All Forums > Gridiron Dynasty Football > Gridiron Dynasty Beta > PBP Debug- How do we interpret it?
2/19/2013 1:41 PM
Posted by katzphang88 on 2/19/2013 11:23:00 AM (view original):
I agree that in real life, upsets can occur, but usually are carried through by our emotionality of being human and rising to the occasion and motivated to play over our heads for a period in time. This is a computer numbers game - the numbers should speak. As a coach in this game - I recruit to get better numbers so I can win. That is how the recruiting is designed.

Ahrens wrote: "However, with the new formations and control, you should be able to take advantage of your strengths and their weaknesses to insure that you win all the game you should.  If you have a good interior line, and theirs is poor, you are now more capable than in the 2.0 to exploit that.  This fact in itself should really help curb the "random" game results."

This actually helps prove my point. The new formations, game planning, and the ability to make attribute match-ups in my advantage SHOULD provide my team with a decision point advantage. BUT any "random" result negates that after the fact of game planning. Some events as penalties, field position, injuries will be somewhat "random" for placement in the game. My biggest concern is that I have seen games where I play similar SIM teams, 50 - 60 points per player average difference, and one I win with my better players performing as they should, and the next they barely scrape by. Since these SIM teams can get emotionally up - why the difference? Game planning is essentially the same, but it is how the 2.0 game treats the attribute match-ups that is off and why gameplanning was not as important in 2.0 as 1.0. I can see how individual plays may be altered by the outcomes of random decision points, but I can't see in the longer view of the whole game that it should be that much different. That is why in the above example of the blocking and tackling code, consistent outcomes of each match-up should be expected when the offense or defense has the advantage of the  DefenseBreakthrough, DefenseStrong, OffenseStrong, OffensePush categories.
I think we are on the same page, but talking up different points.  I'm talking about random game results, as in random winner/random loser.  I think you are talking about each individual play.  There has to be a sense of random.  Even if Alabama is playing Northwestern Southern State, a Bama reciever may drop the ball, which is totally random.  That needs to happen to be a realistic sim.  If Northwestern Southern State beat Bama, it would be the random result I'm talking about.  (this would be an unacceptable one).

Even if your offensive line is way better, there are a couple of times that someone gets beat or whatever and the undermanned unit wins.  That's also football. 

And your point about teams getting emotionally up, I think that is real life randomness.  Bad teams are always up to take down the team that is better.  But sometimes they get slaughtered, and sometimes the other team plays poorly and lets them hang around before beating them.  Honestly, if you want the best way to limit the success of the the bad teams, fatigue needs to matter more.  The reason why good teams wear down bad teams isn't always better players, its better (much better) backups.  

My only point is that random things must happen in order to be sports.  They don't play the game on paper.  I think we are both on the same page in wanting to make this game as great as possible.
2/19/2013 3:01 PM
Posted by katzphang88 on 2/19/2013 11:23:00 AM (view original):
I agree that in real life, upsets can occur, but usually are carried through by our emotionality of being human and rising to the occasion and motivated to play over our heads for a period in time. This is a computer numbers game - the numbers should speak. As a coach in this game - I recruit to get better numbers so I can win. That is how the recruiting is designed.

Ahrens wrote: "However, with the new formations and control, you should be able to take advantage of your strengths and their weaknesses to insure that you win all the game you should.  If you have a good interior line, and theirs is poor, you are now more capable than in the 2.0 to exploit that.  This fact in itself should really help curb the "random" game results."

This actually helps prove my point. The new formations, game planning, and the ability to make attribute match-ups in my advantage SHOULD provide my team with a decision point advantage. BUT any "random" result negates that after the fact of game planning. Some events as penalties, field position, injuries will be somewhat "random" for placement in the game. My biggest concern is that I have seen games where I play similar SIM teams, 50 - 60 points per player average difference, and one I win with my better players performing as they should, and the next they barely scrape by. Since these SIM teams can get emotionally up - why the difference? Game planning is essentially the same, but it is how the 2.0 game treats the attribute match-ups that is off and why gameplanning was not as important in 2.0 as 1.0. I can see how individual plays may be altered by the outcomes of random decision points, but I can't see in the longer view of the whole game that it should be that much different. That is why in the above example of the blocking and tackling code, consistent outcomes of each match-up should be expected when the offense or defense has the advantage of the  DefenseBreakthrough, DefenseStrong, OffenseStrong, OffensePush categories.
One of the things I would suggest is to look at more than one play.  If a player has a disadvantage against a defender in avoiding the tackle but ends up avoiding the tackle, I don't see that in itself as a problem.  What I'd like to know is in 10 similar plays, how often does that happen.  If in 1 out of 10 times, the lower rated guy gets away, then I could see that.  Then I would expect to see a more even match up to see more times the guy avoids the tackle.  But to address the concern, I do think there should probably be some point in the match up, especially considering how the rest of the play has played so far, that there should just be a flat "tackle/no tackle" chance.  I will be reviewing the result code to see if I can shore that up a little to where there is a little more "if x and y then z", but I'm thinking it's going to be more about determining expected ranges of outcomes in most cases.
2/19/2013 3:11 PM
I'll also add that there is a lot of trickle down effect on the results and I will be looking at that more closely.  I think a lot more can be done especially with respect to broken tackles.  I think the chance to break the tackle isn't taking the tackle attempt result and the tackle breaking match up into account as much as it should.  For instance, a tackler trying to tackle a runner should be more likely to make the tackle if the tackle attempt is stronger.  The runner would be more likely to break the tackle the weaker the tackle attempt is.  If you mix the results around how the tackler and runner match up ratings-wise, then you can get a range of the results.  For instance, it might be only possible for a weaker tackler to tackle a stronger runner if the tackle attempt is strong.

You then push this mechanic back to the tackle attempt (avoiding a tackle).  That check would be similar but look more closely at the blocking results as well as the match ups, including the play settings.  It's more difficult to avoid a tackle if there are more defenders in the area.

The good news is that all of these checks are already in the engine.  They just need to be structured a little differently.  The bad news is that when I change these, the play result yardage is going to need to be adjusted as well and we'll have to re-evaluate the overall game results to make sure we are getting reasonable outcomes and stats.

2/19/2013 8:08 PM
Norbert wrote: "One of the things I would suggest is to look at more than one play.  If a player has a disadvantage against a defender in avoiding the tackle but ends up avoiding the tackle, I don't see that in itself as a problem.  What I'd like to know is in 10 similar plays, how often does that happen.  If in 1 out of 10 times, the lower rated guy gets away, then I could see that.  Then I would expect to see a more even match up to see more times the guy avoids the tackle.  But to address the concern, I do think there should probably be some point in the match up, especially considering how the rest of the play has played so far, that there should just be a flat "tackle/no tackle" chance.  I will be reviewing the result code to see if I can shore that up a little to where there is a little more "if x and y then z", but I'm thinking it's going to be more about determining expected ranges of outcomes in most cases."

Game on! I'll be checking. Firing up excel now!

Also - you have "momentum hits" to decrease ball carrier progress. What about "momentum bonus" when fast, elusive guys get outside?
2/20/2013 11:14 AM (edited)

Checked a D1AA team against my DIII team. Starter rating for Iona 519.1, for Manchester 384.4, so a good mis-match to check. Iona won 42 - 13.
Analyzed the BLK inside code line - 95 total plays for this code in this game. Defense Breakthrough (4) ranged from offense-defense differences of -11 points to -25 points (defense is "-" number, percentage < 100%) % differences of 67% to 77% Ave - 72%. Defense Strong (17) range -29 to -6 points, 64% to 86% -Ave 71%, Equal blocking (64) range -31 to +19 points, 61% to 137% Ave 97%, Offense Strong (8) range -8 to +16 points, 82% to 131% Ave 119%, Offense Push (2) range +14 - +17, 128% to 130% 129%. So how do I interpret these numbers. My expectations would be that the various categories would be on a sliding continuum from Defense Breakthrough through Equal Blocking and then to Offense Push. The percentage comparisons were more accurate than the raw scores and take into account the true differences in value (high vs high, low vs low). To a large extent these values followed the expected continuum. Given this sample, my opinion would be that the equal blocking category be tightened up so that the range doesn't overlap into the far ends of the distribution. (top % of 137% and bottom at 61%). So entire average would be -5.6 and Standard deviation was 17.4. Player mismatches (+/- numbers) followed very obvious pattern in offense or defense match-ups. Overall of the 95 scores in this sample, there were about 6 suspect outcomes. Round One goes to: Norbert.

Added question: Do the category names imply that added benefit has been applied to the match-up or are they just names the results fit into?

2/20/2013 5:17 PM
What exactly do you mean by "category names"?
2/20/2013 7:34 PM (edited)
 DefenseBreakthrough, DefenseStrong, EqualBlocking, OffenseStrong, OffensePush categories. As I see so far, EqualBlocking can cover ranges that overlap all other categories. (but I don't know everything that make up the final comparisons).
2/22/2013 1:09 AM
This analysis was of the TKLAvoid playline. The values reflect the comparison of OFF score divided by the DEF score. So match-ups that are above 100% would favor the Offense. All had a (0) player advantage.
Tackle avoided (#30) range 192% to 73% (ave 115%), Weak Tackle (#30) range 206% to 56% (ave 113%), Good Tackle (#42) range 187% to 68% (ave 112%), Strong Tackle (#14) range 138% to 51% (ave 93%).
My expectation would be that a Tackle Avoided would be composed of scores of the highest percentage with some overlap with Weak Tackle and always have the offensive score be much higher that the defensive score. Weak Tackle would overlap Tackle Avoid at the high end and Good tackle at the low end and be primarily offense better than defense. Good Tackle ratings would overlap Weak Tackle at the low end and Strong Tackle at the high end, but should not overlap Tackle avoid scores. Good tackle values should always have defense over offense scores. Strong tackle would overlap Good Tackle at the low end and always have much higher defense scores.
This does not occur in this playline.  Given the full distribution average was 111%, which slightly benefits the defense. Standard deviation was about 30%. So a strict range of Tackle avoid would be above 141%, Strong tackle below 81%, and the middle values divided by the average. This would change the distribution to be 20 Tackle Avoided (from 30), 28 Weak Tackle (from 30), 54 Good Tackle (from 42), and 12 Strong Tackle (from 14).
Some examples of contradictory plays were:
[TKLAvoid:InsideLine 69-36 (0) RESULT:TackleAvoided] compared to [TKLAvoid:Medium 35-48 (0) RESULT:TackleAvoided]
[TKLAvoid:InsideShort 65-37 (0) RESULT:WeakTackleAttempt] compared to  [TKLAvoid:OutsideShort 28-50 (0) RESULT:WeakTackleAttempt]
 [TKLAvoid:InsideLine 56-30 (0) RESULT:GoodTackleAttempt] compared to  [TKLAvoid:Deep 43-63 (0) RESULT:GoodTackleAttempt]
 [TKLAvoid:Medium 63-47 (0) RESULT:StrongTackleAttempt] compared to [TKLAvoid:OutsideShort 38-66 (0) RESULT:StrongTackleAttempt]
Unless there are many other factors not known to the coaches as evaluators, I would say we can’t use the numbers being compared in these playlines as accurate representations of player attributes and expected result. There is too much overlap of results and outcomes.
Also, some playlines do not have results listed, just number comparisons.
2/23/2013 5:57 PM
This analysis was of the TKLBreak playline. The values reflect the comparison of OFF score divided by the DEF score. So match-ups that are above 100% would favor the Offense. I expanded the defensive score x 10% for each momentum hit to give that characterisitc and advantage for comparison.
Tackle break (#33) range 394% to 46% (ave 116%), Weak Tackle (#23) range 365% to 47% (ave 106%), Good Tackle (#30) range 189% to 53% (ave 92%), Strong Tackle (#27) range 189% to 34% (ave 83%).
My expectation would be that a Tackle break would be composed of scores of the highest percentage with some overlap with Weak Tackle and always have the offensive score be much higher that the defensive score. Weak Tackle would overlap Tackle break at the high end and Good tackle at the low end and be primarily offense better than defense. Good Tackle ratings would overlap Weak Tackle at the low end and Strong Tackle at the high end, but should not overlap Tackle break scores. Good tackle values should always have defense over offense scores. Strong tackle would overlap Good Tackle at the low end and always have much higher defense scores.
This does not occur in this playline.  Given the full distribution average was 100%, which balances perfectly for this distribution. Standard deviation was about 55% and ranges for each category was modified to be 37% above and below the mean (as 66% of all scores) as some very high offensive score skewed the results. So a strict range of Tackle break would be above 137%, Strong tackle below 63%, and the middle values divided by the average. This would change the distribution to be 15 Tackle broken (from 33), 21 Weak Tackle (from 23), 54 Good Tackle (from 30), and 22 Strong Tackle (from 27).
Unless there are many other factors not known to the coaches as evaluators, I would say we can’t use the numbers being compared in these playlines as accurate representations of player attributes and expected result. Although the averages place the different categories in the correct distribution, there is too much overlap of results and outcomes.
2/25/2013 12:55 PM
One thing to remember about the tackles is that the rusher moves from zone to zone, shortest to deepest, and in each zone a tackle attempt is checked.  Basically we are saying that given the players involved in a certain area for both offense and defense, factoring in any blocking results, what's the chance the defense makes a tackle attempt on the rusher.  If they do get a tackle attempt on the rusher, then we ask what are the chances the defender(s) that makes the tackle attempt actually brings down the rusher.  Anything other than a WeakTackle, GoodTackle, or StrongTackle continues the play while these results mean the play is dead.

The basic flow of where we check for match ups within a play are:  Blocking (all players in area) -> Tackle Attempt (rusher and selected defenders) -> Tackle (rusher and selected defenders from tackle attempt)  -> PLAY IS DEAD OR REPEAT FOR NEXT FIELD AREA

So when we look at tackle results, we generally will have to consider the Tackle Attempt results, and when looking at Tackle Attempt results we have to look at Blocking Results.  It might be that I can make the Tackle results a lot more fixed based on match up with the randomness of the play (so we don't get all 2-yard rushes) coming from earlier in the play.  I will be looking at this more closely as I work on passing this week.
2/25/2013 1:07 PM
Posted by katzphang88 on 2/20/2013 7:34:00 PM (view original):
 DefenseBreakthrough, DefenseStrong, EqualBlocking, OffenseStrong, OffensePush categories. As I see so far, EqualBlocking can cover ranges that overlap all other categories. (but I don't know everything that make up the final comparisons).
Not sure if I've answered this but if I understand the question correctly, if you are mapping the results back to match ups, then you will find they can overlap.  We don't have many fixed results, as in "Match Up A = Result A" or even "Previous Result A + Match Up A = Result B".  It will be closer to saying "Previous Result A + Match Up A = (10% Result A, 50% Result B, 40% Result C)" whereas a different Previous Result and different Match Up would result in a different distribution.  We also do some things in the engine to help push close to the distribution rather than pure random to help with the "everything falls in the low range" problem we have with the 2.0 engine.

So when I look at results and we see something is off, I typically am adjusting these result distributions and sometimes factoring in things that I hadn't considered.  There may be times where I do set fixed distributions, like a Strong Tackle Attempt with a strong defensive tackle advantage is always a Strong Tackle, or possibly I eliminate some possibilities like you can't have a Weak Tackle with a Strong Tackle Attempt and an advantage.

I could go over a little more detail on each result and match up if you are interested.  Basically it's the Block -> Tackle Attempt -> Tackle cycle.

2/25/2013 10:20 PM
This helps explain some regarding the wide fluctuation of values. I understand better the flow of ingredients that constitute a play. I really like the way the play flows through multiple zones, with different players in that zone being able to take part. This in itself should provide enough variation that computer generated distributions of results would not be needed. I also like the momentum hit concept. I would like to offer this suggestion (let me know if this is how it works). To make each level of decision within the play more identifiable, to use the momentum hits as a score on how the play is progressing from behind the line to the end zone. I know that we see it more down the field, but it would be good to start publishing it from behind the line. Consider this for a running play: At the snap, the play is evaluated for behind the line for blocking as it is now and the results of DefenseBreakthrough, DefenseStrong, EqualBlocking, OffenseStrong, OffensePush are generated and a corresponding mHit is generated and published (say 4mHit for DefenseBreakthrough, 3 for DefenseStrong, 2 for EqualBlocking, 1 for OffenseStrong, 0 for OffensePush or such #'s to make the play acceptable). Make each category a specific % range. Next evaluate if the RB gets through the line with again blocking but adjusted with tackling and RB attributes to break tackle. mHits are evaluated for blocking as above, and + mHits for tackle and - mHits for avoiding and score posted again. So say score for mHits hit 10 - tackle made. Less than 10 play continues, but we as coaches would know how well our offense is doing. Again all scores for various categories are in a specific % range. No probability or predetermined distribution. Best match-ups produce more advantagous scores. Play would continue until 10 mHits are obtained and tackle occurs. 

What has dismayed me thus far is that even in a D1A vs DIII tested game, the Sim was producing very lopsided value match-ups and giving the low score a top result. This is the situation we are in now, with head shaking moments when obvious discrepencies produce illogical (not talking real-life comparisons - this is a computer game) outcomes. When attribute comparisons are taken into consideration, and those numbers are modified by tech, athl, stamina, form IQ, play calling placing different #'s of players in certain positions, I don't see how any of the values are going to remain static and produce a result that will be predicted just by looking at the values before the game. Add chance occurances such as injuries, penalties, field position and it makes it more variable. YET all this could be done without the simulation itself making calls on which array of a set of results will be picked. I fear that the simulation now is going to produce more illogical results, as the talent gap of the teams widen, if the set of results it picks from is not significantly narrowed.
2/26/2013 1:33 PM
Posted by katzphang88 on 2/25/2013 10:20:00 PM (view original):
This helps explain some regarding the wide fluctuation of values. I understand better the flow of ingredients that constitute a play. I really like the way the play flows through multiple zones, with different players in that zone being able to take part. This in itself should provide enough variation that computer generated distributions of results would not be needed. I also like the momentum hit concept. I would like to offer this suggestion (let me know if this is how it works). To make each level of decision within the play more identifiable, to use the momentum hits as a score on how the play is progressing from behind the line to the end zone. I know that we see it more down the field, but it would be good to start publishing it from behind the line. Consider this for a running play: At the snap, the play is evaluated for behind the line for blocking as it is now and the results of DefenseBreakthrough, DefenseStrong, EqualBlocking, OffenseStrong, OffensePush are generated and a corresponding mHit is generated and published (say 4mHit for DefenseBreakthrough, 3 for DefenseStrong, 2 for EqualBlocking, 1 for OffenseStrong, 0 for OffensePush or such #'s to make the play acceptable). Make each category a specific % range. Next evaluate if the RB gets through the line with again blocking but adjusted with tackling and RB attributes to break tackle. mHits are evaluated for blocking as above, and + mHits for tackle and - mHits for avoiding and score posted again. So say score for mHits hit 10 - tackle made. Less than 10 play continues, but we as coaches would know how well our offense is doing. Again all scores for various categories are in a specific % range. No probability or predetermined distribution. Best match-ups produce more advantagous scores. Play would continue until 10 mHits are obtained and tackle occurs. 

What has dismayed me thus far is that even in a D1A vs DIII tested game, the Sim was producing very lopsided value match-ups and giving the low score a top result. This is the situation we are in now, with head shaking moments when obvious discrepencies produce illogical (not talking real-life comparisons - this is a computer game) outcomes. When attribute comparisons are taken into consideration, and those numbers are modified by tech, athl, stamina, form IQ, play calling placing different #'s of players in certain positions, I don't see how any of the values are going to remain static and produce a result that will be predicted just by looking at the values before the game. Add chance occurances such as injuries, penalties, field position and it makes it more variable. YET all this could be done without the simulation itself making calls on which array of a set of results will be picked. I fear that the simulation now is going to produce more illogical results, as the talent gap of the teams widen, if the set of results it picks from is not significantly narrowed.
Well this is really where the burden of the adjustments lie.  If you think of it as for each action there being an array of possible results with each having a weighted chance of being selected, that's pretty close to how it works.  Constructing that array is where we build the results, so we can look at all of the pieces within the play and have it affect the possible outcomes.  Right now, it might be allowing a little too much in favor of a weaker match up, but that's part of the adjusting period is to find where the right mix of results lie considering all the other inputs.   There has to be a little bit of a mix in the final results, but how much mix there has to be in everything leading up to that is still being adjusted.  For instance, I can't have a team that is 5 points higher than the other always tackle for a loss, so there has to be some variation in the match up or results of the match ups so this doesn't happen.  I think we can tighten up the results even more and still have a dynamic game.
2/26/2013 11:48 PM
I think that tightening up the ranges would help. The middle range of results could be larger than the more definite outside ranges, but the outside ranges should produce definite results. The biggest benefit so far in this line of reasoning vs 2.0 is that according to JConte's play design, any decison point could cause the play to be decided. With this flowing method of decision making, a more reliable result will be occuring. As I noted in my examples above (For example:
Tackle break (#33) range 394% to 46% (ave 116%), Weak Tackle (#23) range 365% to 47% (ave 106%), Good Tackle (#30) range 189% to 53% (ave 92%), Strong Tackle (#27) range 189% to 34% (ave 83%)) the ranges of the outcomes were all accross the board, but the averages were about 10% points apart. If the averages were moved apart to about 25 - 30% (say 146%, 116%, 85% and 60%) and the ranges were tighted up to be +/-  30 - 35% points (+the real outliers which could be still 200% further out) It would still give some overlap of sequential values without producing opposite resulting outcomes. At this point of the new game critique, with many teams very even in talent, we should be seeing very close games more than runaway scores (I have seen both on my test games with the same settings against the same teams). If the talent comparisons can produce erratic outcomes when the teams are similar, what will happen when the teams are further seaparated.

For the point of the possible array choices. If they occured as you presented ( It will be closer to saying "Previous Result A + Match Up A = (10% Result A, 50% Result B, 40% Result C)" whereas a different Previous Result and different Match Up would result in a different distribution.) it would not be too undesirable. But for my broken tackle analysis I had the following distributions for my tightened ranges and spread out averages: For Strong Tackle - 45% strong, 15% good, 20% weak, 20 % broken. For Good tackle - 19% strong, 33% good, 19% weak, 28% broken. For weak - 19% strong, 28% Good, 24% weak, 29% broken. For broken - 20% strong, 13% good, 20% weak, 33% broken. The ranges in the games sample couldn't be estimated because they overlapped each other without definite delineation. For my  expanded sample, the expected outcome mode occured where it should have for strong, good and broken tackle categories, but even as I pushed the averages further away for strong tackle and broken tackle they still had overlap (bad news). So if that happened in my spread out sample - you can assume that in the complete current array with ranges overlapping for each category, that no definite pattern would emerge. (the range for strong tackle includes all but 4 values - 2 weak and 2 broken - so that would make 30% strong, 21% good, 26% weak and 23% broken). IF the game engine could produce a 10%, 50%, 40% 0% array that would be better for a good tackle or weak tackle result. For strong tackle or broken the result may be more like 50%, 40%, 10%, 0%. If these results would occur then I'm  , and you are !
of 2
All Forums > Gridiron Dynasty Football > Gridiron Dynasty Beta > PBP Debug- How do we interpret it?

Terms of Use Customer Support Privacy Statement

Popular on WhatIfSports site: Baseball Simulation | College Basketball Game | College Football Game | Online Baseball Game | Hockey Simulation | NFL Picks | College Football Picks | Sports Games

© 1999-2014 WhatIfSports.com, Inc. All rights reserved. WhatIfSports is a trademark of WhatIfSports.com, Inc. SimLeague, SimMatchup and iSimNow are trademarks or registered trademarks of Electronic Arts, Inc. Used under license. The names of actual companies and products mentioned herein may be the trademarks of their respective owners.