Cross-validation software for generalized linear models (R package).
The main and only function estimates k-fold cross-validation prediction error for GLMs. It builds upon boot::cv.glm, providing the following new features: models of class other than "glm" can be evaluated, various cost functions can be efficiently used to estimate the cross-validation prediction error on the same k-fold groups, all k-fold scores are reported (and not just the weighted mean), and k-fold groups can be sampled pro rata with the relative frequencies of response levels.
Do What The Fuck You Want To Public License (version 2).
When using xv.glm for a publication, please cite my doctoral dissertation.
@phdthesis{
author = {Boruta, Luc},
title = {Indicators of Allophony and Phonemehood},
school = {Universit{\'e} Paris Diderot},
year = {2012},
}