APBRmetrics

In my opinion, you shouldn't require forcing to 1230 - that is not guaranteed to improve your error (particularly if you have different variances on each). Also it's up for debate how you'd want to add or subtract the wins (eg multiplier / bias term, both). Just my 2c - maybe 1230 is some rule I don...

Very helpful Mike! Could you add 4141 back in to this post?

@mike - Can we add Pelton's in?

https://www.espn.com/nba/insider/inside ... s-slipping

Also if you could share / show how your target is generated, or just post to google sheets that would help analyze a bit.

LOOCV is nice but it does not give us a sense for the variance in the model, and Random Forest can very easily be tuned to cheat and get "100% accuracy".

Do you have a version of this with test/train accuracy splits?

Friendly reminder to everyone reading the results of this model: feature coefficients do not exist in a vacuum. For example: if you had a regression of *just* turnovers vs RAPM, turnovers would be positive because it's correlated with other successful things (namely, being a high-usage passer). The ...

Random aside - are per-100 stats in the scope of BPM 2.0?

one good interaction that may or may not improve sample size is "turnovers vs expected" i.e. projected turnovers from stat line vs actual turnovers. that has a very good RSQ in general and to me seems pretty important

also (GamesStarted/GamesPlayed)^2 seems to be a good one

2 half-baked ideas: 1) get even more interaction terms (i.e. estimate what proportion of baskets are assisted etc) 2) find mathematically sound way of capping / preventing outlier seasons from breaking the prediction for #2 you could theoretically just generate "possible stat lines" i.e. R...

Hi! This is literally all I care about these days (figuring out why and how some teams beat their Net Rating in terms of win%). Some thoughts: 1. Something is awry if Net Rating has a +.107 correlation. How is that possible? Should be removed from the equation really...just want to measure Net Ratin...

While I appreciate the early love, the spreadsheet linking to my win projections is my *LIVE* win projections which is probably why I'm in first.

Here's my actual pre-season projections: https://twitter.com/bbstats/status/1052331590263742465

RMSE is the standard and should remain unless voted on prior to the contest

*grumbles about prior 2nd place*

Just ran my retrodiction and found the following fit on *Adjusted NetRating* (not wins):

0.573 x my BPM projection
0.375 x Jerry's Multiyear + Aging Curve RPM
0.0583 x Expected Teammate Usage
0.107 x Prior yr Usg% x TS%
-6.5


Definitely overfit, but very interesting.

FWIW, I averaged KP and BMonster's minutes projections for mine.

Long live average error!

My RPM predictions vs Vegas:
2014: 20-9-1
2015: I don't know but I won the Avg Error thing at "6" so "very good"
2016 (projected): 22-8-0

haha - I just meant I missed the frequent colorful chart updates.

thanks for the wins update Mike!

EDIT: Though - this looks like average error. After winning the average error but "losing" the contest last year I'd like to see Squared error if possible - let's stick to it