Evaluate which code can be consolidated into macros across NBA, NFL, NCAAF models.

[matsonj]

The following models should in theory be generic:

• inputs
  • teams
  • schedule
  • actual results
• outputs
  • elo over time
  • predictions
  • end of season standings (blending predictions + actuals) AKA game log

This will allow additional sports to be added very quickly and easily - NHL, MLB, Premier League (?), College bball

non-generic models + reasons

• in season tournament / "champions cups": the games are not fixed, so we have compute projected winners to slot teams into subsequent games.
• end of season seeding: tiebreaking methodology depends on specific league rules
• playoffs: for leagues with playoffs, subsequent games take a dependency on projected outcomes. additionally, some leagues have different criteria for wins (NFL is 1 game, NBA is best of 7, MLB is hybrid, and so on)

However, the most interesting analysis depends on "end of season seeding" (argh), so we will need to figure out how to build models for end of season seeding for each sport. Playoff models I am less certain on, because I am not confident "regular season win totals" which is currently driving ELO ratings is necessarily predictive of playoff success. For all intents and purposes, regular season and post season models should probably behave independently.