RcppEnsmallen package brings to R a Header-Only C++ Mathematical
Optimization Library for Armadillo.
In particular, Ensmallen is a
templated C++ mathematical optimization library (by the MLPACK team) that provides a simple set of abstractions for writing an
objective function to optimize.
The fall update of the ensmallen library has been released by the Ensmallen Development Team. Within this release, we see improvements to gradient stability with the inclusion of a gradient clipping and norm scaling callback based on “On the difficulty of training Recurrent Neural Networks” by Razvan Pascanu, Tomas Mikolov, Yoshua Bengio. Moreover, there were a suite of under-the-hood updates with respect to testing infrastructure. Perhaps the largest user-facing change is a requirement bump for the Armadillo, C++ library for linear algebra & scientific computing, dependency to 9.800, up from the initial 8.400 requirement.
For the complete list of changes, please see the news entry for the package update below.
RcppEnsmallen news file entry for version v0.2.18.0.1 (2021-10-22)
- Upgraded to ensmallen 2.18.0: “Fairmount Bagel” (2021-10-21)
- Add gradient value clipping and gradient norm scaling callback (https://github.com/mlpack/ensmallen/pull/315).
- Remove superfluous CMake option to build the tests (https://github.com/mlpack/ensmallen/pull/313).
- Bump minimum Armadillo version to 9.800 (https://github.com/mlpack/ensmallen/pull/318).
- Update Catch2 to 2.13.7 (https://github.com/mlpack/ensmallen/pull/322).
- Remove redundant template argument for C++20 compatibility (https://github.com/mlpack/ensmallen/pull/324).
- Fix MOEAD test stability (https://github.com/mlpack/ensmallen/pull/327).
- Update GitHub Actions to the standard
- Added pkgdown website and corresponding GitHub Actions.