|
Post by Brad Stevens on Jul 5, 2023 2:40:13 GMT
I started on the offseason spreadsheet to review my player ratings. I've adjusted some based on previous discussions regarding EPM vs RAPTOR, using 0-10 feet scoring for Inside Scoring rather than 0-3 feet, etc. I've only completed the Offense ratings so far & even those could use some additional work. I had to condense the file as well since it was already too big to post. Open for suggestions on what looks like it does or does not work. Especially looking to adjust categories being used that may not align with the sim category. Hard to unpack everything in a post but let me know if anyone wants to dig in to anything specific. D5 RATINGS condensed.xlsx (631.09 KB)
|
|
|
Post by Walt Frazier on Jul 5, 2023 20:42:20 GMT
PASS Rating - What exactly is "Ast-Bad Pass TO per 36"? I mean...I guess I understand actually, but, this is not how NBA Live thinks about the pass rating. - Pass rating for NBA Live is very very very tightly tied to assists per game. - And, I believe, zero effect on TO per game.
My problem with how I think your rating works, is that you could get a pretty high Pass Rating even if you only put up like... 2 ast/36, as long as you don't turn the ball over on bad passes?
And if you had a high pass rating as a 2 ast/36 guy (from RL), you're gonna be an 8-9-10 ast/36 guy in NBA Live.
|
|
|
Post by Brad Stevens on Jul 5, 2023 21:12:03 GMT
The idea behind the PASS rating metric is you are rewarding the players who are getting assists without making "bad" passes. Compare KPJ versus SGA (homer alert). KPJ had 5.7 Assists/game (338 total) while SGA had 5.5 (371 total). KPJ also had 113 bad pass turnovers compared to 84 for SGA. This really only puts SGA's PASS rating 3 points ahead of KPJ but it explains the thought process as he would be behind on assists/game alone. The question becomes do you want to capture assists/game or how good a passer the player is.
Also, the top guys are getting 8 or 9 (Assists-Bad Pass TOs)/36. A player who averages 2 assists with zero turnovers would get a PASS rating in the 20s.
I kind of like this one personally. FG & 3PT are ones that seem to have some holes currently. Also not sure how to truly capture Strength & Quick. I was thinking a combination of converted 2K interior D/strength & perimeter D/quickness.
|
|
|
Post by Walt Frazier on Jul 5, 2023 22:14:38 GMT
The idea behind the PASS rating metric is you are rewarding the players who are getting assists without making "bad" passes. Compare KPJ versus SGA (homer alert). KPJ had 5.7 Assists/game (338 total) while SGA had 5.5 (371 total). KPJ also had 113 bad pass turnovers compared to 84 for SGA. This really only puts SGA's PASS rating 3 points ahead of KPJ but it explains the thought process as he would be behind on assists/game alone. The question becomes do you want to capture assists/game or how good a passer the player is. Also, the top guys are getting 8 or 9 (Assists-Bad Pass TOs)/36. A player who averages 2 assists with zero turnovers would get a PASS rating in the 20s. I kind of like this one personally. FG & 3PT are ones that seem to have some holes currently. Also not sure how to truly capture Strength & Quick. I was thinking a combination of converted 2K interior D/strength & perimeter D/quickness. Gotcha, I didn't fully / critically think through the rate stat on Pass. I suppose that works, then. I at least like that the FG/3PT have some volume built in. Whether or not they should be weighted to some extent, is probably something to talk through? Or maybe you already are? - For example, if the FG Rating equals 80-parts FG% vs 20-parts volume? Or is it 50/50? Or...what? Strength and Quick are definitely difficult. I think that could be a good place to start...sometimes it's easier to adjust than it is to come up with the "right" thing to start. So, let's start somewhere and see what it exposes, maybe?
|
|
|
Post by Brad Stevens on Jul 5, 2023 22:52:29 GMT
For FG I'm weighing PPG, Mid Range%, & Mid Range FGM/game equally (Mid Range being considered as two pointers outside of 0-3feet). You think it should be more weighted towards PPG & FG% (or maybe 2P%)?
3PT weighs 3PT% & 3PTM/game equally. I think the adjustment might be having a low baseline (I think I've heard you say 28% before?) & then weigh how many makes above that level. Below could be some examples.
Steph Curry (.427-.28)x11.4 (3PA/game) = 1.68 3PM/g above baseline Luke Kennard (.494-.28)x4.6 (3PA/game) = 0.98 3PM/g above baseline Jalen Green (.338-.28)x7.3 (3PA/game) = 0.42 3PM/g above baseline Klay Thompson (.412-.28)x10.6 (3pA/game) = 1.40 3PM/g above baseline Jusuf Nurkic (.361-.28)x2.3 (3PA/game) = 0.19 3PM/g above baseline Max Strus (.350-.28)x7.0 (3PA/game) = 0.49 3PM/g above baseline
UPDATE - I think Steph kind of breaks doing it this way.
|
|
|
Post by Walt Frazier on Jul 5, 2023 23:29:45 GMT
For FG I'm weighing PPG, Mid Range%, & Mid Range FGM/game equally (Mid Range being considered as two pointers outside of 0-3feet). You think it should be more weighted towards PPG & FG% (or maybe 2P%)? 1. Didn't you also classify "Inside" as 0-10 feet? So, should mid-range be 10 feet to the 3pt line? 2. I'm not sure the proper distribution, tbh...I like that there are 3 factors here as it's complicated. Currently I am still "fiddling" with this over the months/years of changes, to be honest. I am not entirely convinced that, if we give a big man like Gobert a very high Inside and then a very low FG (because he doesn't really get a lot of points, and doesn't take mid-range shots at all, really), that he will perform in our sim, properly. I think that's a bigger question to solve, first, than this. But, for sake of argument...assuming we got that sorted, just for this thought-exercise... 3. My initial gut says something like: Mid Range % should be about 45% of this rating? PPG should be about 40% (as this also captures some volume on it's own in general. AND, FG rating seems to be involved in shot-distribution in the sim, we think...). And then the last 15% could be the mid-range FGM/game?? Total guesses, just an initial thought. 3PT weighs 3PT% & 3PTM/game equally. I think the adjustment might be having a low baseline (I think I've heard you say 28% before?) & then weigh how many makes above that level. Below could be some examples. Steph Curry (.427-.28)x11.4 (3PA/game) = 1.68 3PM/g above baseline Luke Kennard (.494-.28)x4.6 (3PA/game) = 0.98 3PM/g above baseline Jalen Green (.338-.28)x7.3 (3PA/game) = 0.42 3PM/g above baseline Klay Thompson (.412-.28)x10.6 (3pA/game) = 1.40 3PM/g above baseline Jusuf Nurkic (.361-.28)x2.3 (3PA/game) = 0.19 3PM/g above baseline Max Strus (.350-.28)x7.0 (3PA/game) = 0.49 3PM/g above baseline 1. 28% isn't hard and fast, but generally if players shoot much below that they kind of aren't taking many 3's to begin with. That's a broad statement. I think when I used 28% in the past, it was based on my little guideline that was something like: 90+ = 40% shooter from 3 (I do also add, I try to wait for a 2nd year of a player doing this, AND, at a decent volume) 80-89 = 37-39% shooter from 3 70-79 = 34-36% shooter from 3 60-69 = 31-33% shooter from 3 50-59 = 28-30% shooter from 3 Like I said above, once they shoot less than 28% (really, probably once they're under 32 or something) they start to get discouraged from shooting 3s, USUALLY. Depends if the player is developing and if the team is also developing, among many other factors of course. So I just don't see it as much, basically. And if I do, they shoot so infrequently that they also tend to not have much of a 3pt rating worth worrying about, due to the combined factors. 2. Not sure what I think about your examples yet but I like the general thought behind it. If it does work / everyone likes it, the next question is how to transfer to actual ratings. 2b. And...is there a simpler version that could get the job done? Such as the % itself with some sort of weight for volume / lack of volume? 3. Just thinking further, I'm not sure we should use 28% as a baseline. Don't we want to kind of figure out the "league average" for something like this? Maybe not, I guess it's just two different ways to slice the apple. But, what is the "baseline" 3pt rating. 50? All I know is that the scale / guide I typed out seems to work fairly well in the game. It's mostly figuring out a better way to consistently capture volume and how much of a factor that should be to raise / lower based off of that guide. I think?
|
|
|
Post by Brad Stevens on Jul 5, 2023 23:49:39 GMT
Also, I'll try to explain how I convert a stat category to our rating. I first identify a Max, Median, & Minimum for each rating & stat category. I'll use O Awr as the example which is currently using EPM Offense.
O AWR - Max = 99, Median = 74, Min = 45 EPM OFF - Max = 8.1, Median = -1, Min = -6.5
I essentially use the stat category (EPM OFF in this example) to identify if it is above or below the Median (-1) & then figure the rating based on the difference between the median to the max or the median to the min (if below median). Below are some examples.
Damian Lilliard - 8.1 (EPM) = 99 (O Awr) David Duke - -4.3 = 57 Jayson Tatum 4.9 = 90 Kevin Porter Jr. 1.3 = 80 Mikal Bridges 1.3 = 80 Nikola Vucevic 1.3 = 80
|
|
|
Post by Brad Stevens on Jul 6, 2023 1:31:37 GMT
After looking at 3PT the best option may be to add a 2nd table just like you have for % but for 3PM/g. You could take the average of the two results.
90+ = >3.9 (3PM/g) 80-89 = 3.0-3.9 70-79 = 2.0-2.9 60-69 = 1.0-1.9 50-59 = 0.1-0.99
90+ = 40% shooter from 3 (I do also add, I try to wait for a 2nd year of a player doing this, AND, at a decent volume) 80-89 = 37-39% shooter from 3 70-79 = 34-36% shooter from 3 60-69 = 31-33% shooter from 3 50-59 = 28-30% shooter from 3
|
|
|
Post by Brad Stevens on Jul 6, 2023 23:14:12 GMT
Updated through the Defensive categories.
|
|
|
Post by Brad Stevens on Jul 7, 2023 14:56:31 GMT
D5 RATINGS condensed.xlsx (628.84 KB) The bulk of the work is done. Added the 2K rating & a combined rating to help with normalizing. Willing to discuss any improvement areas. I'll be making a file to enter the data individually from the several sites used that pushes out a rating result.
|
|
|
Post by Brad Stevens on Jul 8, 2023 0:09:24 GMT
D5 RATINGS condensed.xlsx (612.98 KB) Added a tab where you can see the sites/stats for the calculations. The data can be entered to give the ratings results. Also, made corrections to some errors found.
|
|
|
Post by Andrei Kirilenko on Jul 8, 2023 0:23:27 GMT
This is incredibly awesome. I haven’t had a chance to go through the specifics yet but the spreadsheet looks fantastic.
|
|