Public Member Functions

HAPI::Domain Class Reference

A domain is the HUGIN representation of a network. More...

Inheritance diagram for HAPI::Domain:
HAPI::NetworkModel

List of all members.

Public Member Functions

void adapt ()
 Adapts this Domain according to the evidence entered.
void adaptClassTablesUsingFractionalUpdate ()
 Adapts (using the Fractional Update algorithm) the tables of the nodes in the class with the evidence propagated in this runtime Domain.
void adaptClassTablesUsingOnlineEM (double rho)
 Adapts (using the Online EM algorithm) the tables of the nodes in the class with the evidence propagated in this runtime Domain.
void addCases (DataSet *data_set)
 Add all rows of the DataSet to this Domain as cases.
void addCases (DataSet *data_set, size_t start, size_t count)
 Add the specified range of rows of the DataSet to this Domain as cases.
double approximate (double epsilon)
 Remove "near-zero" probabilities from the clique probability tables.
bool cgEvidenceIsPropagated () const
 Test if CG evidence has been propagated for this Domain.
Domainclone () const
 Clone a Domain object.
void compile ()
 Compile this Domain using the default triangulation method.
double compress ()
 Remove the zero entries from the clique and separator tables of the junction trees in this Domain.
void computeDBNPredictions (size_t numberOfTimePoints)
 Computes predictions for numberOfTimePoints time slices beyond the current time window.
void computeSensitivityData (const NodeList &nodes, const std::vector< size_t > &states)
 Compute the constants of the sensitivity functions for the specified output probabilities and all CPT parameters in the network.
 Domain (const std::string &filename)
 Construct a domain by loading the corresponding Hugin Knowledge Base (HKB) file.
 Domain (const std::string &netStringOrFileName, ParseListener *pl)
 Construct a domain from a NET file or a NET description given as a string.
 Domain (const Domain *domain)
 Construct a new domain by cloning an existing domain.
 Domain ()
 Construct a new, empty Domain object.
 Domain (const Class *cls)
 Construct a runtime domain from the given class.
 Domain (const std::string &filename, const std::string &password)
 Construct a domain by loading the corresponding password protected Hugin Knowledge Base (HKB) file.
 Domain (const Class *cls, size_t numberOfSlices)
 Construct a DBN runtime domain from a Class object.
void enterCase (size_t index)
 Enters a case as evidence.
bool equilibriumIs (Equilibrium eq) const
 Test for Equilibrium type.
bool evidenceIsPropagated () const
 Test if evidence has been propagated for this Domain.
bool evidenceModeIs (EvidenceMode ev) const
 Test for evidence mode.
void findMAPConfigurations (const NodeList &nodes, double minProbability)
 Find all configurations of nodes with probability at least minProbability.
void fineTuneNBTables (Node *target)
 Fine-tune a Naive Bayes (NB) model using training data.
double getAIC () const
 Computes the AIC score (Akaike's Information Criterion) of the case data.
double getApproximationConstant () const
 Return the approximation constant.
double getBIC () const
 Computes the BIC score (Bayesian Information Criterion) of the case data.
Number getCaseCount (size_t index) const
 Return the case count associated with case index in this domain.
size_t getConcurrencyLevel () const
 Get the current level of concurrency.
double getConflict () const
 Return the conflict value.
size_t getDBNWindowOffset () const
 Returns the total number of time steps that the time window of ` this DBN runtime domain has been moved.
NodeList getDConnectedNodes (const NodeList &source, const NodeList &hard, const NodeList &soft) const
 Performs a d-separation test and returns a list of d-connected nodes.
NodeList getDConnectedNodes (const NodeList &source, const NodeList &evidence) const
 Performs a d-separation test and returns a list of d-connected nodes.
NodeList getDSeparatedNodes (const NodeList &source, const NodeList &hard, const NodeList &soft) const
 Performs a d-separation test and returns a list of d-separated nodes.
NodeList getDSeparatedNodes (const NodeList &source, const NodeList &evidence) const
 Performs a d-separation test and returns a list of d-separated nodes.
NodeList getEliminationOrder () const
 Return the triangulation order.
size_t getEMConcurrencyLevel () const
 Return the number of threads to be created by the EM algorithm.
Number getExpectedUtility () const
 Return the total expected utility associated with this Domain.
NodeList getExplanation (size_t index) const
 Return the evidence subset associated with the explanation of rank index computed by the most recent call to DiscreteNode::computeExplanationData.
Number getExplanationScore (size_t index) const
 Return the score of the specified explanation.
size_t getGrainSize () const
 Return the current value of the grain size parameter.
JunctionTreeList getJunctionTrees () const
 Return the JunctionTrees of this Domain.
double getLogLikelihood () const
 Computes the log-likelihood of the case data.
double getLogLikelihoodTolerance () const
 Get current setting of the log-likelihood tolerance in this domain.
double getLogNormalizationConstant () const
 Get the logarithm to the normalization constant.
std::vector< size_t > getMAPConfiguration (size_t index) const
 Return a MAP configuration.
TablegetMarginal (const NodeList &nodes) const
 Compute the marginal distribution for the Nodes provided as argument with respect to the (imaginary) joint potential, determined by the current potentials on the junction tree (s) of this Domain.
size_t getMaxNumberOfEMIterations () const
 Retrieve the current maximum number of iterations for the EM algorithm.
size_t getMaxNumberOfSeparators () const
 Retrieve the current maximum number of separators allowed during a triangulation using the H_TM_TOTAL_WEIGHT triangulation method.
size_t getMaxSeparatorSize () const
 Retrieve the current maximum separator size allowed during a triangulation using the H_TM_TOTAL_WEIGHT triangulation method.
size_t getNBFineTuneIterationsLimit () const
 Return the limit for the number of consecutive failed iterations of the Naive Bayes (NB) fine-tuning algorithm.
double getNBFineTuneLearningRate () const
 Return the learning rate of the Naive Bayes (NB) fine-tuning algorithm.
double getNormalDeviate (double mean, double variance)
 Uses the pseudo-random number generator for this Domain to sample a real number from a normal (aka Gaussian) distribution.
double getNormalizationConstant () const
 Retrieve the normalization constant from the most recent propagation.
size_t getNumberOfCases () const
 Return the number of cases currently allocated for this domain.
size_t getNumberOfExplanations () const
 Return the number of explanations.
size_t getNumberOfMAPConfigurations () const
 Return the number of MAP configurations.
double getProbabilityOfMAPConfiguration (size_t index) const
 Return the probability of a MAP configuration.
NodeList getSensitivitySet () const
 Return the sensitivity set computed by the most recent call to DiscreteChanceNode::computeSensitivityData.
NodeList getSensitivitySet (size_t output) const
 Return the sensitivity set computed by the most recent call to Domain::computeSensitivityData.
double getSignificanceLevel () const
 Get current setting of the significance level in this domain.
double getUniformDeviate ()
 Uses the pseudo-random number generator for this Domain to sample a real number from the uniform distribution over the interval [0,1).
bool hasEvidenceToPropagate () const
 Test if evidence has been entered since last propagation.
bool hasTablesToPropagate () const
 Test for new node tables.
void initialize ()
 Establish the initial values for all tables of this Domain (which must be compiled).
void initializeDBNWindow ()
 Moves the time window of this DBN back to its initial position, and removes all evidence.
bool isCompiled () const
 Test whether this Domain is compiled.
bool isCompressed () const
 Test whether this Domain is compressed.
bool isTriangulated () const
 Test whether this Domain is triangulated.
bool isTriangulatedForBK () const
 Test whether this DBN runtime Domain is triangulated for Boyen-Koller approximate inference.
void learnClassTables ()
 Perform EM learning on an OOBN.
void learnHNBStructure (Node *target)
 Learn a Hierarchical Naive Bayes (HNB) model from data.
void learnStructure ()
 Learn the structure (graph) of the Bayesian network from data using the PC algorithm.
void learnTables ()
 Learn the conditional probability tables for each node in this domain that has an experience table.
void learnTreeStructure ()
 Learn a tree-structured network model from data.
void learnTreeStructure (Node *root, Node *target)
 Learn a tree-structured network model from data.
void learnTreeStructure (Node *root)
 Learn a tree-structured network model from data.
bool likelihoodIsPropagated () const
 Test if likelihood eveidence has been propagated for this Domain.
void moveDBNWindow (size_t delta)
 Slides the time window of this DBN delta steps into the future.
size_t newCase ()
 Allocate storage within this domain to a new case.
void parseCase (const std::string &filename, ParseListener *pl)
 Parses the case stored in a file with the given filename and enters the associated findings into this Domain.
void parseCases (const std::string &filename, ParseListener *pl)
 Parses a set of cases, stored in a file with the given filename.
void propagate (Equilibrium eq=H_EQUILIBRIUM_SUM, EvidenceMode ev=H_MODE_NORMAL)
 Establish the specified equilibrium using the evidence mode indicated for incorporation of evidence on all junction trees in this Domain.
void resetInferenceEngine ()
 Establish the initial state of the inference engine.
void retractFindings ()
 Retract (all) findings for all nodes in this Domain.
void saveAsKB (const std::string &filename)
 Save this Domain as a Hugin Knowledge Base (HKB) file.
void saveAsKB (const std::string &filename, const std::string &password)
 Save this Domain as a password protected Hugin Knowledge Base (HKB) file.
void saveCase (const std::string &filename)
 Saves all evidence entered in this Domain in a file with the given file name (if the file exists, it is overwritten).
void saveCases (const std::string &filename, const NodeList &nodes, bool caseCounts, const std::string &separator, const std::string &missingData)
 Saves all cases entered in this Domain in a Hugin data file with the given file name.
void saveCases (const std::string &filename, const NodeList &nodes, const std::vector< size_t > &cases, bool caseCounts, const std::string &separator, const std::string &missingData)
 Save the specified cases entered in this Domain in a Hugin data file with the given file name.
void saveToMemory ()
 Create a copy in memory of the belief and junction tree tables of this Domain.
void seedRandom (unsigned int seed)
 Seeds the random number generator.
void setCaseCount (size_t index, Number count)
 Set the case count associated with case index in this domain to count.
void setConcurrencyLevel (size_t level)
 Set the level of concurrency.
void setEMConcurrencyLevel (size_t level)
 Set the number of threads to be created by the EM algorithm.
void setGrainSize (size_t size)
 Set the grains size parameter.
void setInitialTriangulation (const NodeList &order)
 Specify an initial triangulation for this Domain.
void setLogLikelihoodTolerance (double tolerance)
 Specify the tolerance of the log-likelihood.
void setMaxNumberOfEMIterations (size_t iterations)
 Set the maximal number of iterations allowed for the EM algorithm.
void setMaxNumberOfSeparators (size_t separators)
 Set the maximum number of separators allowed during triangulation.
void setMaxSeparatorSize (size_t size)
 Set the maximum separator size for the H_TM_TOTAL_WEIGHT TriangulationMethod.
void setNBFineTuneIterationsLimit (size_t limit)
 Set the limit for the number of consecutive failed iterations of the Naive Bayes (NB) fine-tuning algorithm.
void setNBFineTuneLearningRate (double learningRate)
 Set the learning rate of the Naive Bayes (NB) fine-tuning algorithm.
void setNumberOfCases (size_t number)
 Adjust the storage capacity for cases in this domain.
void setSignificanceLevel (double significancelevel)
 Specify the Significance Level used for the structure learning aglorithm.
void simulate ()
 Sample a configuration for this Domain with respect to the current distribution.
void triangulate (const NodeList &order)
 Triangulate the graph of this Domain using the specified elimination order.
void triangulate (TriangulationMethod tm=H_TM_BEST_GREEDY)
 Triangulate the graph of this Domain using the specified triangulation method.
void triangulateDBN (TriangulationMethod tm=H_TM_BEST_GREEDY)
 Triangulate a DBN runtime Domain for exact inference.
void triangulateDBNForBK (TriangulationMethod tm=H_TM_BEST_GREEDY)
 Triangulate a DBN runtime Domain for approximate inference.
void uncompile ()
 Uncompiles this Domain.
void updatePolicies ()
 Update the policy tables of this domain.
 ~Domain () throw ()
 Destruct a Domain object.

Detailed Description

A domain is the HUGIN representation of a network.

It is one of the principal structures in HUGIN. It must be constructed before any nodes belonging to the network.

See also:
Node
JunctionTree

Constructor & Destructor Documentation

HAPI::Domain::Domain ( const std::string &  filename ) [explicit]

Construct a domain by loading the corresponding Hugin Knowledge Base (HKB) file.

The HKB file must contain a domain.

Parameters:
filenamethe name of the HKB file
HAPI::Domain::Domain ( const std::string &  filename,
const std::string &  password 
)

Construct a domain by loading the corresponding password protected Hugin Knowledge Base (HKB) file.

The HKB file must contain a domain. If the given password does not match the password stored in the HKB file, an exception is thrown.

Parameters:
filenamethe name of the HKB file
passwordthe password for the file
HAPI::Domain::Domain ( const std::string &  netStringOrFileName,
ParseListener pl 
)

Construct a domain from a NET file or a NET description given as a string.

Parameters:
netStringOrFileNamea string containing a NET description or the name of a NET file
plpointer to object derived from class ParseListener.
HAPI::Domain::Domain ( const Domain domain )

Construct a new domain by cloning an existing domain.

Parameters:
domainThe domain to be cloned.
HAPI::Domain::Domain ( const Class cls )

Construct a runtime domain from the given class.

Parameters:
clsThe Class from which to create the runtime domain.
HAPI::Domain::Domain ( const Class cls,
size_t  numberOfSlices 
)

Construct a DBN runtime domain from a Class object.

The domain is formed by linking (through temporal clones) the specified number of instances (called time slices) of the class.

Parameters:
clsthe Class object describing a single time slice
numberOfSlicesthe number of time slices

Member Function Documentation

void HAPI::Domain::adaptClassTablesUsingFractionalUpdate (  )

Adapts (using the Fractional Update algorithm) the tables of the nodes in the class with the evidence propagated in this runtime Domain.

This method must be called on a runtime domain. The evidence propagated in this domain is used to adapt the tables of the class from which this domain was created. The updated tables are copied back to the nodes of the runtime domain.

void HAPI::Domain::adaptClassTablesUsingOnlineEM ( double  rho )

Adapts (using the Online EM algorithm) the tables of the nodes in the class with the evidence propagated in this runtime Domain.

This method must be called on a runtime domain. The evidence propagated in this domain is used to adapt the tables of the class from which this domain was created. The updated tables are copied back to the nodes of the runtime domain.

Parameters:
rhothe parameter used to determine the "learning rate" of the Online EM algorithm.
void HAPI::Domain::addCases ( DataSet data_set,
size_t  start,
size_t  count 
)

Add the specified range of rows of the DataSet to this Domain as cases.

Parameters:
startthe index of the first row to add
countthe number of rows to add.
double HAPI::Domain::approximate ( double  epsilon )

Remove "near-zero" probabilities from the clique probability tables.

For each Clique object in this domain, a value delta is computed such that the sum of all elements less than delta in the (discrete part) of the clique table is less than epsilon. These elements (less than delta) are then set to 0.

Parameters:
epsilonThe threshold value. Maximal probability mass to eradicate from each clique.
Returns:
A double value which is the sum of all entries in clique probability tables that have been zeroed.
bool HAPI::Domain::cgEvidenceIsPropagated (  ) const

Test if CG evidence has been propagated for this Domain.

Returns:
boolean
Domain* HAPI::Domain::clone (  ) const

Clone a Domain object.

This function returns a pointer to a copy of this domain.

void HAPI::Domain::compile (  )

Compile this Domain using the default triangulation method.

If the domain is already triangulated, nothing is changed. The domain must contain at least one discrete or continuous node.

double HAPI::Domain::compress (  )

Remove the zero entries from the clique and separator tables of the junction trees in this Domain.

Compression can only be applied to (compiled) ordinary belief networks. Continuous nodes are allowed, but compression only applies to configurations of states of the discrete nodes.

Returns:
A double value which indicates a measure of compression achieved. The measure should be less than 1, indicating that the compressed domain requires less space than the uncompressed domain. An output greater than 1 means that the "compressed" domain requires more space than the uncompressed domain.
void HAPI::Domain::computeDBNPredictions ( size_t  numberOfTimePoints )

Computes predictions for numberOfTimePoints time slices beyond the current time window.

This Domain must have been produced by createDBNDomain, and it must have been triangulated using triangulateDBN. The predictions are accessed using getPredictedBelief, getPredictedMean, getPredictedVariance, and getPredictedValue.

Parameters:
numberOfTimePointsthe number of time slices to compute predictions for (this must be a positive number).
void HAPI::Domain::computeSensitivityData ( const NodeList nodes,
const std::vector< size_t > &  states 
)

Compute the constants of the sensitivity functions for the specified output probabilities and all CPT parameters in the network.

The output probabilities are specified using a list of nodes and a list of corresponding states.

Parameters:
nodesthe list of (output) nodes
statesa list of states of the nodes in the nodes list
void HAPI::Domain::enterCase ( size_t  index )

Enters a case as evidence.

A subsequent propagate operation will perform inference using the case.

Parameters:
indexthe index of the case to enter.
bool HAPI::Domain::equilibriumIs ( Equilibrium  eq ) const

Test for Equilibrium type.

If the equilibrium of all junction trees of this Domain is eq, return true.

Parameters:
eqType of Equilibrium to test for.
bool HAPI::Domain::evidenceModeIs ( EvidenceMode  ev ) const

Test for evidence mode.

Test if the equilibrium of all junction trees of this Domain could have been obtained through a propagation using ev as the evidence incorporation mode.

Parameters:
evType of EvidenceMode to test for.
void HAPI::Domain::findMAPConfigurations ( const NodeList nodes,
double  minProbability 
)

Find all configurations of nodes with probability at least minProbability.

This method uses a Monte Carlo algorithm to solve a generalized form of the maximum a posteriori (MAP) configuration problem: The MAP configuration problem is the problem of finding the most probable configuration of a set of nodes given evidence on some of the remaining nodes.

The results of this method are provided by Domain::getNumberOfMAPConfigurations, Domain::getMAPConfiguration, and Domain::getProbabilityOfMAPConfiguration.

Parameters:
nodesa NodeList containing the DiscreteNodes for which to find configurations.
minProbabilityconfigurations with a lower probability than minProbability are ignored.
void HAPI::Domain::fineTuneNBTables ( Node target )

Fine-tune a Naive Bayes (NB) model using training data.

The method makes small adjustments to the conditional probabilities of the attribute nodes in order to obtain more correct classifications. If these adjustments do not result in a better model, the original model is preserved.

The domain must represent a Naive Bayes model with initial conditional probability tables (for example, estimated using the EM algorithm). The network must contain only discrete chance nodes. Case data must be specified in advance.

Parameters:
targetthe class variable of the Naive Bayes model.
double HAPI::Domain::getApproximationConstant (  ) const

Return the approximation constant.

The number returned is based on the most recent (explicit or implicit) approximation operation. An implicit approximation takes place when you change some conditional probability tables of acompressed domain, and then perform a propagation operation. Since some (discrete) state configurations have been removed from a compressed domain, the probability mass of the remaining configurations will typically be less than 1. This probability mass is returned by getApproximationConstant ().

Returns:
A double expressing the probability mass remaining in the approximated domain.
size_t HAPI::Domain::getConcurrencyLevel (  ) const

Get the current level of concurrency.

See also:
setConcurrencyLevel
double HAPI::Domain::getConflict (  ) const

Return the conflict value.

The conflict value is valid for this Domain computed during the most recent propagation. If no propagation has been performed, 1 is returned.

Returns:
A double-precision real value expressing the conflict measure in the domain.
NodeList HAPI::Domain::getDConnectedNodes ( const NodeList source,
const NodeList hard,
const NodeList soft 
) const

Performs a d-separation test and returns a list of d-connected nodes.

Assuming evidence on the specified evidence nodes, this method returns the list of nodes that are d-connected to the specified list of source nodes.

Returns:
the list of d-connected nodes
Parameters:
sourcelist of source nodes
hardlist of nodes assumed to be instantiated
softlist of nodes assumed to have multi-state or likelihood evidence.
NodeList HAPI::Domain::getDConnectedNodes ( const NodeList source,
const NodeList evidence 
) const

Performs a d-separation test and returns a list of d-connected nodes.

Assuming evidence on the specified evidence nodes, this method returns the list of nodes that are d-connected to the specified list of source nodes.

Returns:
the list of d-connected nodes
Parameters:
sourcelist of source nodes
evidencelist of nodes assumed to be instantiated.
NodeList HAPI::Domain::getDSeparatedNodes ( const NodeList source,
const NodeList hard,
const NodeList soft 
) const

Performs a d-separation test and returns a list of d-separated nodes.

Assuming evidence on the specified evidence nodes, this method returns the list of nodes that are d-separated to the specified list of source nodes.

Returns:
the list of d-separated nodes
Parameters:
sourcelist of source nodes
hardlist of nodes assumed to be instantiated
softlist of nodes assumed to have multi-state or likelihood evidence.
NodeList HAPI::Domain::getDSeparatedNodes ( const NodeList source,
const NodeList evidence 
) const

Performs a d-separation test and returns a list of d-separated nodes.

Assuming evidence on the specified evidence nodes, this method returns the list of nodes that are d-separated to the specified list of source nodes.

Returns:
the list of d-separated nodes
Parameters:
sourcelist of source nodes
evidencelist of nodes assumed to be instantiated.
NodeList HAPI::Domain::getEliminationOrder (  ) const

Return the triangulation order.

A NodeList containing the list of nodes in the order used to triangulate the network of this Domain is returned.

Returns:
NodeList containing Nodes representing the elimination order used.
size_t HAPI::Domain::getEMConcurrencyLevel (  ) const

Return the number of threads to be created by the EM algorithm.

See also:
setEMConcurrencyLevel
NodeList HAPI::Domain::getExplanation ( size_t  index ) const

Return the evidence subset associated with the explanation of rank index computed by the most recent call to DiscreteNode::computeExplanationData.

Parameters:
indexspecifies that the indexth best explanation should be retrieved (the best explanation has index 0, the second best has index 1, etc.)
Number HAPI::Domain::getExplanationScore ( size_t  index ) const

Return the score of the specified explanation.

This method returns the score associated with the explanation subset returned by Domain::getExplanation(index).

Parameters:
indexidentifies the explanation.
size_t HAPI::Domain::getGrainSize (  ) const

Return the current value of the grain size parameter.

Returns:
Positive integer.
JunctionTreeList HAPI::Domain::getJunctionTrees (  ) const

Return the JunctionTrees of this Domain.

Returns:
JunctionTreeList.
double HAPI::Domain::getLogLikelihood (  ) const

Computes the log-likelihood of the case data.

std::vector<size_t> HAPI::Domain::getMAPConfiguration ( size_t  index ) const

Return a MAP configuration.

This method returns the configuration identified by index among the configurations with probability at least minProbability — as specified in the most recent successful call to Domain::findMAPConfigurations.

The index argument must be a nonnegative integer less than Domain::getNumberOfMAPConfigurations: 0 requests the most probable configuration, 1 the second-most probable configuration, etc.

Parameters:
indexidentifies the configuration.
Returns:
a vector of state indexes forming the configuration.
Table* HAPI::Domain::getMarginal ( const NodeList nodes ) const

Compute the marginal distribution for the Nodes provided as argument with respect to the (imaginary) joint potential, determined by the current potentials on the junction tree (s) of this Domain.

If nodes contains continuous nodes, they must be last in the list. This operation is not allowed on compressed domains.

Parameters:
nodesNodeList containing the Node objects over which to compute the marginal.
Returns:
A Table which contains the marginal distribution over the nodes provided.
double HAPI::Domain::getNBFineTuneLearningRate (  ) const

Return the learning rate of the Naive Bayes (NB) fine-tuning algorithm.

The default value is 0.01.

double HAPI::Domain::getNormalDeviate ( double  mean,
double  variance 
)

Uses the pseudo-random number generator for this Domain to sample a real number from a normal (aka Gaussian) distribution.

Parameters:
meanthe mean of the distribution
variancethe variance of the distribution
double HAPI::Domain::getNormalizationConstant (  ) const

Retrieve the normalization constant from the most recent propagation.

For sum-propagation, the normalization constant is equal to the probability of the evidence propagated. For max-propagation, the normalization constant is the probability of the most probable configuration with the evidence incorporated.

size_t HAPI::Domain::getNumberOfExplanations (  ) const

Return the number of explanations.

This method returns the number of subsets found by the most recent successful call to DiscreteNode::computeExplanationData.

size_t HAPI::Domain::getNumberOfMAPConfigurations (  ) const

Return the number of MAP configurations.

This method returns the number of configurations found by the most recent successful call to Domain::findMAPConfigurations.

double HAPI::Domain::getProbabilityOfMAPConfiguration ( size_t  index ) const

Return the probability of a MAP configuration.

This method returns the probability of the configuration returned by Domain::getMAPConfiguration (index).

Parameters:
indexidentifies the configuration.
Returns:
the probability of the specified configuration.
NodeList HAPI::Domain::getSensitivitySet (  ) const

Return the sensitivity set computed by the most recent call to DiscreteChanceNode::computeSensitivityData.

If the results produced by that call have been invalidated, a usage exception is thrown.

NodeList HAPI::Domain::getSensitivitySet ( size_t  output ) const

Return the sensitivity set computed by the most recent call to Domain::computeSensitivityData.

If the results produced by that call have been invalidated, a usage exception is thrown.

Parameters:
outputidentifies one of the output probabilities specified in the call to Domain::computeSensitivityData.
bool HAPI::Domain::hasTablesToPropagate (  ) const

Test for new node tables.

Are there any nodes in this Domain having (a conditional probability or utility) table that has changed since the most recent compilation or propagation.

void HAPI::Domain::initialize (  )

Establish the initial values for all tables of this Domain (which must be compiled).

Using this method will erase all evidence previously entered.

void HAPI::Domain::initializeDBNWindow (  )

Moves the time window of this DBN back to its initial position, and removes all evidence.

This Domain must have been produced by createDBNDomain, and it must have been triangulated using triangulateDBN.

bool HAPI::Domain::isTriangulated (  ) const

Test whether this Domain is triangulated.

Being "triangulated" means that the junction forest has been created, but not the associated tables.

void HAPI::Domain::learnClassTables (  )

Perform EM learning on an OOBN.

This requires, that the data matches the domain created from the OOBN, and not the OOBN itself.

void HAPI::Domain::learnHNBStructure ( Node target )

Learn a Hierarchical Naive Bayes (HNB) model from data.

The domain must contain only chance nodes (but continuous nodes are ignored by the HNB algorithm) and no edges. The target variable must be a discrete node. Case data must be specified in advance.

Parameters:
targetthe class variable of the HNB model.
void HAPI::Domain::learnStructure (  )

Learn the structure (graph) of the Bayesian network from data using the PC algorithm.

The domain must contain only chance nodes and no edges. Case data must be specified in advance.

void HAPI::Domain::learnTreeStructure ( Node root )

Learn a tree-structured network model from data.

This method uses the Chow-Liu algorithm for learning a tree model with all edges directed away from root.

The domain must contain only chance nodes and no edges. Case data must be specified in advance.

Parameters:
rootall edges of the tree will be directed away from root
void HAPI::Domain::learnTreeStructure ( Node root,
Node target 
)

Learn a tree-structured network model from data.

This method constructs a Tree-Augmented Naive (TAN) Bayes model for classification. First, a Chow-Liu tree model with root as root is learned. This tree is turned into a TAN model by adding target as parent of all other nodes of this Domain.

The domain must contain only chance nodes and no edges. Case data must be specified in advance.

Parameters:
rootall edges of the Chow-Liu tree will be directed away from root
targetthe class variable of the TAN model.
void HAPI::Domain::learnTreeStructure (  )

Learn a tree-structured network model from data.

This method uses the Rebane-Pearl algorithm for learning a polytree model.

The domain must contain only discrete chance nodes and no edges. Case data must be specified in advance.

void HAPI::Domain::moveDBNWindow ( size_t  delta )

Slides the time window of this DBN delta steps into the future.

This Domain must have been produced by createDBNDomain, and it must have been triangulated using triangulateDBN.

Parameters:
deltathe number of time steps to slide the time window (this must be a positive number).
size_t HAPI::Domain::newCase (  )

Allocate storage within this domain to a new case.

Returns:
New case index.
void HAPI::Domain::parseCase ( const std::string &  filename,
ParseListener pl 
)

Parses the case stored in a file with the given filename and enters the associated findings into this Domain.

All existing evidence in the Domain is retracted before entering the case findings.

Parameters:
filenamethe name of the file containing the case.
plthe ParseListener used for handling parse errors.
void HAPI::Domain::parseCases ( const std::string &  filename,
ParseListener pl 
)

Parses a set of cases, stored in a file with the given filename.

The found cases are entered into the Domain.

Parameters:
filenamethe name of the file containing the case.
plthe ParseListener used for handling parse errors.
void HAPI::Domain::propagate ( Equilibrium  eq = H_EQUILIBRIUM_SUM,
EvidenceMode  ev = H_MODE_NORMAL 
)

Establish the specified equilibrium using the evidence mode indicated for incorporation of evidence on all junction trees in this Domain.

Also, revised beliefs will be computed for all nodes.

Parameters:
eqEquilibrium type. Defaults to H_EQUILIBRIUM_SUM.
evEvidenceMode type. Defaults to H_MODE_NORMAL.
void HAPI::Domain::resetInferenceEngine (  )

Establish the initial state of the inference engine.

: sum-equilibrium with no evidence incorporated. Any propagated findings will thus be removed from the junction tree potentials, but entered findings will still be "registred" (i.e., they will be incorporated in the next propagation).

void HAPI::Domain::saveAsKB ( const std::string &  filename )

Save this Domain as a Hugin Knowledge Base (HKB) file.

Parameters:
filenamename of the HKB file
void HAPI::Domain::saveAsKB ( const std::string &  filename,
const std::string &  password 
)

Save this Domain as a password protected Hugin Knowledge Base (HKB) file.

Parameters:
filenamename of the HKB file
passwordthe password for the HKB file.
void HAPI::Domain::saveCase ( const std::string &  filename )

Saves all evidence entered in this Domain in a file with the given file name (if the file exists, it is overwritten).

Parameters:
filenamethe name of the file in which the case is going to be saved.
void HAPI::Domain::saveCases ( const std::string &  filename,
const NodeList nodes,
bool  caseCounts,
const std::string &  separator,
const std::string &  missingData 
)

Saves all cases entered in this Domain in a Hugin data file with the given file name.

Parameters:
filenameThe name of the file in which the cases will be saved (if the file exists, it is overwritten).
nodesA list of the nodes which are to be included in the file.
caseCountsIf true, include case counts in the data file. If false, case counts will not be included.
separatorThe string used to separate the items in the file.
missingDataThe string used to represent missing data.
void HAPI::Domain::saveCases ( const std::string &  filename,
const NodeList nodes,
const std::vector< size_t > &  cases,
bool  caseCounts,
const std::string &  separator,
const std::string &  missingData 
)

Save the specified cases entered in this Domain in a Hugin data file with the given file name.

Parameters:
filenameThe name of the file in which the cases will be saved (if the file exists, it is overwritten).
nodesA list of the nodes which are to be included in the file.
casesA list of indexes of cases to be included in the file.
caseCountsIf true, include case counts in the data file. If false, case counts will not be included.
separatorThe string used to separate the items in the file.
missingDataThe string used to represent missing data.
void HAPI::Domain::saveToMemory (  )

Create a copy in memory of the belief and junction tree tables of this Domain.

This operation can only be performed if the domain is compiled, the current equilibrium is "sum", and no evidence has been incorporated.

void HAPI::Domain::seedRandom ( unsigned int  seed )

Seeds the random number generator.

The random number generator, used by HUGIN, generates a sequence of numbers which appears random, but are in fact deterministic. However, "seeding" the generator will change the starting point within the sequence.

Parameters:
seeddetermines the starting point of the random number generator.
void HAPI::Domain::setConcurrencyLevel ( size_t  level )

Set the level of concurrency.

The level of concurrency specifies the maximum number of threads to create when performing a specific table operation. Setting the level of concurrency to 1 will cause all table operations to be performed sequentially. The initial parameter value is 1.

void HAPI::Domain::setEMConcurrencyLevel ( size_t  level )

Set the number of threads to be created by the EM algorithm.

The EM algorithm exploits concurrency by partitioning the set of cases into subsets of approximately equal size. Each subset is then processed by a thread that has a copy of the (compiled) domain, enabling the threads to perform inference in parallel.

Setting this parameter to 1 causes the EM algorithm to execute single-threaded. The initial value of this parameter is 1.

Parameters:
levelthe level of concurrency (must be a positive number).
void HAPI::Domain::setGrainSize ( size_t  size )

Set the grains size parameter.

The grain size parameter specifies a lower limit of the tasks to be performed by each thread. The size of a task is approximately equal to the number of floating-point operations needed to perform the task (e.g., the number of elements to sum when performing a marginalization task).

The initial value of the grain size parameter is 10000.

void HAPI::Domain::setInitialTriangulation ( const NodeList order )

Specify an initial triangulation for this Domain.

This triangulation will be used by the H_TM_TOTAL_WEIGHT triangulation method. The purpose is to (1) improve the generated triangulation, and (2) to reduce the run-time of the algorithm. The triangulation must be specified in the form of an elimination sequence.

Parameters:
ordera NodeList containing the Nodes of the network in the order of elimination
void HAPI::Domain::setLogLikelihoodTolerance ( double  tolerance )

Specify the tolerance of the log-likelihood.

Terminate the EM learning when the relative difference between the log-likelihood of two successive iterations becomes less than tolerance.

void HAPI::Domain::setMaxNumberOfEMIterations ( size_t  iterations )

Set the maximal number of iterations allowed for the EM algorithm.

The algorithm termnates when this number is reached or when the relative improvement becomes lower than the log-likelihood tolerance.

Parameters:
iterationsMaximum allowed number of iterations.
void HAPI::Domain::setMaxNumberOfSeparators ( size_t  separators )

Set the maximum number of separators allowed during triangulation.

Parameters:
separatorsMaximum number of separators allowed during triangulation.
void HAPI::Domain::setMaxSeparatorSize ( size_t  size )

Set the maximum separator size for the H_TM_TOTAL_WEIGHT TriangulationMethod.

If a positive maximum separator size is specified, then separators larger than the specified size will be discarded by the separator generation algorithm. However, separators implied by the initial triangulation will be retained (regardless of size). This feature can be used to triangulate much larger graphs (without splitting them into smaller subgraphs). But notice that not all separators up to the specified limit are necessarily generated.

An initial triangulation is required in order to use this feature, see setInitialTriangulation.

If zero is specified for the maximum separator size, then no separators will be discarded.

void HAPI::Domain::setNBFineTuneIterationsLimit ( size_t  limit )

Set the limit for the number of consecutive failed iterations of the Naive Bayes (NB) fine-tuning algorithm.

When limit iterations over the training data, showing no improvements of the classification accuracy, have been performed, the algorithm terminates.

The default value is 1.

Parameters:
limita positive integer.
void HAPI::Domain::setNBFineTuneLearningRate ( double  learningRate )

Set the learning rate of the Naive Bayes (NB) fine-tuning algorithm.

The default value is 0.01.

Parameters:
learningRatea number between 0 and 1.
void HAPI::Domain::setNumberOfCases ( size_t  number )

Adjust the storage capacity for cases in this domain.

Parameters:
numberStorage capacity.
void HAPI::Domain::simulate (  )

Sample a configuration for this Domain with respect to the current distribution.

The current distribution must be in sum-equilibrium and with evidence incorporated in normal mode.

void HAPI::Domain::triangulate ( const NodeList order )

Triangulate the graph of this Domain using the specified elimination order.

The elimination order must contain each discrete and each continuous node of this Domain exactly once, and continuous nodes must appear before discrete nodes.

Parameters:
ordera NodeList containing the Nodes of the network in the order of elimination
void HAPI::Domain::triangulate ( TriangulationMethod  tm = H_TM_BEST_GREEDY )

Triangulate the graph of this Domain using the specified triangulation method.

Parameters:
tmthe TriangulationMethod to use.
void HAPI::Domain::triangulateDBN ( TriangulationMethod  tm = H_TM_BEST_GREEDY )

Triangulate a DBN runtime Domain for exact inference.

Parameters:
tmthe TriangulationMethod to use.
void HAPI::Domain::triangulateDBNForBK ( TriangulationMethod  tm = H_TM_BEST_GREEDY )

Triangulate a DBN runtime Domain for approximate inference.

Boyen-Koller approximate inference is applied to the interfaces between the time slices of the time window.

Parameters:
tmthe TriangulationMethod to use.
void HAPI::Domain::uncompile (  )

Uncompiles this Domain.

The data structures of this Domain produced by a triangulation or a compilation are deleted. Note that pointers to objects within the compiled structure (e.g., Cliques and JunctionTrees) are invalidated. Also note that many of the editing functions automatically performs an uncompile() operation. When this happens, the domain must be compiled again before it can be used for inference.

void HAPI::Domain::updatePolicies (  )

Update the policy tables of this domain.

The policies of all unmade decisions are updated. The new policies maximize the overall expected utility.


Copyright Hugin Expert A/S 1993-2024