Combined kernel and rank selection via pre-screening and LOSO CV

Runs kernel alignment pre-screening followed by LOSO explained-variance cross-validation, applying the one-standard-error rule to pick a kernel/rank pair.

Usage

dkge_cv_kernel_rank(
  B_list,
  X_list,
  K_grid,
  ranks,
  Omega_list = NULL,
  ridge = 0,
  w_method = "mfa_sigma1",
  w_tau = 0.3,
  top_k = 3
)

Arguments

B_list

List of qxP subject beta matrices.

X_list

List of Txq subject design matrices.

K_grid

Named list of candidate kernels.

ranks

Integer vector of ranks to evaluate.

Omega_list

Optional list of spatial weights.

ridge

Optional ridge parameter passed to [dkge_fit()].

w_method

Subject-level weighting scheme passed to [dkge_fit()].

w_tau

Shrinkage parameter toward equal weights passed to [dkge_fit()].

top_k

Number of kernels to keep after pre-screening.

Value

List with the selected `kernel` and `rank`, alignment and CV tables, and a per-kernel summary of scores at the selected rank.

Examples

# \donttest{
toy <- dkge_sim_toy(
  factors = list(cond = list(L = 3)),
  active_terms = "cond", S = 4, P = 15, snr = 5
)
q <- nrow(toy$K)
K_grid <- list(base = toy$K, identity = diag(q))
sel <- dkge_cv_kernel_rank(toy$B_list, toy$X_list, K_grid, ranks = 1:2)
sel$pick
#> $kernel
#> [1] "base"
#> 
#> $rank
#> [1] 2
#> 
# }