drewscottt / cfb-rankings Goto Github PK

I wanted a metric which answers the question: on average, how much does a team over/under-perform in games compared to their opponents typical result? For example, if Team A loses to Team B by 10 points, but Team B typically wins by 20 points, Team A over-performed by 10 (-10 + 20) points. So, for this game Team A would earn 10 points. Then across all games, their results would be average.

Additionally, I wanted a metric to acknowledge that winning (or losing) is also an important aspect (not just margin of victory). So, if Team A beats Team B, and Team B wins 90% of their games, Team A should be awared further. So, the game metric becomes: -10 + 20 + (N * .9), where N is left to be determined.

Each team is ranked based on their average performance from each game (game metric). For each game, the game metric is calculated as:

• margin of victory (negative if a loss) +
• (w * opponent's average margin of victory (not including game(s) played against this team)) +
• (result factor = (x1 * opponent win percent (if win) or -x2 * opponent loss percent (if loss) (not including game(s) played against this team)))

Here, w=.5, x1=10, x2=10, but they are tweakable constants.

Further, the raw score of each game is adjusted by:

• The winning team gets +5 points
• The away team gets +2 points
• If an FBS team loses to an FCS team, the victory margin is multiplied by 3 (after the previous adjustments are made)
• FBS wins vs. FCS teams are completely ignored

Again, each of the mentioned constants are tweakable.

Note: calculating the "result factor" of the game metric is used by counting real-life wins/losses; the point adjustments listed above only affect the first two factors of the game metric.

There are two main programs:

• rank.py: Used to compute a ranking, or compare rankings.

  python3 rank.py <team_results_directory>

  python3 rank.py team_pages/2022-week6-results/

  This has a function to compare a ranking to another ranking (change in Top 25, biggest movers) and another function to filter a ranking based on conference.

  Note: before running for a new week of results, first craate the results directory under team_pages, then it will populate the directory.

• predict_analyze.py: Used to predict a schedule of games based on a previously generated ranking, or analyze a schedule of results and sportsbook predictions in relation to a ranking.

  python3 predict_analyze.py <ranking_filename> <predict_analyze> <optional: espn_schedule_url>

  python3 predict_analyze.py rankings/2022-week6.txt predict

  python3 predict_analyze.py rankings/2022-week6.txt analyze https://www.espn.com/college-football/schedule/_/week/9/year/2022/seasontype/2

  Note: To capture the pre-game view of game pages, just run with the analyze option on the appropriate schedule before the games have started. These will be saved under ./preview_game_pages

1. Georgia (+0)

2. Michigan (+0)

3. Ohio State (+0)

4. Alabama (+1)

5. Utah (+4)

6. Tennessee (+2)

7. Penn State (+0)

8. TCU (-4)

9. Clemson (+1)

10. Kansas State (+1)

11. USC (-5)

12. Florida State (+0)

13. Oregon (+0)

14. Texas (+0)

15. Washington (+1)

16. Oregon State (-1)

17. Tulane (+1)

18. UTSA (+3)

19. UCLA (+0)

20. Illinois (+0)

21. Troy (+1)

22. LSU (-5)

23. South Alabama (+1)

24. Mississippi State (+1)

25. Notre Dame (-2)

    Dropped from last Top 25: None

Notes:

• Games Registered includes only FBS vs. FBS matchups for the week for 2022 Week 6 through 2022 Week 12.
• Sometimes ESPN doesn't include sportsbook data (example) or sportsbook doesn't have a favorite (example).
• 2022, Week 7 was the first week collecting moneyline data and is incomplete due to data collection after some games had been played.

2022, Week 12 Rankings:

• Began ranking FCS teams along with FBS teams
  • Removed 21 point adjusted scoring penalty in FBS vs. FCS games
  • Apply a 0.25 weight to adjusted scoring margins for FCS vs. FCS wins (and 1/0.25 weight for FCS vs. FCS losses)
• Lowered adjusted scoring margin cap from 38 to 28
• Lowered recency bias from 0.05 to 0.03

2022, Week 9 Rankings:

• Major bug fix: neutral site games no longer double counted.

2022, Week 8 Rankings:

• Remove 3x multiplier for FCS losses, replace with 21 point adjusted scoring margin penalty vs. FCS teams
• Consider all FCS games, not just losses
• Apply recency bias 0.05
• Apply adjusted scoring margin cap at 38

Base:

• Ignore FCS wins
• Apply a 3x multiplier to adjusted score margin in losses vs. FCS teams
• Away team receives 2 points in adjusted score
• Winning team receives 5 points in adjusted score
• Win factor: 10, loss factor: 10 in calculating game metric
• Opponent strength weight in calculating game metric: 0.5