Public API

This page lists the stable, supported import paths for CausalRL. Anything not listed here may change without notice.

Core entry points

• crl.evaluate_ope - end-to-end OPE pipeline returning an OpeReport.
• crl.run_spec - run an EvaluationSpec and return an EvaluationResult.
• crl.evaluate - compatibility wrapper (spec usage deprecated).
• crl.ope.evaluate_ope - same entry point, explicit module import.
• crl.ope.OpeReport - aggregate report for an OPE run.
• crl.set_seed - seed Python/NumPy/torch RNGs.

Estimands and assumptions

• crl.PolicyValueEstimand
• crl.PolicyContrastEstimand
• crl.SensitivityPolicyValueEstimand
• crl.ProximalPolicyValueEstimand
• crl.Assumption
• crl.AssumptionSet

Datasets

• crl.data.BanditDataset
• crl.data.LoggedBanditDataset
• crl.data.TrajectoryDataset
• crl.data.TransitionDataset

Behavior policy estimation

• crl.fit_behavior_policy
• crl.BehaviorPolicyFit

Policies

• crl.policies.Policy
• crl.policies.StochasticPolicy
• crl.policies.CallablePolicy
• crl.policies.TabularPolicy
• crl.policies.TorchMLPPolicy

Estimator selection

• crl.select_estimator
• crl.SelectionResult

Dataset adapters

• crl.load_d4rl_dataset
• crl.load_rl_unplugged_dataset

Estimators and reports

• crl.estimators.OPEEstimator
• crl.estimators.EstimatorReport
• crl.estimators.UncertaintySummary
• crl.estimators.DiagnosticsConfig
• crl.estimators.ISEstimator
• crl.estimators.WISEstimator
• crl.estimators.PDISEstimator
• crl.estimators.DoublyRobustEstimator
• crl.estimators.WeightedDoublyRobustEstimator
• crl.estimators.MAGICEstimator
• crl.estimators.MRDREstimator
• crl.estimators.MarginalizedImportanceSamplingEstimator
• crl.estimators.FQEEstimator
• crl.estimators.DualDICEEstimator
• crl.estimators.GenDICEEstimator
• crl.estimators.DoubleRLEstimator
• crl.estimators.DRLEstimator
• crl.estimators.HighConfidenceISEstimator

API namespace

• crl.api re-exports the stable surface for convenience in scripts.