I’m making a game with ml-agents for the Unity Game Jam, and I’d like to be able to start the game without the Academy trying to connect when using inference only behaviours. Is there a way to do this?
Hi,
Sorry for the delay.
So you want to disable the Academy trying to connect to the trainer when all the behaviors are inference, or you want it to successfully connect in that case? The subject line makes it sounds like the second, but the body of the message makes it sound like the first.
In either case, this isn’t currently possible. We could add some global settings to disable the Academy connection attempt. Connecting when there aren’t any agents that are trying to learn is more complicated, because since no messages are being sent between Unity and python, the connection will time out eventually.
disable the Academy trying to connect to the trainer when all the behaviors are inference
This one if possible. The lag time while it’s deciding whether there are agents to train is significant for a small game. I understand that it’s designed so the user can add agents for training after the scene starts (and in fact do that in other projects). Adding a global setting to disable the Academy would be amazing.
By the way, if you want to play the game, it’s here: Secret Battle 