Showing 1–1 of 1 results for author: Polyakova, Y

General notions of depth for functional data

Authors: Karl Mosler, Yulia Polyakova

Abstract: A data depth measures the centrality of a point with respect to an empirical distribution. Postulates are formulated, which a depth for functional data should satisfy, and a general approach is proposed to construct multivariate data depths in Banach spaces. The new approach, mentioned as Phi-depth, is based on depth infima over a proper set Phi of R^d-valued linear functions. Several desirable pr… ▽ More A data depth measures the centrality of a point with respect to an empirical distribution. Postulates are formulated, which a depth for functional data should satisfy, and a general approach is proposed to construct multivariate data depths in Banach spaces. The new approach, mentioned as Phi-depth, is based on depth infima over a proper set Phi of R^d-valued linear functions. Several desirable properties are established for the Phi-depth and a generalized version of it. The general notions include many new depths as special cases. In particular a location-slope depth and a principal component depth are introduced. △ Less

Submitted 30 January, 2018; v1 submitted 9 August, 2012; originally announced August 2012.

Comments: The revision of November 2016 introduces the term Tukey graph depth to distinguish it from half-region depth. In the present version (January 2018) I have dropped an erroneous remark on the band depth, thanks to Stanislav Nagy, who pointed out in his dissertation that the band depth is no infimum depth