Refactor update pass quality #481
Conversation
| However, what if a pass comes from a 'bad' passer? Or what if the pass is from a 'great' passer, but external factors | ||
| (such as the pass being contested) make the pass less effective? We will set the lower bound to -2.5% for now, and 4% for the upper bound. |
There was a problem hiding this comment.
Do we actually need to consider this? Obviously that's how it happens IRL, but when you're just looking at text play-by-play you're never going to notice if this pass gave a 4% bonus but the next one gave a 2% bonus. You would only maybe slightly notice the bonus in overall season stats and team performance being slightly better. But that'd look the same as if both passes had a 3% bonus.
For stuff like that, where it's something realistic but not noticeable in the game, I tend towards just leaving it out of the simulation. Makes things faster and simpler!
There was a problem hiding this comment.
Okay! took your suggestions here and removed some of the complexity for this function. Now its just a flat -0.015 - 0.03, with the passingAtr divided by 5000, so that a 100 passAtr would give 2%, 50 would be 1%, and this is added onto the random number generated between the range
There was a problem hiding this comment.
I am skeptical we even need the random number. Nobody will notice that either. Just make it scale compositeRating.passing between those limits. Then the only thing to do would be testing to make sure it didn't change stats too much, but I can do that.
Besides this, do you think this is done? I should proceed with testing?
There was a problem hiding this comment.
Yup, id classify this as done! I scale it now between the limits. I think the only thing that needs to be checked is if this somehow makes the shot prob negative or positive
| /* | ||
| We want the % chance for better shot to be a higher number, as there are ~ 1.5-2 AST/TOV ratio for NBA teams | ||
| https://www.basketball-reference.com/leagues/NBA_2025.html#per_game-team | ||
| */ |
There was a problem hiding this comment.
Can you replace this comment with a quick explanation (doesn't have to be any longer than what you have already) about the effect of pass quality on FG%, including some source ideally? On Discord you linked to https://www.nba.com/stats/players/passing?dir=D&sort=POTENTIAL_AST but like where are you getting the (-0.015, 0.03) range from?
There was a problem hiding this comment.
Originally I only did a few players on this list and it averaged to about a 4% increase, but as I did more on the list, it jumps up to about 7%. I don't know if 7% seems like a bit too high of a number? And the -0.015, that isn't as easy of a percentage to calculate since this same calculation won't hold with worse passers, since the sample size would be lower
There was a problem hiding this comment.
Okay I see. What is league average, can you do that? Or per team average, does it vary much?
I'll have to do some testing to see, but it could be that we also need to decrease the FG% for non-assisted shots to compensate.
There was a problem hiding this comment.
Ill send you the google sheet I made over discord. I took the player with the highest assist number on the team and used it for my sample. The numbers ranged from 3%-16%, with an average of ~9.71%. Take this as you will, not entirely sure that my methodology was the best for determining the bounds... I could also take the second highest player for each team and then see if this formula holds?
Some other interesting stuff, was the best passers didn't necessarily have the highest conversion rate, and it seems that big men have higher conversion rates overall?
|
Copying some stuff I told you on Discord just to keep everything about this PR in one place... one problem with that methodology (determining magnitude of FG% increase from a pass by comparing total assists and potential assists vs overall FG%) is it doesn't normalize for shot type/quality. like maybe 20% of non-assisted shots are layups, but 30% of assisted shots are layups. that should be modeled differently than if it's just based on openness, like if the % of different shot types is the same but assisted shots tend to be more open. if it's openness, then your PR represents that well. but if it's shot type, it doesn't but, still might be better than nothing.. on the other hand... the "synergy" stuff in game sim code already handles the effect of having good passers, and it does that both by adjusting the distribution of shot types and adjusting the FG%. that's might do good enough to model reality. like you might see similar stats in BBGM about assisted vs non-assisted, just because good passers lead to better shots and good passers tend to produce more assists ( so overall i think it's unclear if further adjustment is needed, and if it is needed it's unclear if this PR is exactly what's needed |
update pass quality. notes r in the commit, ill take them out later