Cutting Edge Football Analytics in Python
summary
Shows how Polars Keras Spektral Kloppy and unravel sports process 7 million frames across 64 matches into one-graph-per-frame inputs and build an expected possession value model using graph neural nets. Labels come from event joins, outputs are per-frame probabilities, and the prototype trains and predicts but is data-limited — about 172 goals total — giving a model that is ~11× better than random yet poorly calibrated without far more matches.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:14Football data types
- 8:53Graph neural networks
- 14:35Tracking data preprocessing
- 18:31Expected possession value labeling
- 22:20Data to graph conversion
- 24:48Training EPV model
- 28:18Making predictions
- 30:26Summary
related talks
51mVideo Analytics for Football games
29mFootball (soccer) data analysis: A pedagogic introduction
42mHow Deep Learning is Revolutionizing Sports Analytics
27mAnalysis of sports scenes using computer vision models
38mDeep Learning plays Handball
45mThe Technical Journey to Define the True Value of a Football Goal
31mFootball is complex, but your code doesn’t have to be — meet DataBallPy and a practical deep dive into pressing
32mIdentifying playstyles in football through spatial networks