alibi.explainers.anchors.anchor_explanation module

class alibi.explainers.anchors.anchor_explanation.AnchorExplanation(exp_type, exp_map)[source]

Bases: object

__init__(exp_type, exp_map)[source]

Class used to unpack the anchors and metadata from the explainer dictionary.

Parameters:
  • exp_type (str) – Type of explainer: tabular, text or image.

  • exp_map (dict) – Dictionary with the anchors and explainer metadata for an observation.

coverage(partial_index=None)[source]
Parameters:

partial_index (Optional[int]) – Get the result coverage until a certain index. For example, if the result has precisions [0.1, 0.5, 0.95] and partial_index=1, this will return 0.5.

Return type:

float

Returns:

coverage – Anchor coverage.

examples(only_different_prediction=False, only_same_prediction=False, partial_index=None)[source]
Parameters:
  • only_different_prediction (bool) – If True, will only return examples where the result makes a different prediction than the original model.

  • only_same_prediction (bool) – If True, will only return examples where the result makes the same prediction than the original model.

  • partial_index (Optional[int]) – Get the examples from the partial result until a certain index.

Return type:

Union[list, ndarray]

Returns:

Examples covered by result.

features(partial_index=None)[source]
Parameters:

partial_index (Optional[int]) – Get the result until a certain index. For example, if the result uses segment_labels=(1, 2, 3) and partial_index=1, this will return [1, 2].

Return type:

list

Returns:

segment_labels – Features used in the result conditions.

names(partial_index=None)[source]
Parameters:

partial_index (Optional[int]) – Get the result until a certain index. For example, if the result is (A=1, B=2, C=2) and partial_index=1, this will return ["A=1", "B=2"].

Return type:

list

Returns:

names – Names with the result conditions.

precision(partial_index=None)[source]
Parameters:

partial_index (Optional[int]) – Get the result precision until a certain index. For example, if the result has precisions [0.1, 0.5, 0.95] and partial_index=1, this will return 0.5.

Return type:

float

Returns:

precision – Anchor precision.