org.openimaj.util.tree

Class LongKDTree.RandomisedBBFMeanSplit

java.lang.Object

org.openimaj.util.tree.LongKDTree.RandomisedBBFMeanSplit

All Implemented Interfaces:

LongKDTree.SplitChooser

Enclosing class:

LongKDTree

public static class LongKDTree.RandomisedBBFMeanSplit
extends Object
implements LongKDTree.SplitChooser

Randomised best-bin-first splitting strategy:

• Nodes with less than a set number of items become leafs.
• Otherwise:
  • a sample of the data is taken and the variance across each dimension is computed.
  • a dimension is chosen randomly from the dimensions with the higest variance.
  • the mean (computed from the variance sample) is taken as the split point.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Constructor Summary

Constructors

Constructor and Description
RandomisedBBFMeanSplit() Construct with the default values of 14 points per leaf (max), 128 samples for computing variance, and the 5 most varying dimensions randomly selected.
RandomisedBBFMeanSplit(int maxLeafSize, int varianceMaxPoints, int randomMaxDims, cern.jet.random.Uniform uniform) Construct with the given values.
RandomisedBBFMeanSplit(cern.jet.random.Uniform uniform) Construct with the default values of 14 points per leaf (max), 128 samples for computing variance, and the 5 most varying dimensions randomly selected.

Method Summary

Modifier and Type	Method and Description
IntLongPair	chooseSplit(long[][] pnts, IntArrayView inds, int depth, long[] minBounds, long[] maxBounds) Choose the dimension and discriminant on which to split the data.

Methods inherited from class java.lang.Object

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

Constructor Detail

RandomisedBBFMeanSplit

public RandomisedBBFMeanSplit()

Construct with the default values of 14 points per leaf (max), 128 samples for computing variance, and the 5 most varying dimensions randomly selected. A new MersenneTwister is created as the source for random numbers.

RandomisedBBFMeanSplit

public RandomisedBBFMeanSplit(cern.jet.random.Uniform uniform)

Construct with the default values of 14 points per leaf (max), 128 samples for computing variance, and the 5 most varying dimensions randomly selected. A new MersenneTwister is created as the source for random numbers.

Parameters:

uniform - the random number source

RandomisedBBFMeanSplit

public RandomisedBBFMeanSplit(int maxLeafSize,
                              int varianceMaxPoints,
                              int randomMaxDims,
                              cern.jet.random.Uniform uniform)

Construct with the given values.

Parameters:

maxLeafSize - Maximum number of items in a leaf.

varianceMaxPoints - Maximum number of points of variance estimation; all points used if <=0.

randomMaxDims - Number of dimensions to consider when randomly selecting one with a big variance.

uniform - the random number source

Method Detail

chooseSplit

public IntLongPair chooseSplit(long[][] pnts,
                               IntArrayView inds,
                               int depth,
                               long[] minBounds,
                               long[] maxBounds)

Description copied from interface: LongKDTree.SplitChooser

Choose the dimension and discriminant on which to split the data.

Specified by:

chooseSplit in interface LongKDTree.SplitChooser

Parameters:

pnts - the raw data

inds - the indices of the data under consideration

depth - the depth of the current data in the tree

minBounds - the minimum bounds

maxBounds - the maximum bounds

Returns:

the dimension and discriminant, or null iff this is a leaf (containing all the points in inds).