def test_meta_create_estimator_inputs_using_valid_input_data(
action_dist,
estimated_rewards_by_reg_model,
description: str,
synthetic_bandit_feedback: BanditFeedback,
) -> None:
"""
Test the _create_estimator_inputs using invalid data
"""
ope_ = OffPolicyEvaluation(
bandit_feedback=synthetic_bandit_feedback, ope_estimators=[ipw]
)
estimator_inputs = ope_._create_estimator_inputs(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
assert set(estimator_inputs.keys()) == set(["ipw"])
assert set(estimator_inputs["ipw"].keys()) == set(
[
"reward",
"action",
"pscore",
"position",
"action_dist",
"estimated_rewards_by_reg_model",
"estimated_pscore",
"estimated_importance_weights",
"p_e_a",
"pi_b",
"context",
"action_embed",
]
), f"Invalid response of _create_estimator_inputs (test case: {description})"
# _create_estimator_inputs function is called in the following functions
_ = ope_.estimate_policy_values(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
_ = ope_.estimate_intervals(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
_ = ope_.summarize_off_policy_estimates(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
_ = ope_.evaluate_performance_of_estimators(
ground_truth_policy_value=0.1,
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
_ = ope_.summarize_estimators_comparison(
ground_truth_policy_value=0.1,
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
def test_meta_create_estimator_inputs_using_invalid_input_data(
action_dist,
estimated_rewards_by_reg_model,
description: str,
synthetic_bandit_feedback: BanditFeedback,
) -> None:
"""
Test the _create_estimator_inputs using valid data
"""
ope_ = OffPolicyEvaluation(
bandit_feedback=synthetic_bandit_feedback, ope_estimators=[ipw]
)
# raise ValueError when the shape of two arrays are different
with pytest.raises(ValueError, match=f"{description}*"):
_ = ope_._create_estimator_inputs(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
# _create_estimator_inputs function is called in the following functions
with pytest.raises(ValueError, match=f"{description}*"):
_ = ope_.estimate_policy_values(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
with pytest.raises(ValueError, match=f"{description}*"):
_ = ope_.estimate_intervals(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
with pytest.raises(ValueError, match=f"{description}*"):
_ = ope_.summarize_off_policy_estimates(
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
with pytest.raises(ValueError, match=f"{description}*"):
_ = ope_.evaluate_performance_of_estimators(
ground_truth_policy_value=0.1,
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
with pytest.raises(ValueError, match=f"{description}*"):
_ = ope_.summarize_estimators_comparison(
ground_truth_policy_value=0.1,
action_dist=action_dist,
estimated_rewards_by_reg_model=estimated_rewards_by_reg_model,
)
def test_meta_create_estimator_inputs_format(
synthetic_bandit_feedback: BanditFeedback,
random_action_dist: np.ndarray) -> None:
"""
Test the _create_estimator_inputs format of OffPolicyEvaluation
"""
# __post_init__ saves the latter estimator when the same estimator name is used
ope_ = OffPolicyEvaluation(bandit_feedback=synthetic_bandit_feedback,
ope_estimators=[ipw])
inputs = ope_._create_estimator_inputs(action_dist=None,
estimated_rewards_by_reg_model=None)
assert set(inputs.keys()) == set([
"reward",
"action",
"pscore",
"position",
"action_dist",
"estimated_rewards_by_reg_model",
]), "Invalid response format of _create_estimator_inputs"
