A framework for learning about and experimenting with reinforcement learning algorithms
Project description
RLFlow
A framework for learning about and experimenting with reinforcement learning algorithms. It is built on top of TensorFlow and TFLearn and is interfaces with the OpenAI gym (universe should work, too). It aims to be as modular as possible so that new algorithms and ideas can easily be tested. I started it to gain a better understanding of core RL algorithms and maybe it can be useful for others as well.
Features
Algorithms (future algorithms italicized):
MDP algorithms
Value iteration
Policy iteration
Temporal Difference Learning
SARSA
Deep Q-Learning
Policy gradient Q-learning
Gradient algorithms
Vanilla policy gradient
Deterministic policy gradient
Natural policy gradient
Gradient-Free algorithms
Cross entropy method
Function approximators (defined by TFLearn model):
Linear
Neural network
RBF
Works with any OpenAI gym environment.
Future Enhancements
Improved TensorBoard logging
Improved model snapshotting to include exploration states, memories, etc.
Any suggestions?
Fixes
Errors / warnings on TensorFlow session save
License
Free software: MIT license
Documentation: https://rlflow.readthedocs.io.
History
0.1.0 (2016-12-15)
First release on PyPI.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.