Hi! I and thanks for this package! I have made a different implementation (ARS.jl) which adds some features which I found necessary as well as significantly faster sampling (from some limited testing. The features I found missing were:
- Allowing different numeric types than
Float64 in the sampler (I needed to use BigFloat).
- Allow the user to choose autodiff backend through
DifferentiationInterface.jl (not all autodiff libraries are compatible with all numeric types).
Instead of having 2 slightly different libraries registered I would prefer merging mine with yours. However, this would probably result in a breaking release and replace most of the internals. Would a PR accomplishing this be welcome?
ARS.jl repo: https://github.com/Eliassj/ARS.jl
Hi! I and thanks for this package! I have made a different implementation (ARS.jl) which adds some features which I found necessary as well as significantly faster sampling (from some limited testing. The features I found missing were:
Float64in the sampler (I needed to useBigFloat).DifferentiationInterface.jl(not all autodiff libraries are compatible with all numeric types).Instead of having 2 slightly different libraries registered I would prefer merging mine with yours. However, this would probably result in a breaking release and replace most of the internals. Would a PR accomplishing this be welcome?
ARS.jl repo: https://github.com/Eliassj/ARS.jl