I am trying to implement the LDA model fit by collapsed Gibbs sampling by myself. I have go through this article. And there is a clear pseudo code (section 5.5), which states that I should check the convergence after each time iterate through all documents, but doesn't say how. Could you please tell me what is the criteria of the convergence in the collapsed Gibbs sampling? Thanks in advance!

I have found this paper, introducing lots of methods to measure the criteria of convergence. But I can't tell which one is suitable for collapsed Gibbs sampling. Could you help me? Thanks.

Thanks for @Greenparker 's help. What I need is the criteria of determining whether the sampler is converged or not.


closed as unclear what you're asking by Xi'an, whuber Apr 10 '16 at 19:27

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ The Cowles+Carlin paper that you refer to does not talk about when to stop sampling, rather it discusses the various convergence diagnostics. There is a difference between assessing convergence and stopping the simulation. $\endgroup$ – Greenparker Apr 12 '16 at 14:15
  • $\begingroup$ @Greenparker Thanks for your reply! So which paper should I refer to? $\endgroup$ – user5779223 Apr 12 '16 at 14:17
  • $\begingroup$ It depends on what your question is? Your question seems to be "how do you assess convergence of the Markov chain in LDA"? In which case this paper is fine. I was just correcting your interpretation of the paper. $\endgroup$ – Greenparker Apr 12 '16 at 14:38
  • $\begingroup$ @Greenparker Indeed I'd like to know how to stop the sampling in the collapsed Gibbs sampling. I had thought it was same with check of convergence. Sorry for the confusing. I'll change the question soon. $\endgroup$ – user5779223 Apr 12 '16 at 14:43
  • $\begingroup$ @Greenparker And I am still not clear about the difference. We have to check the convergence so that we can know when to stop, don't we? $\endgroup$ – user5779223 Apr 12 '16 at 14:46