mllm_shap.shap.neyman package#

Submodules#

mllm_shap.shap.neyman.limited module#

Limited Neyman approximation SHAP explainer implementation.

class mllm_shap.shap.neyman.limited.LimitedComplementaryNeymanShapExplainer(*args: Any, initial_num_samples: int | None = None, initial_fraction: float | None = None, **kwargs: Any)[source]#

Bases: BaseComplementaryNeymanShapExplainer

Limited Neyman SHAP implementation.

allow_mask_duplicates#: Whether to allow duplicate masks during generation.

embedding_model#: The external embedding model to use. If provided, overrides mode.

embedding_reducer#: The embedding reduction strategy to use.

fraction#: Fraction of total possible masks to generate if num_samples is None.

include_minimal_masks = True#: Whether to include minimal masks (single-feature and empty masks) in the sampling.

initial_fraction#: Initial fraction of samples to draw in the first step.

initial_num_samples#: Initial number of samples to draw in the first step.

initial_steps#: Number of initial steps performed in last call.

last_observability_sink = None#: Sink used for the most recent call (if observability was enabled).

mode#: The SHAP mode, either STATIC or CONTEXTUAL. Used if no embedding_model is provided.

normalizer#: The SHAP value normalizer to use.

num_samples#: Number of random masks to generate. If None, uses fraction. -1 stands for minimal number of samples (only single-feature masks and empty mask).

similarity_measure#: The embedding similarity measure to use.

total_n_calls = 0#: Total number of MLLM calls made for last explanation.

use_standard_method = False#: Whether to use the standard method for initial sampling. Default is False, which uses the modified method with pre-defined members.

mllm_shap.shap.neyman.standard module#

Standard Neyman approximation SHAP explainer implementation.

class mllm_shap.shap.neyman.standard.StandardComplementaryNeymanShapExplainer(*args: Any, initial_num_samples: int | None = None, initial_fraction: float | None = None, **kwargs: Any)[source]#

Bases: BaseComplementaryNeymanShapExplainer

Standard Neyman SHAP Explainer.

allow_mask_duplicates#: Whether to allow duplicate masks during generation.

embedding_model#: The external embedding model to use. If provided, overrides mode.

embedding_reducer#: The embedding reduction strategy to use.

fraction#: Fraction of total possible masks to generate if num_samples is None.

include_minimal_masks = True#: Whether to include minimal masks (single-feature and empty masks) in the sampling.

initial_fraction#: Initial fraction of samples to draw in the first step.

initial_num_samples#: Initial number of samples to draw in the first step.

initial_steps#: Number of initial steps performed in last call.

last_observability_sink = None#: Sink used for the most recent call (if observability was enabled).

mode#: The SHAP mode, either STATIC or CONTEXTUAL. Used if no embedding_model is provided.

normalizer#: The SHAP value normalizer to use.

num_samples#: Number of random masks to generate. If None, uses fraction. -1 stands for minimal number of samples (only single-feature masks and empty mask).

similarity_measure#: The embedding similarity measure to use.

total_n_calls = 0#: Total number of MLLM calls made for last explanation.

use_standard_method: bool = True#: Whether to use the standard method for initial sampling. Default is False, which uses the modified method with pre-defined members.

Module contents#

Neyman SHAP explainers module.

LimitedComplementaryNeymanShapExplainer implements a limited Neyman sampling
that samples initial masks of given size with pre-defined member.
StandardComplementaryNeymanShapExplainer does standard Neyman sampling
that samples initial masks of given size randomly.

class mllm_shap.shap.neyman.LimitedComplementaryNeymanShapExplainer(*args: Any, initial_num_samples: int | None = None, initial_fraction: float | None = None, **kwargs: Any)[source]#

Bases: BaseComplementaryNeymanShapExplainer

Limited Neyman SHAP implementation.

allow_mask_duplicates#: Whether to allow duplicate masks during generation.

embedding_model#: The external embedding model to use. If provided, overrides mode.

embedding_reducer#: The embedding reduction strategy to use.

fraction#: Fraction of total possible masks to generate if num_samples is None.

include_minimal_masks = True#: Whether to include minimal masks (single-feature and empty masks) in the sampling.

initial_fraction#: Initial fraction of samples to draw in the first step.

initial_num_samples#: Initial number of samples to draw in the first step.

initial_steps#: Number of initial steps performed in last call.

last_observability_sink = None#: Sink used for the most recent call (if observability was enabled).

mode#: The SHAP mode, either STATIC or CONTEXTUAL. Used if no embedding_model is provided.

normalizer#: The SHAP value normalizer to use.

num_samples#: Number of random masks to generate. If None, uses fraction. -1 stands for minimal number of samples (only single-feature masks and empty mask).

similarity_measure#: The embedding similarity measure to use.

total_n_calls = 0#: Total number of MLLM calls made for last explanation.

use_standard_method = False#: Whether to use the standard method for initial sampling. Default is False, which uses the modified method with pre-defined members.

class mllm_shap.shap.neyman.StandardComplementaryNeymanShapExplainer(*args: Any, initial_num_samples: int | None = None, initial_fraction: float | None = None, **kwargs: Any)[source]#

Bases: BaseComplementaryNeymanShapExplainer

Standard Neyman SHAP Explainer.

allow_mask_duplicates#: Whether to allow duplicate masks during generation.

embedding_model#: The external embedding model to use. If provided, overrides mode.

embedding_reducer#: The embedding reduction strategy to use.

fraction#: Fraction of total possible masks to generate if num_samples is None.

include_minimal_masks = True#: Whether to include minimal masks (single-feature and empty masks) in the sampling.

initial_fraction#: Initial fraction of samples to draw in the first step.

initial_num_samples#: Initial number of samples to draw in the first step.

initial_steps#: Number of initial steps performed in last call.

last_observability_sink = None#: Sink used for the most recent call (if observability was enabled).

mode#: The SHAP mode, either STATIC or CONTEXTUAL. Used if no embedding_model is provided.

normalizer#: The SHAP value normalizer to use.

num_samples#: Number of random masks to generate. If None, uses fraction. -1 stands for minimal number of samples (only single-feature masks and empty mask).

similarity_measure#: The embedding similarity measure to use.

total_n_calls = 0#: Total number of MLLM calls made for last explanation.

use_standard_method: bool = True#: Whether to use the standard method for initial sampling. Default is False, which uses the modified method with pre-defined members.