org.knowceans.dirichlet.sandbox
Class LdaGibbsSamplerHyper

java.lang.Object
  org.knowceans.dirichlet.lda.LdaGibbsSampler
      org.knowceans.dirichlet.sandbox.LdaGibbsSamplerHyper

All Implemented Interfaces:

java.io.Serializable

public class LdaGibbsSamplerHyper
extends LdaGibbsSampler

Gibbs sampler for estimating the best assignments of topics for words and documents in a corpus. The algorithm is introduced in Tom Griffiths' paper "Gibbs sampling in the generative model of Latent Dirichlet Allocation" (2002).

Author:

heinrich

See Also:

Serialized Form

Field Summary

(package private) int[]	alphaParams
(package private) InvGammaArms	alphaSampler
(package private) int[]	betaParams
(package private) InvGammaArms	betaSampler
private static long	serialVersionUID
(package private) LdaMarkovStateHyper	state

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

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

Constructor Summary

LdaGibbsSamplerHyper(ITermCorpus corpus, ExtLdaConfiguration conf, java.util.Random rand)

LdaGibbsSamplerHyper(LdaMarkovStateHyper state, ExtLdaConfiguration conf, java.util.Random rand)

Method Summary

private void	estimateHyperParameters(LdaMarkovStateHyper hyper) Estimate the hyperparameters from the observations (uses vectorial values)
private void	init()
private void	leaveOneOutAlpha(LdaMarkovStateHyper s) Estimate the parameters alpha based on Minka's leave-one-out likeliho
private void	leaveOneOutBeta(LdaMarkovStateHyper s) Estimate the parameters beta based on Minka's leave-one-out likelihood.
protected void	sampleCorpus(LdaMarkovState s) Sample once through the corpus and update the corresponding state.
private void	sampleHyperParameters() Sample scalar hyperparameters from the observations using adaptive rejective metropolis sampling with rasmussen's vague inverse-gamma prior
protected int	sampleLdaFullConditional(LdaMarkovState ms, 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(w_i))/(n_-i,j(.) + sum beta) * (n_-i,j(d_i) + alpha(j))/(n_-i,.

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

getPhi, getState, getTheta, gibbs, gibbs, gibbsHeap, gibbsHeap, initialState, load, main, output, run, 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

alphaSampler

InvGammaArms alphaSampler

betaSampler

InvGammaArms betaSampler

alphaParams

int[] alphaParams

betaParams

int[] betaParams

state

LdaMarkovStateHyper state

serialVersionUID

private static final long serialVersionUID

See Also:

Constant Field Values

Constructor Detail

LdaGibbsSamplerHyper

public LdaGibbsSamplerHyper(ITermCorpus corpus,
                            ExtLdaConfiguration conf,
                            java.util.Random rand)

Parameters:

corpus -

conf -

rand -

LdaGibbsSamplerHyper

public LdaGibbsSamplerHyper(LdaMarkovStateHyper state,
                            ExtLdaConfiguration conf,
                            java.util.Random rand)

Parameters:

state -

conf -

rand -

Method Detail

init

private void init()

sampleCorpus

protected void sampleCorpus(LdaMarkovState s)

Description copied from class: LdaGibbsSampler

Sample once through the corpus and update the corresponding state. The parameter is used to choose the state to be sampled from: query vs. corpus; choice of chain for multichain sampling.

Overrides:

sampleCorpus in class LdaGibbsSampler

sampleLdaFullConditional

protected int sampleLdaFullConditional(LdaMarkovState ms,
                                       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(w_i))/(n_-i,j(.) + sum beta) * (n_-i,j(d_i) + alpha(j))/(n_-i,.(d_i) + sum alpha)

Overrides:

sampleLdaFullConditional in class LdaGibbsSampler

Parameters:

m - document

n - word

estimateHyperParameters

private void estimateHyperParameters(LdaMarkovStateHyper hyper)

Estimate the hyperparameters from the observations (uses vectorial values)

Parameters:

hyper -

leaveOneOutAlpha

private void leaveOneOutAlpha(LdaMarkovStateHyper s)

Estimate the parameters alpha based on Minka's leave-one-out likeliho

Parameters:

s - od.

leaveOneOutBeta

private void leaveOneOutBeta(LdaMarkovStateHyper s)

Estimate the parameters beta based on Minka's leave-one-out likelihood.

Parameters:

s -

sampleHyperParameters

private void sampleHyperParameters()

Sample scalar hyperparameters from the observations using adaptive rejective metropolis sampling with rasmussen's vague inverse-gamma prior

Throws:

java.lang.Exception