Hello! I just realised I am not sure what the online competition platform does. My first impression was that it just takes the uploaded model and evaluates it on a number of differnent experiments. But if that is the case, then why do we submit the policy as well (or the other files)? Does the platform retrain the agent?
Posted by: Team_29 @ Nov. 6, 2019, 11:48 p.m.Our evaluation intends to do no retraining. The policy file is just to serve an `act(observation)` interface so you can call into your model. It can be as simple as loading the model during setup and running inference in the act method call.
Posted by: HuaweiUK @ Nov. 7, 2019, 3:53 a.m.