Merge pull request #125 from st-tech/add-error-meta

fix error of meta

obp/ope/meta.py

@@ -12,7 +12,7 @@
 from pandas import DataFrame
 import seaborn as sns
 
-from .estimators import BaseOffPolicyEstimator
+from .estimators import BaseOffPolicyEstimator, DirectMethod as DM, DoublyRobust as DR
 from ..types import BanditFeedback
 from ..utils import check_confidence_interval_arguments
 
@@ -84,8 +84,11 @@ def __post_init__(self) -> None:
             if key_ not in self.bandit_feedback:
                 raise RuntimeError(f"Missing key of {key_} in 'bandit_feedback'.")
         self.ope_estimators_ = dict()
+        self.is_model_dependent = False
         for estimator in self.ope_estimators:
             self.ope_estimators_[estimator.estimator_name] = estimator
+            if isinstance(estimator, DM) or isinstance(estimator, DR):
+                self.is_model_dependent = True
 
     def _create_estimator_inputs(
         self,
@@ -102,9 +105,7 @@ def _create_estimator_inputs(
                 f"action_dist.ndim must be 3-dimensional, but is {action_dist.ndim}"
             )
         if estimated_rewards_by_reg_model is None:
-            logger.warning(
-                "`estimated_rewards_by_reg_model` is not given; model dependent estimators such as DM or DR cannot be used."
-            )
+            pass
         elif isinstance(estimated_rewards_by_reg_model, dict):
             for estimator_name, value in estimated_rewards_by_reg_model.items():
                 if not isinstance(value, np.ndarray):
@@ -171,6 +172,12 @@ def estimate_policy_values(
             Dictionary containing estimated policy values by OPE estimators.
 
         """
+        if self.is_model_dependent:
+            if estimated_rewards_by_reg_model is None:
+                raise ValueError(
+                    "When model dependent estimators such as DM or DR are used, `estimated_rewards_by_reg_model` must be given"
+                )
+
         policy_value_dict = dict()
         estimator_inputs = self._create_estimator_inputs(
             action_dist=action_dist,
@@ -222,6 +229,12 @@ def estimate_intervals(
             using nonparametric bootstrap procedure.
 
         """
+        if self.is_model_dependent:
+            if estimated_rewards_by_reg_model is None:
+                raise ValueError(
+                    "When model dependent estimators such as DM or DR are used, `estimated_rewards_by_reg_model` must be given"
+                )
+
         check_confidence_interval_arguments(
             alpha=alpha,
             n_bootstrap_samples=n_bootstrap_samples,

obp/ope/meta_continuous.py

@@ -12,7 +12,10 @@
 from pandas import DataFrame
 import seaborn as sns
 
-from .estimators_continuous import BaseContinuousOffPolicyEstimator
+from .estimators_continuous import (
+    BaseContinuousOffPolicyEstimator,
+    KernelizedDoublyRobust as KDR,
+)
 from ..types import BanditFeedback
 from ..utils import check_confidence_interval_arguments
 
@@ -96,8 +99,11 @@ def __post_init__(self) -> None:
             "action"
         ]
         self.ope_estimators_ = dict()
+        self.is_model_dependent = False
         for estimator in self.ope_estimators:
             self.ope_estimators_[estimator.estimator_name] = estimator
+            if isinstance(estimator, KDR):
+                self.is_model_dependent = True
 
     def _create_estimator_inputs(
         self,
@@ -115,9 +121,7 @@ def _create_estimator_inputs(
                 "action_by_evaluation_policy must be 1-dimensional ndarray"
             )
         if estimated_rewards_by_reg_model is None:
-            logger.warning(
-                "`estimated_rewards_by_reg_model` is not given; model dependent estimators such as DM or DR cannot be used."
-            )
+            pass
         elif isinstance(estimated_rewards_by_reg_model, dict):
             for estimator_name, value in estimated_rewards_by_reg_model.items():
                 if not isinstance(value, np.ndarray):
@@ -186,6 +190,12 @@ def estimate_policy_values(
             Dictionary containing estimated policy values by OPE estimators.
 
         """
+        if self.is_model_dependent:
+            if estimated_rewards_by_reg_model is None:
+                raise ValueError(
+                    "When model dependent estimators such as DM or DR are used, `estimated_rewards_by_reg_model` must be given"
+                )
+
         policy_value_dict = dict()
         estimator_inputs = self._create_estimator_inputs(
             action_by_evaluation_policy=action_by_evaluation_policy,
@@ -237,6 +247,12 @@ def estimate_intervals(
             using nonparametric bootstrap procedure.
 
         """
+        if self.is_model_dependent:
+            if estimated_rewards_by_reg_model is None:
+                raise ValueError(
+                    "When model dependent estimators such as DM or DR are used, `estimated_rewards_by_reg_model` must be given"
+                )
+
         check_confidence_interval_arguments(
             alpha=alpha,
             n_bootstrap_samples=n_bootstrap_samples,

tests/ope/test_meta.py

@@ -10,7 +10,7 @@
 import torch
 
 from obp.types import BanditFeedback
-from obp.ope import OffPolicyEvaluation, BaseOffPolicyEstimator
+from obp.ope import OffPolicyEvaluation, BaseOffPolicyEstimator, DirectMethod
 from obp.utils import check_confidence_interval_arguments
 
 
@@ -310,6 +310,32 @@ def test_meta_post_init(synthetic_bandit_feedback: BanditFeedback) -> None:
                 )
 
 
+def test_meta_estimated_rewards_by_reg_model_inputs(
+    synthetic_bandit_feedback: BanditFeedback,
+) -> None:
+    """
+    Test the estimate_policy_values/estimate_intervals functions wrt estimated_rewards_by_reg_model
+    """
+    ope_ = OffPolicyEvaluation(
+        bandit_feedback=synthetic_bandit_feedback, ope_estimators=[DirectMethod()]
+    )
+
+    action_dist = np.zeros(
+        (synthetic_bandit_feedback["n_rounds"], synthetic_bandit_feedback["n_actions"])
+    )
+    with pytest.raises(ValueError):
+        ope_.estimate_policy_values(
+            action_dist=action_dist,
+            estimated_rewards_by_reg_model=None,
+        )
+
+    with pytest.raises(ValueError):
+        ope_.estimate_intervals(
+            action_dist=action_dist,
+            estimated_rewards_by_reg_model=None,
+        )
+
+
 # action_dist, estimated_rewards_by_reg_model, description
 invalid_input_of_create_estimator_inputs = [
     (

tests/ope/test_meta_continuous.py

@@ -12,6 +12,7 @@
 from obp.ope import (
     ContinuousOffPolicyEvaluation,
     BaseContinuousOffPolicyEstimator,
+    KernelizedDoublyRobust,
 )
 from obp.utils import check_confidence_interval_arguments
 
@@ -186,6 +187,32 @@ def test_meta_post_init(synthetic_continuous_bandit_feedback: BanditFeedback) ->
                 )
 
 
+def test_meta_estimated_rewards_by_reg_model_inputs(
+    synthetic_bandit_feedback: BanditFeedback,
+) -> None:
+    """
+    Test the estimate_policy_values/estimate_intervals functions wrt estimated_rewards_by_reg_model
+    """
+    kdr = KernelizedDoublyRobust(kernel="cosine", bandwidth=0.1)
+    ope_ = ContinuousOffPolicyEvaluation(
+        bandit_feedback=synthetic_bandit_feedback,
+        ope_estimators=[kdr],
+    )
+
+    action_by_evaluation_policy = np.zeros((synthetic_bandit_feedback["n_rounds"],))
+    with pytest.raises(ValueError):
+        ope_.estimate_policy_values(
+            action_by_evaluation_policy=action_by_evaluation_policy,
+            estimated_rewards_by_reg_model=None,
+        )
+
+    with pytest.raises(ValueError):
+        ope_.estimate_intervals(
+            action_by_evaluation_policy=action_by_evaluation_policy,
+            estimated_rewards_by_reg_model=None,
+        )
+
+
 # action_by_evaluation_policy, estimated_rewards_by_reg_model, description
 invalid_input_of_create_estimator_inputs = [
     (