Anchor-based DKGE input descriptor — dkge_input

Builds an input specification that projects subject item kernels onto a shared anchor basis before fitting DKGE. Use this object with [dkge_fit_from_input()] or supply it to [dkge_pipeline()] via the `input` argument. The descriptor is immutable; downstream calls may extend its `dkge_args` field with additional DKGE fitting options.

Usage

dkge_input_anchor(
  features_list,
  K_item_list,
  folds = NULL,
  anchors = list(),
  design_kernel = NULL,
  dkge_args = list()
)

Arguments

features_list: List of subject feature matrices (n_s x d).
K_item_list: List of subject item kernels (n_s x n_s PSD).
folds: Optional fold structure passed to [dkge_build_anchor_kernels()].
anchors: Optional list overriding anchor selection defaults (see [dkge_anchor_fit()]).
design_kernel: Optional design kernel supplied to [dkge_fit_from_kernels()]. Defaults to the identity in effect space.
dkge_args: Optional list of arguments forwarded to the DKGE fitter after anchor kernels are constructed (e.g. `w_method`, `cpca_part`).

Value

Object of class `dkge_input_anchor` (inherits from `dkge_input`).

Examples

set.seed(1)
features_list <- list(
  s1 = matrix(rnorm(20 * 4), 20, 4),
  s2 = matrix(rnorm(25 * 4), 25, 4),
  s3 = matrix(rnorm(22 * 4), 22, 4)
)
K_item_list <- lapply(features_list, function(X) tcrossprod(matrix(rnorm(nrow(X) * 3), nrow(X), 3)))
input <- dkge_input_anchor(features_list, K_item_list, anchors = list(L = 6, seed = 1))
class(input)
#> [1] "dkge_input_anchor" "dkge_input"