facet.explanation.base.BaseExplainer#

class facet.explanation.base.BaseExplainer(*args, **kwargs)[source]#

Abstract base class of SHAP explainers, providing stubs for methods used by FACET but not consistently supported by class shap.Explainer across different versions of the shap package.

Provides unified support for the old and new explainer APIs:

The old API uses methods shap_values() and shap_interaction_values() to compute SHAP values and interaction values, respectively. They return numpy arrays for single-output or single-class models, and lists of numpy arrays for multi-output or multi-class models.
The new API introduced in shap 0.36 makes explainer objects callable; direct calls to an explainer object return an Explanation object that contains the SHAP values and interaction values. For multi-output or multi-class models, the array has an additional dimension for the outputs or classes as the last axis.

As of shap 0.36, the old API is deprecated for the majority of explainers while the shap.KernelExplainer still uses the old API exclusively in shap 0.41. We remedy this by adding support for both APIs to all explainers created through an ExplainerFactory object.

Bases

Explainer

Metaclasses

ABCMeta

Parameters

args (Any) – positional arguments passed to the explainer constructor
kwargs (Any) – keyword arguments passed to the explainer constructor

Method summary

`explain_row`	See `shap.explainers.Explainer.explain_row()`
`load`	See `shap.explainers.Explainer.load()`
`save`	See `shap.explainers.Explainer.save()`
`shap_interaction_values`	Estimate the SHAP interaction values for a set of samples.
`shap_values`	Estimate the SHAP values for a set of samples.
`supports_model_with_masker`	See `shap.explainers.Explainer.supports_model_with_masker()`

Attribute summary

supports_interaction

True if the explainer supports interaction effects, False otherwise.

Definitions

__call__(*args, max_evals='auto', main_effects=False, error_bounds=False, batch_size='auto', outputs=None, silent=False, **kwargs)#: See shap.explainers.Explainer.__call__()

explain_row(*row_args, max_evals, main_effects, error_bounds, outputs, silent, **kwargs)#: See shap.explainers.Explainer.explain_row()

classmethod load(in_file, model_loader=<bound method Model.load of <class 'shap.models.Model'>>, masker_loader=<bound method Serializable.load of <class 'shap.maskers.Masker'>>, instantiate=True)#: See shap.explainers.Explainer.load()

save(out_file, model_saver='.save', masker_saver='.save')#: See shap.explainers.Explainer.save()

shap_interaction_values(X, y=None, **kwargs)[source]#

Estimate the SHAP interaction values for a set of samples.

Parameters

X (Union[ndarray[Any, dtype[Any]], DataFrame, Pool]) – matrix of samples (# samples x # features) on which to explain the model’s output
y (Union[ndarray[Any, dtype[Any]], Series, None]) – array of label values for each sample, used when explaining loss functions (optional)
kwargs (Any) – additional arguments specific to the explainer implementation

Return type

Union[ndarray[Any, dtype[float64]], List[ndarray[Any, dtype[float64]]]]

Returns

SHAP values as an array of shape \((n_\mathrm{observations}, n_\mathrm{features}, n_\mathrm{features})\); a list of such arrays in the case of a multi-output model

shap_values(X, y=None, **kwargs)[source]#

Estimate the SHAP values for a set of samples.

Parameters

X (Union[ndarray[Any, dtype[Any]], DataFrame, Pool]) – matrix of samples (# samples x # features) on which to explain the model’s output
y (Union[ndarray[Any, dtype[Any]], Series, None]) – array of label values for each sample, used when explaining loss functions (optional)
kwargs (Any) – additional arguments specific to the explainer implementation

Return type

Union[ndarray[Any, dtype[float64]], List[ndarray[Any, dtype[float64]]]]

Returns

SHAP values as an array of shape (n_observations, n_features); a list of such arrays in the case of a multi-output model

static supports_model_with_masker(model, masker)#: See shap.explainers.Explainer.supports_model_with_masker()

abstract property supports_interaction: bool#

True if the explainer supports interaction effects, False otherwise.

Return type: bool

facet.explanation.base

facet.explanation.base.ExplainerFactory