hex.singlenoderf

Class GiniStatistic

java.lang.Object

hex.singlenoderf.GiniStatistic

public class GiniStatistic
extends java.lang.Object

Computes the gini split statistics. The Gini fitness is calculated as a probability that the element will be misclassified, which is: 1 - \sum(p_i^2) This is computed for the left and right subtrees and added together: gini left * weight left + gini right * weight left -------------------------------------------------- weight total And subtracted from an ideal worst 1 to simulate the gain from previous node. The best gain is then selected. Same is done for exclusions, where again left stands for the rows with column value equal to the split value and right for all different ones.

Field Summary

Fields

Modifier and Type	Field and Description
protected int[][][]	_columnDists Column distributions: column x arity x classes Remembers the number of rows of the given column index, encodedValue, class.
protected float[][][]	_columnDistsRegression
protected int[]	_features
protected java.util.Random	_random

Constructor Summary

Constructors

Constructor and Description
GiniStatistic(Data data, int features, long seed, int exclusiveSplitLimit)

Method Summary

Methods

Modifier and Type	Method and Description
protected hex.singlenoderf.Statistic.Split	eqSplit(int colIndex, Data d, float[] dist, float distWeight, java.util.Random rand)
protected hex.singlenoderf.Statistic.Split	eqSplit(int colIndex, Data d, int[] dist, int distWeight, java.util.Random _)
protected hex.singlenoderf.Statistic.Split	ltSplit(int colIndex, Data d, float[] dist, float distWeight, java.util.Random rand)
protected hex.singlenoderf.Statistic.Split	ltSplit(int col, Data d, int[] dist, int distWeight, java.util.Random _) Returns the best split for a given column

Methods inherited from class java.lang.Object

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

Field Detail

_columnDists

protected final int[][][] _columnDists

Column distributions: column x arity x classes Remembers the number of rows of the given column index, encodedValue, class.

_features

protected final int[] _features

_random

protected java.util.Random _random

_columnDistsRegression

protected final float[][][] _columnDistsRegression

Constructor Detail

GiniStatistic

public GiniStatistic(Data data,
             int features,
             long seed,
             int exclusiveSplitLimit)

Method Detail

ltSplit

protected hex.singlenoderf.Statistic.Split ltSplit(int col,
                                       Data d,
                                       int[] dist,
                                       int distWeight,
                                       java.util.Random _)

Returns the best split for a given column

eqSplit

protected hex.singlenoderf.Statistic.Split eqSplit(int colIndex,
                                       Data d,
                                       int[] dist,
                                       int distWeight,
                                       java.util.Random _)

ltSplit

protected hex.singlenoderf.Statistic.Split ltSplit(int colIndex,
                                       Data d,
                                       float[] dist,
                                       float distWeight,
                                       java.util.Random rand)

eqSplit

protected hex.singlenoderf.Statistic.Split eqSplit(int colIndex,
                                       Data d,
                                       float[] dist,
                                       float distWeight,
                                       java.util.Random rand)