2020-21 RAPM

[shadow]

It's still far too early for this to have much meaning, but here's a link to 2020-21 vanilla RAPM.

I'll try to update it every couple of weeks.

https://docs.google.com/spreadsheets/d/ ... ingle=true

[Crow]

Doncic estimate is slightly negative. 232rd rank.

Tight range. +2 to -2. I assume that will widen with more data.

D'angelo Russell estimated last in league.

Mr. 62 pts, Stephen Curry, 419th of 432.

LaVine, O Porter and Siakam are also huge paycheck guys in bottom 5% on estimated impact so far.


Hot name Herro 415th. Hot rookie name (to some, including Bulls mgt) Patrick Williams, 420th.

DeRozan won't like his 408th ranking. J Butler 407th. Zion Williamson 389. Ayton 374. G Hayward 368. D Fox 370. Beal 364. Adebayo 358. Lillard 352. LaMelo Ball 328. K Hayes 325. Coby White 318. Jerami Grant 320.

Vassell 78. Okoro 51. Mills Bridges 7.

Giannis 28. Westbrook 123. Irving and Durant, 6th and 8th. Anthony Davis 146. Ja Morant 147. Embid and T Harris 2nd and 3rd with Simmons 21. Tatum 165. Ingraham 83. SGA 212. Harden 254.

It is early.

Impact estimates also reflect on Coaches and GMs for rosters, role assignments, lineups, systems, play calls, etc.

[shadow]

Yeah, I leave the lambda at 3000 (standard found by J.E., which usually results in a range of about +/- 5 or 6 by the end of the regular season) for every run, so the early season numbers will be more compressed. There will certainly be some whacky results even with this method, but it's probably far less than what you'd see if the lambda value was prorated based on the number of games played. So I feel like this is the lesser of two evils when it comes to dealing with small sample sizes.

[rainmantrail]

Yeah, I leave the lambda at 3000 (standard found by J.E., which usually results in a range of about +/- 5 or 6 by the end of the regular season) for every run, so the early season numbers will be more compressed. There will certainly be some whacky results even with this method, but it's probably far less than what you'd see if the lambda value was prorated based on the number of games played. So I feel like this is the lesser of two evils when it comes to dealing with small sample sizes.

Are you using Python or R for your RAPM calculations? I'm curious where the lambda of ~3000 is coming from. I've read others say something similar in the past. My lambdas are usually around 0.7 for my code in R, and yields similar ranges of +/- 5 or 6 for players. I'm wondering what the difference is between our forumulas.

[shadow]

I'm using linear_model.Ridge from sklearn in python. Technically the parameter is called alpha in that function. I've never used R so I don't know how it differs from python in regards to calculating RAPM.

It's probably been posted here several times, but this was one case where I saw 3000 referred to as the standard by J.E.

http://apbr.org/metrics/viewtopic.php?f ... =90#p26269

[shadow]

Updated through games played on 1/14:

https://docs.google.com/spreadsheets/d/ ... ingle=true

[Crow]

Steph Curry estimated at -0.75. Because of offensive impact. Draymond barely positive estimate.

LaMelo Ball in bottom 100 at 364. Fell 36 places in last 10 days.

D'angelo Russell still last and by a decent margin.

Michael Porter Jr in bottom 50 and slightly negative on offensive impact estimate.

Chris Paul negative and at 357.

Marc Gasol estimated slightly ahead of Anthony Davis. KCP top, LeBron 2.

[shadow]

Updated through games played on 1/26:

https://docs.google.com/spreadsheets/d/ ... ingle=true

[Crow]

Conley #1 here and on in very top on other metrics.

D Mitchell 74 here. Shaq maybe have been rude but he is imo more right than wrong for top level star.

[shadow]

Updated through games played on 2/13:

https://docs.google.com/spreadsheets/d/ ... ingle=true

[shadow]

Updated through games played on 2/25:

https://docs.google.com/spreadsheets/d/ ... ingle=true

[shadow]

Updated through games played on 3/25:

https://docs.google.com/spreadsheets/d/ ... ingle=true

[shadow]

Updated through all regular season games:

https://docs.google.com/spreadsheets/d/ ... ingle=true

Gobert takes the crown, by a sizable margin as well.

[Crow]

Morant tied for 116th here.

Thunder with 3 of 5 lowest estimates, with Maledon by far the lowest. 2 more in bottom 5%.

Mason Jones, 518. Very limited test but he flubbed it. Too quick releases though.

Oubre and Wiseman both in bottom 20. Light years off expectations.

Patrick Williams 9th lowest.

Collin Sexton, 520. Lonnie Walker, 523. K Hayes, 512.

Low returns to date on some of the hypester favorites.

Hyped D Fox, 498.

Build around Kevin Porter Jr? 513. Build the tank around him.

Go get PJ Tucker? 509 here.

Pay big for Vucevic? 504 this season.

Front office media darling Raptors trade for G Trent. 478.

Clippers brain trust trades and extends Kennard. 485 and not playing.

Haliburton a surprise at 478.

Hyped Okoro at 460.

Prized Mavs acquisition and extended without offer sheet process Porzingis, 448. Will they try to move? Can they get a deal they can swallow?

Who wants 431 Markkanen?

LaMelo Ball ends up 415. But the hypesters are very happy / proud.

DeRozan 378. Still upset at the All-Star "snub".

Hyped SGA, 222 here.

A Davis, 87 when he played.


Good numbers for some guys too obviously. Look them up if interested.