User Pass
Home Sign Up Contact Log In
Forum > Discuss GLB Issues With Catch22 > RESOLVED BUGS > Something is broken with GLB's autoadjust - ISSUE RESOLVED
Page:
 
GMathiasf
offline
Link
 
Originally posted by Catch22
Couple things guys.

1) Need links to the games in question
2) Need links to the AI for the game in question

Not taking this any further until I can actually review the data and information available.


1) Gave you links to my games.
2) Recited the exact AI in question. I'm not some noob that has screwed something up. Feel free to PM me if you want to go through Canton's AI/packages with me.
 
Novus
offline
Link
 
Originally posted by TrevJo
Originally posted by Novus

I don't think it's Auto-Adjust, because I have Auto-Adjust turned OFF entirely, and I'm still getting odd results.


Yeah I think anyone doing these experiments with auto-adjust should also do them with auto adjust off.

That's what made me speak out. I was about to say "LOL AUTO-ADJUST NOOB," but before I could hit Post, I remembered my "WTF?" reaction to the play-distribution in my own game. And since I don't use Auto-Adjust, it got me wondering.

We'll see what comes of it when my next game sims today.
 
Novus
offline
Link
 
Originally posted by GMathiasf
I'm not some noob that has screwed something up.

GAH STUPID STUPID STUPID HOW COULD YOU POST THAT

Do you realize the number of people over the years who have posted something in the Bugs Forum like "I'm not some noob, I didn't screw up my AI," only to either realize 10 minutes later that they did in fact screw up their AI or have a mod with AI access point out publically that they did in fact screw up their AI? Coz it's kinda high.

Just cursed yourself, dude.
 
GMathiasf
offline
Link
 
Originally posted by Novus

GAH STUPID STUPID STUPID HOW COULD YOU POST THAT

Do you realize the number of people over the years who have posted something in the Bugs Forum like "I'm not some noob, I didn't screw up my AI," only to either realize 10 minutes later that they did in fact screw up their AI or have a mod with AI access point out publically that they did in fact screw up their AI? Coz it's kinda high.

Just cursed yourself, dude.


trust me on this one. It's hard to mess up an AI with Input 1st/2nd down all and output 1 package. That package has 10 plays at 10% each. I can promise that this is not an AI error.
 
TrevJo
offline
Link
 
Originally posted by TrevJo
Originally posted by Novus
I don't think it's Auto-Adjust, because I have Auto-Adjust turned OFF entirely, and I'm still getting odd results.


Yeah I think anyone doing these experiments with auto-adjust should also do them with auto adjust off.


That said, the play got called more and more each quarter. First quarter: 5. Second quarter: 9. Third quarter: 14. Fourth quarter: 16

Here is an observation.

Start of the game...
WR Clear Out - 0 yards
WR Clear Out - 6 yards
ignore, 3rd down Shallow Cross - 7 yards
TE Cross Slant - 0 yards
Shotgun Smash - 20.5 yds (1d)
I Pitch Weak - 0.5 yd loss
TE Cross Slant - 4 yd loss
ignore, 3rd downWeak Flag Deep Post - 6.5 yd loss
Shotgun Smash - 14 yds (1d)
Shotgun Smash - 7.5 yds
HB Blast - 10 yds (1d)
HB Blast - 2 yds
Shotgun Smash - 16 yds (1d)

So early on I can see why it would be adjusted up. Run 4 times for 3 big positive results plus another positive result. Of the other 8 plays, 1 was a big positive result, 2 others were positive results, and 3 were losses. That means roughly you have 7 of 12 plays adjusted it up and only 3 of 12 plays adjusted it down.

What I wonder is if plays don't adjust down quickly once they've gone up. Or if other plays can't catch up. Here are the next few plays:

Shotgun Smash - 0 yds
WR Screen Strong - 2 yds
ignore, 3rd down HB Clear - 9 yds (touchdown)
HB Counter Weak - 4 yds
HB Counter Strong - 4 yds
ignore, 3rd down Weak I HB Slam - 1 yd (no first)
Shotgun Smash - 0 yds
Shotgun Smash - 0 yds
ignore, 3rd down Weak I Flag Deep Post - 12 yds (1d)
Shotgun Smash - 0 yds
WR Screen Strong - 42 yds (1d)
HB Counter Strong - 4 yds
WR Screen Strong - 3 yds
ignore, 3rd down Weak I HB Slam - 2 yds (no first)
ignore, 3rd down Weak I HB Slam - 1 yd (touchdown)
Shotgun Smash - 0 yds
HB Counter Weak - 61.5 yds (1d)
FB Slam - 0.5 yds
HB Dive Weak - 2 yds
ignore, 3rd down Weak I HB Slam - 0 yds (touchdown)

(some of these are goal line plays, not sure if that matters)

If I remember correctly, 0 yards doesn't adjust you down. But 5 other plays that got run resulted in first downs or touchdowns, with no losses, so that should have adjusted it down somewhat.

Shotgun Smash - 0 yds
Shotgun Smash - 8 yds
ignore, 3rd down WR Clear Out - 7 yds (1d)
HB Counter Strong - 2.5 yds
Shotgun Smash - 0 yds
ignore, 3rd down Middle Overload - 4.5 yd loss
Shotgun Smash - 0 yds
Shotgun Smash - 0 yds
ignore, 3rd down WR Hook - 0 yds
TE Cross Slant - 0 yds
HB Blast - 10 yds
Shotgun Smash - 20 yds (1d)
Shotgun Smash - 18 yds (1d)
Shotgun Smash - 9 yds

So here is a theory:
-Early big plays from Shotgun Smash pushed it up to the top (no surprise there)
-Early negatives from other plays maybe adjusted Smash up more than it adjusted other plays up, since it was a higher base % at the time?
-The 0 yds (incompletions), especially in the second group where the play was not really successful) didn't adjust it down
-Another loss from another play in the third group, and then a few more big plays from Smash to start the 3rd quarter (end of the 3rd group above) just pushed it way up again.
-Out of the 44 times it was run, Smash lost yards only once (the 25th time it was run)... so the theory is the incompletions didn't hurt the auto adjust value and the losses from other plays and the gains from Smash just pushed it way up
Edited by TrevJo on Sep 16, 2011 11:09:17
Edited by TrevJo on Sep 16, 2011 09:23:23
Edited by TrevJo on Sep 16, 2011 09:22:24
Edited by TrevJo on Sep 16, 2011 09:21:55
 
Novus
offline
Link
 
Originally posted by GMathiasf
Originally posted by Novus


GAH STUPID STUPID STUPID HOW COULD YOU POST THAT

Do you realize the number of people over the years who have posted something in the Bugs Forum like "I'm not some noob, I didn't screw up my AI," only to either realize 10 minutes later that they did in fact screw up their AI or have a mod with AI access point out publically that they did in fact screw up their AI? Coz it's kinda high.

Just cursed yourself, dude.


trust me on this one. It's hard to mess up an AI with Input 1st/2nd down all and output 1 package. That package has 10 plays at 10% each. I can promise that this is not an AI error.


STOP DIGGING

 
Novus
offline
Link
 
Originally posted by TrevJo
If I remember correctly, 0 yards doesn't adjust you down.

Whoa whoa whoa whoa whoa, wait a sec...

Are you telling me that if I call one pass play 5 times in the first quarter for 5 incompletions, Auto-Adjust WON'T ADJUST THAT PLAY DOWN AT ALL?

Man, if that's really how Auto Adjust works, that's just plain stupid.

Please to be providing source for this at earliest conveniences kthxbai
 
TrevJo
offline
Link
 
I don't know that for sure.
Here is some recent discussion/speculation about how auto adjust works with passes: http://goallineblitz.com/game/forum_thread.pl?thread_id=4734964&page=1#43525418

0s not counting as negative would explain it, though, don't you think? In the Bort quote I posted in the above link, there's no mention of incompletions or 0 yardage plays adjusting you down. And it says small gains such as 1-2 yards have no effect. The Bort quote is a little vague though as it mentions losses and 1-2s, but it doesn't mention 0s or 3s, etc.

GMath mentioned the Smash's average, only 4 yards or so. But it doesn't sound like the average matters.
In the case of every play starting with the same %, it might make sense to design the system that you give a play an adjusted % based on the play's average.
BUT if you have say one play at 10%, one play at 20%, and one play at 70%, you can't do it just based on the play's average. If you did, I would think the 20% could never get ahead of the 70% no matter the play results. You just have to adjust them up or down (or not at all) one play at a time, instead of using the average play result. In that one-play-at-a-time system, a bunch of positives and a bunch of zeroes is going to result in a huge net positive eventually.
Edited by TrevJo on Sep 16, 2011 09:44:21
Edited by TrevJo on Sep 16, 2011 09:43:24
Edited by TrevJo on Sep 16, 2011 09:41:41
 
Novus
offline
Link
 
Originally posted by TrevJo
I don't know that for sure.
Here is some recent discussion/speculation about how auto adjust works with passes: http://goallineblitz.com/game/forum_thread.pl?thread_id=4734964&page=1#43525418

0s not counting as negative would explain it, though, don't you think?

That would indeed explain the results.

But man, if an incomplete pass has no effect at all on Auto-Adjust, again... that's just plain stupid. An incomplete pass is a FAILURE. If I'm throwing nothing but incompletions on one particular pass play while using Auto-Adjust, I would want Auto-Adjust to step in at some point and say, "Okay, this play isn't working. Let's dial it down and try something else."

Heck, in a lot of cases, I'd consider a mere 1 or 2-yard gain on a passing play to be a failure as well.

Exhibit #497 why I don't like to use Auto Adjust.
 
Novus
offline
Link
 
Question posed in the Q&A thread:
http://goallineblitz.com/game/forum_thread.pl?thread_id=4731373&page=6#43541951
 
jtrav21
taco
offline
Link
 
Fwiw Iv dealt with this issue since I began OC many many seasons ago, and it seems to be gettin more and more skewed.
 
tautology
offline
Link
 
Originally posted by Catch22
Couple things guys.

1) Need links to the games in question
2) Need links to the AI for the game in question

Not taking this any further until I can actually review the data and information available.


Game in question: http://goallineblitz.com/game/game.pl?game_id=1813341

AI in question: http://goallineblitz.com/game/team_offense_ai.pl?team_id=4451&id=180646

Package in question : http://goallineblitz.com/game/team_package.pl?team_id=4451&type=o&edit=231379


Don't mind GMat, he may be a grumpy n00b but he's right...this is an AI that has been used successfully for quite some time and the playcalls are clearly coming from the package in question, and the package is simple enough to understand.

10 plays, 10% each, Auto Adjust Slowly.

For the first 3 quarters, that package was called on all first and second downs outside of goalline range.

For the 4the quarter that package was called for 70% of the 1st and 2nd downs, and another package was specified for 30% of the 1st and 2nds.


Is it possible for a package to get corrupted in some way?


 
Sal Basss
offline
Link
 
Originally posted by tautology
Is it possible for a package to get corrupted in some way?




Eh, I don't think it has anything to do with packages. I don't really use them in my O AI and I'm seeing it happen to me too. I'll see if I can get some data together...

Edit: No auto-adjust either.
Edit 2: Nope, I've changed the O AI already, so this won't matter.
Edited by Sal Basss on Sep 16, 2011 11:22:25
Edited by Sal Basss on Sep 16, 2011 11:04:10
 
PING72
offline
Link
 
Originally posted by Novus
Originally posted by TrevJo

I don't know that for sure.
Here is some recent discussion/speculation about how auto adjust works with passes: http://goallineblitz.com/game/forum_thread.pl?thread_id=4734964&page=1#43525418

0s not counting as negative would explain it, though, don't you think?

That would indeed explain the results.

But man, if an incomplete pass has no effect at all on Auto-Adjust, again... that's just plain stupid. An incomplete pass is a FAILURE. If I'm throwing nothing but incompletions on one particular pass play while using Auto-Adjust, I would want Auto-Adjust to step in at some point and say, "Okay, this play isn't working. Let's dial it down and try something else."

Heck, in a lot of cases, I'd consider a mere 1 or 2-yard gain on a passing play to be a failure as well.

Exhibit #497 why I don't like to use Auto Adjust.


I agree with you mostly, but not as definitive. Let's say I dial up a SB slant pass and the other team messed up their AI and did not cover the wr3 at all. The first time I call it the guy drops the wide open pass. The second time is a 'bad throw' due to a blitz. I WANT the AI to keep calling this play b/c it's open.

The problem is reading if the play didn't work (incompletions b/c of a great defense), or incompletions b/c of dropped passes, bad throws, sacks, etc.
 
TrevJo
offline
Link
 
Originally posted by PING72
I agree with you mostly, but not as definitive. Let's say I dial up a SB slant pass and the other team messed up their AI and did not cover the wr3 at all. The first time I call it the guy drops the wide open pass. The second time is a 'bad throw' due to a blitz. I WANT the AI to keep calling this play b/c it's open.

The problem is reading if the play didn't work (incompletions b/c of a great defense), or incompletions b/c of dropped passes, bad throws, sacks, etc.


Auto adjust will never be able to account for that. That's up to the coordinators to find the likely successes and give them a high enough base % that they can keep getting called even if they fail a couple of times at the start.

But yeah, it seems to me that a 0 yard gain should be a small downward adjustment. I'm not sure if I believe an incompletion should matter. (Is a 0 yard pass any worse than a 0 yard run?) Ideally, a a 6 yard loss should be a bigger downward adjustment than a 2 yard loss, and a 3.5 yard gain should be better than a 0.5 yard gain. I'm not sure if any of those are the case right now. I do think that 2 yard loss both is and should be a bigger downward adjustment than an incompletion.
Edited by TrevJo on Sep 16, 2011 11:50:17
Edited by TrevJo on Sep 16, 2011 11:42:19
 
Page:
 


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