org.knowceans.sandbox.ilda
Class IldaGibbsSampler

java.lang.Object
  org.knowceans.dirichlet.lda.LdaGibbsSampler
      org.knowceans.sandbox.ilda.IldaGibbsSampler

All Implemented Interfaces:

java.io.Serializable

public class IldaGibbsSampler
extends LdaGibbsSampler
implements java.io.Serializable

Gibbs sampler for estimating the best assignments of topics for words and documents in a corpus. The algorithm based on "Parameter estimation for test analysis" (2005), http://www.arbylon.net/publications/text-est_iv.pdf, which gives a more detailed derivation of a Gibbs sampler for the LDA model than Tom Griffiths' white paper "Gibbs sampling in the generative model of Latent Dirichlet Allocation" (2002) and extends it to the infinite limit on K according to Neal's paper "Monte-Carlo sampling methods for the Dirichlet process".

Author:

heinrich

See Also:

Serialized Form

Field Summary

private int	growstep array grow step
private int[]	ndunrep nubmer of unrepresented words in a document
private int	nwsumunrep total number of unrepresented words
private int[]	nwunrep number of times a term is unrepresented by current topics
private static long	serialVersionUID

Fields inherited from class org.knowceans.dirichlet.lda.LdaGibbsSampler

backupIteration, conf, dispcol, numstats, phisum, rand, state, thetasum

Constructor Summary

IldaGibbsSampler(int[][] documents, int V)
Initialise the Gibbs sampler with data.

Method Summary

private void	addComponent() handle size of componentwise structures.
private void	gibbs() Main method for gibbs sampling
static void	main(java.lang.String[] args)
private void	removeComponent(int j) removes one component from the model
(package private) double	sampleAlpha() sample alpha from a Gam(1,1) distribution using Escobar & West's method
protected int	sampleLdaFullConditional(int m, int n) Sample a topic z_i from the full conditional distribution: p(z_i = j | z_-i, w) = (n_-i,j(w_i) + beta)/(n_-i,j(.) + W * beta) * (n_-i,j(d_i) + alpha)/(n_-i,.

Methods inherited from class org.knowceans.dirichlet.lda.LdaGibbsSampler

getPhi, getState, getTheta, gibbs, gibbsHeap, gibbsHeap, initialState, load, output, run, sampleCorpus, sampleLdaFullConditional, save, saveState, updateParams, updatePhi, updateTheta, writeParameters

Methods inherited from class java.lang.Object

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

Field Detail

serialVersionUID

private static final long serialVersionUID

See Also:

Constant Field Values

growstep

private int growstep

array grow step

nwunrep

private int[] nwunrep

number of times a term is unrepresented by current topics

ndunrep

private int[] ndunrep

nubmer of unrepresented words in a document

nwsumunrep

private int nwsumunrep

total number of unrepresented words

Constructor Detail

IldaGibbsSampler

public IldaGibbsSampler(int[][] documents,
                        int V)

Initialise the Gibbs sampler with data.

Parameters:

V - vocabulary size

data -

Method Detail

addComponent

private void addComponent()

handle size of componentwise structures.

Note: We use arrays for components more readable syntax and possibly speed loss during all cast operations when accessing a Vector. Therefore all loops over components should explicitly use k, not, e.g., mu.length. The problem with this approach is that it is hard to remove unoccupied classes

removeComponent

private void removeComponent(int j)

removes one component from the model

Parameters:

j -

gibbs

private void gibbs()

Main method for gibbs sampling

Overrides:

gibbs in class LdaGibbsSampler

sampleLdaFullConditional

protected int sampleLdaFullConditional(int m,
                                       int n)

Sample a topic z_i from the full conditional distribution: p(z_i = j | z_-i, w) = (n_-i,j(w_i) + beta)/(n_-i,j(.) + W * beta) * (n_-i,j(d_i) + alpha)/(n_-i,.(d_i) + K * alpha)

Parameters:

m - document

n - word

sampleAlpha

double sampleAlpha()

sample alpha from a Gam(1,1) distribution using Escobar & West's method

Returns:

main

public static void main(java.lang.String[] args)