auto_bcv: Use BCV to estimate the optimal k until convergence

In jonalim/mfBiclust: Pipelines for Bi-Clustering Using Matrix Factorization

Description

Repeatedly chooses the k that minimizes the BCV until the the distribution of k's has converged. Often less than 50 iterations are required. The median k is reported as "best".

Usage

auto_bcv(Y, ks = seq_len(min(nrow(Y), ncol(Y)) - 1), holdouts = 3L,
  maxIter = 100L, tol = (10^-4), bestOnly = FALSE, verbose = FALSE,
  interactive = TRUE)

Arguments

Y	the input matrix
ks	a vector of bicluster quantities to consider
holdouts	the number of row and column partitions. The true number of holdouts will be holdouts ^ 2.
maxIter	maximum number of iterations
tol	tolerance used to determine convergence
bestOnly	if FALSE, both the predicted number of biclusters and a table of result counts is returned
verbose	provide output after each iteration
interactive	prompt before running bcv on matrices with missing values

Details

The highest k tested is limited to (\code{holdouts} - 1) / \code{holdouts} * \min(m, n) for Y_{m,n}. A warning will be issued if not all ks can be tested.

Value

if bestOnly = FALSE, the predicted bicluster quantity. if bestOnly = TRUE, a list containing:

best

the predicted bicluster quantity

counts

a named table of result counts

See Also

bcv()

Examples

auto_bcv(yeast_benchmark[[1]])

jonalim/mfBiclust documentation built on May 4, 2019, 4:13 a.m.