xG Model

chickenstats.chicken_nhl includes an Expected Goals (xG) model, which is based off of the prior work of Josh & Luke Younggren of Evolving-Hockey.

The model is an XGBoost classifier, with 38 fields, including distance from net, time since last event, shot angle, number of players on ice, and whether the shot event comes from a high danger scoring area.

Usage

The xg model is available out-of-the-box with chickenstats.chicken_nhl.Scraper and chickenstats.chicken_nhl.Game. xG values are accessed with the pred_goal column in the play-by-play data, or the ixg, xgf, and xga columns in the individual, on-ice, line, and team stats data.

from chickenstats.chicken_nhl import Scraper

game_id = 2023020001

scraper = Scraper(game_id)
play_by_play = scraper.play_by_play

pbp_xg_values = play_by_play.pred_goal
individual_xg = scraper.stats.ixg

Model performance

Information on performance metrics, including AUC-ROC, log-loss, precision, recall, etc

Features and feature performance

Information on model features (e.g., high-danger, distance from net, angle) and contribution to model / model performance

Data and data preparation

Information on how the model was constructed, including code snippets