Behaviour Policy Estimation in Off-Policy Policy Evaluation: Calibration Matters
This addresses a key challenge in OPE for applications like healthcare, where accurate policy evaluation is critical, though it is incremental in improving calibration methods.
The paper tackles the problem of estimating behavior policies in Off-Policy Policy Evaluation (OPE) when the true policy is unknown, showing that accurate OPE strongly depends on calibration of the estimated models, with a k-nearest neighbors model producing better calibrated estimates and superior OPE results compared to neural networks on a real-world medical dataset.
In this work, we consider the problem of estimating a behaviour policy for use in Off-Policy Policy Evaluation (OPE) when the true behaviour policy is unknown. Via a series of empirical studies, we demonstrate how accurate OPE is strongly dependent on the calibration of estimated behaviour policy models: how precisely the behaviour policy is estimated from data. We show how powerful parametric models such as neural networks can result in highly uncalibrated behaviour policy models on a real-world medical dataset, and illustrate how a simple, non-parametric, k-nearest neighbours model produces better calibrated behaviour policy estimates and can be used to obtain superior importance sampling-based OPE estimates.