SHAPInsight

class baybe.insights.shap.SHAPInsight

Bases: object

Class for SHAP-based feature importance insights.

Also supports LIME and MAPLE explainers via the shap package.

Public methods

__init__(explainer, background_data)	Method generated by attrs for class SHAPInsight.
explain([data])	Compute a Shapley explanation for a given data set.
from_campaign(campaign[, explainer_cls, ...])	Create a SHAP insight from a campaign.
from_recommender(recommender, searchspace, ...)	Create a SHAP insight from a recommender.
from_surrogate(surrogate, data[, ...])	Create a SHAP insight from a surrogate.
plot(plot_type[, data, show, explanation_index])	Plot the Shapley values using the provided plot type.

Public attributes and properties

explainer	The explainer instance.
background_data	The background data set used by the explainer.
uses_shap_explainer	Indicates if a SHAP explainer is used or not (e.g. MAPLE, LIME).

__init__(explainer: Explainer, background_data: DataFrame)

Method generated by attrs for class SHAPInsight.

For details on the parameters, see Public attributes and properties.

explain(data: DataFrame | None = None, /)

Compute a Shapley explanation for a given data set.

Parameters:

data (Optional[DataFrame]) – The dataframe for which the Shapley values are to be computed. By default, the background data set of the explainer is used.

Return type:

Explanation

Returns:

The computed Shapley explanation.

Raises:

ValueError – If the columns of the given dataframe cannot be aligned with the columns of the explainer background dataframe.

classmethod from_campaign(campaign: Campaign, explainer_cls: type[Explainer] | str = 'KernelExplainer', *, use_comp_rep: bool = False)

Create a SHAP insight from a campaign.

Uses the measurements of the campaign as background data.

Parameters:

• campaign (Campaign) – A campaign using a surrogate-based recommender.

• explainer_cls (type[Explainer] | str) – See make_explainer_for_surrogate().

• use_comp_rep (bool) – See make_explainer_for_surrogate().

Return type:

SHAPInsight

Returns:

The SHAP insight object.

Raises:

ValueError – If the campaign does not contain any measurements.

classmethod from_recommender(recommender: RecommenderProtocol, searchspace: SearchSpace, objective: Objective, measurements: DataFrame, explainer_cls: type[Explainer] | str = 'KernelExplainer', *, use_comp_rep: bool = False)

Create a SHAP insight from a recommender.

Uses the provided measurements to train the surrogate and as background data for the explainer.

Parameters:

• recommender (RecommenderProtocol) – A recommender using a surrogate model.

• searchspace (SearchSpace) – The searchspace for the recommender.

• objective (Objective) – The objective for the recommender.

• measurements (DataFrame) – The measurements for training the surrogate and the explainer.

• explainer_cls (type[Explainer] | str) – See make_explainer_for_surrogate().

• use_comp_rep (bool) – See make_explainer_for_surrogate().

Return type:

SHAPInsight

Returns:

The SHAP insight object.

Raises:

TypeError – If the recommender has no get_surrogate method.

classmethod from_surrogate(surrogate: SurrogateProtocol, data: DataFrame, explainer_cls: type[Explainer] | str = 'KernelExplainer', *, use_comp_rep: bool = False)

Create a SHAP insight from a surrogate.

For details, see make_explainer_for_surrogate().

plot(plot_type: Literal['bar', 'beeswarm', 'force', 'heatmap', 'scatter'], data: DataFrame | None = None, /, *, show: bool = True, explanation_index: int | None = None, **kwargs: Any)

Plot the Shapley values using the provided plot type.

Parameters:

• plot_type (Literal['bar', 'beeswarm', 'force', 'heatmap', 'scatter']) – The type of plot to be created.

• data (Optional[DataFrame]) – See explain().

• show (bool) – Boolean flag determining if the plot is to be rendered.

• explanation_index (Optional[int]) – Positional index of the data point that should be explained. Only relevant for plot types that can only handle a single data point.

• **kwargs (Any) – Additional keyword arguments passed to the plot function.

Return type:

Axes

Returns:

The plot object.

Raises:

ValueError – If the provided plot type is not supported.

background_data: DataFrame

The background data set used by the explainer.

explainer: Explainer

The explainer instance.

property uses_shap_explainer: bool

Indicates if a SHAP explainer is used or not (e.g. MAPLE, LIME).