argminCS

Argmin Inference over a Discrete Candidate Set

Provides methods to construct frequentist confidence sets with valid marginal coverage for identifying the population-level argmin or argmax based on IID data. For instance, given an n by p loss matrix—where n is the sample size and p is the number of models—the CS.argmin() method produces a discrete confidence set that contains the model with the minimal (best) expected risk with desired probability. The argmin.HT() method helps check if a specific model should be included in such a confidence set. The main implemented method is proposed by Tianyu Zhang, Hao Lee and Jing Lei (2024) "Winners with confidence: Discrete argmin inference with an application to model selection".

Versions across snapshots

Dependencies (latest)