š pip install unravelsports
Iām thrilled to announce today marks the release date of my very own open-source Python package, supported by PySport!
The š®š§š«ššÆšš„š¬š©šØš«šš¬ package is designed to help researchers, analysts and enthusiasts by providing intermediary steps in the complex process of converting raw sports data into meaningful information and actionable insights.
š Its current functionality helps to convert football tracking data from 6 providers, using š¤š„šØš©š©š², into graphs specifically designed for training šš«šš©š” ššš®š«šš„ šššš°šØš«š¤š¬ with Spektral.
It is my aim to add even more functionality in the future, not only for football!
1ļøā£ To get started, simply š„šØšš data and ššØš§šÆšš«š into graphs.
2ļøā£ šš©š„š¢š šš«šš¢š§, ššš¬š and šÆšš„š¢šššš¢šØš§ datasets along match or period with the built in functionality.
3ļøā£ šš¦š©šØš«š and ššØš¦š©š¢š„š the pre-built šš«š²š¬ššš„šš«šš©š”šš„šš¬š¬š¢šš¢šš« (as detailed in the 2023 Sloan Sports Analytics Conference paper by Amod Sahasrabudhe and me), or design your own architecture from scratch.
4ļøā£ Now, š„šØšš the Graphs using the Spektral DisjointLoader and šš¢š the model.
5ļøā£ Finally, ššÆšš„š®ššš and š¬ššØš«š the model and use it to make š©š«ššš¢ššš¢šØš§š¬!
š For complete Jupyter Notebook on how to execute all of the above steps and additional documentation check out the šš§-ššš©šš” ššš„š¤šš”š«šØš®š š” on GitHub or the šš®š¢šš¤š¬ššš«š šš®š¢šš.
š š®š§š«ššÆšš„š¬š©šØš«šš¬==š.š.š
š This new version includes a converter specifically built for converting hashtag #BigDataBowl American Football positional data.
ā” The American Football implementation is lightning fast because it runs on a Polars back-end!
š Here is a ššš¬š¢š ššš„š¤šš”š«šØš®š š” of the new AmericanFootballGraphConverter.