edu.cmu.minorthird.classify.sequential
Class BeamSearcher

java.lang.Object
  edu.cmu.minorthird.classify.sequential.BeamSearcher

All Implemented Interfaces:

SequenceConstants, java.io.Serializable

public class BeamSearcher
extends java.lang.Object
implements SequenceConstants, java.io.Serializable

A conditional markov model classifier.

Author:

William Cohen

See Also:

Serialized Form

Field Summary

Fields inherited from interface edu.cmu.minorthird.classify.sequential.SequenceConstants

HISTORY_FEATURE, NULL_CLASS_NAME

Constructor Summary

BeamSearcher(Classifier classifier, int historySize, ExampleSchema schema)

Method Summary

ClassLabel[]	bestLabelSequence(Instance[] instances) Get the best label sequence, as determined by the beam search
void	doSearch(Instance[] sequence) Do a beam search.
void	doSearch(Instance[] sequence, ClassLabel[] template) Do a beam search, constraining the bestLabel for each classification to match the non-null values in the template.
java.lang.String	explain(Instance[] sequence)
static Instance	getBeamInstance(Instance instance, int historySize)
Explanation	getExplanation(Instance[] sequence)
int	getMaxBeamSize()
int	getNumberOfSolutionsFound() Return the number of solutions found in the beam.
float	score(int k)
void	setMaxBeamSize(int n)
ClassLabel[]	viterbi(int k) Retrieve the k-th best result of the previous beam search.

Methods inherited from class java.lang.Object

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

Constructor Detail

BeamSearcher

public BeamSearcher(Classifier classifier,
                    int historySize,
                    ExampleSchema schema)

Method Detail

getMaxBeamSize

public int getMaxBeamSize()

setMaxBeamSize

public void setMaxBeamSize(int n)

bestLabelSequence

public ClassLabel[] bestLabelSequence(Instance[] instances)

Get the best label sequence, as determined by the beam search

getBeamInstance

public static Instance getBeamInstance(Instance instance,
                                       int historySize)

doSearch

public void doSearch(Instance[] sequence)

Do a beam search.

doSearch

public void doSearch(Instance[] sequence,
                     ClassLabel[] template)

Do a beam search, constraining the bestLabel for each classification to match the non-null values in the template.

This would be better folded in with the one-arg version of doSearch, but is kept separate for backward compatibility.

getNumberOfSolutionsFound

public int getNumberOfSolutionsFound()

Return the number of solutions found in the beam.

viterbi

public ClassLabel[] viterbi(int k)

Retrieve the k-th best result of the previous beam search. To get the best, use viterbi(0), the second best is viterbi(1), etc.

score

public float score(int k)

explain

public java.lang.String explain(Instance[] sequence)

getExplanation

public Explanation getExplanation(Instance[] sequence)