Choose the number of mixture components K for NMF-GMM

This function is experimental and still under development. The interface may change in future versions: argument names, defaults and the contents of the returned object are not yet stable. Code written against it today may need adjusting after an update. The rest of the package does not carry this caveat.

nmf.gmm.select fits nmf.gmm over a vector of \(K\) values and reports the log-likelihood, BIC and ICL for each, selecting K.best by BIC (K.best.icl by ICL). If a vector of known labels is supplied via truth, the adjusted Rand index of the hard clustering is added.

Usage

nmf.gmm.select(Y, A = NULL, rank, K = 1:5, ...)

Arguments

Y, A, rank

As in nmf.gmm.

K

Integer vector of candidate component counts. Default 1:5.

...

Additional arguments passed to nmf.gmm (e.g. cov, X.init, nstart, maxit, seed). Also accepts truth (a length-N vector of known class labels for the ARI column), verbose (logical, print the table; default TRUE), and cores (evaluate the K candidates in parallel; default getOption("mc.cores", 1L)). Parallelism uses a PSOCK cluster on Windows and forking elsewhere; because each K is an independent self-seeded fit and results are returned in order, the table and the selected K are identical for any cores.

Value

An object of class "nmf.gmm.select": a list with the table (data frame over K), K.best (min BIC), K.best.icl (min ICL) and fits (the fitted objects).

See also

nmf.gmm

Examples

# \donttest{
set.seed(1)
Y <- matrix(abs(rnorm(20 * 80)) + 1, 20, 80); A <- rbind(1, rnorm(80))
sel <- nmf.gmm.select(Y, A, rank = 2, K = 1:4, verbose = FALSE)
sel$K.best
#> [1] 2
# }