Warrior General Games

what's the opinion of the masses here? and what setting do yall use? do you use it only in packages or in your AI as well?

my logic has always been if a package has a lot of plays, use very quickly; if fewer plays use a slower setting like medium or below.

similarly if you have a bunch of outputs under one input, a higher AA setting is preferred.

all that said it's p clear I am not very good at this game so I'm probably doing it wrong :|

Originally posted by lardaddy
my logic has always been if a package has a lot of plays, use very quickly; if fewer plays use a slower setting like medium or below.

That's the exact opposite of how I use it.

Each step is 3% ... 3/6/9/12/15.

If you've got a ton of plays in there, and you slap it on Very Quickly, it's probably zero'ing out that play, maybe not even able to fully do so - I never saw a Q&A say if inputs could go negative or not ...

If that's your intended behavior, cool.

But if you've got say 5 plays, each at 20%, and VQuick ... first play gets called, shits the bed, then it's 5% and then 23.75% for the other 4. I probably wouldn't go that extreme, unless it's one that's getting called RARELY at the OAI level. But smaller packages need higher adjust in general, because their starting call is a much higher %.

But if it's a WIDE package with say 10 plays (prob shouldn't be using those, but that's another topic ...) at 10% each, then you pretty much have to go lower. Medium at 9%, would drop it to a 1% call, and 11% for all the others after a bad result.

It's why I actually use the exact % in my packages, instead of letting the sim do the math. So my brain can easily do the math on how hard AA% is going to hit them. If all your packages are 100%, 200%, for all the play calls, like a lot of folks use, it's hard to keep track of what 1-2 bad calls is doing to them.

I'm a fan of mild OAI level adjust (pass/run balance, etc.), and stronger package level adjust to your second question.

My headcannon (which I think is loosely based on Bort quotes + my guesses)
Very Slowly = +/- up to 3% based on play results
Slowly = +/- up to 6% based on play results
Medium = +/- up to 9% based on play results
Quickly = +/- up to 12% based on play results
Very Quickly = +/- up to 15% based on play results

I have Auto Adjust in a few places.

OAI Run/Pass ratios on 1st/2nd down: Adjust Slowly. Since these inputs happen frequently, you don't want a few positive/negative results to take things on off the rails. More of a guiding force for what's working. I separate my runs and passes to separate packages.

Within packages: High frequency packages (1st/2nd down plays, 3rd and long plays): Adjust Quickly

Within packages: Low frequency packages (3rd and medium, 3rd and short): Adjust Very Quickly

By and large this has served me well. Keep in mind however that many OC's use WAY TOO MANY PLAYS, and that if you have way too many plays, it not only gives away all gameplan control, but it also makes it impossible for auto adjust to do it's job to hone in on good/bad plays.

You only have around 80 plays ran per game. Roughly 30 1st down, 30 2nd down, and 20 3rd down. On average you want your plays to be called 2-3x per game each, so that means your 1st/2nd down should only have around 20-25 total run+pass plays, and your 3rd down shouldn't have more than 10 plays (I normally do 5 or so on 3rd and long, 2-3 on 3rd and medium, 2-3 on 3rd and short).

Follow-up questions welcome. Good luck!

Also, kind of a tangent - but related. Be careful where you put screens. They're considered successful to the sim, if they get completed. Even if it's for like -5 yards. Looking at you, 99% of FB screens ...

Originally posted by Gambler75

But if it's a WIDE package with say 10 plays (prob shouldn't be using those, but that's another topic ...) at 10% each, then you pretty much have to go lower. Medium at 9%, would drop it to a 1% call, and 11% for all the others after a bad result.

It's why I actually use the exact % in my packages, instead of letting the sim do the math. So my brain can easily do the math on how hard AA% is going to hit them. If all your packages are 100%, 200%, for all the play calls, like a lot of folks use, it's hard to keep track of what 1-2 bad calls is doing to them.

I'm a fan of mild OAI level adjust (pass/run balance, etc.), and stronger package level adjust to your second question.

I agree with the first half of your post, so zero'ing in on the 2nd half.

To the OP:
-Keep in mind, in Gamble's scenario, that 1% is then boosted back up every time a different play is unsuccessful

-These playcalls are not pass/fail. So the range of outcomes is not simply +9% or -9%. Bort has some kind of arbitrary play result grading on the back-end, but for example a catch for 0 yards might be neutral, a catch for 4 yards might be +4%, and a catch for 10+ yards might be 9%.

-Do not do the thing of manually editing every playcall and adding them up to 100%, just leave them as 100% as a base, what Gambler is doing is an exhausting waste of time

-10 to 14 play packages work perfectly fine with heavier auto adjust, if it's a high frequency package. Example I use a 1st/2nd down pass package, that works without issue. Ideally you want Auto Adjust to "work hard" once, because if you have it auto adjusting between a ton of packages and auto adjusting between a bunch of plays in those packages, you are trying to make the game find a needle in the haystack of the "good plays"

Originally posted by Gambler75
Also, kind of a tangent - but related. Be careful where you put screens. They're considered successful to the sim, if they get completed. Even if it's for like -5 yards. Looking at you, 99% of FB screens ...

Agreed screens are kinda fucky.

I used to separate but now just lump them together with passes (or occasionally runs). Key is avoiding screens that will go for negative yards, then it won't break your AI.

thanks very much guys. I keep my packages pretty small [pause] so avoided going quick or very quick for fear of the repeat play penalty. maybe that's not a thing?

never have I done anything less than 100% for each play bc I was generally aware of the % changes relative to AA setting. I thought that AA worked like this: if a play loses yards it's called less, if it gains yards it's called more.

this absolutely blew my mind and just seems so dumb

Originally posted by Gambler75
Also, kind of a tangent - but related. Be careful where you put screens. They're considered successful to the sim, if they get completed. Even if it's for like -5 yards. Looking at you, 99% of FB screens ...

and yall talking about screens has me spiraling bc I love me some screens, so please elaborate here

Originally posted by WiSeIVIaN
Agreed screens are kinda fucky.

I used to separate but now just lump them together with passes (or occasionally runs). Key is avoiding screens that will go for negative yards, then it won't break your AI.

like should I not have screens as their own package? is that dumb and I'm doing it wrong?

I question how exactly the AI works because I have put some tests through and I'm skeptical.

I have put a bunch of plays in a package and run those in the first quarter. Those plays are in other packages for the remainder of the game, in theory (at least how I would have coded AA if I was a programmer) is whatever plays worked well in the first quarter package, would be called more the rest of the game and it just doesn't work like that. Makes me think that AA is package specific, and doesn't recognize the same play in multiple packages.

Originally posted by Robbnva
Makes me think that AA is package specific, and doesn't recognize the same play in multiple packages.

Correct

Yeah AA is package level, or OAI level. There's no play based AA.

Now if you had something like package A - inside runs. And package B - outside runs. And package C - 'safe' passes. And you set up 1stQ - something like 1st + 2ndD, call A 50%, B 25%, and C 25% ... and reused those packages elsewhere in your OAI, the adjustments to the PLAYS inside A/B/C packages that hit during the 1stQ, would carry over. What would be lost is any adjustments made between the 50/25/25% for A/B/C at the OAI level, as soon as the 1st quarter is over.

So if it ended up shifting to like 10/5/85 by the end of the 1st (extreme example), because the other team was wildly selling out to stop the run, and getting gashed by the pass ... that data ends up lost, but your safe pass package is probably nicely tuned for the 2ndQ. Hope that helps, rather further obscures.

And FWIW, that's always been my hesitance with slow starter QBs. That if you really tailor the 1stQ to minimize risks, you end up burning 1/4 of the 'intel' you've gained on the D. YMMV.

Originally posted by Robbnva
I question how exactly the AI works because I have put some tests through and I'm skeptical.

I have put a bunch of plays in a package and run those in the first quarter. Those plays are in other packages for the remainder of the game, in theory (at least how I would have coded AA if I was a programmer) is whatever plays worked well in the first quarter package, would be called more the rest of the game and it just doesn't work like that. Makes me think that AA is package specific, and doesn't recognize the same play in multiple packages.

What you can do is run your 1st quarter package by itself so the AI has time to evaluate those plays and adjust to them. Then, in the later quarters, add your second package alongside it in the AI.

That way the first package gets a clean sample early in the game, and afterward you’re effectively running it against the second package. If certain plays from the first package performed well, the AI should continue favoring them while the second package introduces additional options.

In other words, you’re staging the packages instead of running everything at once, which gives the AI clearer feedback on what is actually working.