lux.metrics.stability

lux.metrics.stability(rules_and_instances, dataset, features, categorical_indicator)

Calculate stability metrics for a set of rules and instances.

Parameters:

  • rules_and_instances (pandas.DataFrame) – pandas.DataFrame A DataFrame containing rules and instances. It should include at least the following columns: - ‘true_class’: The true class label for each instance. - ‘explain_instance’: The explanation instance corresponding to each rule.

  • dataset (pandas.DataFrame) – pandas.DataFrame The dataset on which the rules are applied. It should be a pandas DataFrame.

  • features (list) – list A list of feature names in the dataset.

  • categorical_indicator (list) – list A list indicating whether each feature is categorical or not. Each element of the list corresponds to a feature in features, with True indicating the feature is categorical and False indicating it is not.

Returns:

tuple of floats A tuple containing the following stability metrics: - Mean stability of rules for each true class label. - Mean standard deviation of stability for each true class label. - Mean feature Jaccard similarity coefficient between rules. - Standard deviation of feature Jaccard similarity coefficient between rules.

Return type:

tuple

Notes:

  • Stability is calculated based on the Jaccard similarity coefficient between rules applied to similar instances.

  • Feature Jaccard similarity coefficient measures the similarity of features between rules.

  • Large stability and low variance are desired for stable explanations.