Python Framework for reinforcement learning. Contains both tabular (Q-learning, n-strep Tree backup) and approximate (deep Q-learning) methods, to deal with finite and infinite state spaces, respectively.
The implementation is general and works with any type of environments, although 'train' and 'play' methods, implemented for convinience, work with gym-like environments. However, gym-like wrappers are easily defined around other types of environments, and a minimal class example of such a wrapper is given.
Some example reinforcement learning tasks are presented to illustrate the framework's capabilities, namely the frozen lake (finite state space) and cart pole (infinite state space) gym tasks.
Work in progress...
Olivier C. Pasche 2022