Alternative explanations and ensured filters

Alternative explanations show how feature changes may change predictions.

Semantics and non-guarantees are defined in Calibrated interval semantics.

Generate alternatives

alternatives = explainer.explore_alternatives(X_query)
alt0 = alternatives[0]

For probabilistic regression:

alternatives = explainer.explore_alternatives(X_query, threshold=150.0)

Ensured filters

The alternative object supports these filters:

  • super_explanations()

  • semi_explanations()

  • counter_explanations()

  • ensured_explanations()

  • pareto_explanations()

Plotting and conjunctions

alt0.add_conjunctions(n_top_features=5, max_rule_size=2)
alt0.plot(style="triangular", show=False)