lti::gaussianMixtureModel< T > Class Template Reference

Functor which estimates a gaussian mixture model of a given data set. More...

#include <ltiGaussianMixtureModel.h>

Inheritance diagram for lti::gaussianMixtureModel< T >:

[legend]

Collaboration diagram for lti::gaussianMixtureModel< T >:

[legend]

List of all members.

Classes
class	parameters
	The parameters for the class gaussianMixtureModel. More...
Public Member Functions
	gaussianMixtureModel ()
	gaussianMixtureModel (const gaussianMixtureModel &other)
virtual	~gaussianMixtureModel ()
virtual const char *	getTypeName () const
bool	apply (const matrix< T > &src)
bool	estimate (const matrix< T > &src)
const std::vector< matrix< T > > &	getComponentSigmas () const
const std::vector< vector< T > > &	getComponentMus () const
const vector< T > &	getAlpha () const
void	getComponentSigmas (std::vector< matrix< T > > &sigs) const
void	getComponentMus (std::vector< vector< T > > &ms) const
void	getAlpha (vector< T > &a) const
gaussianMixtureModel &	copy (const gaussianMixtureModel &other)
virtual functor *	clone () const
const parameters &	getParameters () const
void	setProgressInfo (progressInfo &pi)

---

Detailed Description

template<class T>
class lti::gaussianMixtureModel< T >

Functor which estimates a gaussian mixture model of a given data set.

See documentation of the apply() methods for details. Please note that, even though this is a template class, it does not make much sense to use it with anything but floating-point types. The higher the precision, the better. So, if you use gaussianMixtureModel<float> and get crappy results, you should not wonder. If you get reasonable results, you're just lucky. Other template arguments than float or double probably won't even compile.

---

Constructor & Destructor Documentation

template<class T >

lti::gaussianMixtureModel< T >::gaussianMixtureModel

(

)

default constructor

template<class T >

lti::gaussianMixtureModel< T >::gaussianMixtureModel

(

const gaussianMixtureModel< T > &

other

)

copy constructor

Parameters:

other

the object to be copied

template<class T >

virtual lti::gaussianMixtureModel< T >::~gaussianMixtureModel

(

)

[virtual]

destructor

---

Member Function Documentation

template<class T >

bool lti::gaussianMixtureModel< T >::apply

(

const matrix< T > &

src

)

[inline]

Performs estimation of a gaussian mixture model for the given data matrix.

The matrix is interpreted as having a data point in each row. The method uses the Expectation Maximization method; for a really nice explanation of the algorithm, see the paper A Gentle Tutorial of the EM Algorithm and its Application to Parameter Estimation for Gaussian Mixture and Hidden Markov Models by Jeff Bilmes. (published as Technical Report TR-97-021 by the International Computer Science Institute). If the useSM parameter is true, the split-and-merge algorithm by Split-and-Merge EM algorithm by Ueda, Nakano, Ghahramani, and Hinton (Neural Computation 12(9), pp. 2109-2128) is used instead.

Parameters:

src

matrix<T> with the source data, each row is considered a data point.

Returns:

true if the estimation was successfull.

References lti::gaussianMixtureModel< T >::estimate().

template<class T >

virtual functor* lti::gaussianMixtureModel< T >::clone

(

)

const [virtual]

returns a pointer to a clone of this functor.

Reimplemented from lti::statisticsFunctor.

template<class T >

gaussianMixtureModel& lti::gaussianMixtureModel< T >::copy

(

const gaussianMixtureModel< T > &

other

)

copy data of "other" functor.

Parameters:

other

the functor to be copied

Returns:

a reference to this functor object

Reimplemented from lti::statisticsFunctor.

template<class T >

bool lti::gaussianMixtureModel< T >::estimate

(

const matrix< T > &

src

)

Equivalent to apply.

Parameters:

src

matrix<T> with the source data, each row is considered a data point.

Returns:

true if the estimation was successfull.

Referenced by lti::gaussianMixtureModel< T >::apply().

template<class T >

void lti::gaussianMixtureModel< T >::getAlpha

(

vector< T > &

a

)

const [inline]

Copies the mixing coefficients to the given parameter.

Each vector element is the a-priori probability of a model component.

Parameters:

a

will receive the component mixture coefficients of the last estimation

template<class T >

const vector<T>& lti::gaussianMixtureModel< T >::getAlpha

(

)

const [inline]

Returns the mixing coefficient vector.

Each element is the mixing coefficient of a model component. It can be viewed as a

Returns:

a reference to the mixing coefficients of the last estimation

template<class T >

void lti::gaussianMixtureModel< T >::getComponentMus

(

std::vector< vector< T > > &

ms

)

const [inline]

Copies a std::vector of mean vectors of the components to the given parameter.

Each vector element is the mean vector of a model component.

Parameters:

ms

will receive the component means of the last estimation

template<class T >

const std::vector< vector<T> >& lti::gaussianMixtureModel< T >::getComponentMus

(

)

const [inline]

Returns a std::vector of mean vectors of the components.

Each vector element is the mean vector of a model component.

Returns:

a reference to the component means of the last estimation

template<class T >

void lti::gaussianMixtureModel< T >::getComponentSigmas

(

std::vector< matrix< T > > &

sigs

)

const [inline]

Copies a std::vector of covariance matrices of the components to the given parameter.

Each vector element is the covariance matrix of a model component.

Parameters:

sigs

will receive the component covariance matrices of the last estimation

template<class T >

const std::vector< matrix<T> >& lti::gaussianMixtureModel< T >::getComponentSigmas

(

)

const [inline]

Returns a std::vector of covariance matrices of the components.

Each vector element is the covariance matrix of a model component.

Returns:

a reference to the component covariance matrices of the last estimation

template<class T >

const parameters& lti::gaussianMixtureModel< T >::getParameters

(

)

const

returns used parameters

Reimplemented from lti::statisticsFunctor.

template<class T >

virtual const char* lti::gaussianMixtureModel< T >::getTypeName

(

)

const [virtual]

returns the name of this type ("gaussianMixtureModel")

Reimplemented from lti::statisticsFunctor.

template<class T >

void lti::gaussianMixtureModel< T >::setProgressInfo

(

progressInfo &

pi

)

Sets the progress info object.

---

The documentation for this class was generated from the following file:

• ltiGaussianMixtureModel.h