The Technical Journey to Define the True Value of a Football Goal
summary
Builds a 1350×1350 Markov chain of minute-by-score states from scraped match data (Selenium/WebDriverKit), converts it to sparse matrices, and avoids a full matrix inverse to compute the B matrix. Uses Erlang and Haskell prototypes but settles on Julia for speed: computing per-season B matrices takes ~32 seconds and full aggregation ~174 seconds. Produces player rankings where Lukaku falls from 2nd to 8th (Lukaku drops six places), invents a champagne ratio, and finds a close, slightly Messi-favouring goal+assist value in their overlapping years.
AI-written summary · about
chapters (13)
- 0:00Intro
- 2:08Football means a lot
- 2:32October 1989
- 5:47Tennis
- 11:46Football
- 19:21Bummer time
- 26:25Autumn 2023
- 28:45Spring 2024
- 29:52Late summer 2024
- 37:32Lukaku at Everton 2016/17
- 40:21Messi vs Ronaldo
- 42:20Learnings
- 44:27Outro
related talks
18mis delighting fans with game-changing insights
32mCutting Edge Football Analytics in Python
24mFootbal meets graphs
19mFootball Analytics with the Soccermatics Package
29mFootball (soccer) data analysis: A pedagogic introduction
33mUsing Data Science to Better Evaluate American Football Players
41mThe evolution of basketball with data science
38mDeep Learning plays Handball