Closed glmcdona closed 2 years ago
I wanted to try the idea when I saw that code but I don't know how to add such a callback to use rules in certain conditions. (Could be used when stacked units all try to go in the same direction too)
On Thu, Sep 23, 2021, 21:54 Geoff McDonald @.***> wrote:
Just noting here, that the winning imitation learning solution only applies actions to Units, not CityTiles: https://www.kaggle.com/shoheiazuma/lux-ai-with-imitation-learning
The cities have hardcoded actions of building workers if possible, otherwise research. Adding support for this would approximately reduce the number of action decisions by ~33%, and reduce the number of actions by a couple.
Suggest we add more agent callbacks for hardcoded logic to apply that can give units actions without creating an observation. Thoughts? I think we can make it backwards compatible.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/glmcdona/LuxPythonEnvGym/issues/79, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFZQYJ6OBB2AW24SL5WRW6TUDQAADANCNFSM5EVDZZSQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.
Just noting here, that the winning imitation learning solution only applies actions to Units, not CityTiles: https://www.kaggle.com/shoheiazuma/lux-ai-with-imitation-learning
The cities have hardcoded actions of building workers if possible, otherwise research. Adding support for this would approximately reduce the number of action decisions by ~33%, and reduce the number of actions by a couple.
Suggest we add more agent callbacks for hardcoded logic to apply that can give units actions without creating an observation. Thoughts? I think we can make it backwards compatible.