[Topic-models] hLDA question

Laura Dietz dietz at informatik.hu-berlin.de
Fri May 30 11:41:19 EDT 2008


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi David,

MCMC theory says: you can create an MCMC chain by interleaving two MCMC
(sub-)chains in an arbitrary way and the chain will converge.

Thus, you do not have to train a one of the subchains until convergence
before proceeding with the next iteration of the other subchain.

For details have a look at the chapter about interleaving chains in
Robert&Casella: Monte Carlo Statistical Methods, 2004.

Hope this helps,
	Laura

David Blei wrote:
| i have run this sampler both ways, and have not seen any problems with
| the former.  both schedules yield a valid markov chain.
|
|
| On May 29, 2008, at 2:21 PM, David André Broniatowski wrote:
|> Hello,
|>
|> Does one hLDA iteration consist of one LDA iteration followed by one
|> nCRP iteration? Or do more LDA iterations need to happen first such
|> that the LDA Markov chain converges?

- --
- ------------------------------------------------------
Laura Dietz, Dipl.-Inform.
Max Planck PhD scholarship holder
Research Group 2 (Machine Learning)
Max Planck Institut für Informatik
Stuhlsatzenhausweg 85, 66123 Saarbrücken, Germany
Room 429
Phone: +49 681 9325 529
E-mail: dietz at mpi-inf.mpg.de
http://www.mpi-inf.mpg.de/~dietz
http://www.mpi-sb.mpg.de/departments/rg2/
- ------------------------------------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFIQCAfu7dZjvX3AX8RAg9pAKDF9YtZfseu5z9pdjOH+dgqaHaT4QCfdjWf
OxYzGYiq0P/nzhZO/DLi8Kg=
=/8sJ
-----END PGP SIGNATURE-----


More information about the Topic-models mailing list