NEWS (original) (raw)

origami 1.0.7

Fixes to stratified cross-validation bugs.
More informative warnings when n (sample size) is less than or equal to V (number of folds).

Addition of glmnet to package Suggestssince used in the vignette.
Resolve issue stemming from the partial matching of argument names (https://github.com/tlverse/origami/issues/56) by specifying throughout.

Fix the incorrect use of future.apply by includingfuture.seed = TRUE, as per https://github.com/tlverse/origami/issues/48.
Fixes to folds_rolling_origin_pooled andfolds_rolling_window_pooled, as per https://github.com/tlverse/origami/pull/50.
Sped up all pooled time-series fold functions withlapply.
All pooled time-series fold functions (folds_rolling_origin_pooled,folds_rolling_window_pooled,folds_vfold_rolling_origin_pooled,folds_vfold_rolling_window_pooled) now allow for variability in the number of observations for each independent unit (i.e., the subject).
Added test for pooled time-series cross-validation with multi-unit time-series to test the update above. In this test, the total number of observations and the time-points for which there are observations varies across the units.

A maintenance release addressing reported issues, including changes to unit tests that relied on Suggested packages dependencies.

Adds new functionality for cross-validation with time-series, especially functionality for pooling time-series data. This includes the new functions folds_rolling_origin_pooled,folds_rolling_window_pooled,folds_vfold_rolling_origin_pooled,folds_vfold_rolling_window_pooled.

Adds a simple check of the current R version in the unit tests in the file tests/testthat/test-overall.R to use appropriate hard-coded values that depend on the R version. Note that these differ between R < 3.6.0 and R > 3.6.0 due to an important change in the default PRNG.

Adds a new function folds2foldvec for easy conversion between the folds structure used by origami and other packages (e.g., glmnet).
Adds an option to use lapply instead offuture_lapply, for compatibility with other packages in a custom parallelization framework.
Calls future_lapply from the new more modular packagefuture.apply rather than from future, matching the author’s recommendation.