jamac22/ddqn

Deep Double Q-Learning implementation introduced by Hasselt et al in this paper: https://arxiv.org/abs/1509.06461. It's interfacing with openAI Gym. WIP.

Watch it in action at the Gym here:

https://gym.openai.com/evaluations/eval_GFtDBmuyRjCzcAkBibwYWQ#reproducibility

The algorithm is based on the great research of such great minds like David Silver, Hado van Hasselt, Vlad Minh and many more in particular (bust not exclusively) on Double DQN.

I also wrote about the algorithm on my blog, if you want to know more:

https://davidsanwald.github.io/2016/12/11/Double-DQN-interfacing-OpenAi-Gym.html

If you want to reproduce the exact results from the Gym please use the one file Gist, sometimes I feel like doing stupid things with every master branch I can get my hands on (;