jbnc.util
Class BNTools

java.lang.Object
  jbnc.util.BNTools

public final class BNTools

extends java.lang.Object

Utilities for Bayesian networks.

Since:

June 1, 1999

Author:

Jarek Sacha

Field Summary

static double

beta_ijk A some small value larger than zero.

Constructor Summary

BNTools()

Method Summary

static double	gammaLn(double xx) Returns the value ln[ gamma(xx)] for xx > 0 Implementation based on W.H.
static double	getASBMParamComponent(BayesianNetworks.BayesNet net, DatasetInt dataset, boolean usePriors, double alphaK) Return network parameters component of the asymptotic standard Bayesian measure (ASBM).
static int	getNetworkDimension(BayesianNetworks.BayesNet net) Returns dimension of a Bayesian network.
protected static InferenceGraphs.InferenceGraphNode[]	getNodes(Dataset dataset, InferenceGraphs.InferenceGraph graph) Get node names from a graph in an order they appear in the dataset.
protected static void	learnParameters_old(BayesianNetworks.BayesNet net, Dataset data, boolean useDirihlet, double alphaK) Learns parameters for the current network structure.
static void	learnParameters(BayesianNetworks.BayesNet net, DatasetInt data, boolean useDirihlet, double alphaK) Learns parameters for the current network structure.
static void	learnParameters(BayesianNetworks.BayesNet net, FrequencyCalc fc, boolean useDirihlet, double alphaK) Learns parameters for the current network structure.
static void	main(java.lang.String[] args) Description of the Method

Methods inherited from class java.lang.Object

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

Field Detail

beta_ijk

public static final double beta_ijk

A some small value larger than zero.

See Also:

Constant Field Values

Constructor Detail

BNTools

public BNTools()

Method Detail

getNetworkDimension

public static int getNetworkDimension(BayesianNetworks.BayesNet net)
                               throws java.lang.Exception

Returns dimension of a Bayesian network.

Dimension of a Bayesian network: Let X be a set of random variables and B be a Bayesian network defined over X. The dimension of this network, Dim(B), is the number of free parameters required to completely specify the joint probability distribution of X.

E. Castillo, J. M. Gutierrez and A. S. Hadi, Expert Systems and Probabilistic Network Models , Springer, 1997. p.486.

Parameters:

net - Description of Parameter

Returns:

The NetworkDimension value

Throws:

java.lang.Exception - .

getASBMParamComponent

public static final double getASBMParamComponent(BayesianNetworks.BayesNet net,
                                                 DatasetInt dataset,
                                                 boolean usePriors,
                                                 double alphaK)
                                          throws java.lang.Exception

Return network parameters component of the asymptotic standard Bayesian measure (ASBM).

q = sum_{i =1...n} sum_{j =1... qi} sum_{k =1...ri} N_ijk log N_ijk / N_ij

where N_ijk means that variable X_i is in configuration k and parents of variable X_i are in configuration j .

E. Castillo, J. M. Gutierrez and A. S. Hadi, Expert Systems and Probabilistic Network Models , Springer, 1997. p.494, eq.(11.28).

Parameters:

net - Description of Parameter

dataset - Description of Parameter

usePriors - Description of Parameter

alphaK - Description of Parameter

Returns:

The ASBMParamComponent value

Throws:

java.lang.Exception - .

gammaLn

public static double gammaLn(double xx)
                      throws java.lang.Exception

Returns the value ln[ gamma(xx)] for xx > 0 Implementation based on W.H. Press et al. Numerical Recipes in C , 2nd Ed., Cambridge University Press, 1992.

Parameters:

xx -

Returns:

Throws:

java.lang.Exception - When xx <= 0.

learnParameters

public static void learnParameters(BayesianNetworks.BayesNet net,
                                   FrequencyCalc fc,
                                   boolean useDirihlet,
                                   double alphaK)
                            throws java.lang.Exception

Learns parameters for the current network structure. Existing network parameters are replaced with the new ones. This method can use "uniform" Dirihlet priors.

Parameters:

net - Bayesian network.

useDirihlet - Indicates whether Dirihlet priors should be used for network parameters.

alphaK - alpha_k parameter for Dirihlet priors. All alpha_k are assumed to be the same and greater than zero.

fc - Description of Parameter

Throws:

java.lang.Exception

learnParameters

public static void learnParameters(BayesianNetworks.BayesNet net,
                                   DatasetInt data,
                                   boolean useDirihlet,
                                   double alphaK)
                            throws java.lang.Exception

Learns parameters for the current network structure. Existing network parameters are replaced with the new ones. This method can use "uniform" Dirihlet priors.

Parameters:

net - Bayesian network.

useDirihlet - Indicates whether Dirihlet priors should be used for network parameters.

alphaK - alpha_k parameter for Dirihlet priors. All alpha_k are assumed to be the same and greater than zero.

data - Description of Parameter

Throws:

java.lang.Exception

main

public static void main(java.lang.String[] args)

Description of the Method

Parameters:

args - Description of Parameter

getNodes

protected static InferenceGraphs.InferenceGraphNode[] getNodes(Dataset dataset,
                                                               InferenceGraphs.InferenceGraph graph)
                                                        throws java.lang.Exception

Get node names from a graph in an order they appear in the dataset.

Parameters:

dataset - Description of Parameter

graph - Description of Parameter

Returns:

The Nodes value

Throws:

java.lang.Exception - Description of Exception

learnParameters_old

protected static void learnParameters_old(BayesianNetworks.BayesNet net,
                                          Dataset data,
                                          boolean useDirihlet,
                                          double alphaK)
                                   throws java.lang.Exception

Learns parameters for the current network structure. Existing network parameters are replaced with the new ones. This method can use "uniform" Dirihlet priors.

Parameters:

net - Bayesian network.

useDirihlet - Indicates whether Dirihlet priors should be used for network parameters.

alphaK - alpha_k parameter for Dirihlet priors. All alpha_k are assumed to be the same and greater than zero.

data - Description of Parameter

Throws:

java.lang.Exception