wk04_d01

Still needs work, this is an early draft.

Transcript

1. Basketball Performance Predictor - Real-time dashboard for coaches  Coach has to rely on intuition to make real-time decisions.  Can quantify performance by looking at data.  Unique play-by-play data.
2. Data and Algorithms 1. Download individual files from archive in csv 2. Scrape a public basketball reference website (beautiful soup) 3. Free account on premium Basketball API site (queries limited to 2000). 1 game.csv 1 game.csv 1 game.csv • Read in > 4000 files • Munge, produce clean pandas DataFrames • Store tables in MySQL database • Run analytics and produce different SQL tables • Web app: Make SQL queries, run analysis, update display. 1 game.csv 1 game.csv Event-based data: Every single play is logged
3. Data and Algorithms Performance might appear to be stochastic. However, there are patterns in basketball games. Exploit these patterns. When a player has been performing well, or performing poorly, it affects their next few plays. Feedback loop for positive performance: • Positive gain: Momentum, increased confidence, team-mates will pas • Negative gain: opponents play better defence
4. Evidence of player streaks Ray Allen
5. Evidence of player streaks Ray Allen
6. Evidence of player streaks Ray Allen
7. Evidence of player streaks (3) (5) (4) (3) (3) (5) (4) Ray Allen
8. Data Story Decaying exponential, but with unique features. Features to exploit in week 4
9. Data Story Decaying exponential, but with unique features. Features to exploit in week 4 Count[n] = count of all streaks of length n. Count[>n] = count of all streaks greater than n Performance[n+1] = Count[>n]/(Count[n]+Count[>n]
10. Data Story Plotted for 262 players
11. Data Story Plotted for 89 players (played over 75% of games).