30
4

On the consistency of the Kozachenko-Leonenko entropy estimate

Abstract

We revisit the problem of the estimation of the differential entropy H(f)H(f) of a random vector XX in RdR^d with density ff, assuming that H(f)H(f) exists and is finite. In this note, we study the consistency of the popular nearest neighbor estimate HnH_n of Kozachenko and Leonenko. Without any smoothness condition we show that the estimate is consistent (E{HnH(f)}0E\{|H_n - H(f)|\} \to 0 as nn \to \infty) if and only if E{log(X+1)}<\mathbb{E} \{ \log ( \| X \| + 1 )\} < \infty. Furthermore, if XX has compact support, then HnH(f)H_n \to H(f) almost surely.

View on arXiv
Comments on this paper