Prepare subject-to-medoid transport operators for reuse

Computes and caches subject-to-medoid transport matrices so downstream routines (e.g. bootstraps) can reuse a fixed consensus mapping without re-solving the transport problem on every call.

Usage

dkge_prepare_transport(
  fit,
  centroids = NULL,
  loadings = NULL,
  betas = NULL,
  sizes = NULL,
  mapper = "sinkhorn",
  medoid = 1L,
  ...
)

Arguments

fit: A `dkge` object.
centroids: List of subject centroid matrices. Defaults to the centroids stored on `fit` or `fit$input`.
loadings: Optional list of subject loadings (`P_s x r`). When omitted, they are recomputed from `fit$Btil` or the supplied `betas`.
betas: Optional list of subject betas used to recompute loadings when `loadings` is `NULL`.
sizes: Optional list of cluster masses.
mapper: Mapper specification or shorthand passed to [dkge_mapper_spec()].
medoid: Index (1-based) of the reference subject.
...: Additional mapper arguments such as `epsilon` or `lambda_spa`.

Value

A list containing cached transport objects: `operators`, `mapper_spec`, `feature_list`, `size_list`, `feature_ref`, `size_ref`, `centroids`, and `medoid`.