lda_fgs_st: LDA Full Gibbs sampler (FGS) with Serial Tempering and Tuning

In clintpgeorge/ldamcmc: Markov chain Monte Carlo Algorithms for the Latent Dirichlet Allocation Model

Description

This a Markov chain on (z, β, θ)

Usage

lda_fgs_st(K, V, wid, doc.N, h.grid, st.grid, st.grid.nbrs, init.st.grid.index,
  init.st.grid.zetas, max.iter, burn.in, spacing, tuning.iter, save.z,
  save.beta, save.theta, save.st.grid.index, save.lp, save.hat.ratios,
  save.tilde.ratios, verbose, max.iter.final = max.iter)

Arguments

K	the number of topics in the corpus
V	the vocabulary size
wid	the vocabulary ids of every word instance in each corpus document (1 X total.N vector). We assume vocabulary id starts with 1
doc.N	the document lengths
h.grid	the grid of hyperparameters h = (η, α), 2 x G matrix, where G is the number of grid points and the first row is for α points and the second row is for η points
st.grid	the helper grid of hyperparameters h = (η, α), 2 x G' matrix, for Serial Tempering, where G' is the number of grid points and the first row is for α points and the second row is for η points
st.grid.nbrs	the neighbor indices, from [0, G'], of each st.grid point
init.st.grid.index	the initial h index from st.grid for Serial Tempering
init.st.grid.zetas	initial guess for normalization constants
max.iter	the max number of Gibbs iterations to be performed
burn.in	the burn in period of the Gibbs sampler
spacing	the spacing between the stored samples (to reduce correlation)
tuning.iter	the number of tuning iterations
save.z	if 0 the function does not save z samples
save.beta	if 0 the function does not save β samples
save.theta	if 0 the function does not save θ samples
save.st.grid.index	if 0 the function does not save selected indices of h in the grid
save.lp	if 0 the function does not save computed log posterior for iterations
save.hat.ratios	if 1 sub grid and main grid ratios are saved
save.tilde.ratios	if 1 normalized main grid ratios are saved. It's mainly used for estimating posterior expectations.
verbose	values: 1, 0
max.iter.final	the max number of Gibbs iterations to be performed for the final tuning run

Details

Last modified on: January 28, 2016

Value

the Gibbs sampling output

See Also

Other Gibbs sampling methods: get_grid_neighbors, lda_acgs_hs, lda_acgs, lda_fgs_hs, lda_fgs

clintpgeorge/ldamcmc documentation built on Feb. 22, 2020, 12:39 p.m.