Gaussian Mixture Model (GMM) optimization using maximal likelihood estimation via the Expection-Maximization (EM) algorithm. More...

#include <mdsEM.h>

Inheritance diagram for mds::math::CMaxLikelihoodByEM< C, N >:

[legend]

List of all members.

Public Types
enum	{ NO_ITERATOR_IS_DECLARED_FOR_TEMPLATE_PARAMETER = C::ITERATOR_DECLARED }
	Check that C has an iterator declared. More...
enum	{ NUM_OF_DIMENSIONS = N }
	Dimensionality of input data. More...
typedef CGaussianFunc< double, N >	tComponent
	Gaussian function.
typedef C::tConstIterator	tConstIterator
	Container iterator type.
typedef C	tContainer
	Container type.
typedef CVector< double >	tVector
	Used internal vector type.
Public Member Functions
CHasOptions &	addOption (int Flag)
	Adds a given flag using bitwise OR operation.
bool	checkAllOptions (int Flags) const
	Checks if all specified flags are set.
bool	checkAnyOption (int Flags) const
	Checks if any of specified flags is set.
int	checkOption (int Flag) const
	Checks if a single flag is set.
CHasOptions &	clearOption (int Flag)
	Removes a given flag.
	CMaxLikelihoodByEM (int Flags=MaxLikelihoodByEM::DEFAULT_FLAGS)
	Default constructor.
bool	execute (const tContainer &Input)
	Estimates GMM of input data.
bool	execute (const tContainer &Input, tSize NumOfClasses)
	Estimates GMM of input data.
const tComponent &	getComponent (tSize i) const
	Returns reference to the i-th gaussian component.
tComponent &	getComponent (tSize i)
bool	getMembership (tSize i, tVector &Membership)
	Returns soft membership function of the i-th input sample.
tSize	getMembership (tSize i)
	Returns index of the component into which the i-th input vector was classified (hard membership function).
tSize	getNumOfComponents () const
	Returns the number of components.
tSize	getNumOfDimensions () const
	Returns the number od dimensions.
int	getOptions () const
	Returns all flags.
CMaxLikelihoodByEM &	setMinAddChange (double dValue)
	Changes the minimal required change of the log-likelihood function which results in a new component addition.
CHasOptions &	setOption (int Flag)
	Adds a given flag using bitwise OR operation.
CHasOptions &	setOptions (int Flags)
	Sets complete flags.
virtual	~CMaxLikelihoodByEM ()
	Virtual destructor.
Protected Types
typedef std::vector< tComponent >	tComponents
	Gaussian functions representing mixture components/segments.
typedef tComponent::tVector	tCov
	Diagonal of the gaussian component covariance matrix.
typedef CMatrix< double >	tMaps
	Support maps.
typedef tComponent::tVector	tMean
	Gaussian component mean.
Protected Member Functions
void	addRandomNoise (double dNoiseCoeff)
	Add random noise to i-th component.
bool	areSeparated (tSize i, tSize j, int c)
	Checks degree of separation of two given components.
bool	checkForSimilarComponents ()
	Checks for similar components.
void	clearSupportMaps ()
	Clears all support maps.
double	computeLogLikelihood (const tContainer &Input)
	Returns value of the objective function.
double	computeMDL (double dLogLikelihood, tSize NumOfSamples)
	Compute the MDL (Minimum Descriptive Length) coefficient.
tSize	findComponent ()
	Finds component which can be further splitted.
int	getDegreeOfSeparation (tSize NumOfComponents)
	Returns degree of separation.
void	initComponentCov (tSize i)
	Intialization of a component covariance matrix.
void	initComponentMean (tSize i, const tContainer &Input)
	Random intialization of i-th component mean value.
void	initComponentWeight (tSize i, double dWeight)
	Intialization of a component weight.
bool	initFirstComponent (const tContainer &Input)
	Initialization of the first mixture component.
double	iterateEM (const tContainer &Input, double dMinChange)
	Iterates both E-step and M-step.
void	processEStep (const tContainer &Input)
	E-step of the algorithm.
void	processMStep (const tContainer &Input)
	M-step of the algorithm.
void	splitComponent (tSize Index, tSize NewIndex)
	Splits a given component.
Protected Attributes
tComponents	m_Components
	Mixture model components.
double	m_dMinAddChange
	Minimal required change of the log-likelihood function which results in component addition.
tMaps	m_Maps
	Support maps.
CNormalPRNG	m_Normal
	Random number generator with normal distribution.
int	m_Options
	Optional flags.
CUniformPRNG	m_Uniform
	Uniform random number generator.

Detailed Description

template<class C, tSize N>
class mds::math::CMaxLikelihoodByEM< C, N >

Gaussian Mixture Model (GMM) optimization using maximal likelihood estimation via the Expection-Maximization (EM) algorithm.

Parameter C is a container type whose items will be clustered.
Only static vectors and fundamental types of container items are allowed.
Parameter N specifies dimensionality of input data.

Member Typedef Documentation

template<class C , tSize N>

typedef CGaussianFunc<double,N> mds::math::CMaxLikelihoodByEM< C, N >::tComponent

Gaussian function.

template<class C , tSize N>

typedef std::vector<tComponent> mds::math::CMaxLikelihoodByEM< C, N >::tComponents [protected]

Gaussian functions representing mixture components/segments.

template<class C , tSize N>

typedef C::tConstIterator mds::math::CMaxLikelihoodByEM< C, N >::tConstIterator

Container iterator type.

template<class C , tSize N>

typedef C mds::math::CMaxLikelihoodByEM< C, N >::tContainer

Container type.

template<class C , tSize N>

typedef tComponent::tVector mds::math::CMaxLikelihoodByEM< C, N >::tCov [protected]

Diagonal of the gaussian component covariance matrix.

template<class C , tSize N>

typedef CMatrix<double> mds::math::CMaxLikelihoodByEM< C, N >::tMaps [protected]

Support maps.

template<class C , tSize N>

typedef tComponent::tVector mds::math::CMaxLikelihoodByEM< C, N >::tMean [protected]

Gaussian component mean.

template<class C , tSize N>

typedef CVector<double> mds::math::CMaxLikelihoodByEM< C, N >::tVector

Used internal vector type.

Member Enumeration Documentation

template<class C , tSize N>

anonymous enum

Check that C has an iterator declared.

You will see name of this enum somewhere in compiler error message if the type C can't be used with iterators.

Enumerator:

NO_ITERATOR_IS_DECLARED_FOR_TEMPLATE_PARAMETER

template<class C , tSize N>

anonymous enum

Dimensionality of input data.

Enumerator:

NUM_OF_DIMENSIONS

Constructor & Destructor Documentation

template<class C , tSize N>

CMaxLikelihoodByEM::CMaxLikelihoodByEM ( int Flags = MaxLikelihoodByEM::DEFAULT_FLAGS )

Default constructor.

template<class C , tSize N>

virtual mds::math::CMaxLikelihoodByEM< C, N >::~CMaxLikelihoodByEM ( ) [virtual]

Virtual destructor.

Member Function Documentation

CHasOptions& mds::base::CHasOptions::addOption ( int Flag ) [inherited]

Adds a given flag using bitwise OR operation.

template<class C , tSize N>

void CMaxLikelihoodByEM::addRandomNoise ( double dNoiseCoeff ) [protected]

Add random noise to i-th component.

template<class C , tSize N>

bool CMaxLikelihoodByEM::areSeparated	(	tSize	i,
		tSize	j,
		int	c
	)		`[protected]`

Checks degree of separation of two given components.

bool mds::base::CHasOptions::checkAllOptions ( int Flags ) const [inherited]

Checks if all specified flags are set.

bool mds::base::CHasOptions::checkAnyOption ( int Flags ) const [inherited]

Checks if any of specified flags is set.

template<class C , tSize N>

bool CMaxLikelihoodByEM::checkForSimilarComponents ( ) [protected]

Checks for similar components.

If there are two similar components, one of them is re-initialized.

int mds::base::CHasOptions::checkOption ( int Flag ) const [inherited]

Checks if a single flag is set.

Returns zero on false.

CHasOptions& mds::base::CHasOptions::clearOption ( int Flag ) [inherited]

Removes a given flag.

template<class C , tSize N>

void CMaxLikelihoodByEM::clearSupportMaps ( ) [protected]

Clears all support maps.

template<class C , tSize N>

double CMaxLikelihoodByEM::computeLogLikelihood ( const tContainer & Input ) [protected]

Returns value of the objective function.

template<class C , tSize N>

double CMaxLikelihoodByEM::computeMDL	(	double	dLogLikelihood,
		tSize	NumOfSamples
	)		`[protected]`

Compute the MDL (Minimum Descriptive Length) coefficient.

template<class C , tSize N>

bool CMaxLikelihoodByEM::execute ( const tContainer & Input )

Estimates GMM of input data.

Return false on failure.

template<class C , tSize N>

bool CMaxLikelihoodByEM::execute	(	const tContainer &	Input,
		tSize	NumOfClasses
	)

Estimates GMM of input data.

Fixed number of mixture components.
Return false on failure.

template<class C , tSize N>

tSize CMaxLikelihoodByEM::findComponent ( ) [protected]

Finds component which can be further splitted.

Returns index of such componnet.

template<class C , tSize N>

const tComponent& mds::math::CMaxLikelihoodByEM< C, N >::getComponent ( tSize i ) const

Returns reference to the i-th gaussian component.

Method can be called after successfull clustering!

template<class C , tSize N>

tComponent& mds::math::CMaxLikelihoodByEM< C, N >::getComponent ( tSize i )

template<class C , tSize N>

int CMaxLikelihoodByEM::getDegreeOfSeparation ( tSize NumOfComponents ) [protected]

Returns degree of separation.

template<class C , tSize N>

bool CMaxLikelihoodByEM::getMembership	(	tSize	i,
		tVector &	Membership
	)

Returns soft membership function of the i-th input sample.

Method can be called after successfull clustering!

template<class C , tSize N>

tSize CMaxLikelihoodByEM::getMembership ( tSize i )

Returns index of the component into which the i-th input vector was classified (hard membership function).

Method can be called after successfull clustering!

template<class C , tSize N>

tSize mds::math::CMaxLikelihoodByEM< C, N >::getNumOfComponents ( ) const

Returns the number of components.

Method can be called after successfull clustering!

template<class C , tSize N>

tSize mds::math::CMaxLikelihoodByEM< C, N >::getNumOfDimensions ( ) const

Returns the number od dimensions.

int mds::base::CHasOptions::getOptions ( ) const [inherited]

Returns all flags.

template<class C , tSize N>

void CMaxLikelihoodByEM::initComponentCov ( tSize i ) [protected]

Intialization of a component covariance matrix.

template<class C , tSize N>

void CMaxLikelihoodByEM::initComponentMean	(	tSize	i,
		const tContainer &	Input
	)		`[protected]`

Random intialization of i-th component mean value.

template<class C , tSize N>

void CMaxLikelihoodByEM::initComponentWeight	(	tSize	i,
		double	dWeight
	)		`[protected]`

Intialization of a component weight.

template<class C , tSize N>

bool CMaxLikelihoodByEM::initFirstComponent ( const tContainer & Input ) [protected]

Initialization of the first mixture component.

template<class C , tSize N>

double CMaxLikelihoodByEM::iterateEM	(	const tContainer &	Input,
		double	dMinChange
	)		`[protected]`

Iterates both E-step and M-step.

Returns final value of the log-likelihood function

template<class C , tSize N>

void CMaxLikelihoodByEM::processEStep ( const tContainer & Input ) [protected]

E-step of the algorithm.

template<class C , tSize N>

void CMaxLikelihoodByEM::processMStep ( const tContainer & Input ) [protected]

M-step of the algorithm.

template<class C , tSize N>

CMaxLikelihoodByEM< C, N > & CMaxLikelihoodByEM::setMinAddChange ( double dValue )

Changes the minimal required change of the log-likelihood function which results in a new component addition.

CHasOptions& mds::base::CHasOptions::setOption ( int Flag ) [inherited]

Adds a given flag using bitwise OR operation.

CHasOptions& mds::base::CHasOptions::setOptions ( int Flags ) [inherited]

Sets complete flags.

template<class C , tSize N>

void CMaxLikelihoodByEM::splitComponent	(	tSize	Index,
		tSize	NewIndex
	)		`[protected]`

Splits a given component.

Member Data Documentation

template<class C , tSize N>

tComponents mds::math::CMaxLikelihoodByEM< C, N >::m_Components [protected]

Mixture model components.

template<class C , tSize N>

double mds::math::CMaxLikelihoodByEM< C, N >::m_dMinAddChange [protected]

Minimal required change of the log-likelihood function which results in component addition.

template<class C , tSize N>

tMaps mds::math::CMaxLikelihoodByEM< C, N >::m_Maps [protected]

Support maps.

template<class C , tSize N>

CNormalPRNG mds::math::CMaxLikelihoodByEM< C, N >::m_Normal [protected]

Random number generator with normal distribution.

int mds::base::CHasOptions::m_Options [protected, inherited]

Optional flags.

template<class C , tSize N>

CUniformPRNG mds::math::CMaxLikelihoodByEM< C, N >::m_Uniform [protected]

Uniform random number generator.

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

Public Types

Public Member Functions

Protected Types

Protected Member Functions

Protected Attributes

Detailed Description

template<class C, tSize N> class mds::math::CMaxLikelihoodByEM< C, N >

Member Typedef Documentation

Member Enumeration Documentation

Constructor & Destructor Documentation

Member Function Documentation

Member Data Documentation

template<class C, tSize N>
class mds::math::CMaxLikelihoodByEM< C, N >