End-to-end DKGE workflow — dkge

Fits DKGE (if needed), computes cross-fitted contrasts, optionally transports them to a medoid parcellation, and performs sign-flip inference.

Usage

dkge_pipeline(
  fit = NULL,
  input = NULL,
  betas = NULL,
  designs = NULL,
  kernel = NULL,
  omega = NULL,
  contrasts,
  transport = NULL,
  inference = list(),
  classification = NULL,
  method = c("loso", "kfold", "analytic"),
  ridge = 0,
  ...
)

Arguments

fit: Optional pre-computed `dkge` object. If `NULL`, provide `betas`, `designs`, and `kernel` to fit inside the pipeline.
input: Optional DKGE input descriptor created with [dkge_input_anchor()] or future helpers. When supplied (and `fit` is `NULL`), `dkge_pipeline()` will build the fit via [dkge_fit_from_input()].
betas, designs, kernel: Inputs passed to [dkge()] when neither `fit` nor `input` is supplied.
omega: Optional spatial weights forwarded to [dkge()].
contrasts: Contrast specification as accepted by [dkge_contrast()].
transport: Either a transport specification/service or `NULL`.
inference: Either an inference specification/service or `NULL`.
classification: Optional specification passed to [dkge_classify()].
method: Cross-fitting strategy for contrasts (default "loso").
ridge: Optional ridge added during held-out decompositions.
...: Additional arguments passed to [dkge()] when fitting inside the pipeline, or to [dkge_contrast()].

Value

List containing the fit, diagnostics, raw contrast values, transported maps (if requested), and inference results.

Examples

# Simulate toy data
toy <- dkge_sim_toy(
  factors = list(A = list(L = 2), B = list(L = 3)),
  active_terms = c("A", "B"), S = 5, P = 25, snr = 5
)

# Run pipeline with LOSO contrasts
result <- dkge_pipeline(
  betas = toy$B_list,
  designs = toy$X_list,
  kernel = toy$K,
  contrasts = c(1, rep(0, 4)),  # first effect
  method = "loso"
)
names(result)
#> [1] "fit"            "diagnostics"    "contrasts"      "transport"     
#> [5] "inference"      "classification"