This is my fork of RLPy.
RLPy is a great reinforcement learning library with some good features like custom representations and value function viewers. However, it is no longer maintained and the PyPI version of RLPy is not compatible with the latest numpy C-API, which causes runtime errors.
So I decided to fork the repository and provide a maintained version of RLPy only for Python>=3.6, named RLPy3. You can download wheels of RLPy3 from PyPI.
In addition, I pushed lots of refactorings, bug fixes, new environments such as FixedRewardGridWorld, and new algorithms such as PSRL. I also removed some modules (like parallel-job utilities) to keep this library maintainable by sufficiently low efforts.
To see what was changed, see the new CHANGELOG.
The latest version is 2.0.0b1, which you can install by:
pip3 install rlpy3 -U --pre
An agent that explores GridWorld.
Value function of the agent.
RLPy is a framework to conduct sequential decision making experiments. The current focus of this project lies on value-function-based reinforcement learning. The project is distributed under the 3-Clause BSD License.
- Official source code repository: http://github.com/rlpy/rlpy
- Bitbucket mirror: http://bitbucket.org/rlpy/rlpy
- Documentation: http://rlpy.readthedocs.org
- Issue tracker: https://github.com/rlpy/rlpy/issues
Installation instructions can be found at http://rlpy.readthedocs.org/en/latest/install.html