Testing Calibration in Nearly-Linear Time
This work addresses the problem of efficiently measuring calibration for researchers and practitioners in machine learning, providing a near-linear time algorithm that is a significant improvement over existing black-box solvers, though it is incremental in advancing algorithmic efficiency for a known bottleneck.
The paper tackles the algorithmic challenge of efficiently testing whether a binary prediction model is calibrated, by reformulating the empirical smooth calibration linear program as a minimum-cost flow problem and solving it with a dynamic programming-based algorithm that runs in O(n log^2 n) time, which is information-theoretically optimal and improves upon prior methods requiring Ω(n^ω) time.
In the recent literature on machine learning and decision making, calibration has emerged as a desirable and widely-studied statistical property of the outputs of binary prediction models. However, the algorithmic aspects of measuring model calibration have remained relatively less well-explored. Motivated by [BGHN23], which proposed a rigorous framework for measuring distances to calibration, we initiate the algorithmic study of calibration through the lens of property testing. We define the problem of calibration testing from samples where given $n$ draws from a distribution $\mathcal{D}$ on $(predictions, binary outcomes)$, our goal is to distinguish between the case where $\mathcal{D}$ is perfectly calibrated, and the case where $\mathcal{D}$ is $\varepsilon$-far from calibration. We make the simple observation that the empirical smooth calibration linear program can be reformulated as an instance of minimum-cost flow on a highly-structured graph, and design an exact dynamic programming-based solver for it which runs in time $O(n\log^2(n))$, and solves the calibration testing problem information-theoretically optimally in the same time. This improves upon state-of-the-art black-box linear program solvers requiring $Ω(n^ω)$ time, where $ω> 2$ is the exponent of matrix multiplication. We also develop algorithms for tolerant variants of our testing problem improving upon black-box linear program solvers, and give sample complexity lower bounds for alternative calibration measures to the one considered in this work. Finally, we present experiments showing the testing problem we define faithfully captures standard notions of calibration, and that our algorithms scale efficiently to accommodate large sample sizes.