User Pass
Home Sign Up Contact Log In
BP
offline
Link
 
I run the Offense for Black Sea Storm. For this scrimmage

http://goallineblitz.com/game/game.pl?game_id=882496


Here is the input

Between our 0 and their 20 on first down

11% Medium Passes (I classify long passes out of 2 WR sets as "medium" too)
16% Short Passes
21% Inside Run
26% Outside Run
26% Long Pass



Inside of their 20, last 2 minutes of the half, if we are up with less then 4 minutes left...inputs are different.

During this scrimmage we had 25 such first downs...here are the results

11% medium passes called for in AI - Fired 5 times = 20% +9
16% short passes called for in AI - Fired 5 Times = 20% +4
21% Inside Run called for in AI - fired 9 Times = 36% +15
26% Long Passes called for in AI - fired 5 times = 20% -6
26% outside run called for in AI - FIRED 1 TIME! = 4% -22%



The chances of a 26% output only firing once in 25 tries is

0.001337855

or 1 time out of 1000...or 4 times out of every 3000.

Inputs/outputs are fine. Auto adjust is off on the basic, package auto adjust is off.

I had another OC in a pro league tell me that it's gotten so bad that he has gone back to inputting specific plays in his AI and doesn't use the packages at all.

Is something wrong with the AI roller bort?
Edited by BP on Nov 7, 2009 12:33:18
 
Sixx
offline
Link
 
This is my understanding to how the sim works.

You have to think of it as drawing (like drawing numbers from a hat) and the percentages are how many times that certain play is placed in the drawing.

Play A = 25%
Play B = 25%
Play C = 50%

100 pieces in the draw.
Play "A" and "B" would be put in 25 times a piece and "C" put in 50 times.

Now the sim draws from those 100 choices. Even though there are more "C" choices in the draw doesn't mean it's necessarily going to draw it.

I've actually had it work the other way too. Play A gets 1 call, B none and C gets called 20 times.
It sucks but that's how it goes sometimes.
Edited by Sixx on Nov 7, 2009 12:52:40
 
Octowned
offline
Link
 
Not sure how OP did his math, but here is how I believe it should have been done:

> x=pbinom(1,25,0.26)
> x
[1] 0.005263806

The odds of observing something as extreme as "only called once out of 25 trials with a 26% chance each time" should also include the event of having zero outcomes. Such an event has a binomial distribution (coinflip 25 times, each with 26% chance of "heads", count how many "heads"), and the odds are shown above as .00526.

Now, in a season, you have 32 teams playing 16 games each.. that's 512 AIs that get run through.

The odds of this happening to somebody in a league is then

> 1-pbinom(0,512,x)
[1] 0.9329413

93%!!

And this is only looking at something "this rare," I mean there are more than 1st down outputs, there are tons of leagues, there are tons of scrimmages, etc.

Stranger things have happened, be sure of it! The problem is that OP is looking at one game and trying to make generalizations about the population (well, about the way plays are called, but that applies to all games so the population is every game that has been run). You'd have a point if you could put together several examples. But when you see a .005, or 1 in 200 chance... well come on, how many games have been played? A heck of a lot more than 200.
Edited by Octowned on Nov 7, 2009 12:55:28
 
BP
offline
Link
 
alright..thank you for helping the math, figured it wasn't perfect.

So the chances of it happening to one team in one game per season is 93%?


Funny thing is, he and I were TALKING about it on the messageboard before the game....and he brought it up. I told him what the %'s were and he guaranteed it'd be off if I counted after the game...and it was, first game I observed it.


I'm leaving the same exact offense in for next game cuz it's against a gutted squad....I'll show you the results. In my eyes, I was told it was going it happen and it did....that's more then a 5 out of 1000 times chance as your math stated.
Edited by BP on Nov 7, 2009 12:59:12
 
Octowned
offline
Link
 
Yeah I mean doing hard math on this is silly because every game is so different. Plus, you're talking about 25 of your offense play calls. How about the other 50? Did they go "as planned"? Setting up an AI with a lot of different outputs and packages and plays you simply wont get what you want, it is that simple.

All I'm trying to say is from a statistical point of view what you saw isn't a big deal. Seems lame, I know, but if you want to see something ran a lot, you need to put it higher than 26%.
 
BP
offline
Link
 
the 1st down thing was easiest to check because I have 2nd and short/medium/long...3rd short/medium/long...and that would have just taken forever. Figured this was the best way, first thing I check after I was told...it was there.

Next game is on Monday with same offense, I'll let you know what happens...but if it's another 5 in 1000 chance thing happening I think there might be an issue.
 
Link
 
I lol'd at the title!!

Prematurely maybe but I laughed none the less!!



DMP
 
Jaxon Voom
offline
Link
 
Originally posted by Dirty Mo Posse
I lol'd at the title!!


Yeah me too.
 
ThaOutlaw
offline
Link
 
Thread does not deliver.
 
AD_SoDope
offline
Link
 
Originally posted by Dirty Mo Posse
I lol'd at the title!!

Prematurely maybe but I laughed none the less!!



DMP


 
RMiller517
offline
Link
 
Thread locked by moderator.
 


You are not logged in. Please log in if you want to post a reply.