Welcome to ElegantRL!¶
ElegantRL is an open-source massively parallel library for deep reinforcement learning (DRL) algorithms, implemented in PyTorch. We aim to provide a next-generation framework that leverage recent techniques, e.g., massively parallel simulations, ensemble methods, population-based training, and showcase exciting scientific discoveries.
ElegantRL features strong scalability, elasticity and lightweightness, and allows users to conduct efficient training on either one GPU or hundreds of GPUs:
Scalability: ElegantRL fully exploits the parallelism of DRL algorithms at multiple levels, making it easily scale out to hundreds or thousands of computing nodes on a cloud platform, say, a SuperPOD platform with thousands of GPUs.
Elasticity: ElegantRL can elastically allocate computing resources on the cloud, which helps adapt to available resources and prevents over/under-provisioning/under-provisioning.
Lightweightness: The core codes <1,000 lines (check elegantrl_helloworld).
Efficient: in many testing cases, it is more efficient than Ray RLlib.
ElegantRL implements the following DRL algorithms:
DDPG, TD3, SAC, A2C, PPO, REDQ for continuous actions
DQN, DoubleDQN, D3QN, PPO-Discrete for discrete actions
QMIX, VDN; MADDPG, MAPPO, MATD3 for multi-agent RL
For beginners, we maintain ElegantRL-HelloWorld as a tutorial. It is a lightweight version of ElegantRL with <1,000 lines of core codes. More details are available here.
Installation¶
ElegantRL generally requires:
Python>=3.6
PyTorch>=1.0.2
gym, matplotlib, numpy, pybullet, torch, opencv-python, box2d-py.
You can simply install ElegantRL from PyPI with the following command:
1pip3 install erl --upgrade
Or install with the newest version through GitHub:
1git clone https://github.com/AI4Finance-Foundation/ElegantRL.git
2cd ElegantRL
3pip3 install .
HelloWorld
Tutorials
RLSolver
API Reference
Other