org.openimaj.ml.gmm

Class GaussianMixtureModelEM

java.lang.Object

org.openimaj.ml.gmm.GaussianMixtureModelEM

public class GaussianMixtureModelEM
extends Object

Gaussian mixture model learning using the EM algorithm. Supports a range of different shapes Gaussian through different covariance matrix forms. An initialisation step is used to learn the initial means using K-Means, although this can be disabled in the constructor.

Implementation was originally inspired by the SciPy's "gmm.py".

Author:

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

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	GaussianMixtureModelEM.CovarianceType Different forms of covariance matrix supported by the GaussianMixtureModelEM.
protected static class	GaussianMixtureModelEM.EMGMM
static class	GaussianMixtureModelEM.UpdateOptions Options for controlling what gets updated during the initialisation and/or iterations.

Constructor Summary

Constructors

Constructor and Description
GaussianMixtureModelEM(int nComponents, GaussianMixtureModelEM.CovarianceType ctype) Construct with the given arguments.
GaussianMixtureModelEM(int nComponents, GaussianMixtureModelEM.CovarianceType ctype, double thresh, double minCovar, int nIters, int nInit, EnumSet<GaussianMixtureModelEM.UpdateOptions> iterOpts, EnumSet<GaussianMixtureModelEM.UpdateOptions> initOpts) Construct with the given arguments.

Method Summary

Modifier and Type	Method and Description
GaussianMixtureModelEM	clone()
MixtureOfGaussians	estimate(double[][] X) Estimate a new MixtureOfGaussians from the given data.
MixtureOfGaussians	estimate(Jama.Matrix X) Estimate a new MixtureOfGaussians from the given data.
boolean	hasConverged() Get's the convergence state of the algorithm.
protected void	mstep(GaussianMixtureModelEM.EMGMM gmm, double[][] X, double[][] responsibilities)

Methods inherited from class java.lang.Object

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

Constructor Detail

GaussianMixtureModelEM

public GaussianMixtureModelEM(int nComponents,
                              GaussianMixtureModelEM.CovarianceType ctype,
                              double thresh,
                              double minCovar,
                              int nIters,
                              int nInit,
                              EnumSet<GaussianMixtureModelEM.UpdateOptions> iterOpts,
                              EnumSet<GaussianMixtureModelEM.UpdateOptions> initOpts)

Construct with the given arguments.

Parameters:

nComponents - the number of gaussian components

ctype - the form of the covariance matrices

thresh - the threshold at which to stop iterating

minCovar - the minimum value allowed in the diagonal of the estimated covariance matrices to prevent overfitting

nIters - the maximum number of iterations

nInit - the number of runs of the algorithm to perform; the best result will be kept.

iterOpts - options controlling what is updated during iteration

initOpts - options controlling what is updated during initialisation. Enabling the GaussianMixtureModelEM.UpdateOptions.Means option will cause K-Means to be used to generate initial starting points for the means.

GaussianMixtureModelEM

public GaussianMixtureModelEM(int nComponents,
                              GaussianMixtureModelEM.CovarianceType ctype)

Construct with the given arguments.

Parameters:

nComponents - the number of gaussian components

ctype - the form of the covariance matrices

Method Detail

hasConverged

public boolean hasConverged()

Get's the convergence state of the algorithm. Will return false if estimate(double[][]) has not been called, or if the last call to estimate(double[][]) failed to reach convergence before running out of iterations.

Returns:

true if the last call to estimate(double[][]) reached convergence; false otherwise

estimate

public MixtureOfGaussians estimate(Jama.Matrix X)

Estimate a new MixtureOfGaussians from the given data. Use hasConverged() to check whether the EM algorithm reached convergence in the estimation of the returned model.

Parameters:

X - the data matrix.

Returns:

the generated GMM.

estimate

public MixtureOfGaussians estimate(double[][] X)

Estimate a new MixtureOfGaussians from the given data. Use hasConverged() to check whether the EM algorithm reached convergence in the estimation of the returned model.

Parameters:

X - the data array.

Returns:

the generated GMM.

mstep

protected void mstep(GaussianMixtureModelEM.EMGMM gmm,
                     double[][] X,
                     double[][] responsibilities)

clone

public GaussianMixtureModelEM clone()

Overrides:

clone in class Object