Kernel alignment pre-screening — dkge_kernel_prescreen • dkge

Ranks candidate kernels by their alignment with the pooled design-space covariance produced by [dkge_pooled_cov_q()].

Usage

dkge_kernel_prescreen(K_grid, C, normalize_k = TRUE, top_k = 3)

Arguments

K_grid: Named list of qxq kernels.
C: Pooled covariance matrix.
normalize_k: Logical; if `TRUE`, kernels are scaled to unit trace before alignment.
top_k: Number of kernels to retain.

Value

Data frame sorted by decreasing alignment; the `top` attribute carries the names of the retained kernels.

Examples

toy <- dkge_sim_toy(
  factors = list(cond = list(L = 3)),
  active_terms = "cond", S = 3, P = 10, snr = 5
)
pooled <- dkge_pooled_cov_q(toy$B_list, toy$X_list)
q <- nrow(toy$K)
K_grid <- list(base = toy$K, identity = diag(q))
dkge_kernel_prescreen(K_grid, pooled$C, top_k = 1)
#>     kernel     align
#> 2 identity 0.5095343
#> 1     base 0.3682251