Subject-level projection bootstrap in medoid space — dkge_bootstrap

Resamples transported subject vectors (already aligned in the medoid parcellation) to quantify between-subject variability without recomputing the group basis.

Usage

dkge_bootstrap_projected(
  values_medoid,
  B = 1000L,
  aggregate = c("mean", "median"),
  weights = NULL,
  seed = NULL,
  voxel_operator = NULL,
  return_samples = TRUE
)

Arguments

values_medoid: List of length `S` where each element is a numeric vector defined on the medoid parcellation (e.g. LOSO contrast values).
B: Number of bootstrap replicates.
aggregate: Aggregation function applied to the resampled subjects (`"mean"` or `"median"`).
weights: Optional subject weights applied when computing the resampled mean. Only used when `aggregate = "mean"`.
seed: Optional random seed for reproducibility.
voxel_operator: Optional matrix that maps medoid vectors to voxel space (columns = voxels). When supplied, summaries in voxel space are also returned.
return_samples: Logical; when `TRUE` the matrix of bootstrap samples is returned in the output bundle.

Value

A list containing bootstrap summaries (`mean`, `sd`, `z`, confidence intervals), and optionally the raw bootstrap draws (medoid and voxel space).

Examples

# \donttest{
toy <- dkge_sim_toy(
  factors = list(A = list(L = 2), B = list(L = 3)),
  active_terms = c("A", "B"), S = 5, P = 20, snr = 5
)
fit <- dkge_fit(toy$B_list, toy$X_list, toy$K, rank = 2)
# Bootstrap requires transport setup - example shows API
# }