IBM's "Deep Blue" computer beat the chess champ in '97,but we can't have adaptive AI?

[lint]

Quote: Originally Posted by lint

I remember from last year I got Ian or someone to confirm they were in agile. I will attempt to find the thread.

Found it http://www.operationsports.com/forum...-using-ea.html

[woowoo782]

theres adapted A.I. in NHL 09

[shotgun styles]

Quote: Originally Posted by woowoo782

theres adapted A.I. in NHL 09

Someone else mentioned this, but I don't play hockey games. How exactly did it work?

[roadman]

Quote: Originally Posted by shotgun styles

I'm not denying that. But the proof is in the pudding, and none of us have seen the game yet. I hope you and Adembroski are right. I hold Adembroski's opinion on these things in high regard and have been reading his posts for many years. I really hope you guys are right. I'm not out here just calling people out liars or anything like that. I just think that we suspect that improvement has been made, but we do not know for sure. Am I wrong to be just a teeny bit uncertain after the last 4 years?

You can be a teeny bit uncertain, but no one can change the past. Everyone is new in 10 and it appears they are going to a sim style.

All we can do is wait and not prejudge(good or bad) before it comes out.(not saying you are, either.)

[kcarr]

Quote: Originally Posted by lint

I am trying to wrap my head around an algorithm that only looks 4 or 5 plays ahead in football. How do you calculate how powerful a play will be in a situation. Oh boy I feel for anyone who tries to take on that challenge. There are so many variables to just calculate which play is the best play and then try and calculate what the defense is going to do that it just made my head spin.

It doesn't really have to look ahead which is good because unlike chess there ien't a definite outcome to each move made. All it really has to look at is previous playcalls and predicted strength of possible playcalls in a given situation.

Head coach had calculated success for plays in given situations based on previous preformance. They also calculated tendencies for coaches playcalling in various situations. If head coach could do that I don't see it as being impossible in madden.

From there writing an algorythem that makes a playcall based on those successes along with opponent tendencies and coach tendencies along with a random factor to keep the user guessing isn't that tough.

Off the top of my head for offensive playcalling I would use something along the lines of:

Take the top 10 most successful plays, number them one through 10 and weight each of them in order opposit of success. Give play 1 10 points, 9 for play 2 etc.

Then look at the set playcalling for the coach in a given situation. Maybe for 3rd and short, QB run 10%, FB run 10%, HB inside run 40%, PA pass 20%, HB outside run 10%, HB counter 10%. Now go through the 10 plays and multiply each of the 10 plays by the percentage.

Now add up all those numbers, pick a random number between 0 and the sum and whatever range the number is in, pick that play.

Defensively it takes a little more reatcion to the opponent. As far as Double teaming, pressing certain players, playing off of certain players, rolling a safety over to help on a receiver, etc you just have to be able to analyze the ratings and determine mismatches. This ia a simple matter of looking at the ratings and if there is a certain amount of difference making the predetermined adjustment. Also, you could have them after every few plays look back at those matchups and if a player has had some level of success maybe adjust the requirements to make the adjustments.

As far as moving extra players into the box, it is a matter of assesing their own run defense and your HBs ratings. Maybe also figure in the user's playcall tendencies in the situation and the user's success so far that game running the ball.

All of those are kinda yes or no things and easy to write a code for. As far as playcalling it gets a little tougher. First when it comes to calling a formation, just match the personel that the user uses. As far as the play call head coach was able to determine success of a play against a certain type of offensive play. Ok, knowing user tendencies and the strength of each of your plays against each of those possibilities an algorythem can be determined. Off the top of my head it should go something like:

For user's top tendency order the top 10 plays in the same way as with offensive plays. Then multiply each by the percent chance of that tendency.

Next go to the second highest tendency and do the same thing. Do this for all possible play call types within the user tendencies and then add up all the results.

Next go to the CPU coaches tendencies and adjust each of the possible playcalls in the same way descibes in the offensive playcall description. From there you can determine a random number and pick the play as described before.

[Clarke_311]

Quote: Originally Posted by shotgun styles

Someone else mentioned this, but I don't play hockey games. How exactly did it work?

It's basically like if you keep on doing the same thing over and over again the computer will eventually catch on and attempt to take it away.

For example, if you keep passing it to the high slot, there will be a player there to take away the pass.

[IndyColts2]

Quote: Originally Posted by shotgun styles

We have a saying in the Army- "Trust but verify" I mean Ian no offense when I say this, but "fan interaction" on these boards could just be another gimmick. You're stating the situation as if EA has not made lofty promises before, or dropped conspicous hints before, and then not delivered. They do not have a good track record, going back 10 years. Ian and his team may be different, time will tell. Trust, but verify. Agreed 100%. And that's all I was trying to say.

My teacher was just talking about that saying in my American History class. The Soviet Union had that saying or somethin like that, this girl in my class called her Russian friend to translate it but he didn't understand what trust but verify meant cause it might be said differently. Anyway, just thought that was cool to see that.

[lint]

Quote: Originally Posted by kcarr

It doesn't really have to look ahead which is good because unlike chess there ien't a definite outcome to each move made. All it really has to look at is previous playcalls and predicted strength of possible playcalls in a given situation. Head coach had calculated success for plays in given situations based on previous preformance. They also calculated tendencies for coaches playcalling in various situations. If head coach could do that I don't see it as being impossible in madden. From there writing an algorythem that makes a playcall based on those successes along with opponent tendencies and coach tendencies along with a random factor to keep the user guessing isn't that tough. Off the top of my head for offensive playcalling I would use something along the lines of: Take the top 10 most successful plays, number them one through 10 and weight each of them in order opposit of success. Give play 1 10 points, 9 for play 2 etc. Then look at the set playcalling for the coach in a given situation. Maybe for 3rd and short, QB run 10%, FB run 10%, HB inside run 40%, PA pass 20%, HB outside run 10%, HB counter 10%. Now go through the 10 plays and multiply each of the 10 plays by the percentage. Now add up all those numbers, pick a random number between 0 and the sum and whatever range the number is in, pick that play. Defensively it takes a little more reatcion to the opponent. As far as Double teaming, pressing certain players, playing off of certain players, rolling a safety over to help on a receiver, etc you just have to be able to analyze the ratings and determine mismatches. This ia a simple matter of looking at the ratings and if there is a certain amount of difference making the predetermined adjustment. Also, you could have them after every few plays look back at those matchups and if a player has had some level of success maybe adjust the requirements to make the adjustments. As far as moving extra players into the box, it is a matter of assesing their own run defense and your HBs ratings. Maybe also figure in the user's playcall tendencies in the situation and the user's success so far that game running the ball. All of those are kinda yes or no things and easy to write a code for. As far as playcalling it gets a little tougher. First when it comes to calling a formation, just match the personel that the user uses. As far as the play call head coach was able to determine success of a play against a certain type of offensive play. Ok, knowing user tendencies and the strength of each of your plays against each of those possibilities an algorythem can be determined. Off the top of my head it should go something like: For user's top tendency order the top 10 plays in the same way as with offensive plays. Then multiply each by the percent chance of that tendency. Next go to the second highest tendency and do the same thing. Do this for all possible play call types within the user tendencies and then add up all the results. Next go to the CPU coaches tendencies and adjust each of the possible playcalls in the same way descibes in the offensive playcall description. From there you can determine a random number and pick the play as described before.

I see where you are going but I guess I am seeing it a little differently. If someone truly wants to have the cpu choose a play that can be unpredictable or what not you have to have an opportunity to have every play in your playbook become available. I would want the cpu to rank each play depending on situation.

I agree that there is only one move that you would have to code for to an extent, but what can you use to determine what play? what calculates the run vs the pass? should I go for the first down, run or go to the homerun. There are a lot of variables that have to come in to make a seemingly simple decision.

So the cpu could have run three times to get you to bite on the playaction, but what determined the cpu to run the first 3 times?

I am probably over complicating this, but there is no easy algorithm for this in my eyes.