Universal distribution of the empirical coverage in split conformal prediction

IF 0.9 4区数学 Q3 STATISTICS & PROBABILITY

Statistics & Probability Letters Pub Date : 2025-01-02 DOI:10.1016/j.spl.2024.110350

Paulo C. Marques F.

引用次数: 0

Abstract

When split conformal prediction operates in batch mode with exchangeable data, we determine the exact distribution of the empirical coverage of prediction sets produced for a finite batch of future observables. This distribution is universal, being determined solely by the batch size, the nominal miscoverage level, and the calibration sample size. The exact distribution of the almost sure limit of the empirical coverage as the batch size goes to infinity is also identified, leading to a criterion for choosing the minimum required calibration sample size in applications.

查看原文本刊更多论文

劈裂保形预测中经验覆盖率的普遍分布

当分割保形预测在具有可交换数据的批处理模式下操作时，我们确定了有限批未来可观测值产生的预测集的经验覆盖的确切分布。这种分布是通用的，仅由批量大小、标称误覆盖水平和校准样本量决定。当批量大小趋于无穷大时，经验覆盖的几乎肯定限制的确切分布也被确定，从而导致在应用中选择最小所需校准样本量的标准。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Statistics & Probability Letters 数学-统计学与概率论

CiteScore

1.60

自引率

0.00%

发文量

173

审稿时长

6 months

期刊介绍： Statistics & Probability Letters adopts a novel and highly innovative approach to the publication of research findings in statistics and probability. It features concise articles, rapid publication and broad coverage of the statistics and probability literature. Statistics & Probability Letters is a refereed journal. Articles will be limited to six journal pages (13 double-space typed pages) including references and figures. Apart from the six-page limitation, originality, quality and clarity will be the criteria for choosing the material to be published in Statistics & Probability Letters. Every attempt will be made to provide the first review of a submitted manuscript within three months of submission. The proliferation of literature and long publication delays have made it difficult for researchers and practitioners to keep up with new developments outside of, or even within, their specialization. The aim of Statistics & Probability Letters is to help to alleviate this problem. Concise communications (letters) allow readers to quickly and easily digest large amounts of material and to stay up-to-date with developments in all areas of statistics and probability. The mainstream of Letters will focus on new statistical methods, theoretical results, and innovative applications of statistics and probability to other scientific disciplines. Key results and central ideas must be presented in a clear and concise manner. These results may be part of a larger study that the author will submit at a later time as a full length paper to SPL or to another journal. Theory and methodology may be published with proofs omitted, or only sketched, but only if sufficient support material is provided so that the findings can be verified. Empirical and computational results that are of significant value will be published.