keel.Algorithms.Discretizers.Khiops

Class Khiops

java.lang.Object

keel.Algorithms.Discretizers.Basic.Discretizer

keel.Algorithms.Discretizers.Khiops.Khiops

public class Khiops
extends Discretizer

Khiops Discretizer Implemented by Julian Luengo, March 2010 julianlm@decsai.ugr.es Based on the work of Marc Boullé M. Boulle. Khiops: A Statistical Discretization Method of Continuous Attributes Machine Learning 55:1 (2004) 53-69

Since:

JDK 1.5

Version:

0.2

Author:

Written by Julián Luengo Martín 18/03/2010

Field Summary

Fields inherited from class keel.Algorithms.Discretizers.Basic.Discretizer

classOfInstances, cutPoints, iClassIndex, realAttributes, realValues

Constructor Summary

Constructors

Constructor and Description
Khiops() Default constructor.

Method Summary

Modifier and Type	Method and Description
double	chi2RowValues(java.util.Vector disc, int[] values) Creates the initial chi square value of the initial discretization scheme.
java.util.Vector	createCP(java.util.Vector intervals) Construct an array of cutpoints from the set of intervals.
protected java.util.Vector	discretizeAttribute(int attribute, int[] values, int begin, int end) This abstract method creates the cut points of the attribute given using its values for each instances given.
double	mergeCostVariation(java.util.ArrayList<java.lang.Double> na, int indexna, double naChi2, java.util.ArrayList<java.lang.Double> nb, int indexnb, double nbChi2, int[] values) Computes the cost derived form merging two adjacent intervals na and nb
double	mergedRowChi2Value(java.util.ArrayList<java.lang.Double> mergedInterval, int index, int[] values) This method calculates the contribution to the global chi square value of a new interval (produced by merging two adjacent ones).

Methods inherited from class keel.Algorithms.Discretizers.Basic.Discretizer

applyDiscretization, buildCutPoints, discretize, getCutPoint, getNumIntervals, sortValues

Methods inherited from class java.lang.Object

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

Constructor Detail

Khiops

public Khiops()

Default constructor.

Method Detail

discretizeAttribute

protected java.util.Vector discretizeAttribute(int attribute,
                                               int[] values,
                                               int begin,
                                               int end)

Description copied from class: Discretizer

This abstract method creates the cut points of the attribute given using its values for each instances given. Every discretizer should implement this method with its particular scheme.

Specified by:

discretizeAttribute in class Discretizer

Parameters:

attribute - given attribute to discretize.

values - given attribute values in the dataset.

begin - First position of the section to discretize.

end - Last position of the section to discretize.

Returns:

the cut points of the attribute given

mergeCostVariation

public double mergeCostVariation(java.util.ArrayList<java.lang.Double> na,
                                 int indexna,
                                 double naChi2,
                                 java.util.ArrayList<java.lang.Double> nb,
                                 int indexnb,
                                 double nbChi2,
                                 int[] values)

Computes the cost derived form merging two adjacent intervals na and nb

Parameters:

na - Interval to the left to merge

indexna - Index of the first element of na in the whole list of real values

naChi2 - number of intervals

nb - Right interval to merge

indexnb - Index of the first element of nb in the whole list of real values

nbChi2 - Current number of intervals (the total intervals prior to the merging)

values - Array in which position i there is the number of instance which explanatory (real) value has rank i after sorting

Returns:

The cost variation produced by the merge operation

chi2RowValues

public double chi2RowValues(java.util.Vector disc,
                            int[] values)

Creates the initial chi square value of the initial discretization scheme. It also initialize the contribution to this value of each interval, so the combination of intervals can be quickly evaluated.

Parameters:

disc - the initial discretization scheme (one interval for each different value)

values - the global array of values (sorted)

Returns:

the chi square value for the initial discretization scheme

mergedRowChi2Value

public double mergedRowChi2Value(java.util.ArrayList<java.lang.Double> mergedInterval,
                                 int index,
                                 int[] values)

This method calculates the contribution to the global chi square value of a new interval (produced by merging two adjacent ones).

Parameters:

mergedInterval - the new interval

index - the index of the first element (left-most one) value in the global array of values

values - the global array of values

Returns:

the contribution of this new interval to the global chi square value

createCP

public java.util.Vector createCP(java.util.Vector intervals)

Construct an array of cutpoints from the set of intervals.

Parameters:

intervals - Vector which contains the intervals in ArrayList format

Returns:

A Vector with double formatted cutpoints, computed as the midterm between two adjacent intervals.