I'd need to dig into the source code of this project more to be convinced, but I trust the implementation and algorithm choice in core R way more than any other stats package except possibly SAS: among other reasons it has far more of the right eyeballs looking at the code. Picking numerically reliable implementations of most estimators is not trivial without the right training. Now, I don't know how much needed to be rewritten for this project, so that's just my general opinion.
My exact thought. R has great tools but the language is wonky. That's not a dig against them. It wasn't designed by language designers. However, its interactive nature is great for data scientists. So it seems like Clojure (possibly Scala) is a better fit than Java.
The ecosystem didnt come into existence by magic. The early adopters did have a choice into which language to invest their time. Obviously the R syntax alone was enough of an incentive to starts building useful libraries in it, instead of lisp, python, etc.
