shapiq.explainer.AgnosticExplainer¶

class shapiq.explainer.AgnosticExplainer(game, *, n_players=None, index='SV', max_order=1, approximator='auto', random_state=None)[source]¶

Bases: Explainer

Agnostic Explainer for shapiq.

This explainer is used to explain models that do not have a specific implementation in shapiq. It uses the game-based approach to explain the model’s predictions.

Initialize the AgnosticExplainer.

Parameters:

  • game (Game | Callable[[ndarray], ndarray]) – The cooperative game to be explained. This can be an instance of shapiq.games.base.Game or a callable value function that expects a one-hot matrix of coalitions as input and returns the value of the coalition.

  • n_players (int | None) – The number of players in the game. If not provided, it will be inferred from the shapiq.games.base.Game.

  • index (Literal['SV', 'SII', 'k-SII', 'STII', 'FSII', 'BV', 'BII', 'FBII']) – The type of game-theoretic index to be used for the explanation. Defaults to "SV".

  • max_order (int) – The maximum order of interactions to be computed. Set to 1 for no interactions (i.e, for Shapley values "SV" or Banzhaf values "BV").

  • approximator (Union[Approximator[Literal['SV', 'SII', 'k-SII', 'STII', 'FSII', 'BV', 'BII', 'FBII']], Literal['auto'], Literal['spex', 'montecarlo', 'svarm', 'permutation', 'regression']]) – The approximator to use for the game-based approach. Defaults to “auto”, which will automatically select the appropriate approximator based on the index and max_order. Other options include “regression”, “spex”, “svarm”, “montecarlo”, and “permutation”. Can also be an instance of shapiq.approximator._base.Approximator.

  • random_state (int | None) – The random state to use for reproducibility. Defaults to None.

Raises:

ValueError – If the game is not an instance of shapiq.games.base.Game and n_players is not specified.

explain_function(budget, *, x=None, random_state=None, **kwargs)[source]¶

Explain the function using the game-based approach.

Parameters:

  • budget (int) – The budget used for the approximation / computation of interaction values.

  • x (ndarray | None) – An optional data point to explain. This is only usable if the game is a Imputer. If provided, the imputer will be fitted to this data point before computing the interaction values. Defaults to None.

  • random_state (int | None) – An optional random state for reproducibility. Defaults to None. If None, no random state is set for the game or approximator.

  • **kwargs (Any) – Additional keyword arguments (not used, only for compatibility).

Returns:

The computed interaction values.

Return type:

InteractionValues

game: Game | Callable[[ndarray], ndarray]¶

The cooperative game to be explained, either as a Game instance or a callable value function.