Learns a source->target mapping with grouped ridge domain adaptation and evaluates target-domain representational geometry (feature-RSA style) on held-out target folds. This model is intended for settings where source rows have higher SNR/coverage and target rows are the analysis domain of interest.
Usage
feature_rsa_da_model(
dataset,
design,
mode = c("stacked", "coupled"),
lambdas,
alpha_recall = 0.2,
alpha_target = NULL,
rho = 5,
recall_folds = NULL,
target_folds = NULL,
recall_nfolds = 5L,
target_nfolds = NULL,
recall_gap = 0L,
target_gap = NULL,
target_nperm = 0L,
target_perm_strategy = c("circular_shift", "block_shuffle"),
target_perm_block = NULL,
rsa_simfun = c("spearman", "pearson"),
return_diagnostics = FALSE,
...
)Arguments
- dataset
An `mvpa_dataset` with external `test_data`.
- design
A `feature_sets_design` carrying `X_train` plus either fixed `X_test` predictors or a fold-aware `target_builder`.
- mode
`"stacked"` or `"coupled"`.
- lambdas
Named ridge penalties per feature set.
- alpha_recall
Non-negative scalar weighting target-domain rows.
- alpha_target
Optional alias for `alpha_recall`; if provided, overrides.
- rho
Coupling strength for `mode = "coupled"`.
- recall_folds
Optional explicit test-domain folds (list of `train`/`test` indices).
- target_folds
Optional alias for `recall_folds`; if provided, overrides.
- recall_nfolds
Number of contiguous folds for single-run targets.
- target_nfolds
Optional alias for `recall_nfolds`; if provided, overrides.
- recall_gap
Non-negative purge gap (TRs) for single-run contiguous folds.
- target_gap
Optional alias for `recall_gap`; if provided, overrides.
- target_nperm
Number of single-run target permutations for a null model.
- target_perm_strategy
`"circular_shift"` or `"block_shuffle"` for target permutations.
- target_perm_block
Optional block size (TRs) for `"block_shuffle"`.
- rsa_simfun
Similarity for target RDM correlation, `"spearman"` or `"pearson"`.
- return_diagnostics
If TRUE, store fold/parameter diagnostics in fits.
- ...
Additional fields stored on the model spec.