver
extends OptionHandler

Interface for observing the class data distribution for an attribute. This observer monitors the class distribution of a given attribute. Used in naive Bayes and decision trees to monitor data statistics on leaves.

Version:

$Revision: 7 $

Author:

Richard Kirkby (rkirkby@cs.waikato.ac.nz)

Method Summary

Methods

Modifier and Type	Method and Description
AttributeSplitSuggestion	getBestEvaluatedSplitSuggestion(SplitCriterion criterion, double[] preSplitDist, int attIndex, boolean binaryOnly) Gets the best split suggestion given a criterion and a class distribution
void	observeAttributeClass(double attVal, int classVal, double weight) Updates statistics of this observer given an attribute value, a class and the weight of the instance observed
void	observeAttributeTarget(double attVal, double target)
double	probabilityOfAttributeValueGivenClass(double attVal, int classVal) Gets the probability for an attribute value given a class

Methods inherited from interface moa.options.OptionHandler

copy, getCLICreationString, getOptions, getPurposeString, prepareForUse, prepareForUse

Methods inherited from interface moa.MOAObject

getDescription, measureByteSize

Method Detail

observeAttributeClass

void observeAttributeClass(double attVal,
                         int classVal,
                         double weight)

Updates statistics of this observer given an attribute value, a class and the weight of the instance observed

Parameters:

attVal - the value of the attribute

classVal - the class

weight - the weight of the instance

probabilityOfAttributeValueGivenClass

double probabilityOfAttributeValueGivenClass(double attVal,
                                           int classVal)

Gets the probability for an attribute value given a class

Parameters:

attVal - the attribute value

classVal - the class

Returns:

probability for an attribute value given a class

getBestEvaluatedSplitSuggestion

AttributeSplitSuggestion getBestEvaluatedSplitSuggestion(SplitCriterion criterion,
                                                       double[] preSplitDist,
                                                       int attIndex,
                                                       boolean binaryOnly)

Gets the best split suggestion given a criterion and a class distribution

Parameters:

criterion - the split criterion to use

preSplitDist - the class distribution before the split

attIndex - the attribute index

binaryOnly - true to use binary splits

Returns:

suggestion of best attribute split

observeAttributeTarget

void observeAttributeTarget(double attVal,
                          double target)