Train best RL algorithm on a cloud service

🚀 Feature Request

Part of #42. Depends on #45. Once the best RL algorithm has been determined, it should be neatly packaged and uploaded to a cloud service for production training, where several hours of training can take place. By this point, hopefully, the visualization platform in #32 should be at a point where the results can be visualized for publicity, and for iterating on the next network. The focus is on 2019: creating a generic tool for this is not essential, but it will be very beneficial for future years, and the task of #52.