SplitCriteria (Fiji Javadocs 0-SNAPSHOT API)

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

All Methods Static Methods Concrete Methods
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

Class SplitCriteria

Method Summary

Methods inherited from class java.lang.Object

Method Detail

entropyConditionedOnRows

entropyOverColumns

fastLog2