Kaladin

This machine learning tool is aimed at automating cheat detection on Lichess using insights (example).

It is built using CNNs on Keras/TensorFlow.

Setup

You will need:

• Linux OS (tested on Ubuntu 20.04 LTS)
• Docker
• MongoDB

Docker container setup for Tensorflow with CPU or GPU

Pre-requisites

Install Docker using your favorite package manager, or for example you can follow this guide.

Create custom image and container

Run $./docker.sh gpu|cpu [dev|prod] with the needed target, it will create/update the image and start the container. dev (default) will open bash, while prod will directly launch the queue manager: python3 queue_manager.py

Useful commands

To restart the container: docker restart kaladin

To view the logs: docker logs -f kaladin

Configuration

For the list of options and default values used by Kaladin, see src/.env.base. You can override these either by setting environmental variables or create a src/.env file.

Acknowledgments

The Kaladin repository was re-created when transitioning to open source to ensure that user data was not made public. Git history was expunged during that transition. A record of the commits prior to the transition can be found here:
Special thanks to:

• kraktus for your work on the queue manager, Docker config, error handling, lila integration, and integration testing.
• michael1241 for your domain expertise, design discussions, initial queue manager and mongo and deployment support.
• ornicar for your support, your mongo wizardry, and your lila integration work.
• the others around the globe who helped by validating the model output, generating ideas, and providing valuable feedback.