- New argument
fold_info = FALSE in create_folds() to return fold information as attributes “fold” and “repeat”, see 33.
- Update code coverage version.
- Documentation update.
- Remove dependency {ranger}.
- Documentation update
- Push the package to mature version 1.0.0.
- Replacing
importFrom by ::.
New Functionality
- Sometimes, one need to stratify on multiple columns. The new function
multi_strata() provides a vector of stratification groups based on a data frame that can be then passed to partition() or create_folds(). Each stratification group will contain “similar” data rows, where similarity is either based on a kmeans cluster analysis or forming all combinations of binned columns. Thanks to kapsner for the idea and the help with the implementation.
Maintenance
- Set up github actions, thanks to L. Kapsner.
This is documentation and maintenance update only with the following changes:
- Updated documentation to clarify that
create_folds() creates in-sample indices by default. If out-of-sample indices are to be generated, set invert = TRUE.
- Got rid of a CRAN check notes about LazyData.
- Changed to better way of updating/generating the package.
New Functionality
-
create_folds and partition have received a shuffle option to shuffle rows within folds/partitions. The default is FALSE.
Breaking change for tiny data sets
-
create_folds and partition cannot return empty folds/partitions anymore. This impacts only extremely small data sets.
Other
- Unit tests have been added.
New Functionality
-
create_timefolds now allows also moving windows training data, not just extending windows data.
Other
Reduced minimally required R version from 3.5 to 3.1.
This is the initial CRAN release.