Construct an ITEM design for trial-wise decoding

Creates a design object for ITEM-style decoding where the trial-wise design matrix (`X_t`) and supervised trial targets (`T_target`) are decoupled from TR-level observations used in searchlight/ROI extraction.

Usage

item_design(
  train_design = NULL,
  X_t,
  T_target,
  run_id,
  C_transform = NULL,
  V = NULL,
  v_type = c("cov", "precision"),
  trial_id = NULL,
  trial_hash = NULL,
  split_by = NULL,
  Z = NULL,
  Nuisance = NULL,
  meta = list(),
  ...
)

Arguments

train_design: Data frame/tibble describing TR-level observations. If `NULL`, a minimal design with one row per TR is created.
X_t: Numeric trial-wise design matrix (`n_time x n_trials`).
T_target: Trial-level supervised targets (matrix/data.frame/vector/factor). Length/rows must equal `n_trials`.
run_id: Trial-level run/session identifiers (length `n_trials`).
C_transform: Optional trial-to-condition transform matrix.
V: Optional temporal covariance/precision passed to `fmrilss::item_compute_u()`.
v_type: One of `"cov"` or `"precision"`.
trial_id: Optional trial identifiers.
trial_hash: Optional trial hash used by alignment guards.
split_by: Optional split variable passed to `mvpa_design()`.
Z: Optional nuisance matrix (`n_time x q`) used by LS-A (`fmrilss::lsa`).
Nuisance: Alias nuisance matrix for LS-A; if both provided, `Z` is used.
meta: Optional metadata list.
...: Reserved for forward compatibility.

Value

An object of class `c("item_design", "mvpa_design", "list")`.