talkingpractice wrote:After some consideration, we're planning to throw up an open source "RAPM light" (defense only, no prior) over the next week or two.
Would this have a box-score influence?
Regardless, that's pretty awesome, considering there's nothing of the sort out there ATM, with 2013 xRAPM having been taken down.
talkingpractice wrote:After some consideration, we're planning to throw up an open source "RAPM light" (defense only, no prior) over the next week or two.
Would this have a box-score influence?
Regardless, that's pretty awesome, considering there's nothing of the sort out there ATM, with 2013 xRAPM having been taken down.
Why was it taken down?
For Synergy stats, they should adjust the PPP defensive stats to be based on who you're guarding. If a guy averages 1.1 PPP in a certain situation and you hold him to 1.0 PPP throughout the game, you should get credit for "saving" 0.1 PPP for however many possessions that was. Some guys are hidden on defense and guard crappy players, pumping up their Synergy defensive stats. That's pretty flawed.
v-zero wrote:Will it be multi-year? Since single-year non-prior-informed RAPM is roughly useless.
sideshowbob wrote:Would this have a box-score influence?
We're flexible. Maybe weighted avg of last 3 years (weighted by a combo of recency and MP)? Tbh, I'm fine if you guys want to take the floor and let us know what you'd most like. We have the info so it's basically just a matter of putting it together into however we're going to define "light".
Has it been confirmed that J.E. got hired by a team? He's been on here a few times in the last couple of weeks, and its not like everything was taken down on his site, just 2013.
talkingpractice wrote:
v-zero wrote:Will it be multi-year? Since single-year non-prior-informed RAPM is roughly useless.
sideshowbob wrote:Would this have a box-score influence?
We're flexible. Maybe weighted avg of last 3 years (weighted by a combo of recency and MP)? Tbh, I'm fine if you guys want to take the floor and let us know what you'd most like. We have the info so it's basically just a matter of putting it together into however we're going to define "light".
Yeah either that or single-year prior-informed (non-box prior), akin to what J.E. had up on his site before xRAPM, would be my vote.
I'm sure sooner or later someone will start publishing it again. It's not like the basic form is rocket science.
I wish I were better with computers, I would like to do it, but I don't have any idea how to do any of it (parsing the PbP or doing the regularization). Hopefully, someone else puts it back up soon, so I have some hard data to confirm my suspicions about who's good on defense.
talkingpractice wrote:We're flexible. Maybe weighted avg of last 3 years (weighted by a combo of recency and MP)? Tbh, I'm fine if you guys want to take the floor and let us know what you'd most like. We have the info so it's basically just a matter of putting it together into however we're going to define "light".
How about a WLS(Ridge) estimate using, say, six years of past data, with exponentially decayed weights, with time in days and a decay constant of 0.003?
