org.openimaj.ml.clustering.spectral

Class ChangeDetectingEigenChooser

java.lang.Object

org.openimaj.ml.clustering.spectral.EigenChooser

org.openimaj.ml.clustering.spectral.ChangeDetectingEigenChooser

public class ChangeDetectingEigenChooser
extends EigenChooser

Attempts to automatically choose the number of eigen vectors based on the relative gap between eigen values. In spectral clustering the gap between the eigen values of "good" clusters jumps. This class ignores the gap between 0 and the next item because 0s represent completely isolated objects and in all but the trivial case we must stop after we have run out of 0s.

Author:

Sina Samangooei (ss@ecs.soton.ac.uk)

Constructor Summary

Constructors

Constructor and Description
ChangeDetectingEigenChooser(double relativeGap, double maxSelect)

Method Summary

Modifier and Type	Method and Description
int	nEigenVectors(Iterator<DoubleObjectPair<Vector>> vals, int totalEigenVectors)
FewEigenvalues	prepare(SparseMatrix laplacian) Make a coarse decision of the number of eigen vectors to extract in the first place with the knowledge of the eigen values that will likely be important

Methods inherited from class java.lang.Object

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

Constructor Detail

ChangeDetectingEigenChooser

public ChangeDetectingEigenChooser(double relativeGap,
                                   double maxSelect)

Parameters:

relativeGap - the gap between previous and current (treated as absolute if previous value == 0)

maxSelect -

Method Detail

nEigenVectors

public int nEigenVectors(Iterator<DoubleObjectPair<Vector>> vals,
                         int totalEigenVectors)

Specified by:

nEigenVectors in class EigenChooser

totalEigenVectors - the total number of eigen vectors

Returns:

count the eigen vectors

prepare

public FewEigenvalues prepare(SparseMatrix laplacian)

Description copied from class: EigenChooser

Make a coarse decision of the number of eigen vectors to extract in the first place with the knowledge of the eigen values that will likely be important

Specified by:

prepare in class EigenChooser

Parameters:

laplacian - the matrix to be decomposed

Returns:

the prepared eigen values