action annotations

[wendlerc]

is there any way to get good action annotations for your game datasets? i have a similar library for wm training but i need the actions i'd love to expand to your games since i think some of them would be nice testbeds for memory augmented / game state augmented / long context architecture exploration https://github.com/wendlerc/toy-wm

[AlmondGod]

hey, whole purpose of genie is no need for action annotations!

you can try to get some perhaps by annotating yourself, wouldn't take too long for a 2-min clip to just to test your idea out!

[wendlerc]

I know I know. Just wanted to check whether I can save some hustle over here.

We should record some of these games -- you did a nice selection -- with action annotations. That would also make for nice latent action model evaluations as well. You could automatically compute whether your learnt actions align with the real ones.

Best,
Chris