rgcca_stability: Stability selection for SGCCA

In Tenenhaus/RGCCA: Regularized (or Sparse) Generalized Canonical Correlation Analysis (R/SGCCA) for multi-block data analysis

Stability selection for SGCCA

Description

This function can be used to identify the most stable variables identified as relevant by SGCCA. A Variable Importance in the Projection (VIP) based criterion is used to identify the most stable variables.

Usage

rgcca_stability(
  rgcca_res,
  keep = vapply(rgcca_res$a, function(x) mean(x != 0), FUN.VALUE = 1),
  n_boot = 100,
  n_cores = 1,
  verbose = TRUE,
  balanced = TRUE,
  keep_all_variables = FALSE
)

Arguments

rgcca_res	A fitted RGCCA object (see rgcca)
keep	numeric vector indicating the proportion of top variables per block.
n_boot	Number of bootstrap samples (Default: 100).
n_cores	Number of cores for parallelization.
verbose	Logical value indicating if the progress of the procedure is reported.
balanced	A boolean indicating if a balanced bootstrap procedure is performed or not (default is TRUE).
keep_all_variables	A boolean indicating if all variables have to be kept even when some of them have null variance for at least one bootstrap sample (default is FALSE).

Value

top	indicator on which variables are ranked.
keepVar	indices of the top variables.
bootstrap	block-weight vectors for ech bootstrap sample.
rgcca_res	an RGCCA object fitted on the most stable variables.

Examples

## Not run: 
###########################
# stability and bootstrap #
###########################

data("ge_cgh_locIGR", package = "gliomaData")
blocks <- ge_cgh_locIGR$multiblocks
Loc <- factor(ge_cgh_locIGR$y)
levels(Loc) <- colnames(ge_cgh_locIGR$multiblocks$y)
blocks[[3]] <- Loc


fit.sgcca <- rgcca(blocks,
  sparsity = c(.071, .2, 1),
  ncomp = c(1, 1, 1),
  scheme = "centroid",
  verbose = TRUE, response = 3
)

boot.out <- rgcca_bootstrap(fit.sgcca, n_boot = 100, n_cores = 1)

fit.stab <- rgcca_stability(fit.sgcca,
  keep = sapply(fit.sgcca$a, function(x) mean(x != 0)),
  n_cores = 1, n_boot = 10,
  verbose = TRUE
)

boot.out <- rgcca_bootstrap(
  fit.stab, n_boot = 500, n_cores = 1, verbose = FALSE
)

## End(Not run)

Tenenhaus/RGCCA documentation built on March 16, 2023, 2:04 p.m.