Package edu.cmu.tetrad.bayes

Class BdeMetricCache

java.lang.Object

edu.cmu.tetrad.bayes.BdeMetricCache

public final class BdeMetricCache
extends Object

Provides a method for computing the score of a model, called the BDe metric (Bayesian Dirchlet likelihood equivalence), given a dataset (assumes no missing values) and a Bayes parameterized network (assumes no latent variables).> 0

This version has a method that computes the score for a given factor of a model, where a factor is determined by a node and its parents. It stores scores in a map whose argument is an ordered pair consisting of 1) a node and 2) set of parents. The score for the entire model is the product of the scores of its factors. Since the log of the gamma function is used here the sum of the logs is computed as the score. Compare this with the score method in the BdeMetric class which computes the score for the entire model in one pass. The advantage of the approach in this class is that it is more efficient in the context of a search algorithm where different models are scored but where many of them will have the same factors. This class stores the score (relative to the dataset) for any [node, set of parents] pair and thus avoids the expensive log gamma function calls. Instead, it looks in the map scores to see if it has already computed the score and, if so, returns the previously computed value.> 0

See "Learning Bayesian Networks: The Combination of Knowledge and Statistical Data" by David Heckerman, Dan Geiger, and David M. Chickering. Microsoft Technical Report MSR-TR-94-09.> 0

Version:

$Id: $Id

Author:

Frank Wimberly

Constructor Summary

Constructors

Constructor	Description
BdeMetricCache(DataSet dataSet, BayesPm bayesPm)	Constructs a BdeMetricCache object for a given dataset and BayesPm.

Method Summary

Modifier and Type	Method	Description
double[][]	getObservedCounts(Node node, BayesPm bayesPm, BayesIm bayesIm)	This method is used in testing and debugging and not in the BDe metric calculations.
int	getScoreCount(Node node, Set<Node> parents)	This is just for testing the operation of the inner class and the map from nodes and parent sets to scores.
double	scoreLnGam(Node node, Set<Node> parents, BayesPm bayesPmMod, BayesIm bayesIm)	Computes the BDe score, using the logarithm of the gamma function, relative to the data, of the factor determined by a node and its parents.

Methods inherited from class java.lang.Object

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

Constructor Details

BdeMetricCache

public BdeMetricCache(DataSet dataSet,
 BayesPm bayesPm)

Constructs a BdeMetricCache object for a given dataset and BayesPm.

Parameters:

dataSet - The dataset for which the BDe metric is to be computed.

bayesPm - The BayesPm that determines the list of variables (nodes) and the structure of the graph.

Method Details

scoreLnGam

public double scoreLnGam(Node node,
 Set<Node> parents,
 BayesPm bayesPmMod,
 BayesIm bayesIm)

Computes the BDe score, using the logarithm of the gamma function, relative to the data, of the factor determined by a node and its parents.

Parameters:

node - The node of the factor.

parents - The parents of the node.

bayesPmMod - The BayesPm that determines the list of variables (nodes) and the structure of the graph.

bayesIm - The BayesIm that determines the observed counts.

Returns:

The score of the factor.

getObservedCounts

public double[][] getObservedCounts(Node node,
 BayesPm bayesPm,
 BayesIm bayesIm)

This method is used in testing and debugging and not in the BDe metric calculations.

Parameters:

node - The node for which the observed counts are to be returned.

bayesPm - The BayesPm that determines the list of variables (nodes) and the structure of the graph.

bayesIm - The BayesIm that determines the observed counts.

Returns:

The observed counts for the given node.

getScoreCount

public int getScoreCount(Node node,
 Set<Node> parents)

This is just for testing the operation of the inner class and the map from nodes and parent sets to scores.

Parameters:

node - The node of the factor.

parents - The parents of the node.

Returns:

The score of the factor.