formulas for nba awards

[nicidob]

Just some linear regressions from NBA results. Added a new stat (frac_ws) for fractional win shares, to reward players who are clearly the best player on their team, not just part of a duo or big 3.

not sure if I truly believe or if I did all the frac_ws stuff right. But posted for reference

[dumbmatter]

If it's only used in awards, I'd probably compute frac_ws on demand rather than store it in the database. You don't have to do it, I can change that if I decide to merge the PR, I'm just talking for the sake of talking.

But are the awards in BBGM actually a problem now? Will anybody notice much of a difference with this new method? I'll try to play around with it soon and see if I can tell, but if you have some examples already, please share :)

Also if you happen to have any details about the regression you did, would be good to put it in a comment or something, otherwise it will look very magical 6 months from now.

[dumbmatter]

Probably should scale by games played, like in my old formula. Otherwise you could get a ROY with very few games played. Maybe same thing for SMOY.

[dumbmatter]

These coefficients for blk and stl are so low they barely do anything, should we just get rid of them if that's what they data really suggests?

[nicidob]

i screwed up. those used totals not per-game numbers. also why i thought no scaling

[dumbmatter]

https://github.com/zengm-games/zengm/compare/nicidob-awards?expand=1 is my version of this. Differences:

• Dynamically computes fracWS
• ROY formula scales pts by gp

Overall it's a pretty minor change from the old award formulas, and I think hardly anyone would notice. But in my unscientific testing, when there is a difference, it tends to look reasonable. So I'll probably merge that branch, I just wanted to give you a chance to say something if you want.

[dumbmatter]

I'm going to put it in the next update.

[nicidob]

@dumbmatter so yeah my bad. Here's a script that generates those coefficients from a "all historical data" league file.

https://github.com/nicidob/bbgm/blob/master/historical-gen.ipynb

I had a minor mistake in the DPOY one where I didn't use per-game BLK and STL. This script should output dividing numbers for the coefficients

First Team All-League
winp        1.0
ewa        22.0
vorp       32.3
frac_ws    10.2
dtype: float64

Sixth
winp     1.0
pts_g    9.9
ewa      5.5
vorp     2.3
ws       4.9
dtype: float64

Rookie of
ewa      2.1
vorp     1.0
pts_g    2.0
dtype: float64

First Team All-Defensive
dws      1.3
blk_g    3.0
stl_g    1.0
dtype: float64