Removing akima and geoR dependencies.
Fixed predict problems when using by variables.
Full support for scaleable distributional regression models.
Bug fixing/C code.
Improved batchwise backfitting algorithm. Data is not copied anymore.
New setup for neural network model terms.
Bug fixing.
New naming convention for optimizer and sampling functions. Optimizer are now indicated with prefix opt_, sampler functions with prefix sam_. Old optimizer and sampler functions are still supported for reverse compatibility.
New function engines(), which lists available optimizer and sampling functions for families.
New function CRPS() for computing the continuous rank probability score.
Bug fixing.
Bug fixing.
Removed bit package depends.
More stable implementation of the batchwise backfitting optimizer.
Bug fixing.
Support new features for generating weights for neural networks.
New features.
Bug fixing.
New paper introducing bamlss at https://arxiv.org/abs/1909.11784
In addition, there is now a new website http://www.bamlss.org/ with technical examples, textbook examples, as well as examples from publications. The website will be expanded in the future and will serve as the main source for presenting new features of the bamlss package.
Adding a vignette to CRAN which links to the new website.
Bug fixing.
First neural network implementations, the n()
constructor.
Support for very large data sets using the ff and ffbase package.
Experimental batchwise backfitting algorithm
bbfit().
Bug fixing.
More model fitting engines: lasso(),
stabsel()
Fixed some issues with tx() for BayesX.
Added tx3() for BayesX.
Fixed environment saving problem when using
light = TRUE in bamlss().
Dropped some parts of the jm.mode() return
value.
Solved PROTECT() problems in the C code.
Many other small edits …