crossValRepWithGs3: Replicated K-fold cross-validation
In INRA/rgs3: R package wrapping the GS3 program for genomic selection

Description Usage Arguments Value Author(s) See Also

Perform replicated K-fold cross-validation with GS3, i.e. call crossValWithGs3 several times, with different seeds.

crossValRepWithGs3(genos, dat, config, task.id = "GS3",
  binary.trait = FALSE, ped.file = "", afs = NULL, nb.reps = 50,
  seed = NULL, nb.cores.rep = 1, nb.folds = 10, remove.files = "some",
  nb.cores.fold = 1, cl = NULL, verbose = 1)

`genos`	matrix of SNP genotypes
`dat`	data frame with phenotypes
`config`	list containing the configuration for GS3
`task.id`	character containing the task identifier used as prefix for the output files (for each fold, its index will be added)
`binary.trait`	logical
`ped.file`	path to the file containing the pedigree (if not used, use NA or "" instead of NULL)
`afs`	vector of allele frequencies which names are SNP identifiers (column names of `genos`); if NULL, will be estimated from `genos`; used to compute the variance of additive genotypic values from the variance of additive SNP effects (see Vitezica et al, 2013), then used to compute narrow-sense heritability
`nb.reps`	number of replicates (a set of folds will be sampled for each replicate)
`seed`	if not NULL, this seed for the pseudo-random number generator will be used to sample as many seeds as the number of replicates, these new seeds being used to shuffle genotypes before partitioning per fold
`nb.cores.rep`	number of cores to launch replicates in parallel (via `mclapply`, on Unix-like computers)
`nb.folds`	number of folds
`remove.files`	remove files per fold (none/some/all); use `"some"` in real-life applications in order to keep estimates of SNP effects per fold, thereby allowing to perform genomic prediction afterwards by averaging them
`nb.cores.fold`	number of cores to launch folds in parallel (via `mclapply`, on Unix-like computers, or `parLapply` on Windows); you can also use `detectCores`; will be set to 1 if `nb.cores.rep` is bigger than 1
`cl`	object returned by `makeCluster`, necessary only if `nb.cores.fold > 1` and the computer runs Windows; if NULL in such cases, will be created silently
`verbose`	verbosity level (0/1/2); there will be a progress bar only for `verbose=1`