jsat.classifiers.trees

Class ImpurityScore

java.lang.Object

jsat.classifiers.trees.ImpurityScore

All Implemented Interfaces:

java.lang.Cloneable

public class ImpurityScore
extends java.lang.Object
implements java.lang.Cloneable

ImpurityScore provides a measure of the impurity of a set of data points respective to their class labels. The impurity score is maximized when the classes are evenly distributed, and minimized when all points belong to one class.
The gain in purity can be computed using the static gain methods of the class. However, not all impurity measures can be used for arbitrary data and splits. Some may only support binary splits, and some may only support binary target classes.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	ImpurityScore.ImpurityMeasure Different methods of measuring the impurity in a set of data points based on nominal class labels

Constructor Summary

Constructors

Constructor and Description
ImpurityScore(int classCount, ImpurityScore.ImpurityMeasure impurityMeasure) Creates a new impurity score that can be updated

Method Summary

Modifier and Type	Method and Description
void	addPoint(DataPoint dp, int targetClass) Adds one more point to the impurity score
void	addPoint(double weight, int targetClass) Adds one more point to the impurity score
static double	gain(ImpurityScore wholeData, double wholeScale, ImpurityScore... splits) Computes the gain in score from a splitting of the data set
static double	gain(ImpurityScore wholeData, ImpurityScore... splits) Computes the gain in score from a splitting of the data set
ImpurityScore.ImpurityMeasure	getImpurityMeasure() Returns the impurity measure being used
CategoricalResults	getResults() Obtains the current categorical results by prior probability
double	getScore() Computes the current impurity score for the points that have been added.
double	getSumOfWeights() Returns the sum of the weights for all points currently in the impurity score
void	removePoint(DataPoint dp, int targetClass) Removes one point from the impurity score
void	removePoint(double weight, int targetClass) Removes one point from the impurity score

Methods inherited from class java.lang.Object

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

Constructor Detail

ImpurityScore

public ImpurityScore(int classCount,
                     ImpurityScore.ImpurityMeasure impurityMeasure)

Creates a new impurity score that can be updated

Parameters:

classCount - the number of target class values

impurityMeasure -

Method Detail

removePoint

public void removePoint(DataPoint dp,
                        int targetClass)

Removes one point from the impurity score

Parameters:

dp - the data point to add

targetClass - the class of the point to add

removePoint

public void removePoint(double weight,
                        int targetClass)

Removes one point from the impurity score

Parameters:

weight - the weight of the point to add

targetClass - the class of the point to add

addPoint

public void addPoint(DataPoint dp,
                     int targetClass)

Adds one more point to the impurity score

Parameters:

dp - the data point to add

targetClass - the class of the point to add

addPoint

public void addPoint(double weight,
                     int targetClass)

Adds one more point to the impurity score

Parameters:

weight - the weight of the point to add

targetClass - the class of the point to add

getScore

public double getScore()

Computes the current impurity score for the points that have been added. A higher score is worse, a score of zero indicates a perfectly pure set of points (all one class).

Returns:

the impurity score

getSumOfWeights

public double getSumOfWeights()

Returns the sum of the weights for all points currently in the impurity score

Returns:

the sum of weights

getImpurityMeasure

public ImpurityScore.ImpurityMeasure getImpurityMeasure()

Returns the impurity measure being used

Returns:

the impurity measure being used

getResults

public CategoricalResults getResults()

Obtains the current categorical results by prior probability

Returns:

the categorical results for the current score

gain

public static double gain(ImpurityScore wholeData,
                          ImpurityScore... splits)

Computes the gain in score from a splitting of the data set

Parameters:

wholeData - the score for the whole data set

splits - the scores for each of the splits

Returns:

the gain for the values given

gain

public static double gain(ImpurityScore wholeData,
                          double wholeScale,
                          ImpurityScore... splits)

Computes the gain in score from a splitting of the data set

Parameters:

wholeData - the score for the whole data set

wholeScale - a constant to scale the wholeData counts and sums by, useful for handling missing value cases

splits - the scores for each of the splits

Returns:

the gain for the values given