shapiq.approximator.moebius_converter#

Classes

MoebiusConverter(N, moebius_coefficients)

Computes exact Shapley Interactions using the (sparse) Möbius representation.
class shapiq.approximator.moebius_converter.MoebiusConverter(N, moebius_coefficients)[source]#

Bases: object

Computes exact Shapley Interactions using the (sparse) Möbius representation. Much faster than exact computation, if Möbius representation is sparse.

Parameters:

  • N (set) – The set of players.

  • moebius_coefficients (InteractionValues) – An InteractionValues objects containing the (sparse) Möbius coefficients

N#

The set of players

n#

The number of players.

moebius_coefficients#

The InteractionValues object containing all non-zero (sparse) Möbius coefficients

n_interactions#

A pre-computed array containing the number of interactions up to the size of the index, e.g. n_interactions[4] is the number of interactions up to size 4.

base_aggregation(base_interactions, order)[source]#

Transform Base Interactions into Interactions satisfying efficiency, e.g. SII to k-SII

Parameters:

  • base_interactions (InteractionValues) – InteractionValues object containing interactions up to order “order”

  • order (int) – The highest order of interactions considered

Returns:

InteractionValues object containing transformed base_interactions

fsii_routine(order)[source]#

Computes STII. Routine to distribute the Moebius coefficients onto all interactions for FSII. The higher-order interactions (size > order) are distributed onto all FSII interactions (size <= order).

Parameters:

order (int) – The order of the explanation

Returns:

An InteractionValues object containing the FSII interactions

get_moebius_distribution_weight(moebius_size, interaction_size, order, index)[source]#

Return the distribution weights for the Möbius coefficients onto the lower-order interaction indices.

Parameters:

  • moebius_size (int) – The size of the Möbius coefficient

  • interaction_size (int) – The size of the interaction

  • order (int) – The order of the explanation

  • index (str) – The interaction index, e.g. SII, k-SII, FSII…

Returns:

A distribution weight for the given combination

moebius_to_base_interaction(order, index)[source]#

Computes a base interaction index, e.g. SII or BII

Parameters:

  • order (int) – The order of the explanation

  • index (str) – The base interaction index, e.g. SII, BII

Returns:

An InteractionValues object containing the base interactions

moebius_to_shapley_interaction(index, order)[source]#

Converts the Möbius coefficients to Shapley Interactions up to order k

Parameters:

  • index (str) – The Shapley Interaction index, e.g. k-SII, STII, FSII

  • order (int) – The order of the explanation

Returns:

An InteractionValues object containing the Shapley interactions

stii_routine(order)[source]#

Computes STII. Routine to distribute the Moebius coefficients onto all interactions for STII. The lower-order interactions are equal to their Moebius coefficients, whereas the top-order interactions contain the distributed higher-order interactions.

Parameters:

order (int) – The order of the explanation

Returns:

An InteractionValues object containing the STII interactions