The per-word variable that indicates which topic it comes from.
Add a document to the LDA model.
The prior over per-document topic distribution
The prior over per-topic word distribution
Infer doc.
Infer doc.theta. If the document is not already part of this LDA, do not add it and do not collapse anything that would effect this LDA.
Run a collapsed Gibbs sampler to estimate the parameters of the LDA model.
The per-topic distribution over words.
The per-topic distribution over words. FiniteMixture is a Seq of Dirichlet-distributed Proportions.
Typical recommended value for alpha1 is 50/numTopics.