clustKL: Choose cluster number by KL method

In maoyinan/BayesPC: R package for Bayesian Projection Clustering

Description

To find minimal cluster number which reduces Kullback-Leibler divergence no larger than a customized threshold. KL monotonically decreases with number of clusters.

Usage

clustKL(id_var, ls_par, dat, ls_idxA, nIter, thKL, regQ, seed)

Arguments

id_var	Character of id variable
ls_par	Posterior mean of LMM parameters generated from postMean output
dat	Longitudinal data input
ls_idxA	List of random effect indices to project on
nIter	Number of iterations used in clustering optimization
thKL	Proportion of max KL, used as upper threshold of Kullback-Leibler divergence corresponding to chosen number of clusters
regQ	Positive regularization value to add to the diagonal of matrix to be inverted
seed	Random seed to initialize cluster centers

Value

list(nClusters,KLs,cluster0)

nClusters	Vector of optimised cluster numbers corresponding to sets of random effects in ls_idxA
KLs	List of Kullback-Leibler divergence across grid of cluster numbers corresponding to ls_idxA
cluster0	List of matrix, column k indicates cluster membership for k clusters corresponding to ls_idxA

See Also

Other BayesPC main functions: modelStan(), pcFit(), postMean()

Other Cluster number choices: clustBoot()

Examples

data(df_of_draws)
ls_par <- postMean(df_of_draws, paste0("Z", 1:10), "ID", DATASET)
ls_idxA <- list(
  seq(10),
  1:4,
  5:7,
  8:10
)
out_KL <- clustKL("ID", ls_par, DATASET, ls_idxA, 10, .1, 1e-6, 1)

maoyinan/BayesPC documentation built on Dec. 21, 2021, 1:48 p.m.