NEWS

SDModels 2.0.2 (2025-12-14)

Switch all the parallelization to futures. See vignette("Runtime")
Switch all the progress updates to progressr. Progress updates are now also available for parallel processing and are customizable.
Process are much more RAM efficient now.

Fix bug in SDTree and SDForest where an error occurred, if X had columns with only one unique value.

Removal of data.tree dependence. Trees are now saved as a matrix.
This results in a substantial reduction of RAM usage and space needed to save an SDTree or SDForest. It also results in increased computational speed.
Removal of copy, fromList, and toList. Remove the copy arguments from all pruning involving functions.
New plotting of SDTree objects.
Improved/Robust parallelization.
Remove gpu support and dependence on GPUmatrix due to it being scheduled for archiving.

In case of parallel processing use random number generator "L'Ecuyer-CMRG" for reproducibility

Add option to plot SDForests. The plot shows the out-of-bag performance against the number of trees. This helps to evaluate whether enough trees were used.

Added feature to select some predictors not to be regularized closes option to use some covariates not regularized in SDAM #4
Fix the length of the coefficient list to the number of predictors and name the elements
change predict_individual_j to expect a numeric new data vector instead of a whole data.frame

Fix various bugs on edge cases with just one variable or just one tree
SDForest, regPath.SDTree, regPath.SDForest, predict.SDForest, prune.SDForest, varImp.SDTree

Fix bug in estimation of SDTree when using only one covariate (did stop splitting after one split)
Add support to predict with an SDForest using multiple cores in parallel