opendial.inference

Class SwitchingAlgorithm

java.lang.Object

opendial.inference.SwitchingAlgorithm

All Implemented Interfaces:

InferenceAlgorithm

public class SwitchingAlgorithm
extends java.lang.Object
implements InferenceAlgorithm

Switching algorithms that alternates between an exact algorithm (variable elimination) and an approximate algorithm (likelihood weighting) depending on the query.

The switching mechanism is defined via two thresholds:

• one threshold on the maximum branching factor of the network
• one threshold on the maximum number of combination of values in a node factor

If one of these threshold is exceeded or if the Bayesian network contains a continuous distribution, the selected algorithm will be likelihood weighting. Variable elimination is selected in the remaining cases.

Field Summary

Fields

Modifier and Type	Field and Description
static int	MAX_BRANCHING_FACTOR
static int	MAX_NBVALUES

Constructor Summary

Constructors

Constructor and Description
SwitchingAlgorithm()

Method Summary

Modifier and Type	Method and Description
MultivariateDistribution	queryProb(Query.ProbQuery query) Selects the best algorithm for performing the inference on the provided probability query and return its result.
UtilityTable	queryUtil(Query.UtilQuery query) Selects the best algorithm for performing the inference on the provided utility query and return its result.
BNetwork	reduce(Query.ReduceQuery query) Reduces a Bayesian network to a subset of variables.
InferenceAlgorithm	selectBestAlgorithm(Query query)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface opendial.inference.InferenceAlgorithm

queryProb, queryProb, queryProb, queryProb, queryUtil, queryUtil, queryUtil, queryUtil, reduce, reduce

Field Detail

MAX_BRANCHING_FACTOR

public static int MAX_BRANCHING_FACTOR

MAX_NBVALUES

public static int MAX_NBVALUES

Constructor Detail

SwitchingAlgorithm

public SwitchingAlgorithm()

Method Detail

queryProb

public MultivariateDistribution queryProb(Query.ProbQuery query)

Selects the best algorithm for performing the inference on the provided probability query and return its result.

Specified by:

queryProb in interface InferenceAlgorithm

Parameters:

query - the probability query

Returns:

the inference result

queryUtil

public UtilityTable queryUtil(Query.UtilQuery query)

Selects the best algorithm for performing the inference on the provided utility query and return its result.

Specified by:

queryUtil in interface InferenceAlgorithm

Parameters:

query - the utility query

Returns:

the inference result

reduce

public BNetwork reduce(Query.ReduceQuery query)

Reduces a Bayesian network to a subset of variables. The method is divided in three steps:

• The method first checks whether inference is necessary at all or whether the current network can be returned as it is.
• If inference is necessary, the algorithm divides the network into cliques and performs inference on each clique separately.
• Finally, if only one clique is present, the reduction selects the best algorithm and return the result of the reduction process.

Specified by:

reduce in interface InferenceAlgorithm

Parameters:

query - the reduction query

Returns:

the reduced network

selectBestAlgorithm

public InferenceAlgorithm selectBestAlgorithm(Query query)