JORLDY (Beta)
Hello WoRLd!!
π₯ Features
- 20+ RL Algorithms and various RL environment are provided
- Algorithms and environment are customizable
- New algorithms and environment can be added
- Distributed RL algorithms are provided using ray
- Benchmark of the algorithms is conducted in many RL environment
βοΈ Tested
Python | Windows | Mac | Linux |
---|---|---|---|
3.8 3.7 |
Windows Server 2022 | macOS Big Sur 11 macOS Catalina 10.15 |
Ubuntu 20.04 Ubuntu 18.04 |
β¬οΈ Installation
git clone https://github.com/kakaoenterprise/JORLDY.git
cd JORLDY
pip install -r requirements.txt
# linux
apt-get update
apt-get -y install libgl1-mesa-glx # for opencv
apt-get -y install libglib2.0-0 # for opencv
apt-get -y install gifsicle # for gif optimize
π³ To use docker
(customize if necessary)
cd JORLDY
docker pull jorldy/jorldy
# mac, linux
docker run -it --rm --name jorldy -v `pwd`:/JORLDY jorldy/jorldy /bin/bash
# windows
docker run -it --rm --name jorldy -v %cd%:/JORLDY jorldy/jorldy /bin/bash
β To use additional environments
- Atari and Super Mario Bros
atari and super-mario-bros need to be installed manually due to licensing issues
# To use atari
pip install --upgrade gym[atari,accept-rom-license]
# To use super-mario-bros
pip install gym-super-mario-bros
- Mujoco (Mac and Linux only)
Mujoco is supported in docker. However, if you don't use docker, several subprocesses should be done. Please refer to the mujoco-py github installation
π Getting started
cd jorldy
# Examples: python main.py [run mode] --config [config path]
python main.py --config config.dqn.cartpole
python main.py --async --config config.ape_x.cartpole
# Examples: python main.py [run mode] --config [config path] --[optional parameter key] [parameter value]
python main.py --config config.rainbow.atari --env.name breakout
python main.py --sync --config config.ppo.cartpole --train.num_workers 8
ποΈ Release
Version | Release Date | Source | Release Note |
---|---|---|---|
0.5.0 | April 18, 2022 | Source | Release Note |
0.4.0 | April 01, 2022 | Source | Release Note |
0.3.0 | March 10, 2022 | Source | Release Note |
0.2.0 | January 23, 2022 | Source | Release Note |
0.1.0 | December 23, 2021 | Source | Release Note |
π How to
- How to use
- How to customize config
- How to customize agent
- How to customize environment
- How to customize network
- How to customize buffer
π Documentation
- Algorithm Descriptions
- Benchmark
- Distributed Architecture
- List of Contents
- Naming Convention
- Reference
- Role of Managers
π₯ Contributors
leonard.q (Kyushik Min) |
ramanuzan.lee (Hyunho Lee) |
kan.s (Kwansu Shin) |
erinn.lee (Taehak Lee) |
link.lee (Hojoon Lee) |
royce.choi (Jinwon Choi) |
crest.son (Sungho Son) |
lisa.ekkim (Eunkyeong Kim) |
π·οΈ Citation
@article{min2022jorldy,
title={JORLDY: a fully customizable open source framework for reinforcement learning},
author={Min, Kyushik and Lee, Hyunho and Shin, Kwansu and Lee, Taehak and Lee, Hojoon and Choi, Jinwon and Son, Sungho},
journal={arXiv preprint arXiv:2204.04892},
year={2022}
}
Β©οΈ License
π« Disclaimer
Installing in JORLDY and/or utilizing algorithms or environments not provided KEP may involve a use of third partyβs intellectual property. It is advisable that a user obtain licenses or permissions from the right holder(s), if necessary, or take any other necessary measures to avoid infringement or misappropriation of third partyβs intellectual property rights.