Multithreading

[Byter09]

Hey, so far I really like this project!

Am just playing around, but one thing I noticed is that only ever one thread is being used.

From the looks of it rayon could be used in various places to speed things up.

Just leaving this suggestion here, and I hope you keep developing this. I love playing around with NEAT and doing it in Rust is amazing 😄

[Fr4ctlize]

I completely agree, Genome evaluation is one of the more obvious places where rayon could be used (obviously depending on what the evaluation strategy is).

I haven't had as much time to work on this as I'd like, and the currently more pressing issue, I think, is updating all the docs and doc tests to reflect the move to Genome as a trait (which I think could be used to apply NEAT to other things which are not neural networks, which might be cool).

When I find the time I will finish that up and maybe work on parallelization. However, I've also started sketching some ideas for implementing rt-NEAT in a wrapper for Population, with variable genome count replacement and dynamically modifiable population sizes (for experiments like open-ended artificial life simulations). I think that may result in some modifications to how Population itself functions, so until I figure that out I'm not committing to parallelizing anything.

Of course, any ideas are welcome. I may push what I've sketched so far, in case you want to take a look at it. I'm glad you like the project!

[Byter09]

Wow, sounds amazing. I'm just playing around and found this one issue that I could see with htop haha. If I have any more ideas I'll leave an issue for sure!

I didn't know you had this much planned. Maybe keep this issue on the back-burner then and finish the other amazing stuff you got going. I would love to see where this is going.

Thank you so much for responding so fast! Love it. Gonna star this repo and follow your progress, maybe leave a comment here and there :)

[Byter09]

Tiny update: for my own purposes I cloned the repo, went back to before the Genome trait commit and just put par_iter() and par_iter_mut() everywhere I could and it now successfully uses all my cores haha.

So for my personal application this is fine. Just wanted to let you know in case you're curious.