model.interpreter.explanation_aggregator module

class ExplanationAggregator(confidence_threshold=0.8)

Bases: object

Class for explanation aggregator. It aggregates the explanations based on classes, feature and scores.

feed(explanation: Dict)

Feed explanation into the aggregator for further analysis

Parameters

explanation – dict, the pre-defined format as the output in xai.explainer.utils.explanation_to_json

get_feature_names(list_explanations: List[Dict]) → Set

Get feature names for an explanation, plus schema validation :param list_explanations: List of explanations :type list_explanations: list

Returns

(set) feature names

get_statistics(stats_type: str = 'top_k', k: int = 5) → Tuple[Dict[int, Dict], int]

return statistics of explanations in the aggregator based on the type

Parameters

• stats_type –

  str, not None. The pre-defined types of statistics. For now, it supports 3 types:

  • top_k: how often a feature appears in the top K features in the explanation

  • average_score: average score for each feature in the explanation

  • average_ranking: average ranking for each feature in the explanation

  Default type is top_k.

• k – int, not None. the k value for top_k method and average_ranking. It will be ignored if the stats type are not top_k or average_ranking. Default value of k is 5.

Returns

A dictionary maps the label to its aggregated statistics. An integer to indicate the total number of explanations to generate the statistics.