il2.inf.edgedeletion
Class EDMPAlgorithm

Object
  extended by EDAlgorithm
      extended by EDMPAlgorithm

public class EDMPAlgorithm
extends EDAlgorithm


Nested Class Summary
 
Nested classes/interfaces inherited from class EDAlgorithm
EDAlgorithm.RankingHeuristic
 
Field Summary
static Algorithm ALGORITHM_DEFAULT
           
 
Fields inherited from class EDAlgorithm
doConvergenceUpdates, DOUBLE_THRESHOLD_DEFAULT, ecTimes, edgeConvergenceScores, INT_MAX_ITERATIONS_DEFAULT, LONG_TIMEOUT_MILLIS_DEFAULT
 
Constructor Summary
EDMPAlgorithm(BayesianNetwork bn, int[][] ed, int mi, long tm, double ct, Algorithm alg, Map<Algorithm.Setting,?> settings)
           
 
Method Summary
 double getLogZeroMiCorrection()
           
 double[] getZeroMiCorrections()
           
 double logCompMPE()
          This computes an approximation to log Pr(MPE).
 double logPrMPE()
          This computes an approximation to log Pr(MPE), where: = MPE is computed in simplified network = the value log Pr(MPE) is computed in original network
 double logPrMPE(int[] mpe)
          This computes an approximation to log Pr(MPE), where: = MPE is given = the value log Pr(MPE) is computed in original network Use this if you call EDMPAlgorithm.mpe() yourself, as the method EDMPAlgorithm.logPrMPE() also calls EDMPAlgorithm.mpe(), which performs a non-trivial amount of work.
 double logPrpMPE()
          This computes an approximation to log Pr(MPE), where: = MPE is given = the value log Pr'(MPE) is computed in simplfied network This gives another approximation to the value of the MPE in the original network.
 int[] mpe()
           
 
Methods inherited from class EDAlgorithm
cloneConditional, converged, convergenceSummary, die, edgeCorrelations, edgeMis, edgesDeleted, edgeTargetedMis, edNet, getCloneCPT, getClusterStats, getCompilationTime, getConvThreshold, getCorrelationTime, getEdgeRankingTime, getGeneralEdgeCorrection, getGeneralEdgeCorrections, getMaxIterations, getPropagationTime, getSeparatorStats, getSoftEvidenceCPT, getTimeoutMillis, getZeroMiCorrection, iterations, iterationStatusOk, kl, limitsExceeded, logPrEvidence, oneMoreIteration, prEvidence, rankEdgesByConvergence, rankEdgesByCorrelation, rankEdgesByMi, rankEdgesByScore, rankEdgesByTargetMi, rankEdgesRandomly, setConvThreshold, setEvidence, setMaxIterations, setTable, setTimeoutMillis, skl, timeMillis, varConditional
 
Methods inherited from class Object
equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Field Detail

ALGORITHM_DEFAULT

public static final Algorithm ALGORITHM_DEFAULT
Constructor Detail

EDMPAlgorithm

public EDMPAlgorithm(BayesianNetwork bn,
                     int[][] ed,
                     int mi,
                     long tm,
                     double ct,
                     Algorithm alg,
                     Map<Algorithm.Setting,?> settings)
Method Detail

getZeroMiCorrections

public double[] getZeroMiCorrections()
Overrides:
getZeroMiCorrections in class EDAlgorithm

getLogZeroMiCorrection

public double getLogZeroMiCorrection()
Overrides:
getLogZeroMiCorrection in class EDAlgorithm

logCompMPE

public double logCompMPE()
This computes an approximation to log Pr(MPE). via Compensation. Analagous to the approximation for Pr(e) and the Bethe free energy.


logPrMPE

public double logPrMPE()
This computes an approximation to log Pr(MPE), where: = MPE is computed in simplified network = the value log Pr(MPE) is computed in original network


logPrMPE

public double logPrMPE(int[] mpe)
This computes an approximation to log Pr(MPE), where: = MPE is given = the value log Pr(MPE) is computed in original network Use this if you call EDMPAlgorithm.mpe() yourself, as the method EDMPAlgorithm.logPrMPE() also calls EDMPAlgorithm.mpe(), which performs a non-trivial amount of work.


logPrpMPE

public double logPrpMPE()
This computes an approximation to log Pr(MPE), where: = MPE is given = the value log Pr'(MPE) is computed in simplfied network This gives another approximation to the value of the MPE in the original network. However, we recommend using EDMPAlgorithm.logCompMPE() instead.


mpe

public int[] mpe()


Copyright 2010 UCLA Automated Reasoning Group