cc.mallet.topics
Class TopicInferencer

java.lang.Object
  cc.mallet.topics.TopicInferencer

All Implemented Interfaces:

java.io.Serializable

public class TopicInferencer
extends java.lang.Object
implements java.io.Serializable

See Also:

Serialized Form

Field Summary

protected double[]	alpha
protected double	beta
protected double	betaSum
protected int	numTopics
protected int	numTypes
protected Randoms	random
protected int[]	tokensPerTopic
protected int	topicBits
protected int	topicMask
protected int[][]	typeTopicCounts

Constructor Summary

TopicInferencer(int[][] typeTopicCounts, int[] tokensPerTopic, Alphabet alphabet, double[] alpha, double beta, double betaSum)

Method Summary

double[]	getSampledDistribution(Instance instance, int numIterations, int thinning, int burnIn) Use Gibbs sampling to infer a topic distribution.
static TopicInferencer	read(java.io.File f)
void	setRandomSeed(int seed)
void	writeInferredDistributions(InstanceList instances, java.io.File distributionsFile, int numIterations, int thinning, int burnIn, double threshold, int max) Infer topics for the provided instances and write distributions to the provided file.

Methods inherited from class java.lang.Object

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

Field Detail

numTopics

protected int numTopics

topicMask

protected int topicMask

topicBits

protected int topicBits

numTypes

protected int numTypes

alpha

protected double[] alpha

beta

protected double beta

betaSum

protected double betaSum

typeTopicCounts

protected int[][] typeTopicCounts

tokensPerTopic

protected int[] tokensPerTopic

random

protected Randoms random

Constructor Detail

TopicInferencer

public TopicInferencer(int[][] typeTopicCounts,
                       int[] tokensPerTopic,
                       Alphabet alphabet,
                       double[] alpha,
                       double beta,
                       double betaSum)

Method Detail

setRandomSeed

public void setRandomSeed(int seed)

getSampledDistribution

public double[] getSampledDistribution(Instance instance,
                                       int numIterations,
                                       int thinning,
                                       int burnIn)

Use Gibbs sampling to infer a topic distribution. Topics are initialized to the (or a) most probable topic for each token. Using zero iterations returns exactly this initial topic distribution.

This code does not adjust type-topic counts: P(w|t) is clamped.

writeInferredDistributions

public void writeInferredDistributions(InstanceList instances,
                                       java.io.File distributionsFile,
                                       int numIterations,
                                       int thinning,
                                       int burnIn,
                                       double threshold,
                                       int max)
                                throws java.io.IOException

Infer topics for the provided instances and write distributions to the provided file.

Parameters:

instances -

distributionsFile -

numIterations - The total number of iterations of sampling per document

thinning - The number of iterations between saved samples

burnIn - The number of iterations before the first saved sample

threshold - The minimum proportion of a given topic that will be written

max - The total number of topics to report per document]

Throws:

java.io.IOException

read

public static TopicInferencer read(java.io.File f)
                            throws java.lang.Exception

Throws:

java.lang.Exception