interpret_community.common.blackbox_explainer module

Defines the black box explainer API, which can either take in a black box model or function.

class interpret_community.common.blackbox_explainer.BlackBoxExplainer(model, is_function=False, model_task=<ModelTask.Unknown: 'unknown'>, **kwargs)

Bases: interpret_community.common.base_explainer.BaseExplainer, interpret_community.common.blackbox_explainer.BlackBoxMixin

The base class for black box models or functions.

Parameters:
  • model (model that implements sklearn.predict or sklearn.predict_proba or function that accepts a 2d ndarray) – The model to explain or function if is_function is True.
  • is_function (bool) – Default is false. Set to True if passing sklearn.predict or sklearn.predict_proba function instead of model.
class interpret_community.common.blackbox_explainer.BlackBoxMixin(model, is_function=False, model_task=<ModelTask.Unknown: 'unknown'>, **kwargs)

Bases: interpret_community.common.chained_identity.ChainedIdentity

Mixin for black box models or functions.

Parameters:
  • model (model that implements sklearn.predict or sklearn.predict_proba or function that accepts a 2d ndarray) – The model to explain or function if is_function is True.
  • is_function (bool) – Default is False. Set to True if passing sklearn.predict or sklearn.predict_proba function instead of model.
interpret_community.common.blackbox_explainer.add_prepare_function_and_summary_method(cls)

Decorate blackbox explainer to allow aggregating local explanations to global.

Adds two protected methods _function_subset_wrapper and _prepare_function_and_summary to the blackbox explainer. The former creates a wrapper around the prediction function for explaining subsets of features in the evaluation samples dataset. The latter calls the former to create a wrapper and also computes the summary background dataset for the explainer.

interpret_community.common.blackbox_explainer.init_blackbox_decorator(init_func)

Decorate a constructor to wrap initialization examples in a DatasetWrapper.

Provided for convenience for tabular data explainers.

Parameters:init_func (Initialization constructor.) – Initialization constructor where the second argument is a dataset.