This collection of Python modules implements some Reinforcement Learning algorithms, most notably Deep Q Networks (DQN) and Prioritized Experience Replay (PER), where the proportional prioritization variant has been implemented.. It has been built to solve OpenAI Gym environments, although it has only been tested on classic control environments with discrete action sets.
The code supports a variety of hyper parameters, that are usually tuned to particular environments. Bayesian optimization with Scikit-Optimize is a simple way of tuning those hyper parameters.
The code uses Tensorflow to model a value function for a Reinforcement Learning agent. I've run it with Tensorflow 1.0 on Python 3.5 under Windows 7.
- Deep Learning tutorial, David Silver, Google DeepMind.
- Prioritized Experience Replay, T. Schaul., J. Quan and D. Silver. Feb 2016.
- Deep Reinforcement Learning with Double Q-learning, Hado van Hasselt, Arthur Guez, David Silver. Dec 2015.