Blocked cross-validated reconstruction error under genpca geometry

Evaluates reconstruction error by holding out contiguous time blocks and projecting hold-out data onto the k-dimensional column space learned on the training block, all under the whitened geometry.

Usage

blocked_cv_recon_error(X, A, M, k, nfold = 5L, block_frac = 0.2)

Arguments

X, A, M

Data and metrics as in whitened_matrix().

k

Number of components.

nfold

Number of folds (hold-out blocks).

block_frac

Fraction of time points per hold-out block.

Value

Data frame with per-fold errors and the mean.

See also

choose_rank_gd() for Gavish-Donoho approach, choose_rank_pa() for parallel analysis approach

Other rank selection: choose_rank_gd(), choose_rank_pa()

Examples

# \donttest{
set.seed(1)
X <- matrix(rnorm(500), nrow = 50, ncol = 10)
A <- Matrix::Diagonal(ncol(X))
M <- Matrix::Diagonal(nrow(X))
cv <- blocked_cv_recon_error(X, A, M, k = 3)
# }