Working off of my initial python/SQL based AHL xG model, I decided implement it as a linear model using tensorflow.
Prior to running the linear model, we’ll need to scrape data. The ahl_scraper.py script gets all x,y shot & goal locations at a certain strength from the AHL website and exports that to a csv file. This script should be run once to generate training data and a second time to generate the testing data.
command: Usage: python ahl_scraper.py [start game ID] [last game ID] [output csv name]
Where the game ID can be found by going to theahl.com website, finding the most recent completed game and extracting the game ID from the URL
for example, the game ID for this URL: https://theahl.com/stats/game-center/1019145 is 1019145
Scraping richer data (data with more features) would allow one to predict xG not only for a (x,y,strength) combination but also additional features to train, evaluate and predict on.
Robin Wisniewski – LinkedIn – wisniewski.ro@gmail.com