This is the repository for the baseline solution to the stated Kaggle competition. It is hosted by Kaggle InClass and contains data from the Bay12 Games forums, where I have played a lot of mafia (werewolf) and variants.
The solution itself can be found in this notebook. It requires the libraries to be installed, though.
In order to use this effectively, you need to install the library bay12_solution_eposts and its dependencies from the src directory.
You may want to activate a new virtual environment before running (example with Conda, using the environment name from the file):
conda env create -f environment.yaml
activate kaggle-mafia-gametype
Then you can install (better, "develop") from source:
cd src
python setup.py develop
You may use python setup.py install instead, but it won't update with any changes
you will make, so I suggest develop.
Side note: The references to "eposts" are due to the initial name for the competition
being "extended posts", which is how I called it during the internet scraping phase.
I set the data to be in the directories:
- Train:
data/train/... - Test:
data/test/... - Label map:
data/label_map.csv.
In order for the loading funtions from bay12_solution_eposts.prepare to work, you should
place these files and folders there. Alternatively, you can just set the path_data variable in the function call (if the path is
somewhere else) or even change the DEFAULT_PATH_DATA variable in the library.
Use the jupyter notebook command from this directory, then navigate to the notebooks.
If you want to do some of your own experiments, I suggest making a copy first. ;)