hr.irb.fastRandomForest

Class SplitCriteria

java.lang.Object

hr.irb.fastRandomForest.SplitCriteria

public final class SplitCriteria
extends Object

Functions used for finding best splits in FastRfTree. Based on parts of weka.core.ContingencyTables, revision 1.7, by Eibe Frank

Version:

0.9

Author:

Eibe Frank (eibe@cs.waikato.ac.nz) - original code, Fran Supek (fran.supek[AT]irb.hr) - adapted code

Method Summary

Modifier and Type	Method and Description
static double	entropyConditionedOnRows(double[][] matrix) Similar to weka.core.ContingencyTables.entropyConditionedOnRows.
static double	entropyOverColumns(double[][] matrix) Similar to weka.core.ContingencyTables.entropyOverColumns Does not output entropy, output is modified to make routine faster: the result is not divided by "total", as the total is a constant in all operations (subtraction, comparison) performed as a part of splitting in FastRfTree.
static float	fastLog2(float val) A fast approximation of log base 2, in single precision.

Methods inherited from class java.lang.Object

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

Method Detail

entropyConditionedOnRows

public static double entropyConditionedOnRows(double[][] matrix)

Similar to weka.core.ContingencyTables.entropyConditionedOnRows. Does not output entropy, output is modified to make routine faster: the result is not divided by "total", as the total is a constant in all operations (subtraction, comparison) performed as a part of splitting in FastRfTree. Also, we don't have to divide by Math.log(2) as the logarithms provided by fastLog2() are already base 2.

Parameters:

matrix - the contingency table

Returns:

the conditional entropy of the columns given the rows

entropyOverColumns

public static double entropyOverColumns(double[][] matrix)

Similar to weka.core.ContingencyTables.entropyOverColumns Does not output entropy, output is modified to make routine faster: the result is not divided by "total", as the total is a constant in all operations (subtraction, comparison) performed as a part of splitting in FastRfTree. Also, we don't have to divide by Math.log(2) as the logarithms provided by fastLog2() are already base 2.

Parameters:

matrix - the contingency table

Returns:

the columns' entropy

fastLog2

public static float fastLog2(float val)

A fast approximation of log base 2, in single precision. Approximately 4 times faster than Java's Math.log() function. Inspired by C code by Laurent de Soras: http://www.flipcode.com/archives/Fast_log_Function.shtml