8th International Conference of the ERCIM WG on Computational and Methodological Statistics (CMStatistics 2015), Date: 2015/12/12 - 2015/12/14, Location: Londen, UK

Publication date: 2015-12-01

Author:

Hubert, Mia
Rousseeuw, Peter ; Segaert, Pieter ; Vakili, Kaveh

Abstract:

The concept of depth first appeared in the context of multivariate data with the halfspace depth. It is a way to order the data from the centre outwards, and can, e.g. be used to visualise the distribution of bivariate data by means of the bagplot. Several other notions of multivariate depth have also been introduced including the simplicial depth, the projection depth and the skew-adjusted projected depth based on the adjusted outlyingness. The idea of depth has been extended to the regression setting and more recently to functional data. We have developed an R-package ``mrfDepth'' which contains implementations of many of these depth-based estimators as well as several graphical representations of the data. User-friendly R-code is combined with C++ implementations of several fast algorithms for the halfspace depth, the Stahel-Donoho outlyingness, the adjusted outlyingness, the depth median and depth contours. Also corresponding Matlab functions are available as part of the LIBRA toolbox.