Baconian [beˈkonin] is a toolbox for model-based reinforcement learning with user-friendly experiment setting-up, logging and visualization modules developed by CAP. We aim to develop a flexible, re-usable and modularized framework that can allow the user's to easily set-up a model-based rl experiments by reuse modules we offered.
- 2019.7.30 Release the v0.2.0, updated full API documentations, added visualization module.
- 2019.6.23 Released the v0.1.5, added data pre-processing module, added support for more Gym environment (Roboschool, Atari) and deep mind control suit.
For previous news, please go here
We support python 3.5, 3.6, and 3.7 with Ubuntu 16.04 or 18.04. Documentation is available at http://baconian-public.readthedocs.io/
Currently, the project is under activate development. We are working towards a stable 1.0 version. Details of the road map and future plan will be released as soon as possible.
Currently we are working on
- Simplified flow module
- Latent-space method supporting.
- General Ensemble method
Thanks to the following open-source projects:
- garage: https://github.com/rlworkgroup/garage
- rllab: https://github.com/rll/rllab
- baselines: https://github.com/openai/baselines
- gym: https://github.com/openai/gym
- trpo: https://github.com/pat-coady/trpo
- PILCO: https://github.com/nrontsis/PILCO
If you find Baconian is useful for your research, please consider cite our demo paper here:
@article{
linsen2019baconian,
title={Baconian: A Unified Opensource Framework for Model-Based Reinforcement Learning},
author={Linsen, Dong and Guanyu, Gao and Yuanlong, Li and Yonggang, Wen},
journal={arXiv preprint arXiv:1904.10762},
year={2019}
}
If you find any bugs on issues during your usage of the package, please open an issue or send an email to me (linsen001@e.ntu.edu.sg) with detailed information. I appreciate your help!