In Deepmind's RL lectures there is also a fair comparison between several methods from a statistical inference point-of-view and less of a empirical one. (it is also probably backup up with their hands-on experience).
I would love a clarification on how you see it in connection to the hackathon.