Huh? What are you trying to build? Is your primary interest RL or building a simple NN? What task are you trying to solve? I programmed in Java daily from 2002 - 2015, then switched to Python in 2015. I'd suggest you not use Java because you can get practice w/ the same fundamentals without having to write so much verbose code.

If your goal is to learn fundamentals, I'd suggest you start w/ just numpy and no additional libraries. If you're already very comfortable w/ PyTorch, then I'd suggest using JAX.

I'll extend this: for the majority of real computer science/ engineering work (especially back-end stuff), *nix is the way to go. It's a huge reason Macs are the standard for CS folks (ever since the terminal became part of the MacOS). At every university I attended and taught, and at every company I've worked for, linux/macs have been used by essentially 100% of the people. Roughly once a year I'll hear/see someone using Windows.