A bot playing Farkle (a dice game) in the Kingdom Come: Deliverance
It uses the (ostensibly) optimal strategy, which is a simplified variation of Markov Decision Process (MDP), see mdp.py.
The dice position detection is done using OpenCV
and some image processing / contour detection algorithms exposed by the library, see detection.py.
The dice value recognition is handled by a simple convolutional neural network, see recognition.py.
The instructions are written for Windows command line, but it's almost identical for bash
- Clone the project
git clone https://github.com/keenua/farkle
- Switch to the project directory
cd farkle
- (Optional) Create a virtual environment with Python 3.7 (using
venv
only as an example)
python -m venv env
- (Optional) Activate the environment
.\env\Scripts\activate
- Install the dependencies
pip install -r requirements.txt
- Switch to
src
folder
cd src
- Train the MDP model. It'll probably take a lot of time depending on your specs. The output file (
4000_parallel.pkl.npy
) would be ~1GB.
python -m farkle -a train
- Run the bot
python -m farkle
- Launch the
Kingdom Come: Deliverance
game and sit at the Farkle table. The bot should start printing suggestions as soon as it's your move. Until then it should be showingWaiting for hero's turn
message
If you're having troubles running it or understanding something, just create a github issue in this repo