|
Abstract:
|
Vectors are at the core of everything that R does. We have created a new extensible framework for defining custom implementations of atomic R vectors compatible with base R and package code. The implications of this system and our applications of it within R itself are vast, for analysts and package developers alike. These include: nigh-instantaneously importing massive vectors from in-memory column stores, subsetting and reordering data.frames instantly with no data duplication, vectors with dramatically reduced memory footprints, vectors which know whether they are sorted or have NAs, improvements to R's sorting and matching capabilities which make use of them, and more. I will show examples of improvements in the performance of normal R code, as well as how to use the framework to help improve R even more from base or package code. These changes are currently slated to go live in R-devel by Summer 2017 and is slated to be included in the 2018 release of R.
|
