relabel_true_log: Relabel membership vector by minimizing KL-algorithm to true...
In baruuum/relabelKL: Relabeling MCMC Output

relabel_true_log

R Documentation

Relabel membership vector by minimizing KL-algorithm to true labels

Description

Relabels the membership vectors of a mixed membership model or mixed membership stochastic blockmodel, by minimizing the KL-divergence to a priori known true labels, where probabilities are entered on the log-scale.

Usage

relabel_true_log(
  lphi,
  lphi_true,
  renormalize = FALSE,
  nthreads = 0L,
  verbose = FALSE
)

Arguments

`lphi`	cube of length `S`, each slice of which is a matrix of dimension `N` times `K`, and where `N` is the number of individuals and K is the number of extreme types (or classes). Elements of `lphi` should be log-probabilities.
`lphi_true`	a `N` times `K` matrix containing the "true" values of latent-class/mixed-membership probabilities on the log-scale
`renormalize`	whether to renormalize the rows of `lphi`
`nthreads`	number of threads to use if OPENMP is enabled
`verbose`	if true, KL-divergence from true labels is calculated and printed before and after relabeling

Details

OpenMP is enabled if available. Due to overhead, the inner-most loop is parallelized only if the number of latent classes/types is larger than 3.

Value

Returns a Rcpp::List with two elements: 1) A arma::cube, permuted, of the same dimensions as phi but with the labels permuted; 2) perms is a S times K matrix containing the permutations necessary to produce permuted from phi (i.e., the mapping from phi to permuted)