Papers Video GitHub In-the-Loop EXPEKCTATION Release Notes About

Time Limit

Some explanations are computationally expensive to compute. For this reason, most explanation methods accept a time_limit parameter (in seconds). If the time limit is reached, the method stops and returns either an approximation of the desired explanation or [] if no approximation was found. By default, time_limit is set to None, giving an unlimited amount of time.

All explainers expose an elapsed_time attribute set to the time in seconds taken by the last explanation method call. It is equal to Explaining.TIMEOUT if the time limit was reached.

The following methods accept a time_limit parameter:

Method	DT	RF	BT	BT (regression)
`sufficient_reason`	✓	✓		✓
`minimal_sufficient_reason`	✓	✓
`preferred_sufficient_reason`	✓
`n_sufficient_reasons`	✓
`n_sufficient_reasons_per_attribute`	✓
`anchored_reason`	✓
`minimal_contrastive_reason`		✓	✓
`majoritary_reason`		✓
`preferred_majoritary_reason`		✓
`minimal_majoritary_reason`		✓
`most_anchored_reason`		✓
`tree_specific_reason`			✓	✓
`minimal_tree_specific_reasons`			✓
`range_for_partial_instance`				✓

The following example illustrates the usage of time_limit and elapsed_time:

from pyxai import Learning, Explaining

learner = Learning.Scikitlearn("../../dataset/iris.csv", problem_type=Learning.CLASSIFICATION)
model = learner.evaluate(splitting_method=Learning.HOLD_OUT, model_type=Learning.DT)
instance, prediction = learner.get_instances(model, n=1, is_correct=True)
explainer = Explaining.initialize(model, instance)

sufficient_reason = explainer.sufficient_reason(time_limit=10)
if explainer.elapsed_time == Explaining.TIMEOUT:
    print("Timeout reached.")
    print(f"sufficient_reason: {sufficient_reason} (approximation)")
else:
    print(f"Time to compute: {explainer.elapsed_time:.3f}s")
    print(f"sufficient_reason: {sufficient_reason}")