def test_mean_decrease_impurity_importance_evaluator_with_infinite(
inf_value: float) -> None:
# The test ensures that trials with infinite values are ignored to calculate importance scores.
n_trial = 10
seed = 13
# Importance scores are calculated without a trial with an inf value.
study = create_study(sampler=RandomSampler(seed=seed))
study.optimize(objective, n_trials=n_trial)
evaluator = MeanDecreaseImpurityImportanceEvaluator(seed=seed)
param_importance_without_inf = evaluator.evaluate(study)
# A trial with an inf value is added into the study manually.
study.add_trial(
create_trial(
value=inf_value,
params={
"x1": 1.0,
"x2": 1.0,
"x3": 3.0
},
distributions={
"x1": FloatDistribution(low=0.1, high=3),
"x2": FloatDistribution(low=0.1, high=3, log=True),
"x3": FloatDistribution(low=2, high=4, log=True),
},
))
# Importance scores are calculated with a trial with an inf value.
param_importance_with_inf = evaluator.evaluate(study)
# Obtained importance scores should be the same between with inf and without inf,
# because the last trial whose objective value is an inf is ignored.
assert param_importance_with_inf == param_importance_without_inf
def test_mean_decrease_impurity_importance_evaluator_max_depth() -> None:
# Assumes that `seed` can be fixed to reproduce identical results.
study = create_study()
study.optimize(objective, n_trials=3)
evaluator = MeanDecreaseImpurityImportanceEvaluator(max_depth=1, seed=0)
param_importance = evaluator.evaluate(study)
evaluator = MeanDecreaseImpurityImportanceEvaluator(max_depth=2, seed=0)
param_importance_different_max_depth = evaluator.evaluate(study)
assert param_importance != param_importance_different_max_depth
def test_mean_decrease_impurity_importance_evaluator_n_trees() -> None:
# Assumes that `seed` can be fixed to reproduce identical results.
study = create_study(sampler=RandomSampler(seed=0))
study.optimize(objective, n_trials=3)
evaluator = MeanDecreaseImpurityImportanceEvaluator(n_trees=10, seed=0)
param_importance = evaluator.evaluate(study)
evaluator = MeanDecreaseImpurityImportanceEvaluator(n_trees=20, seed=0)
param_importance_different_n_trees = evaluator.evaluate(study)
assert param_importance != param_importance_different_n_trees
def test_mean_decrease_impurity_importance_evaluator_seed() -> None:
study = create_study()
study.optimize(objective, n_trials=3)
evaluator = MeanDecreaseImpurityImportanceEvaluator(seed=2)
param_importance = evaluator.evaluate(study)
evaluator = MeanDecreaseImpurityImportanceEvaluator(seed=2)
param_importance_same_seed = evaluator.evaluate(study)
assert param_importance == param_importance_same_seed
evaluator = MeanDecreaseImpurityImportanceEvaluator(seed=3)
param_importance_different_seed = evaluator.evaluate(study)
assert param_importance != param_importance_different_seed
def test_mean_decrease_impurity_importance_evaluator_with_target() -> None:
# Assumes that `seed` can be fixed to reproduce identical results.
study = create_study(sampler=RandomSampler(seed=0))
study.optimize(objective, n_trials=3)
evaluator = MeanDecreaseImpurityImportanceEvaluator(seed=0)
param_importance = evaluator.evaluate(study)
param_importance_with_target = evaluator.evaluate(
study,
target=lambda t: t.params["x1"] + t.params["x2"],
)
assert param_importance != param_importance_with_target