This is the repo containing the code used in our paper titled "Emergent Password Signalling in the Game of Werewolf" accepted at EmeCom 2022. It is based on the code and work by Brandizzi et al. available at this link.
For details of our modifications and code we refer to our paper on OpenReview
To install follow the instructions in the Installation markdown.
To run you will have to move the script train.sh from the scripts dir into the root of this repo.
Parameters such as the number of rounds or the required threshold can be adjusted in Params.py