scientific article; zbMATH DE number 3969896

From MaRDI portal
Publication:3736722





[For the entire collection see Zbl 0572.00015.] In \textit{C. J. Stone}, Ann. Stat. 10, 1040-1053 (1982; Zbl 0511.62048), a nonparametric estimator of the regression curve \[ m(x)=E(Y| X=x)\in \{g\in C^ k: | g^{(k)}(x)-g^{(k)}(x')| \leq H| x- x'|^{\beta}\} \] was constructed, which achieves the optimal, in the \(L_ 2\)-sense, rate \(n^{-r}\) with \(r=p/(2p+d)\), \(p=k+\beta\), \(Y\in R^ d\), \(X\in R^ 1\). But that construction required the knowledge of k and \(\beta\). In this paper it is shown that kernel estimators with a bandwidth selected by a cross-validatory technique achieve the optimal rate independently of r.




Cited in
(25)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3736722)