Test for non-negligible adverse shifts

Abstract

Statistical tests for dataset shift are susceptible to false alarms: they are sensitive to minor differences when there is in fact adequate sample coverage and predictive performance. We propose instead a framework to detect adverse dataset shifts based on outlier scores, $\text{D-SOS}$ for short. $\text{D-SOS}$ holds that the new (test) sample is not substantively worse than the reference (training) sample, and not that the two are equal. The key idea is to reduce observations to outlier scores and compare contamination rates at varying weighted thresholds. Users can define what $worse$ means in terms of relevant notions of outlyingness, including proxies for predictive performance. Compared to tests of equal distribution, our approach is uniquely tailored to serve as a robust metric for model monitoring and data validation. We show how versatile and practical $\text{D-SOS}$ is on a wide range of real and simulated data.

arXiv commentAccepted at UAI 2022

References in corpus (11)

• A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
• Can You Trust Your Model's Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift
• Challenges in Deploying Machine Learning: a Survey of Case Studies
• Small sample methods for cluster-robust variance estimation and hypothesis testing in fixed effects models
• To Trust Or Not To Trust A Classifier
• Failing Loudly: An Empirical Study of Methods for Detecting Dataset Shift
• WAIC, but Why? Generative Ensembles for Robust Anomaly Detection
• Testing for Outliers with Conformal p-values
• OpenML: An R Package to Connect to the Machine Learning Platform OpenML
• OpenML Benchmarking Suites
• Tracking the risk of a deployed model and detecting harmful distribution shifts