In order to learn subgoals from trajectories of expert play, we hope to achieve must faster learning for RL agents. We use segmented, infinite HMMs, and variants thereof to segment the data in an unsupervised fashion, to eventually build up to more functionality for learning.
subgoal-inf relies on a few modules derived from pyhsmm, which are included with pip:
pip install pyhsmm
TODO: Write usage instructions