More support for deep distributional networks.
Small touch ups for ff based smooth terms.
Removing maptools dependencies.
Fixes in JM code.
Improvements on estimating deep distributional neural networks using TensorFlow.
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.
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.
Removed bit package depends.
More stable implementation of the batchwise backfitting optimizer.
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.