Rmlx and RmlxStats

[hughjonesd]

Rmlx is an R interface to mlx. It mostly translates the core C++ API, making a few changes to fit R conventions, such as 1-based indexing. Then it adds methods for R generics, including arithmetic and logical comparisons. As a result, mlx objects can often work with standard R code.

RmlxStats uses Rmlx to implement some basic statistical models including linear models (mlxs_lm) and generalized linear models (mlxs_glm). As far as I know, it is currently the fastest way for R Apple Silicon users to run a regression on a large dataset.

Both are very early stage, but I hope to get them on to CRAN eventually. One interesting nugget is that both packages were vibe-coded with Claude and Codex. The results seem OK, though there have been bumps along the way.