Misc code associated with exercises in Sutton's 'Reinforcement Learning'. Suitable environment provided for conda in envirnment.yml

Hadn't started this repo at this point

Go back and implement at least a bandit framework

Implemented MDP framework in a minimal-dependencies manner, with reusable abstractions for probability distributions. The only dependencies are numpy and matplotlib (the latter just to generate chart results for the exercises)

Should try porting to a probabilistic programming framework such as Pyro

Added code for (4.5) and (4.9) using the MDP framework. Interesting results in regard to stability of policy where multiple optimal (or near optimal?) policies exist