Approximate Data Depth Revisited
Abstract
Halfspace depth and -skeleton depth are two types of depth functions in nonparametric data analysis. The halfspace depth of a query point with respect to is the minimum portion of the elements of which are contained in a halfspace which passes through . For , the -skeleton depth of with respect to is defined to be the total number of \emph{-skeleton influence regions} that contain , where each of these influence regions is the intersection of two hyperballs obtained from a pair of points in . The -skeleton depth introduces a family of depth functions that contain \emph{spherical depth} and \emph{lens depth} if and , respectively. The main results of this paper include approximating the planar halfspace depth and -skeleton depth using two different approximation methods. First, the halfspace depth is approximated by the -skeleton depth values. For this method, two dissimilarity measures based on the concepts of \emph{fitting function} and \emph{Hamming distance} are defined to train the halfspace depth function by the -skeleton depth values obtaining from a given data set. The goodness of this approximation is measured by a function of error values. Secondly, computing the planar -skeleton depth is reduced to a combination of some range counting problems. Using existing results on range counting approximations, the planar -skeleton depth of a query point is approximated in , . Regarding the -skeleton depth functions, it is also proved that this family of depth functions converge when . Finally, some experimental results are provided to support the proposed method of approximation and convergence of -skeleton depth functions.
Keywords
Cite
@article{arxiv.1805.07373,
title = {Approximate Data Depth Revisited},
author = {Rasoul Shahsavarifar and David Bremner},
journal= {arXiv preprint arXiv:1805.07373},
year = {2018}
}
Comments
This paper is submitted to CCCG2018