giovanni-cutri/NeRa

Seamless integration of sport rating systems into graph neural networks in the PyTorch environment

PyTorch based package for incorporating rating systems to neural networks.

Prerequisities

Python >= 3.11

Installation

$ pip install --upgrade pip
$ pip install torch==2.1.2
$ pip install torch-scatter -f https://data.pyg.org/whl/torch-2.1.2+cpu.html
$ pip install torch-sparse -f https://data.pyg.org/whl/torch-2.1.2+cpu.html
$ pip install torch-geometric==2.4.0
$ pip install torch-geometric-temporal==0.52.0
$ pip install git+https://github.com/kubosis/NeRa.git

Nera - Neural rating

This package implements seamless integration of statistical rating systems into graph neural network in the PyTorch environment. This project was developed as my Bachelor's thesis.

Implemented rating layers and recurrent graph neural network architectures

• Elo rating
• Berrar rating
• Pi rating
• GConvElman
• RatingRGNN

Showcases of predictive validation accuracy on collected datasets:

Note: the RatingRGNN was fine-tuned only on the NBL dataset and then applied across the other.

Note: the accuracy is across time snapshots. These snapshots represent seasons. They do not represents epochs of iterating the whole dataset. The training was done only for one epoch.