Posted by norbert on 3/7/2012 4:17:00 PM (view original):
I dusted off the old code.  I was wrong.  It actually did go through a progression, sort of.  It actually picks the first read randomly from the distribution and then if not thrown to, it picks randomly again from the distribution, each time shortening the distance of the throw.  It's odd how it does it, but it is in there.

We could bring back that style of progression and simply base it on distribution numbers.  What about distribution in different situations?  The old style just had one distribution for each formation.  Would you ever want that progression to change if it were 1st and 10 versus 3rd and 4?

We haven't discussed adding some player specific settings.  It might make sense to add the fatigue settings per player, like we do in HD.  Perhaps there could be other settings per player that set or adjust their tendencies.

Oops! Guess I should've read the whole thread before responding.

I hate, hate, hate that we were lied to about how "progression" worked in the old engine.  We were told - using Trips as our formation - that the progression went from first look, then down the depth chart for WR and over to TE, then back to WR.  Or if first look was TE, then over to WR and down the depth chart.

So imagine the following OLD distribution/depth:

#1 WR = 0
#2 WR = 0
#3 WR = 1
#1 TE = 1

So that means #3 WR and #1 TE each had a 50/50 chance of being first look.  From this, would arise only two progressions:

#3 WR -> #1 TE -> #1 WR -> #2 WR -> #3 WR.... etc
#1 TE -> #1 WR -> #2 WR -> #3 WR -> #1 TE... etc

Under the old system, supposedly.  And RBs were only ever dump-offs/check-downs.

But now you're telling us that this is not at all the case.  Though I can now see, after what you've told us about how it really worked, how I would've gotten the results I got, using the distributions I used.  It kinda makes sense.

Things I like about the old way
=========================
- Setting a specific distribution for each target
- That the distance down the field was directly related to your Aggresiveness (not a *chance* of distance)
- There was a progression, based on depth chart and first look (or so we were TOLD... I liked the idea of it, anyway)

Things I didn't like about the old way
===============================
- No way to directly involve the RBs in the passing game
- That it didn't actually use a progression
- That it wasn't flexible in terms of sending different receiving targets to different depths on the same play

So, I ask you, can we build the perfect passing system?  :)

I believe adding RB into the distribution choices will help one of your dislikes, Aaron.

It sounds like Norbert could easily make the new engine use a progression, as he says GD2.0 has it now.

#3 is the hard one, but when/why would you want the receivers running different depths?

Posted by norbert on 3/7/2012 4:17:00 PM (view original):
I dusted off the old code.  I was wrong.  It actually did go through a progression, sort of.  It actually picks the first read randomly from the distribution and then if not thrown to, it picks randomly again from the distribution, each time shortening the distance of the throw.  It's odd how it does it, but it is in there.

We could bring back that style of progression and simply base it on distribution numbers.  What about distribution in different situations?  The old style just had one distribution for each formation.  Would you ever want that progression to change if it were 1st and 10 versus 3rd and 4?

We haven't discussed adding some player specific settings.  It might make sense to add the fatigue settings per player, like we do in HD.  Perhaps there could be other settings per player that set or adjust their tendencies.

Norbert:  this is a great example of why most coaches are in your corner!!!  You are WORKING FOR US!!!  In this case you had some things mixed up, but you went back, checked the facts, admitted that you had it incorrect, and told us what was really going on.   WOW...that is really cool.  You definitely earn your paycheck and the gratitude of most GD coaches.

Thanks 

We can look at progressions in several ways.

The distribution method: Set percentages. This would just pick the targets based on percentage chance.  I think second and third looks would use the distribution and exclude any player that was already looked at.  The benefits are that you could use to get a variety of targets which would mean a variety of plays it could be used on. The downside is that depending on how we use aggressiveness, you still don't have much control on who is primary target and how deep those targets are.

The hard progression: Set a hard progression list.  The benefits are you have more control over your progressions.  The downside is you'd have to have some way to mix it up for different plays, as you probably don't want the same exact progression list in every play.  It would be too much to set up for every single play and too inflexible to set it up once for all plays.

The other thing to consider is that if the progression is tied to a slot, how does that work when a backup comes in.  If you have your stud WR set to be the primary target in most cases, how do we handle when a backup comes in?  I can't imagine you'd want him to share the same progression settings.

Let me throw something out for ideas.  What if we had some sort of setting on a Player Settings page that would indicate some sort of distribution for Short, Medium, and Long passes.  Then when setting up a play you could set the desired progression of Long->Short or something like that that would then pick the primary target based on their progression settings.  Not sure how this would work when you set your RB to Long passes, but that's something we'd have to address in any situation.

Posted by bhouska on 3/7/2012 9:20:00 PM (view original):
I believe adding RB into the distribution choices will help one of your dislikes, Aaron.

It sounds like Norbert could easily make the new engine use a progression, as he says GD2.0 has it now.

#3 is the hard one, but when/why would you want the receivers running different depths?

Because I don't want all five (in Trips - three WR, one TE, one RB) running deep patterns at the same time.  Maybe I want my two outside WR to run deep routes, with my slot WR dragging underneath and my TE settling in the intermediate, while my RB flares.  I know this is complex, but it's more to explain the concept than it is to describe what exactly I want to happen.

I know there are certain realities due to coding complexities, but this is just an example of the idea behind the idea of having different receivers at different "levels" of the field.  As in real life.  Perhaps that makes defense too complex, so it would have to be all-or-nothing.  I don't know.  Just throwing it out there.

And after thinking more about GD 1.0 and the way passing worked, it turns out that NOT having progressions - as we were told they worked - actually made my method of distribution work even more perfectly than I had imagined and was probably a very large part of the reason I had that Trips-only offense working so well.

Yes... I'm very upset that it doesn't work so well anymore.  

Posted by norbert on 3/8/2012 11:42:00 AM (view original):
We can look at progressions in several ways.

The distribution method: Set percentages. This would just pick the targets based on percentage chance.  I think second and third looks would use the distribution and exclude any player that was already looked at.  The benefits are that you could use to get a variety of targets which would mean a variety of plays it could be used on. The downside is that depending on how we use aggressiveness, you still don't have much control on who is primary target and how deep those targets are.

The hard progression: Set a hard progression list.  The benefits are you have more control over your progressions.  The downside is you'd have to have some way to mix it up for different plays, as you probably don't want the same exact progression list in every play.  It would be too much to set up for every single play and too inflexible to set it up once for all plays.

The other thing to consider is that if the progression is tied to a slot, how does that work when a backup comes in.  If you have your stud WR set to be the primary target in most cases, how do we handle when a backup comes in?  I can't imagine you'd want him to share the same progression settings.

Let me throw something out for ideas.  What if we had some sort of setting on a Player Settings page that would indicate some sort of distribution for Short, Medium, and Long passes.  Then when setting up a play you could set the desired progression of Long->Short or something like that that would then pick the primary target based on their progression settings.  Not sure how this would work when you set your RB to Long passes, but that's something we'd have to address in any situation.

I think if you set distributions on a per-player basis, you solve the problem of back-ups coming into the game.  And I think I like your suggestion of setting separate distributions for Long/Med/Short situations and then setting a progression on a per-play basis.

At any rate, I think *more specific* controls over our ability to primarily (try to) get the ball into the hands of our best playmakers.  In GD 1.0, stud receiving TEs were huge weapons that we could specifically target to exploit a weak LB/DB corps.  They were not unstoppable - I had a few shrewd coaches adjust and shut down my TE, as well as having done it to my opponent's stud TE on occasion - but then that furthers the "chess game," right?   When that happened, I adjusted my distribution to begin targeting a WR and changed my aggressiveness, which then exploited their weakened defense that was paying too much attention to my TE.  Even if the exact mechanisms didn't work like I thought they did in my head, the result was pretty much what I was intending.

And that was, I think, the beauty of GD 1.0 in its passing game - you pretty much got the results that you intended, when you accounted for talent differentials that may have existed.  You had more controls and those controls lead to predictable outcomes... not just when you look a season worth of results, but when you looked at individual games.  Sure, there were a few outliers, which I always attributed to a shrewd gameplan by my opponent.  But that was fun!

Does anyone see any reason why you would want different personal distribution numbers for players on different plays?  For example, Player A is set to 5% Short, 10% Medium, and 50% Long.  Not his breakdown, but the distribution relative to everyone else.  Would there ever be a reason why you wouldn't want him to get 50% of the long passes or get more than 5% of the short passes?

Keep in mind this is just an example.  Actual setting we end up with may be different.

I can't think of a reason why it would change formation to formation, play to play.  Or at least not a reason that needs to be programmed in.....unless it'd be easy and not cause us hours and hours to set up a team's playbook.

Posted by norbert on 3/9/2012 10:17:00 AM (view original):
Does anyone see any reason why you would want different personal distribution numbers for players on different plays?  For example, Player A is set to 5% Short, 10% Medium, and 50% Long.  Not his breakdown, but the distribution relative to everyone else.  Would there ever be a reason why you wouldn't want him to get 50% of the long passes or get more than 5% of the short passes?

Keep in mind this is just an example.  Actual setting we end up with may be different.

From formation to formation, I could see a reason, but from play to play... I can't really think of any reason.

Posted by gt_deuce on 3/9/2012 11:51:00 AM (view original):

Posted by norbert on 3/9/2012 10:17:00 AM (view original):
Does anyone see any reason why you would want different personal distribution numbers for players on different plays?  For example, Player A is set to 5% Short, 10% Medium, and 50% Long.  Not his breakdown, but the distribution relative to everyone else.  Would there ever be a reason why you wouldn't want him to get 50% of the long passes or get more than 5% of the short passes?

Keep in mind this is just an example.  Actual setting we end up with may be different.

From formation to formation, I could see a reason, but from play to play... I can't really think of any reason.

I agree. Distribution may change from formation to formation depending on the number of receivers, but not within one formation. If the skills needed are different to get open at the different distances (short - GI + hands + ath > others, medium - Elus + athl + hands > others, long - speed + elus + hands > others), then each receiver would be best at certain distances. Coaches decision then is who to put in for the play call. If the passing game plays off the distribution % and position of the WR/TE/RB then that could be another way of setting a team's passing style. Variations in receivers settings could be made within the depth charts at halftime to alter distribution of passes to take advantage of weakness in the defense.


POPULATE HEISMAN!

Maybe I am not understanding the distribution completely, but if my best WR is also my fastest so for 80% of the field his distribution looks like it does in norberts example with him at 50% for long passes and 5% on short passes, but I want to throw inside the red zone at the 5 I might want him to the main target on the short and medium looks as well.

I guess if we take a step back...ideally in the new engine if we could come up with some way to tell the general location of a player first and then the chance he will be the primary or next target, the easier it would be, and leaves it in a better spot to set up defenses, since we want to see more effects of match ups. If we pick location first and then target, it makes it more difficult to pick defenders in on the play.  It can be done though, so probably best to continue with this train of discussion and look at easiest way to set up game plans and gain control.

If we were looking at the most ideal way of setting up plays with the most control, I would say I would build a list of plays by setting a formation, setting the routes for each player, or assignments in the case of blocking versus running routes.  Setting routes could just mean instead of setting slant, post, etc. that we just set general location of routes, like long, medium, short.    Then pick the plays from that list to use in each situation, like 1st and long, as we do now.  However, I don't want to set up 100 plays and I certainly don't want to have to try to remember what each one is when I'm trying to set my plays for 3rd and short. So how do we simplify the settings while maintaining control?

The problem with putting distribution and depth charts on formations, like the old way, is that I'd like to add 5-6 new formation options and that's that many more distributions and depth charts to set and maintain.  I'll play around with some of the ideas we've brought up here and see if I can get some mock of some settings.  It's usually easier to see with an example.  I know we've been hitting the distribution issue pretty heavy, but it's pretty much the basis of a lot of the offense, so I really want to work it out.  If you have any other ideas, keep them coming.

5-6 new formations? Most guys practice for 2-3 OF and 2 for DF because of the amount of practice time aloted for formation and player practice. Will there be more time aloted for more formations? No use having more if you can not really use it.

Now do not get me wrong. I would welcome more formations if I am abule to practice them with out spreading my time too thin.

WOW! norbert. Just took a look at your Otterbin team. Good rebuild and congrats on a job well done. :-)