@DavidAFrench TLDR, looking ahead to 2028 and we all gotta get paid. If that means taking money from actual commies or jihadists, that’s cool. Don’t hate the player, hate the game!
Este es el mejor video que vi en mucho tiempo.
Soldados israelíes se hacen pasar por terroristas de Hamás en EEUU y comienzan a detallar con orgullo las atrocidades del 7 de octubre a personas que apoyan a Hamás.
Miren las reacciones, es imperdible.
I get the OLS thing here. Y = observed score difference in game, one X column for every FBS team, one column for home field. Each row a game with one constraining row for uniqueness. Team A beats Team B by 14 at Team B. Team A = Team B + 14 + home field. So for that row code Y=14, the Team A column=1, Team B column=-1, the home field advantage column =-1, all other team columns for that row=0. Fill out the rest of the rows for other games similarly. then OLS B=inv(X'X)*X'Y, beta vector contains your power ratings. This is basically what Sagarin does.
Now for my questions:
1. You said you are predicting the *point spread* in the games. Does this mean you are literally modeling Vegas point spreads as Y, and not the actual outcomes of the games? If not, then how do you account for these being so wildly at odds with ratings estimated with convential OLS or Maximum Likelihood methods like Sagarin?
2. Yes or no - do recruiting rankings play a part in preseason FPI? If so exactly how?
3. Do you have a forgetting function? I.e., a lambda parameter (l < 1) to discount older Y values? If so, how is it determined?
4. Do you test forward / backward, with test/training/validation hold out?
5. Why is it "proprietary," treated as if it's some sort of huge trade secret? Trust me, no one's trying to steal it, just understand it. Why not open source the code?
@infantrydort@JimHansonDC If these guys are going to play such a critical role in defining proper conduct, surely they should be embedded in front line units much like medics or chaplains are.