rlgym-learn-algos 0.1.1

Rust backend for the more expensive parts of the rlgym-learn-algos python module

RLGym-Learn-algos

A set of standard implementations of common algorithms for use with rlgym-learn

Features

• PPO agent controller
• Flexible metrics logging
• File-based configuration

Installation

1. install RLGym via pip install rlgym. If you're here for Rocket League, you can use pip install rlgym[rl-rlviser] instead to get the RLGym API as well as the Rocket League / Sim submodules and rlviser support.
2. If you would like to use a GPU install PyTorch with CUDA
3. Install rlgym-learn via pip install rlgym-learn
4. Install this project via pip install rlgym-learn-algos
5. If pip installing fails at first, install Rust by following the instructions here

Usage

See the RLGym website for complete documentation and demonstration of functionality [COMING SOON]. For now, you can take a look at quick_start_guide.py and speed_test.py in the rlgym-learn repo to get a sense of what's going on.

Credits

This code was separated out from rlgym-learn, which itself was created using rlgym-ppo as a base. The following files were largely written by Matthew Allen, the creator of rlgym-ppo: ppo/basic_critic.py ppo/continuous_actor.py ppo/critic.py ppo/discrete_actor.py ppo/multi_discrete_actor.py ppo/experience_buffer.py util/*

In general his support for this project has allowed it to become what it is today.