edu.umass.cs.mallet.base.fst.confidence
Class NBestViterbiConfidenceEstimator

java.lang.Object
  extended byedu.umass.cs.mallet.base.fst.confidence.TransducerSequenceConfidenceEstimator
      extended byedu.umass.cs.mallet.base.fst.confidence.NBestViterbiConfidenceEstimator

public class NBestViterbiConfidenceEstimator
extends TransducerSequenceConfidenceEstimator

Estimates the confidence of an entire sequence by the probability that one of the the Viterbi paths rank 2->N is correct. Note that this is a strange definition of confidence, and is mainly used for MultipleChoiceCRFActiveLearner, where we want to find Instances that are mislabeled, but are likely to have a correct labeling in the top N Viterbi paths.


Constructor Summary
NBestViterbiConfidenceEstimator(Transducer model, int N)
           
 
Method Summary
 double estimateConfidenceFor(Instance instance, java.lang.Object[] startTags, java.lang.Object[] inTags)
          Calculates the confidence in the tagging of a Instance.
 
Methods inherited from class edu.umass.cs.mallet.base.fst.confidence.TransducerSequenceConfidenceEstimator
rankInstancesByConfidence
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

NBestViterbiConfidenceEstimator

public NBestViterbiConfidenceEstimator(Transducer model,
                                       int N)
Method Detail

estimateConfidenceFor

public double estimateConfidenceFor(Instance instance,
                                    java.lang.Object[] startTags,
                                    java.lang.Object[] inTags)
Calculates the confidence in the tagging of a Instance.

Specified by:
estimateConfidenceFor in class TransducerSequenceConfidenceEstimator