I made a Neural Network library from scratch in Rust trying to solve a regression problem for my university's ML course

Hello everyone!

For the last 3 weeks I've been learning about ML & Neural Networks and coding my own library from scratch. It has been a blast doing it in Rust as it was both my most challenging and interesting Rust (and maybe even programming) project so far. I would love to get some feedback from the community on the design or even better from seasoned ML scientists who may know some Rust as I'm a newbie to ML who had never used any ML library before.

You can see the results and some examples on the Github https://github.com/AnicetNgrt/neural_networks_rust

Obviously making it on-par with Python libs even limited on a subset of NNs is a non-goal, but I'd love next to implement my own webGPU backend for matrices, as I've already abstracted all the matrix code away and added both an nalgebra and my custom vec+rayon backend. I'd also love to dip my toes into RNNs, CNNs, but I feel that my cpu-bound lib will first need to be made gpu-bound for the usually large models using CNNs to be testable. Or maybe there is a faster cpu-bound backend I could use other than nalgebra (which seems as fast as it can get to me already).

Also, what do you think about ML in rust? I'd love to hear from people who tried that before me.

Thanks for reading and have a nice day :)