The base class for estimating parameters of CPTs. More...

#include <agrum/BN/learning/paramUtils/paramEstimator.h>

Inheritance diagram for gum::learning::ParamEstimator:

Collaboration diagram for gum::learning::ParamEstimator:

Public Member Functions
Constructors / Destructors
	ParamEstimator (const DBRowGeneratorParser &parser, const Prior &external_prior, const Prior &_score_internal_prior, const std::vector< std::pair< std::size_t, std::size_t > > &ranges, const Bijection< NodeId, std::size_t > &nodeId2columns=Bijection< NodeId, std::size_t >())
	default constructor
	ParamEstimator (const DBRowGeneratorParser &parser, const Prior &external_prior, const Prior &_score_internal_prior, const Bijection< NodeId, std::size_t > &nodeId2columns=Bijection< NodeId, std::size_t >())
	default constructor
	ParamEstimator (const ParamEstimator &from)
	copy constructor
	ParamEstimator (ParamEstimator &&from)
	move constructor
virtual ParamEstimator *	clone () const =0
	virtual copy constructor
virtual	~ParamEstimator ()
	destructor
Accessors / Modifiers
virtual void	clear ()
	clears all the data structures from memory
virtual void	setNumberOfThreads (Size nb)
	sets the number max of threads that can be used
virtual Size	getNumberOfThreads () const
	returns the current max number of threads of the scheduler
virtual bool	isGumNumberOfThreadsOverriden () const
	indicates whether the user set herself the number of threads
virtual void	setMinNbRowsPerThread (const std::size_t nb) const
	changes the number min of rows a thread should process in a multithreading context
virtual std::size_t	minNbRowsPerThread () const
	returns the minimum of rows that each thread should process
void	setRanges (const std::vector< std::pair< std::size_t, std::size_t > > &new_ranges)
	sets new ranges to perform the counts used by the parameter estimator
void	clearRanges ()
	reset the ranges to the one range corresponding to the whole database
const std::vector< std::pair< std::size_t, std::size_t > > &	ranges () const
	returns the current ranges
std::vector< double >	parameters (const NodeId target_node)
	returns the CPT's parameters corresponding to a given target node
std::pair< std::vector< double >, double >	parametersAndLogLikelihood (const NodeId target_node)
	returns the parameters of a CPT as well as its log-likelihood
virtual std::vector< double >	parameters (const NodeId target_node, const std::vector< NodeId > &conditioning_nodes)=0
	returns the CPT's parameters corresponding to a given nodeset
virtual std::pair< std::vector< double >, double >	parametersAndLogLikelihood (const NodeId target_node, const std::vector< NodeId > &conditioning_nodes)=0
	returns the parameters of a CPT as well as its log-likelihood
template<typename GUM_SCALAR>
double	setParameters (const NodeId target_node, const std::vector< NodeId > &conditioning_nodes, Tensor< GUM_SCALAR > &pot, const bool compute_log_likelihood=false)
	sets a CPT's parameters and, possibly, return its log-likelihhod
const Bijection< NodeId, std::size_t > &	nodeId2Columns () const
	returns the mapping from ids to column positions in the database
const DatabaseTable &	database () const
	returns the database on which we perform the counts
template<typename GUM_SCALAR>
void	setBayesNet (const BayesNet< GUM_SCALAR > &new_bn)
	assign a new Bayes net to all the counter's generators depending on a BN

Protected Member Functions
ParamEstimator &	operator= (const ParamEstimator &from)
	copy operator
ParamEstimator &	operator= (ParamEstimator &&from)
	move operator

Protected Attributes
Prior *	external_prior_ {nullptr}
	an external a priori
Prior *	score_internal_prior_ {nullptr}
	if a score was used for learning the structure of the PGM, this is the priori internal to the score
RecordCounter	counter_
	the record counter used to parse the database
const std::vector< NodeId >	empty_nodevect_
	an empty vector of nodes, used for empty conditioning

Detailed Description

The base class for estimating parameters of CPTs.

Definition at line 67 of file paramEstimator.h.

Constructor & Destructor Documentation

◆ ParamEstimator() [1/4]

gum::learning::ParamEstimator::ParamEstimator	(	const DBRowGeneratorParser &	parser,
		const Prior &	external_prior,
		const Prior &	_score_internal_prior,
		const std::vector< std::pair< std::size_t, std::size_t > > &	ranges,
		const Bijection< NodeId, std::size_t > &	nodeId2columns = Bijection< NodeId, std::size_t >() )

default constructor

Parameters

parser	the parser used to parse the database
external_prior	An prior that we add to the computation of the score
score_internal_prior	The prior within the score used to learn the data structure (might be a NoPrior)
ranges	a set of pairs {(X1,Y1),...,(Xn,Yn)} of database's rows indices. The counts are then performed only on the union of the rows [Xi,Yi), i in {1,...,n}. This is useful, e.g, when performing cross validation tasks, in which part of the database should be ignored. An empty set of ranges is equivalent to an interval [X,Y) ranging over the whole database.
nodeId2Columns	a mapping from the ids of the nodes in the graphical model to the corresponding column in the DatabaseTable parsed by the parser. This enables estimating from a database in which variable A corresponds to the 2nd column the parameters of a BN in which variable A has a NodeId of 5. An empty nodeId2Columns bijection means that the mapping is an identity, i.e., the value of a NodeId is equal to the index of the column in the DatabaseTable.

Warning: If nodeId2columns is not empty, then only the scores over the ids belonging to this bijection can be computed: applying method score() over other ids will raise exception NotFound.

References ranges().

Referenced by ParamEstimator(), ParamEstimator(), clone(), operator=(), and operator=().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ ParamEstimator() [2/4]

gum::learning::ParamEstimator::ParamEstimator	(	const DBRowGeneratorParser &	parser,
		const Prior &	external_prior,
		const Prior &	_score_internal_prior,
		const Bijection< NodeId, std::size_t > &	nodeId2columns = Bijection< NodeId, std::size_t >() )

default constructor

Parameters

parser	the parser used to parse the database
external_prior	An prior that we add to the computation of the score
score_internal_prior	The prior within the score used to learn the data structure (might be a NoPrior)
nodeId2Columns	a mapping from the ids of the nodes in the graphical model to the corresponding column in the DatabaseTable parsed by the parser. This enables estimating from a database in which variable A corresponds to the 2nd column the parameters of a BN in which variable A has a NodeId of 5. An empty nodeId2Columns bijection means that the mapping is an identity, i.e., the value of a NodeId is equal to the index of the column in the DatabaseTable.

Warning: If nodeId2columns is not empty, then only the scores over the ids belonging to this bijection can be computed: applying method score() over other ids will raise exception NotFound.

◆ ParamEstimator() [3/4]

gum::learning::ParamEstimator::ParamEstimator ( const ParamEstimator & from )

copy constructor

References ParamEstimator().

Here is the call graph for this function:

◆ ParamEstimator() [4/4]

gum::learning::ParamEstimator::ParamEstimator ( ParamEstimator && from )

move constructor

References ParamEstimator().

Here is the call graph for this function:

◆ ~ParamEstimator()

virtual gum::learning::ParamEstimator::~ParamEstimator ( )

virtual

destructor

Member Function Documentation

◆ clear()

virtual void gum::learning::ParamEstimator::clear ( )

virtual

clears all the data structures from memory

Referenced by gum::learning::DAG2BNLearner::createBNwithEM(), and gum::learning::DAG2BNLearner::createBNwithEM().

Here is the caller graph for this function:

◆ clearRanges()

void gum::learning::ParamEstimator::clearRanges ( )

reset the ranges to the one range corresponding to the whole database

◆ clone()

virtual ParamEstimator * gum::learning::ParamEstimator::clone ( ) const

pure virtual

virtual copy constructor

Implemented in gum::learning::ParamEstimatorML.

References ParamEstimator().

Here is the call graph for this function:

◆ database()

const DatabaseTable & gum::learning::ParamEstimator::database ( ) const

returns the database on which we perform the counts

◆ getNumberOfThreads()

virtual Size gum::learning::ParamEstimator::getNumberOfThreads ( ) const

virtual

returns the current max number of threads of the scheduler

Implements gum::IThreadNumberManager.

◆ isGumNumberOfThreadsOverriden()

virtual bool gum::learning::ParamEstimator::isGumNumberOfThreadsOverriden ( ) const

virtual

indicates whether the user set herself the number of threads

Implements gum::IThreadNumberManager.

◆ minNbRowsPerThread()

virtual std::size_t gum::learning::ParamEstimator::minNbRowsPerThread ( ) const

virtual

returns the minimum of rows that each thread should process

◆ nodeId2Columns()

const Bijection< NodeId, std::size_t > & gum::learning::ParamEstimator::nodeId2Columns ( ) const

returns the mapping from ids to column positions in the database

Warning: An empty nodeId2Columns bijection means that the mapping is an identity, i.e., the value of a NodeId is equal to the index of the column in the DatabaseTable.

◆ operator=() [1/2]

ParamEstimator & gum::learning::ParamEstimator::operator= ( const ParamEstimator & from )

protected

copy operator

References ParamEstimator().

Here is the call graph for this function:

◆ operator=() [2/2]

ParamEstimator & gum::learning::ParamEstimator::operator= ( ParamEstimator && from )

protected

move operator

References ParamEstimator().

Here is the call graph for this function:

◆ parameters() [1/2]

std::vector< double > gum::learning::ParamEstimator::parameters ( const NodeId target_node )

returns the CPT's parameters corresponding to a given target node

Referenced by gum::learning::ParamEstimatorML::operator=().

Here is the caller graph for this function:

◆ parameters() [2/2]

virtual std::vector< double > gum::learning::ParamEstimator::parameters	(	const NodeId	target_node,
		const std::vector< NodeId > &	conditioning_nodes )

pure virtual

returns the CPT's parameters corresponding to a given nodeset

The vector contains the parameters of an n-dimensional CPT. The distribution of the dimensions of the CPT within the vector is as follows: first, there is the target node, then the conditioning nodes (in the order in which they were specified).

Implemented in gum::learning::ParamEstimatorML.

◆ parametersAndLogLikelihood() [1/2]

std::pair< std::vector< double >, double > gum::learning::ParamEstimator::parametersAndLogLikelihood ( const NodeId target_node )

returns the parameters of a CPT as well as its log-likelihood

◆ parametersAndLogLikelihood() [2/2]

virtual std::pair< std::vector< double >, double > gum::learning::ParamEstimator::parametersAndLogLikelihood	(	const NodeId	target_node,
		const std::vector< NodeId > &	conditioning_nodes )

pure virtual

returns the parameters of a CPT as well as its log-likelihood

The vector contains the parameters of an n-dimensional CPT. The distribution of the dimensions of the CPT within the vector is as follows: first, there is the target node, then the conditioning nodes (in the order in which they were specified).

Parameters

target_node	the node on the left side of the CPT's conditioning bar
conditioning_nodes	thes nodes on the right side of the conditioning bar

Returns: a pair containing i) the vector of parameters and ii) the log-likelihood

Implemented in gum::learning::ParamEstimatorML.

◆ ranges()

const std::vector< std::pair< std::size_t, std::size_t > > & gum::learning::ParamEstimator::ranges ( ) const

returns the current ranges

Referenced by ParamEstimator(), and gum::learning::ParamEstimatorML::ParamEstimatorML().

Here is the caller graph for this function:

◆ setBayesNet()

template<typename GUM_SCALAR>

void gum::learning::ParamEstimator::setBayesNet ( const BayesNet< GUM_SCALAR > & new_bn )

assign a new Bayes net to all the counter's generators depending on a BN

Typically, generators based on EM or K-means depend on a model to compute correctly their outputs. Method setBayesNet enables to update their BN model.

◆ setMinNbRowsPerThread()

virtual void gum::learning::ParamEstimator::setMinNbRowsPerThread ( const std::size_t nb ) const

virtual

changes the number min of rows a thread should process in a multithreading context

When computing score, several threads are used by record counters to perform counts on the rows of the database, the MinNbRowsPerThread method indicates how many rows each thread should at least process. This is used to compute the number of threads actually run. This number is equal to the min between the max number of threads allowed and the number of records in the database divided by nb.

◆ setNumberOfThreads()

virtual void gum::learning::ParamEstimator::setNumberOfThreads ( Size nb )

virtual

sets the number max of threads that can be used

Parameters

nb	the number max of threads to be used. If this number is set to 0, then it is defaulted to aGrUM's max number of threads

Implements gum::IThreadNumberManager.

Referenced by gum::learning::IBNLearner::createParamEstimator_().

Here is the caller graph for this function:

◆ setParameters()

template<typename GUM_SCALAR>

double gum::learning::ParamEstimator::setParameters	(	const NodeId	target_node,
		const std::vector< NodeId > &	conditioning_nodes,
		Tensor< GUM_SCALAR > &	pot,
		const bool	compute_log_likelihood = false )

sets a CPT's parameters and, possibly, return its log-likelihhod

The tensor (CPT) is assumed to be a conditional probability, the first variable of its variablesSequence() being the target variable, the other ones being on the right side of the conditioning bar.

Parameters

target_node	the node on the left side of the CPT's conditioning bar
conditioning_nodes	the set of nodes on the right side of the conditioning bar
pot	the tensor (CPT) that is filled
compute_log_likelihood	a Boolean indicating whether we wish to compute the log-likelihood or not. Computing it is needed by the EM algorithm

Returns: a double which corresponds to the log-likelihood (w.r.t. the CPT) if compute_log_likelihood=true, else the method returns 0

Referenced by gum::learning::DAG2BNLearner::createBNwithEM().

Here is the caller graph for this function:

◆ setRanges()

void gum::learning::ParamEstimator::setRanges ( const std::vector< std::pair< std::size_t, std::size_t > > & new_ranges )

sets new ranges to perform the counts used by the parameter estimator

Parameters

ranges a set of pairs {(X1,Y1),...,(Xn,Yn)} of database's rows indices. The counts are then performed only on the union of the rows [Xi,Yi), i in {1,...,n}. This is useful, e.g, when performing cross validation tasks, in which part of the database should be ignored. An empty set of ranges is equivalent to an interval [X,Y) ranging over the whole database.

Referenced by gum::learning::IBNLearner::createParamEstimator_().

Here is the caller graph for this function:

Member Data Documentation

◆ counter_

RecordCounter gum::learning::ParamEstimator::counter_

protected

the record counter used to parse the database

Definition at line 273 of file paramEstimator.h.

◆ empty_nodevect_

const std::vector< NodeId > gum::learning::ParamEstimator::empty_nodevect_

protected

an empty vector of nodes, used for empty conditioning

Definition at line 276 of file paramEstimator.h.

◆ external_prior_

Prior* gum::learning::ParamEstimator::external_prior_ {nullptr}

protected

an external a priori

Definition at line 266 of file paramEstimator.h.

266{nullptr};

◆ score_internal_prior_

Prior* gum::learning::ParamEstimator::score_internal_prior_ {nullptr}

protected

if a score was used for learning the structure of the PGM, this is the priori internal to the score

Definition at line 270 of file paramEstimator.h.

270{nullptr};

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

agrum/BN/learning/paramUtils/paramEstimator.h

Public Member Functions

Protected Member Functions

Protected Attributes

Detailed Description

Constructor & Destructor Documentation

◆ ParamEstimator() [1/4]

◆ ParamEstimator() [2/4]

◆ ParamEstimator() [3/4]

◆ ParamEstimator() [4/4]

◆ ~ParamEstimator()

Member Function Documentation

◆ clear()

◆ clearRanges()

◆ clone()

◆ database()

◆ getNumberOfThreads()

◆ isGumNumberOfThreadsOverriden()

◆ minNbRowsPerThread()

◆ nodeId2Columns()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ parameters() [1/2]

◆ parameters() [2/2]

◆ parametersAndLogLikelihood() [1/2]

◆ parametersAndLogLikelihood() [2/2]

◆ ranges()

◆ setBayesNet()

◆ setMinNbRowsPerThread()

◆ setNumberOfThreads()

◆ setParameters()

◆ setRanges()

Member Data Documentation

◆ counter_

◆ empty_nodevect_

◆ external_prior_

◆ score_internal_prior_