RE: Week 2 NFL Daily Fantasy Football Results - My Bayesian Inspired Lineups Took a Hit This Week (-$50.00)
Don't forget the rookies! Tajae Sharpe and Will Fuller were just too cheap this week, and a few guys with situations that are improved via injury should have been a bit higher like Travis Benjamin. It's a tough thing to model since the samples are so small for bench players, even smaller for rookies, and people are always getting injured. Next week make sure to tweak a couple guys that are benefiting from injuries.
But tough break with Antonio, crushes every week...except one a season. Last year's dud game made sense: he went up against Jason Verrett who is one of the best corners in the league and got shut down. Sunday he just didn't seem to get targeted much, while Sammie Coates (who I think is awful) was targeted a bunch. Weird day for him. Same for Brees, how does he not murder vs the Giants? I projected him for 330 yards haha, one of the highest I can ever remember.
Have you ever tried DFS baseball? It's my worst sport by far, but I think these models would be really interesting for it cause the samples are so huge.
Also really interested if a future version of your model can start beating human projections. Models crush in NBA and MLB, do ok in golf, but seem to be giving up edge in football and mixed martial arts. Lack of sample sizes make eyeball numbers do a bit better, but there has to be a way to model it better than human brains are capable of.
Thanks for the feedback! The model here takes a huge hit for not having a projection on a player until they hit at least two sample points in a season. Even then, their gamma curves are really weird looking and inconsistent until they've hit their 8-9 game mark. In backtesting this is around where the model started winning consistently, hoping that trend continues.
In news of the positive, this upcoming week is my first with rookies present. Although as I will write soon, their estimates tend to be very conservative to start because of the lack of a body of work.
One thing I am tweaking a bit this week is the statistical significance of defenses and the impact that has on point projections. Modeling has it count for a little more of the variance in points projections than what I have already accounted for.
I hope no one hates me for saying this, but I haven't tried applying the model to baseball yet because I'm not the biggest fan of baseball. Casual watcher at best. I am though a huge NBA fan and am currently looking into the applicability of a Bayesean algorithm there!
I completely get the lack of love for baseball. I played for 13 years and even I have trouble watching at times. But the statistics part of it is fascinating. Events seem to occur randomly at certain frequencies, the sample sizes are enormous, and the game can be modeled with simple markov chains. You would fall in love with baseball stats. But basketball is still more fun so really interested to see what you can put together there