Fit several models and keep the ones on the quality frontier

With random initialization the variational objective is multimodal, so the standard workflow (cf. stm::selectModel) is to fit many models and keep those on the semantic-coherence / exclusivity frontier, then choose among them. faSTM fits the candidates in parallel.

Usage

select_model(
  corpus,
  K,
  N = 10L,
  prevalence = NULL,
  content = NULL,
  init.type = "Random",
  cores = 1L,
  M = 10L,
  frexw = 0.7,
  seed = 1L,
  ...
)

Arguments

corpus

A faSTM_corpus.

K

Number of topics.

N

Number of candidate models (distinct random inits).

prevalence, content

Optional covariate formulas.

init.type

Initialization; "Random" (the point of selecting) or "Spectral" (deterministic — then all N are identical).

cores

Candidates to fit in parallel.

M

Top words for coherence/exclusivity scoring.

frexw

Exclusivity FREX weight.

seed

Base RNG seed (candidate i uses seed + i - 1).

...

Passed to stm().

Value

A faSTM_selectmodel: models (the fits), semcoh, exclusivity, and frontier (indices of non-dominated models).