torch-rating 0.0.1b0

Seamless integration of sports rating systems as layers into pytorch environment

PyTorch based package for incorporating rating systems to neural networks. This package provides model rating layers. The resulting RatingRGNN can be found here

Prerequisities

Python >= 3.10

Installation

pip install --upgrade pip
pip install torch-rating

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

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.