search

boluoweifenda / werewolf

MIT License
16 stars 1 forks source link

The implementation details of the agent #3

Open LYH-YF opened 7 months ago

LYH-YF commented 7 months ago

Up to now, you should have only released the Werewolf environment. Could you provide some details on the implementation of the agent and the training details of its modules?

boluoweifenda commented 7 months ago

@LYH-YF Thanks for reaching out about the Werewolf agent and training details. We're still prepping that info, but you can get some background from our paper in the meantime. It covers key concepts we've applied. We plan to share more soon, so keep an eye out for updates!

yuezhao238 commented 7 months ago

@boluoweifenda Hi, I am interested in the implementation details of agents too. Can your clarify how the "181", "1019", "523" are caculated in F.2 part of your paper? I tried the scripts provided in this repo, but I got a "feature" with a length of 17. Thanks for your precious time!

boluoweifenda commented 7 months ago

@Heathcliff-Zhao The keys and values from the dataset are raw feature, mostly speech texts and voting results. You should get full game states from game environment and encoding them in to input feature for the model networks, just as many other RL enviroments such as GYM. https://github.com/boluoweifenda/werewolf/blob/main/processor/env9_en.py#L613

yuezhao238 commented 7 months ago

@boluoweifenda Thanks for your reply. I will learn to implement it like in gym.