edu.ksu.cis.bnj.bbn.learning.score
Class WeightedBDEScore

java.lang.Object
  edu.ksu.cis.bnj.bbn.learning.LearnerScore
      edu.ksu.cis.bnj.bbn.learning.score.BDEScore
          edu.ksu.cis.bnj.bbn.learning.score.WeightedBDEScore

public class WeightedBDEScore

extends BDEScore

Field Summary

static double	lnPi
static double	oneThird

Fields inherited from class edu.ksu.cis.bnj.bbn.learning.score.BDEScore

arityCache, curNode, lnFactorialCache, nodeArity, nodes, result, ri_minus_one, ri_minus_one_fact, tupleSize

Fields inherited from class edu.ksu.cis.bnj.bbn.learning.LearnerScore

NO_CANDIDATES, owner, tally

Constructor Summary

WeightedBDEScore()

WeightedBDEScore(Learner l)

Method Summary

double	getScore(int curNode, int candidate, java.util.Set[] parentTable) BDE scoring is: g(vi, parents(vi)) = \Pi_{j=1}^{qi} ((ri-1)!
void	initialize() This is just to reset the cache.
protected void	iterateUniqueInstantiation(int depth, TableTally tally) Recursively filter out the data according to the current node (Roby Joehanes)
protected static double	stirlingGosperApprox(double n) Stirling-Gosper approximation for ln n!

Methods inherited from class edu.ksu.cis.bnj.bbn.learning.score.BDEScore

iterateUniqueInstantiation, precache

Methods inherited from class edu.ksu.cis.bnj.bbn.learning.LearnerScore

getTallyer, setData, setOwner

Methods inherited from class java.lang.Object

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

Field Detail

oneThird

public static final double oneThird

See Also:

Constant Field Values

lnPi

public static final double lnPi

Constructor Detail

WeightedBDEScore

public WeightedBDEScore()

WeightedBDEScore

public WeightedBDEScore(Learner l)

Method Detail

getScore

public double getScore(int curNode,
                       int candidate,
                       java.util.Set[] parentTable)

Description copied from class: BDEScore

BDE scoring is:
g(vi, parents(vi)) = \Pi_{j=1}^{qi} ((ri-1)!/(Nij+ri-1)!) \Pi_{k=1}^{ri} Nijk!

With:
vi is the node being queried
qi is the unique instantiation of vi's parents
ri is vi's arity
Nij is the total number of unique cases of one unique instantiation
Nijk is the total number of unique cases of one unique instantiation that contains the value of r_ik.

Then, we do a logarithm out of g.

In short, UTSL or RTFP.

Overrides:

getScore in class BDEScore

See Also:

edu.ksu.cis.bnj.bbn.learning.LearnerScore#score(java.lang.String, java.util.Set)

stirlingGosperApprox

protected static double stirlingGosperApprox(double n)

Stirling-Gosper approximation for ln n! (Roby Joehanes) ln n! = 0.5*(ln(2*n+ 1/3) + ln pi) + n ln n - n I did this because the weights may be real valued

iterateUniqueInstantiation

protected void iterateUniqueInstantiation(int depth,
                                          TableTally tally)

Recursively filter out the data according to the current node (Roby Joehanes)

Parameters:

depth -

tally -

initialize

public void initialize()

This is just to reset the cache. But since it is not dependent on tuple numbers, we can simply ignore this.

Overrides:

initialize in class LearnerScore

See Also:

LearnerScore.initialize()