com.wcohen.ss
Class JensenShannonDistance

java.lang.Object
  com.wcohen.ss.AbstractStringDistance
      com.wcohen.ss.AbstractTokenizedStringDistance
          com.wcohen.ss.JensenShannonDistance

All Implemented Interfaces:

StringDistance, StringDistanceLearner

Direct Known Subclasses:

DirichletJS, JelinekMercerJS, UnsmoothedJS

public abstract class JensenShannonDistance
extends AbstractTokenizedStringDistance

Distance metrics based on Jensen-Shannon distance of two smoothed unigram language models.

Field Summary

Fields inherited from class com.wcohen.ss.AbstractTokenizedStringDistance

tokenizer

Constructor Summary

JensenShannonDistance()

JensenShannonDistance(Tokenizer tokenizer)

Method Summary

protected double	backgroundProb(Token tok) Probability of token in the background language model
java.lang.String	explainScore(StringWrapper s, StringWrapper t) This method needs to be implemented by subclasses.
StringWrapper	prepare(java.lang.String s) Preprocess a string by finding tokens and giving them weights W such that W is the smoothed probability of the token appearing in the document.
double	score(StringWrapper s, StringWrapper t) Jensen-Shannon distance between distributions.
protected abstract double	smoothedProbability(Token tok, double freq, double totalWeight) Smoothed probability of the token with frequency freq in a bag with the given totalWeight
void	train(StringWrapperIterator i) Accumulate statistics on how often each token occurs.

Methods inherited from class com.wcohen.ss.AbstractTokenizedStringDistance

asBagOfTokens, prepare, setStringWrapperPool

Methods inherited from class com.wcohen.ss.AbstractStringDistance

addExample, doMain, explainScore, getDistance, hasNextQuery, nextQuery, prepare, score, setDistanceInstancePool

Methods inherited from class java.lang.Object

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

Constructor Detail

JensenShannonDistance

public JensenShannonDistance(Tokenizer tokenizer)

JensenShannonDistance

public JensenShannonDistance()

Method Detail

train

public final void train(StringWrapperIterator i)

Accumulate statistics on how often each token occurs.

Specified by:

train in class AbstractTokenizedStringDistance

prepare

public final StringWrapper prepare(java.lang.String s)

Preprocess a string by finding tokens and giving them weights W such that W is the smoothed probability of the token appearing in the document.

Specified by:

prepare in interface StringDistance

Overrides:

prepare in class AbstractStringDistance

smoothedProbability

protected abstract double smoothedProbability(Token tok,
                                              double freq,
                                              double totalWeight)

Smoothed probability of the token with frequency freq in a bag with the given totalWeight

backgroundProb

protected double backgroundProb(Token tok)

Probability of token in the background language model

score

public final double score(StringWrapper s,
                          StringWrapper t)

Jensen-Shannon distance between distributions.

Specified by:

score in interface StringDistance

Specified by:

score in class AbstractStringDistance

explainScore

public final java.lang.String explainScore(StringWrapper s,
                                           StringWrapper t)

Description copied from class: AbstractStringDistance

This method needs to be implemented by subclasses.

Specified by:

explainScore in interface StringDistance

Specified by:

explainScore in class AbstractStringDistance