Aggregated Hold-Out Cross Validation

The 'agghoo' procedure is an alternative to usual cross-validation. Instead of choosing the best model trained on V subsamples, it determines a winner model for each subsample, and then aggregates the V outputs. For the details, see "Aggregated hold-out" by Guillaume Maillard, Sylvain Arlot, Matthieu Lerasle (2021) published in Journal of Machine Learning Research 22(20):1--55.

agghoo

R package for model selection based on aggregation. Alternative to standard cross-validation.

Install the package

From GitHub: devtools::install_github("yagu0/agghoo")

Locally, in a terminal: R CMD INSTALL .

Use the package

library(agghoo)
?agghoo