|国家预印本平台
首页|The weak-feature-impact effect on the NPMLE in monotone binary regression

The weak-feature-impact effect on the NPMLE in monotone binary regression

The weak-feature-impact effect on the NPMLE in monotone binary regression

来源:Arxiv_logoArxiv
英文摘要

The nonparametric maximum likelihood estimator (NPMLE) in monotone binary regression models is studied when the impact of the features on the labels is weak. Here, weakness is colloquially understood as ``close to flatness'' of the feature-label relationship $x \mapsto \mathbb{P}(Y=1 | X=x)$. Statistical literature provides the analysis of the NPMLE for the two extremal cases: If there is a non-vanishing derivative of the regression function, then the NPMLE possesses a nonparametric rate of convergence with Chernoff-type limit distribution, and it converges at the parametric $\sqrt{n}$-rate if the underlying function is (locally) flat. To investigate the transition of the NPMLE between these two extremal cases, we introduce a novel mathematical scenario. New pointwise and $L^1$-rates of convergence and corresponding limit distributions of the NPMLE are derived. They are shown to exhibit a sharp phase transition solely characterized by the level of feature impact.

Dario Kieffer、Angelika Rohde

数学

Dario Kieffer,Angelika Rohde.The weak-feature-impact effect on the NPMLE in monotone binary regression[EB/OL].(2025-04-13)[2025-05-01].https://arxiv.org/abs/2504.09564.点此复制

评论