modularl 0.1.5

A modular reinforcement learning library

ModulaRL

🚧 This library is still under construction. 🚧

ModulaRL is a highly modular and extensible reinforcement learning library built on PyTorch. It aims to provide researchers and developers with a flexible framework for implementing, experimenting with, and extending various RL algorithms.

Features

• Modular architecture allowing easy component swapping and extension
• Efficient implementations leveraging PyTorch's capabilities
• Integration with TorchRL for optimized replay buffers
• Clear documentation and examples for quick start
• Designed for both research and practical applications in reinforcement learning

TODO

• Add new algorithms
• Add exploration modules
• Add experiment wrapper modules

Installation

pip install modularl

Algorithms Implemented

Algorithm	Type	Paper	Continuous Action	Discrete Action
SAC (Soft Actor-Critic)	Off-policy	Haarnoja et al. 2018	✅	Not implemented YET
TD3 (Twin Delayed DDPG)	Off-policy	Fujimoto et al. 2018	✅	Not implemented YET
DDPG (Deep Deterministic Policy Gradient)	Off-policy	Lillicrap et al. 2015	✅	Not implemented YET

Citation

@software{modularl2024,
  author = {zakaria narjis},
  title = {ModulaRL: A Modular Reinforcement Learning Library},
  year = {2024},
  url = {https://github.com/zakaria-narjis/modularl}
}