ageas.n_iter_boost_selection

ageas.n_iter_boost_selection(hangar: Hangar, operation_name: str = 'boost_selection', accelerator: str = 'cpu', cuda_devices: list = None, query_dataset=None, test_dataset=None, max_boost_iter: int = 1, extract_ratio: float = 0.5, extract_top_n: int = 1000, seed: int = 42, verbose: bool = None, selection_args: dict = None, explain_args: dict = None) → tuple

N-iteration boost selection driven by per-class explanation scores.

Each iteration runs n_kfold_selection(), debriefs the surviving deck, and trims query_dataset (and test_dataset if provided) to the top features per class. The number of features kept in iteration i is roughly extract_top_n / extract_ratio**(max_boost_iter - 1 - i), converging to extract_top_n in the final round. After the iterative pruning a final selection is run on the trimmed feature set.

Parameters:

  • hangar – Source hangar from which the operating squad is generated.

  • operation_name – Operation name under which per-iteration reports are stored.

  • accelerator – Accelerator hint forwarded to selection ('cpu' or 'cuda').

  • cuda_devices – Optional GPU device indices.

  • query_dataset – Required dataset whose feature axis is iteratively pruned.

  • test_dataset – Optional held-out test dataset, pruned in lock-step with query_dataset.

  • max_boost_iter – Number of boost iterations.

  • extract_ratio – Geometric ratio of feature shrinkage between iterations.

  • extract_top_n – Total number of features kept after the last boost iteration.

  • seed – Random seed forwarded to selection.

  • verbose – If True, emit per-iteration feature count logs.

  • selection_args – Extra keyword arguments forwarded to n_kfold_selection().

  • explain_args – Extra keyword arguments forwarded to debrief().

Returns:

(deck, fea_keep) — the final deck after the post-boost selection and the list of feature IDs retained.