Run Global (Whole-Brain) MVPA Analysis

Train a single classifier on ALL features (parcels or voxels) with cross-validation, and compute per-feature importance via Haufe et al. (2014) activation patterns.

Usage

run_global(model_spec, ...)

# S3 method for class 'mvpa_model'
run_global(
  model_spec,
  X = NULL,
  summary_fun = NULL,
  return_fits = FALSE,
  aggregation = c("mean", "sum", "maxabs"),
  preflight = c("warn", "error", "off"),
  ...
)

# Default S3 method
run_global(model_spec, ...)

# S3 method for class 'pattern_model'
run_global(
  model_spec,
  return_fits = isTRUE(model_spec$keep_fold_fits),
  refit = model_spec$refit,
  preflight = c("warn", "error", "off"),
  ...
)

Arguments

model_spec

An mvpa_model specification.

...

Additional arguments (currently unused).

X

Optional pre-computed T x P feature matrix. If NULL, extracted from model_spec$dataset via get_feature_matrix.

summary_fun

Function to summarize activation pattern matrix rows into a scalar importance per feature. Default: L2 norm.

return_fits

Retain the per-fold pattern_fit objects.

aggregation

How to aggregate multi-basis feature importance (default "mean").

preflight

Preflight validation level ("warn", "error", or "off"); the pattern model runs a lightweight specification check.

refit

Also fit the model on all training rows (descriptive fit).

Value

A global_mvpa_result object.

A global_mvpa_result object.

Architecture TODO

run_global currently uses a dedicated global CV/training pipeline rather than dispatching through fit_roi. This is intentional for now; future cleanup may unify global and ROI fitting interfaces.

Examples

# \donttest{
  ds <- gen_sample_dataset(c(5,5,5), 40, nlevels=2, blocks=3)
  cval <- blocked_cross_validation(ds$design$block_var)
  mdl <- load_model("sda_notune")
  mspec <- mvpa_model(mdl, ds$dataset, ds$design,
    "classification", crossval=cval)
  result <- run_global(mspec)
# }