Simple tabular reinforcement learning in Python.
This is a very simple project for tabular reinforcement learning (TabulaRL) that I wrote during my PhD studies. There are already several great open source packages for reinforcement learning, including:
- RL Glue
- RLPY
- PyBrain
- OpenAI Gym
- Microsoft minecraft (to come)
Out of these, it looks like OpenAI Gym is currently the most promising... if I was starting this all again maybe I would have focused my attention here!
That said, I found that for studying simple tabular RL problems these frameworks were all a little bit of overkill. If you are experimenting with small, simple, tabular MDPs or just learning about RL for the first time it can be nice to have a really simple bit of code to get started. The nice thing about this code instead of something like Gym is:
- All written in simple Python+Numpy, no complicated dependencies beyond standard Anaconda
- Very few lines of code, potentially more accessible to novice programmers.
- Focus on very simple problems.
But the main reason I'm putting this code up is so that people can more easily implement/verify some of the ideas in some of my papers!