search

lda-project / lda

Topic modeling with latent Dirichlet allocation using Gibbs sampling
https://lda.readthedocs.io/
Mozilla Public License 2.0
1.24k stars 390 forks source link

Is log-likelihood calculation irrelavant with gibbs sampling? #96

Closed Hins closed 4 years ago

Hins commented 4 years ago

IMHO, log-likelihood calculation is irrelavant with Gibbs sampling, finanlly LDA can get topic-word distribution, document-topic distribution and topic distribution by sampling w/o log-likelihood calculation.

ariddell commented 4 years ago

Are you talking about held-out log likelihood calculations (i.e., the probability of an unseen document)? In this case, we need to estimate the document-topic distribution for the unseen document. We do not know that in advance.