pymc3_nba

Using PYMC3 to model NBA matches

Get data

• get_daily_player_gamelogs.py: data by player/game req:api (train)
• get_seasonal_games.py: scheduled games for the season req:api
• manually download odds from https://sportsbookreviewsonline.com/scoresoddsarchives/nba/nbaoddsarchives.htm

---

• parse_daily_player_gamelogs: all player/games in one clean table (train)
• parse_seasonal_games.py: all games from the season
• parse_odds.py: write clean csv with odds

---

Model and test

• shared_reg: feature engineering + bayesian model
• odds: calculate return on test data using odds.

---

Predict live matches

• get_daily_lineup.py: expected lineup per game
• parse_game_lineup.py: write clean csv with lineups
• predict_one: predict one particular game
• predict_batch: predict all matches for selected dates

others

• benchmark: results from a model with one intercept (dummy) per player (not bayesian)
• out_of_training.ipynb: notebook showing what happens if you include a player that wasn't seen in training