Base class for Learning a Mixture mixed model when some classes are known. More...

#include <STK_MixtureSemiLearner.h>

Inheritance diagram for STK::MixtureSemiLearner:

Public Member Functions
	MixtureSemiLearner (int nbSample, int nbCluster)
	Constructor.

	MixtureSemiLearner (MixtureSemiLearner const &model)
	copy constructor.

virtual	~MixtureSemiLearner ()
	destructor

virtual MixtureSemiLearner *	create () const
	Create a composer, but reinitialize the mixtures parameters.

virtual MixtureSemiLearner *	clone () const
	Create a clone of the current model, with mixtures parameters preserved.

std::vector< int > const &	knownLabels () const

std::vector< int > const &	unknownLabels () const

virtual int	cStep ()
	Replace tik by zik.

virtual int	sStep ()
	Simulate zi accordingly to tik and replace tik by zik by calling cStep().

virtual Real	eStep ()
	compute the zi, the lnLikelihood of the current estimates and the next value of the tik.

virtual void	mapStep ()
	Compute zi using the Map estimate.

template<class RowVector >
void	setLabels (ExprBase< RowVector > const &zi)
	Set the known labels.

Public Member Functions inherited from STK::MixtureComposer
	MixtureComposer (int nbSample, int nbCluster)
	Constructor.

	MixtureComposer (MixtureComposer const &composer)
	copy constructor.

virtual	~MixtureComposer ()
	The registered mixtures will be deleted there.

virtual Real	lnComponentProbability (int i, int k) const

virtual void	writeParameters (ostream &os) const
	write the parameters of the model in the stream os.

virtual void	randomInit ()
	initialize randomly the parameters of the components of the model

virtual void	paramUpdateStep ()
	Compute the proportions and the model parameters given the current tik mixture parameters.

virtual void	imputationStep ()
	Impute the missing values.

virtual void	samplingStep ()
	Simulation of all the latent variables and/or missing data excluding class labels.

virtual void	storeIntermediateResults (int iteration)
	This step can be used to signal to the mixtures that they must store results.

virtual void	releaseIntermediateResults ()
	This step can be used to signal to the mixtures that they must release the stored results.

virtual void	setParametersStep ()
	Utility method allowing to signal to a mixture to set its parameters.

virtual void	finalizeStep ()
	This step can be used by developer to finalize any thing.

Public Member Functions inherited from STK::IMixtureComposer
virtual	~IMixtureComposer ()
	destructor

Clust::modelState	state () const

void	setState (Clust::modelState state)
	set the state of the model : should be used by any strategy

virtual void	pStep ()
	Compute proportions using the ML estimates, default implementation.

virtual void	initializeStep ()
	Initialize the model before its first use.

void	randomClassInit ()
	Initialize randomly the labels zi of the model.

void	randomFuzzyInit ()
	Initialize randomly the posterior probabilities tik of the model, then compute the zi values with mapStep, compute the initial parameter values using paramUpdateStep, and compute the tik using eStep().

Public Member Functions inherited from STK::IMixtureStatModel
virtual	~IMixtureStatModel ()
	destructor

int	nbCluster () const

CPointX const &	pk () const

CArrayXX const &	tik () const

CPointX const &	tk () const

CVectorXi const &	zi () const

std::vector< IMixture * > const &	v_mixtures () const

Real	computeLnLikelihood (int i) const

Real	computeLikelihood (int i) const

Real	computeLnLikelihood () const

Real	computeICL () const

IMixture *	getMixture (String const &idData) const
	Utility lookup function allowing to find a Mixture from its idData.

void	registerMixture (IMixture *p_mixture)
	register a mixture to the composer.

void	releaseMixture (String const &idData)
	release a mixture from the composer.

int	computeNbFreeParameters () const
	compute the number of free parameters of the model.

int	computeNbMissingValues () const
	compute the missing values of the model.

template<class Array >
void	setMixtureParameters (Array const &tik)
	set the mixture parameters using an array of posterior probabilities.

template<class Array , class RowVector >
void	setMixtureParameters (Array const &tik, RowVector const &pk)
	set the mixture parameters giving the posterior probabilities and the proportions.

template<class RowVector >
void	setProportions (RowVector const &pk)
	Set proportions of each classes.

template<class Manager , class Parameters >
void	setParameters (IMixtureManager< Manager > const &manager, String const &idData, Parameters const &param)
	Utility method allowing to set the parameters to a specific mixture.

template<class Manager >
void	createMixture (IMixtureManager< Manager > &manager)
	Utility method allowing to create all the mixtures registered in the data handler of a mixture manager and to register them.

template<class Manager >
IMixture *	createMixture (IMixtureManager< Manager > &manager, String const &idData)
	Utility method allowing to create a mixture with a given data set and register it.

template<class Manager >
void	removeMixture (IMixtureManager< Manager > &manager, String const &idData)
	Utility method allowing to release completely a mixture with its data set.

template<class Manager , class Parameters >
void	getParameters (IMixtureManager< Manager > const &manager, String const &idData, Parameters &param) const
	Utility method allowing to get the parameters of a specific mixture.

template<class Manager , class MissingValues >
void	getMissingValues (IMixtureManager< Manager > const &manager, String const &idData, MissingValues &missing) const
	Utility method allowing to get the missing values of a specific mixture.

template<class DataHandler >
void	createMixture (IMixtureManager< DataHandler > &manager)

Public Member Functions inherited from STK::IStatModelBase
int	nbSample () const

Real	lnNbSample () const

int	nbVariable () const

Real	lnLikelihood () const

Real	likelihood () const

int	nbFreeParameter () const

Real	computeBIC () const

Real	computeAIC () const

Real	computeML () const

Protected Member Functions
virtual void	initializeMixtureParameters ()
	Create the mixture model parameters tik_ and pk_.

virtual int	randomTik ()
	generate random tik_

virtual int	randomZi ()
	generate random zi_

void	createSemiLearner (MixtureSemiLearner const *const p_learner)
	Copy the ingredient of the semi-learner in the creation process.

Protected Member Functions inherited from STK::MixtureComposer
void	createComposer (std::vector< IMixture * > const &v_mixtures_)
	Create the composer using existing data handler and mixtures.

Protected Member Functions inherited from STK::IMixtureComposer
	IMixtureComposer (int nbSample, int nbCluster)
	Constructor.

	IMixtureComposer (IMixtureComposer const &model)
	copy constructor.

void	sStep (int i)
	Simulate zi accordingly to tik.

void	cStep (int i)
	Replace tik by zik.

Real	eStep (int i)
	compute one zi and the next value of the tik for i fixed

void	mapStep (int i)
	Compute zi using the Map estimate for i fixed.

Protected Member Functions inherited from STK::IMixtureStatModel
	IMixtureStatModel (int nbSample, int nbCluster)
	Constructor.

	IMixtureStatModel (IMixtureStatModel const &model)
	copy constructor.

void	setNbCluster (int nbCluster)
	set the number of cluster of the model

Protected Member Functions inherited from STK::IStatModelBase
	IStatModelBase ()
	Default constructor.

	IStatModelBase (int nbSample)
	Constructor with specified dimension.

	IStatModelBase (int nbSample, int nbVariable)
	Constructor with specified dimension.

	IStatModelBase (IStatModelBase const &model)
	Copy constructor.

	~IStatModelBase ()
	destructor

void	setNbFreeParameter (int const &nbFreeParameter)
	set the number of free parameters of the model

void	setNbSample (int const &nbSample)
	set the number of samples of the model

void	setNbVariable (int const &nbVariable)
	set the number of variables of the model

void	setLnLikelihood (Real const &lnLikelihood)
	set the log-likelihood of the model

void	initialize (int nbSample, int nbVariable)
	set the dimensions of the parameters of the model

Protected Attributes
std::vector< int >	knownLabels_
	indexes of the known labels

std::vector< int >	unknownLabels_
	indexes of the unknown labels

Protected Attributes inherited from STK::MixtureComposer
Real	meanlnLikelihood_
	averaged lnLikelihood values.

Protected Attributes inherited from STK::IMixtureStatModel
int	nbCluster_
	number of cluster.

CPointX	pk_
	The proportions of each mixtures.

CArrayXX	tik_
	The tik probabilities.

CPointX	tk_
	The sum of the columns of tik_.

CVectorXi	zi_
	The zi class label.

std::vector< IMixture * >	v_mixtures_
	vector of pointers to the mixtures components

Additional Inherited Members
Public Types inherited from STK::MixtureComposer
typedef std::vector< IMixture * >::const_iterator	ConstMixtIterator

typedef std::vector< IMixture * >::iterator	MixtIterator

Public Types inherited from STK::IMixtureStatModel
typedef std::vector< IMixture * >::const_iterator	ConstMixtIterator

typedef std::vector< IMixture * >::iterator	MixtIterator

Detailed Description

Base class for Learning a Mixture mixed model when some classes are known.

This class extend the MixtureComposer class and allow to handle the case where some labels are known (semi-supervised learning).

See also: MixtureComposer, MixtureLearner

Definition at line 52 of file STK_MixtureSemiLearner.h.

Constructor & Destructor Documentation

◆ MixtureSemiLearner() [1/2]

STK::MixtureSemiLearner::MixtureSemiLearner	(	int	nbSample,
		int	nbCluster
	)

Constructor.

Parameters

nbCluster,nbSample number of clusters and samples

Definition at line 48 of file STK_MixtureSemiLearner.cpp.

                                      : MixtureComposer(nbSample, nbCluster)
                                      , knownLabels_()
                                      , unknownLabels_(nbSample)
{
  int i = zi_.begin();
  std::vector<int>::iterator it;
  for (it = unknownLabels_.begin(); it != unknownLabels_.end(); ++it, ++i)
  { *it = i;}
}

References unknownLabels_, and STK::IMixtureStatModel::zi_.

◆ MixtureSemiLearner() [2/2]

STK::MixtureSemiLearner::MixtureSemiLearner ( MixtureSemiLearner const & model )

copy constructor.

Parameters

model the model to clone

Definition at line 61 of file STK_MixtureSemiLearner.cpp.

                                      : MixtureComposer(model)
                                      , knownLabels_(model.knownLabels_)
                                      , unknownLabels_(model.unknownLabels_)
{}

◆ ~MixtureSemiLearner()

STK::MixtureSemiLearner::~MixtureSemiLearner ( )

virtual

destructor

Definition at line 67 of file STK_MixtureSemiLearner.cpp.

67{}

Member Function Documentation

◆ clone()

MixtureSemiLearner * STK::MixtureSemiLearner::clone ( ) const

virtual

Create a clone of the current model, with mixtures parameters preserved.

Reimplemented from STK::MixtureComposer.

Reimplemented in STK::MixtureSemiLearnerFixedProp.

Definition at line 70 of file STK_MixtureSemiLearner.cpp.

71{ return new MixtureSemiLearner(*this);}

STK::MixtureSemiLearner::MixtureSemiLearner

MixtureSemiLearner(int nbSample, int nbCluster)

Constructor.

Definition STK_MixtureSemiLearner.cpp:48

◆ create()

MixtureSemiLearner * STK::MixtureSemiLearner::create ( ) const

virtual

Create a composer, but reinitialize the mixtures parameters.

Reimplemented from STK::MixtureComposer.

Reimplemented in STK::MixtureSemiLearnerFixedProp.

Definition at line 73 of file STK_MixtureSemiLearner.cpp.

{
  // set dimensions
  MixtureSemiLearner* p_composer = new MixtureSemiLearner(nbSample(), nbCluster());
  p_composer->createSemiLearner(this);
  p_composer->createComposer(v_mixtures());
  return p_composer;
}

References STK::MixtureComposer::createComposer(), createSemiLearner(), STK::IMixtureStatModel::nbCluster(), STK::IStatModelBase::nbSample(), and STK::IMixtureStatModel::v_mixtures().

◆ createSemiLearner()

void STK::MixtureSemiLearner::createSemiLearner ( MixtureSemiLearner const *const p_learner )

protected

Copy the ingredient of the semi-learner in the creation process.

Definition at line 178 of file STK_MixtureSemiLearner.cpp.

{
  knownLabels_   = p_learner->knownLabels();
  unknownLabels_ = p_learner->unknownLabels();
}

References knownLabels_, and unknownLabels_.

Referenced by create(), and STK::MixtureSemiLearnerFixedProp::create().

◆ cStep()

int STK::MixtureSemiLearner::cStep ( )

virtual

Replace tik by zik.

Returns: the minimal value of individuals in a class

Reimplemented from STK::IMixtureComposer.

Definition at line 84 of file STK_MixtureSemiLearner.cpp.

{
  std::vector<int>::const_iterator it;
  for (it=unknownLabels_.begin(); it != unknownLabels_.end(); it++)
  { MixtureComposer::cStep(*it);}
  // count the number of individuals in each class
  tk_= Stat::sum(tik_);
  return tk_.minElt();
}

References STK::IMixtureComposer::cStep(), STK::ExprBase< Derived >::minElt(), STK::Stat::sum(), STK::IMixtureStatModel::tik_, STK::IMixtureStatModel::tk_, and unknownLabels_.

Referenced by randomZi(), and sStep().

◆ eStep()

Real STK::MixtureSemiLearner::eStep ( )

virtual

compute the zi, the lnLikelihood of the current estimates and the next value of the tik.

Returns: the minimal value of tk

Reimplemented from STK::IMixtureComposer.

Definition at line 105 of file STK_MixtureSemiLearner.cpp.

{
#ifdef STK_MIXTURE_DEBUG
  stk_cout << _T("Entering MixtureSemiLearner::eStep()\n");
#endif
  Real sum = 0.; tk_ =0.;
//#ifdef _OPENMP
//#pragma omp parallel for reduction (+:sum)
//#endif
  std::vector<int>::const_iterator it;
  for(it=unknownLabels_.begin(); it != unknownLabels_.end(); it++)
  { sum += MixtureComposer::eStep(*it);}
  for(it=knownLabels_.begin(); it != knownLabels_.end(); it++)
  { sum += lnComponentProbability(*it, zi_[*it]) + std::log(pk_[zi_[*it]]);}
  // update log-likelihood
  setLnLikelihood(sum);
  // compute proportions
  tk_ = Stat::sumByCol(tik_);
#ifdef STK_MIXTURE_DEBUG
  stk_cout << _T("MixtureSemiLearner::eStep() done\n");
  stk_cout << _T("lnLikelihood =") << sum << _T("\n");
#endif
  return tk_.minElt();
}

References _T, STK::IMixtureComposer::eStep(), knownLabels_, STK::MixtureComposer::lnComponentProbability(), STK::ExprBase< Derived >::minElt(), STK::IMixtureStatModel::pk_, STK::IStatModelBase::setLnLikelihood(), stk_cout, STK::sum(), STK::Stat::sumByCol(), STK::IMixtureStatModel::tik_, STK::IMixtureStatModel::tk_, unknownLabels_, and STK::IMixtureStatModel::zi_.

◆ initializeMixtureParameters()

void STK::MixtureSemiLearner::initializeMixtureParameters ( )

protectedvirtual

Create the mixture model parameters tik_ and pk_.

Reimplemented from STK::IMixtureComposer.

Definition at line 140 of file STK_MixtureSemiLearner.cpp.

{
  std::vector<int>::const_iterator it;
  for (it=unknownLabels_.begin(); it != unknownLabels_.end(); it++)
  { tik_.row(*it) = 1./nbCluster();}
  for (it=knownLabels_.begin(); it != knownLabels_.end(); it++)
  { tik_.row(*it) = 0.; tik_(*it, zi_[*it]) = 1.;}
}

References knownLabels_, STK::IMixtureStatModel::nbCluster(), STK::ICArray< Derived >::row(), STK::IMixtureStatModel::tik_, unknownLabels_, and STK::IMixtureStatModel::zi_.

◆ knownLabels()

std::vector< int > const & STK::MixtureSemiLearner::knownLabels ( ) const

inline

Returns: the vector of the known labels

Definition at line 70 of file STK_MixtureSemiLearner.h.

70{ return knownLabels_;}

References knownLabels_.

◆ mapStep()

void STK::MixtureSemiLearner::mapStep ( )

virtual

Compute zi using the Map estimate.

Reimplemented from STK::IMixtureComposer.

Definition at line 131 of file STK_MixtureSemiLearner.cpp.

{
  std::vector<int>::const_iterator it;
  for (it=unknownLabels_.begin(); it != unknownLabels_.end(); it++)
  { MixtureComposer::mapStep(*it);  }
}

References STK::IMixtureComposer::mapStep(), and unknownLabels_.

◆ randomTik()

int STK::MixtureSemiLearner::randomTik ( )

protectedvirtual

generate random tik_

Reimplemented from STK::IMixtureComposer.

Definition at line 150 of file STK_MixtureSemiLearner.cpp.

{
  tk_ = 0.;
  std::vector<int>::const_iterator it;
  for (it = unknownLabels_.begin(); it != unknownLabels_.end(); ++it)
  {
    // create a reference on the i-th row
    CPointX tikRowi(tik_.row(*it), true);
    tikRowi.randUnif();
    tikRowi = tikRowi * pk_;
    tikRowi /= tikRowi.sum();
    tk_ += tikRowi;
  }
  return tk_.minElt();
}

References STK::ExprBase< Derived >::minElt(), STK::IMixtureStatModel::pk_, STK::ICArray< Derived >::row(), STK::IMixtureStatModel::tik_, STK::IMixtureStatModel::tk_, and unknownLabels_.

◆ randomZi()

int STK::MixtureSemiLearner::randomZi ( )

protectedvirtual

generate random zi_

Reimplemented from STK::IMixtureComposer.

Definition at line 167 of file STK_MixtureSemiLearner.cpp.

{
  Law::Categorical law(pk_);
  std::vector<int>::const_iterator it;
  for (it = unknownLabels_.begin(); it != unknownLabels_.end(); ++it)
  { zi_[*it] = law.rand();}
  // create tik and compute nk
  return cStep();
}

References cStep(), STK::IMixtureStatModel::pk_, STK::Law::Categorical::rand(), unknownLabels_, and STK::IMixtureStatModel::zi_.

◆ setLabels()

template<class RowVector >

void STK::MixtureSemiLearner::setLabels ( ExprBase< RowVector > const & zi )

Set the known labels.

This method is expecting a vector of size nbSample with NA values for unknown labels.

Parameters

zi	the vector with the labels

Definition at line 113 of file STK_MixtureSemiLearner.h.

{
  if (nbSample() != zi.size())
  { STKRUNTIME_ERROR_NO_ARG(MixtureSemiLearner::setKnownLabels,zi must be of size nbSample);}
  int nbUnknown = zi.isNA().count();
#ifdef STK_MIXTURE_DEBUG
  stk_cout << _T("Entering MixtureSemiLearner::setLabels()\n");
  stk_cout << _T("nbUnknown=") << nbUnknown << _T("\n");
#endif
  // clear previous initialization
  unknownLabels_.clear();
  knownLabels_.clear();
  zi_= baseIdx;
  // reserve and set lables
  unknownLabels_.reserve(nbUnknown);
  knownLabels_.reserve(nbSample() - nbUnknown);
  for (int i=zi.begin(); i < zi.end(); ++i)
  {
    if (isNA(zi[i]))
    { unknownLabels_.push_back(i);}
    else
    {
      knownLabels_.push_back(i); zi_[i] = zi[i];
      tik_.row(i) = 0; tik_(i, zi_[i]) = 1.;
    }
  }
#ifdef STK_MIXTURE_DEBUG
  stk_cout << _T("Terminating MixtureSemiLearner::setLabels()\n");
  stk_cout << _T("unknownLabels_.size()=") << unknownLabels_.size() << _T("\n");
  stk_cout << _T("knownLabels_.size()=") << knownLabels_.size() << _T("\n");
#endif
}

References _T, STK::baseIdx, STK::ExprBase< Derived >::isNA(), STK::isNA(), knownLabels_, STK::IStatModelBase::nbSample(), STK::ICArray< Derived >::row(), stk_cout, STKRUNTIME_ERROR_NO_ARG, STK::IMixtureStatModel::tik_, unknownLabels_, STK::IMixtureStatModel::zi(), and STK::IMixtureStatModel::zi_.

◆ sStep()

int STK::MixtureSemiLearner::sStep ( )

virtual

Simulate zi accordingly to tik and replace tik by zik by calling cStep().

Returns: the minimal value of individuals in a class

Reimplemented from STK::IMixtureComposer.

Definition at line 95 of file STK_MixtureSemiLearner.cpp.

{
  // simulate zi
  std::vector<int>::const_iterator it;
  for (it=unknownLabels_.begin(); it != unknownLabels_.end(); it++)
  { MixtureComposer::sStep(*it);}
  return cStep();
}

References cStep(), STK::IMixtureComposer::sStep(), and unknownLabels_.

◆ unknownLabels()

std::vector< int > const & STK::MixtureSemiLearner::unknownLabels ( ) const

inline

Returns: the vector of the unknown labels

Definition at line 72 of file STK_MixtureSemiLearner.h.

72{ return unknownLabels_;}

References unknownLabels_.

Member Data Documentation

◆ knownLabels_

std::vector<int> STK::MixtureSemiLearner::knownLabels_

protected

indexes of the known labels

Definition at line 106 of file STK_MixtureSemiLearner.h.

Referenced by createSemiLearner(), eStep(), initializeMixtureParameters(), knownLabels(), and setLabels().

◆ unknownLabels_

std::vector<int> STK::MixtureSemiLearner::unknownLabels_

protected

indexes of the unknown labels

Definition at line 108 of file STK_MixtureSemiLearner.h.

Referenced by createSemiLearner(), cStep(), eStep(), initializeMixtureParameters(), mapStep(), MixtureSemiLearner(), randomTik(), randomZi(), setLabels(), sStep(), and unknownLabels().

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

Public Member Functions

Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ MixtureSemiLearner() [1/2]

◆ MixtureSemiLearner() [2/2]

◆ ~MixtureSemiLearner()

Member Function Documentation

◆ clone()

◆ create()

◆ createSemiLearner()

◆ cStep()

◆ eStep()

◆ initializeMixtureParameters()

◆ knownLabels()

◆ mapStep()

◆ randomTik()

◆ randomZi()

◆ setLabels()

◆ sStep()

◆ unknownLabels()

Member Data Documentation

◆ knownLabels_

◆ unknownLabels_