RL Matrix - ML Agents alternative is (almost) here - Pure C# deep reinforcement learning

Dear Community,

I first tried reaching out to Miguel privately via linkedin but that didn't work out so I thought I'd show you some of my work and see what you all think.

https://github.com/asieradzk/RL_Matrix/

For the last year or so I've been developing a pure-C# deep reinforcement learning library, using TorchSharp for backend/tensors.

If you ever wanted to do deep reinforcement learning in console/asp.net core application - now you can.


Right now I am offering distributed deep reinforcement learning with PPO and mods of DQN all the way up to DQN-Rainbow.

It's been going pretty well and I had a lot of good time getting it to work in Godot (since Godot has .net core api). My objective is to gear it towards easy use with game engines so that beginners can start making DRL agents but also make modifications to underlying algorithms as easy as possible.

To do that I am looking away from python bloat hell and try my best to apply SOLID principles when developing it (went through 2 painful re-writes so far).

To be honest I am not a programmer by trade... and I'm using RL Matrix to do my PhD in chemical engineering so it would be nice to have a community join in and make suggestions and advice. I'd accept any mentor :)

Currently it's still under heavy development and not yet working with .net standard 2.0 - so you can't use it with Unity in shared memory mode - only networked mode. The port to .net standard 2.0 will be here eventually..

What do you all think?