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:
 
tautology
offline
Link
 
Originally posted by TrevJo
Well the way I think it works (based on vague recollection of Bort answers) is that if you only have 10 plays at 1% then when the sim begins it will fix them all to 10% each at the start so it's 100% total. In which case it wouldn't matter if you had 10 plays starting at 10% or 10 plays starting at 1%.

I agree with your point about multiplicative being good for avoiding zero though. I also think multiplicative is necessary so that if you have an output at 0%, it can stay at 0%. (Although it's possibly coded just to remove that output all together at the start of the sim if it's set to 0%.) tbh I have always figured it was multiplicative...

Lots of mysteries!


The other relevant question is how it adjusts other plays up or down when a play gets bumped up or down from success or failure.

Does it do this proportionally to their original weighting, or does it add or subtract a flat bonus?






 
tautology
offline
Link
 
Originally posted by TrevJo
Well the way I think it works (based on vague recollection of Bort answers) is that if you only have 10 plays at 1% then when the sim begins it will fix them all to 10% each at the start so it's 100% total. In which case it wouldn't matter if you had 10 plays starting at 10% or 10 plays starting at 1%.

I agree with your point about multiplicative being good for avoiding zero though. I also think multiplicative is necessary so that if you have an output at 0%, it can stay at 0%. (Although it's possibly coded just to remove that output all together at the start of the sim if it's set to 0%.) tbh I have always figured it was multiplicative...

Lots of mysteries!


There's also the question of whether Bort just adds the total and uses each input's percentage as a ratio against that base, or if he actually scales the proportions to a 100% base before performing additive or multiplicative changes.

This is particularly important if the changes are additive.


If you have 10 plays at 10% and one is bumped by 9% due to success, then the result is presumably 1 play with 19% and 9 plays with 9%... a pretty dramatic swing.

If you have 10 plays at 100% each (for a 1000% total) and Bort is just using ratios against the 1000% base for play determination, then the change of 9% would result in 109% for one play and 99% for the rest...not a very dramatic change.


Edited by tautology on Sep 16, 2011 13:09:37
Edited by tautology on Sep 16, 2011 13:09:06
Edited by tautology on Sep 16, 2011 13:08:49
 
johnbarber
offline
Link
 
Here's a quick and easy one-

Defensive auto adjust-
Package- 1st down weak (all yardages vs weak formation exclusively) excluding 2 min and last 6 minutes when down by x points.
13 plays in package- 7 at 10%, 6 at 5%

Plays called:
1- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4596979 - 2wr- se 1
2- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4597525 - 2wr- se 1
3- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4599683 - 2wr- se 1
4- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4601020 - 2wr- weak 2


1-3 is the bad play- I accidentally pasted cb1 to cover wr2- shit happens, I'm fine with the mistake. That's a mega failure though so next time the situation comes up I expect auto adjust (very quickly) to choose a different play. Nope, same play- long td. Third time the situation comes up- same play- long td.
Edited by johnbarber on Sep 16, 2011 15:47:24
Edited by johnbarber on Sep 16, 2011 15:47:05
 
tragula
title
offline
Link
 
A clear proof that AA doesn't work as it should

Game:
http://goallineblitz.com/game/game.pl?game_id=1792958

play in question
HB Counter Sweep Strong

Links
AI - http://goallineblitz.com/game/team_offense_ai.pl?team_id=1955&id=184385
Package - http://goallineblitz.com/game/team_package.pl?team_id=1955&type=o&edit=233188

No change was made since the game. And no change will be made. (I will make a copy if I necessary)

Story:
- The 1st 4 appearance of the play were all losses.
- AA is set to very quickly
- All other play in that package gained positive yards (until 1:41 to end of the 4th Q, which is too late)
- initial probability is 100/6 = 16.7%

1st time the play called :
http://goallineblitz.com/game/replay.pl?game_id=1792958&pbp_id=4545381
16.7% adjusted to 2%

2nd time the play called :
http://goallineblitz.com/game/replay.pl?game_id=1792958&pbp_id=4546038
2% adjusted to 0%

3rd time the play is called :
http://goallineblitz.com/game/replay.pl?game_id=1792958&pbp_id=4546625
should never been called

4th time the play is called :
http://goallineblitz.com/game/replay.pl?game_id=1792958&pbp_id=4547279
should never been called

The play actually get called 5th time, but another plays in the packages fail in the mean time to this may explain it.

package summary
play #1 (play in question) : called 5 times
play #2 : called 2 times
play #3 : called 2 times
play #4 : called 4 times
play #5 : called 2 times
play #6 : called 3 times

Final note:
There is no statistic involved, if my fact are correct (and I think they are) there is a bug. The AI and package are very small so it should not be too hard to validate this.
Edited by tragula on Sep 16, 2011 17:31:50
 
TrevJo
offline
Link
 
Originally posted by tragula
16.7% adjusted to 2%
...
2% adjusted to 0%
...
should never been called
...
should never been called


Where do these adjustment numbers come from? Are you just dividing by 6? Are you saying that this play was called the first 4 times this package was used? Without knowing when your package gets called it's too hard to determine if that's the case just from looking at the replays.
Edited by TrevJo on Sep 16, 2011 18:11:10
 
tragula
title
offline
Link
 
Originally posted by TrevJo
Where do these adjustment numbers come from? Are you just dividing by 6? Are you saying that this play was called the first 4 times this package was used? Without knowing when your package gets called it's too hard to determine if that's the case just from looking at the replays.


The package is called on 1st and 2nd down (not solely).
100 / 6 = 16.7%

17%-15% = 2% (doesn;t really matter how you run things)

2% - 15% = 0

The plays was not called one after the other, they just was never adjusted up by anything. plays in the package are:
100% SI - HB dive weak
100% I - HB slam
100% SB - HB dive weak
100% WI - HB dive weak
100% SB - HB outside handoff
100% WI - HB counter sweep strong

I am sure the play in question is not present in other packages, I didn't checked if that is true to the other (it is not relevant).

-------------

Honestly, a bug mod need to check the fact and move this to the admins. Unless I have an error there is no much room to wiggle out of a bug here.


 
TrevJo
offline
Link
 
I understand the 100 / 6 part.
But where the heck do you get 17%-15% from? (Same question for 2% - 15%.)

(If you expect any loss of yards to be -15% with adjust very quickly on, I'd love to know where you got that information.)
Edited by TrevJo on Sep 16, 2011 18:37:46
 
GMathiasf
offline
Link
 
lol same 1st/2nd down AI was run for a tourney game we just had. The results are hysterical.

http://goallineblitz.com/game/game.pl?game_id=1816713

Only 8 of the 10 plays got called this time at all. The only change I made was switching the WI WR screen to the SB WR screen for lulz.

HB Dive Weak 19
HB Counter Weak 18
HB Blast 10
TE Cross Slant 8
WR Screen 4
Shotgun Smash 2
I Pitch Weak 1
HB Out 1
WR Clear Out 0
sb counter 0

It just doesn't make any sense. Even if auto adjust is the culprit, this CANNOT be how Bort imagines it to be working out. 10 plays at 10% run 63 times should not end up with any zeros. The odds are like 1 in 1000. Having 2 zeros is like 1 in one million. So, it's obvious that auto-adjust even on slow is just taking over in an non-sensical way.
 
Weather
offline
Link
 
Originally posted by Thread title
Something is broken with GLB


/thread
 
dbreeze
offline
Link
 
The crux of this and all "bugs" is that Bort gets all the controls we want, and all we get is to try and figure out how Bort wants to play football. We should all be able to define the AA parameters that work for us. LIKE IN THE REAL GAME! I'm about fed up with it. Ya hear me Bort! No more stinkin' flex purchases for you! Hire some more damn coders to get in the guts of the sim with you and make this thing work more like real football. I ain't payin' for friggin' bizarro ball......
i've been talking about the insanity of one person trying to stay on top of this much complexity for a long time and just don't see anything changing.
What use is a free player when the game play is getting this poor?
 
dbreeze
offline
Link
 
Bort
February 6, 2011 Question I assume that the 5 settings (very quickly, quickly, medium, slowly, very slowly) give a different change to the % given in a certain play based on if its successful or unsuccessful. Could you share how much % each of these settings will change a plays %? Also, could you share what is considered a successful/unsuccessful run/pass play?
Answer "Very Quickly" is 5 times faster than "Very Slowly." Very slowly will adjust an output by between -3 to 3 depending on what happened (meaning "very quickly" will do from -15 to 15). For runs, a gain of 6 or more is a great gain, more than 3-6 is a good gain, 0-3 is neutral, and a loss is negative. For a pass, a 15 yd gain is considered really good, 6 yds is good, a loss is bad, and the rest is neutral. If the play was a turnover or a score, that's an additional + or -.

Bort
February 6, 2011 Question Can we dispel the myths of the auto adjust and would you be willing to finally tell us how auto adjust works. It's the 2nd most important part of the game behind player building, and yet we only have speculation on how the adjust works. Is it multiplicitive or additive in nature? So, if "adjust quickly" is say a 15% adjustment based on the outcome of the play, is it adding/subtracting 15% as a % of the actual % inputed (15%*X)? Or is it adding/subtracting 15% as a whole number (15%+X)?
Answer I explained the actual values in a previous Q/A. It is very straightforward. It's simply adding/subtracting points to/from the bias you enter. If you put in 50, and it's raising the score of an output by 5, that output goes up by 5, and the others get -5 split up amongst them.

What more does anyone need or need to know? Enjoy your free players and hush......
Edited by dbreeze on Sep 16, 2011 20:32:42
 
GMathiasf
offline
Link
 
So a 7 yard run is equal to a 70 yard run, makes sense.
 
TrevJo
offline
Link
 
thanks db, I had missed those
 
johnbarber
offline
Link
 
Originally posted by johnbarber
Here's a quick and easy one-

Defensive auto adjust-
Package- 1st down weak (all yardages vs weak formation exclusively) excluding 2 min and last 6 minutes when down by x points.
13 plays in package- 7 at 10%, 6 at 5%

Plays called:
1- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4596979 - 2wr- se 1
2- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4597525 - 2wr- se 1
3- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4599683 - 2wr- se 1
4- http://goallineblitz.com/game/replay.pl?game_id=1815918&pbp_id=4601020 - 2wr- weak 2


1-3 is the bad play- I accidentally pasted cb1 to cover wr2- shit happens, I'm fine with the mistake. That's a mega failure though so next time the situation comes up I expect auto adjust (very quickly) to choose a different play. Nope, same play- long td. Third time the situation comes up- same play- long td.


using Dbreeze's Bort quote- then that play should have been shut off or at least minimized significantly. It's at 10% and adjust quickly (-15% according to the quote) so it's either shut off or it's reduced to a lesser degree and the others should be improved % wise- then it comes up in the next two situations? It didn't just fail it was crap of the highest order- the reason auto adjust was invented- to get out of a bad play and try something different.

Here's the only thing that I can think of- It's a 1 package input in the AI (100% 1st down weak) so the auto adjust in the ai to adjust packages isn't on (1 package 100% doesn't need adjusting like a multi package input would). Inside the package auto adjust is turned on to very quickly.
 
Dr. E
offline
Link
 
No idea how it would be determined if Auto Adjust is broken, but in this game http://goallineblitz.com/game/game.pl?game_id=1793100 I had a package lock on to AA to use it 16 times. I only have the one package with this play, there are 5 plays in the package, one set at zero, the play locked onto, (sb flanker Drag) was set at 30%, the other three at 20 to 25% and even after giving up an interception TD with flanker drag, the play kept getting called. Package name is "short passing" and the AI is "imported offense". Auto Adjust is only on inside the package though.
Edited by Dr. E on Sep 18, 2011 07:46:04
 
Page:
 


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