viLDA.stoch: viLDA.stoch

In cvraut/viLDA: variational inference Latent Dirichlet Allocation

Description

viLDA.stoch

Usage

viLDA.stoch(
  dat,
  K,
  d = NULL,
  v = NULL,
  maxIterVal = NULL,
  maxVBiterVal = NULL,
  alphaWords = 0.2,
  alphaTopics = 0.2,
  rho = 0.001,
  tol = 0.1,
  seed = 19890419,
  ...
)

Arguments

dat	The data in long format. This is a data.frame with 3 columns: $doc = doc ids $word = word ids $N = counts An example of this format can be generated through viLDA::data_gen
K	Integer, the number of clusters
d	Intger, the number of documents (default = NULL) if NULL then this is imputed from the dat object
v	Integer, the size of the vocabulary (default = NULL) if NULL then this is also imputed from the dat object
maxIterVal	Integer, the maximum number of outer iterations for the global optima search (default = NULL) if NULL then this imputed to be log(v)
alphaWords	Float, the prior for the word-cluster distributions (default = 0.2)
alphaTopics	Float, the prior for the document-cluster distributions (default = 0.2)
rho	Float, the learning rate for the parameter updates (default = 0.001)
tol	Float, the tolerance for assessing convergence (default = 0.1)
seed	Float, the seed run at the begining (default = 19890418)
...	extra arguments can be passed, but they don't do anything yet 🛠
maxVBIterVal	Integer, the maximum number of inner iterations for the local optima search (default = NULL) if NULL then this is imputed to be ceiling(sqrt(v))

Value

list([1],[2],[3])

• [1]: k x v matrix of the expected values for the word-cluster probabilities

• [2]: I x k matrix of the complete stored values of the parameters through all iterations

• [3]: n length vector of the true value

cvraut/viLDA documentation built on Dec. 19, 2021, 7:05 p.m.