Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> Events >> Flexible time control for the challenge match
        
(Message started by: omar on Oct 14th, 2006, 10:37am)
Title: Flexible time control for the challenge match
Post by omar on Oct 14th, 2006, 10:37am
Deciding on exactly what time control should be used for the Arimaa challenge match has been a bit difficult. Some of the issues involved are:
 * Human and computer play quality improves with slower time controls. Thus the quality of the games (and thereby the relevance of the match) will be better if the games are played at a slower time control.
 * Computers can handle most any time control, the human spectators and players can't.
 * Spectators become very bored if the time control is too slow.
 * Some human players also become bored if the time control is too slow.
 * Human players may not be able to fit the games in their busy life if the games are too slow.
 * It becomes difficult to find human players for the match if the time control is too slow.
 * I would like to fix the rules pertaining to time controls for all future challenge matches.

The first Arimaa challenge match used a time control of 3/3/100/15/8 and was played by me. The second and third used a time control of 2/2/100/10/8 since the players would not be able to fit the games into their schedule otherwise.

It has recently been requested that the time controls be made even faster since human players currently have such a dominance that playing at the current time controls is boring them. See reply #4 and #6 of 2007 Computer Championship and Challenge:
http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=talk;action=display;num=1160206013

It has also been suggested recently and in the past that the players be allowed to fully manage their own time within the alloted game time. And that the ability to manage ones time is part of the intelligence required for sucessfully playing the game. See reply #10 of 2007 Computer Championship and Challenge:
http://arimaa.com/arimaa/forum/cgi/YaBB.cgi?board=talk;action=display;num=1160206013

I am considering the possibility of allowing the human player to select the time control before each game in the Arimaa challenge match with the requirement that the average time-per-move (ATPM) be in the range of 1 to 3 minutes and the max time-per-move (MTPM) be in the range of 3 to 5 minutes.

This allows for a wide range of time controls. This would also alleviate the need to specify a fixed time control in the match rules, ensure that the games are played slow enough to have quality games and ensure that the games are played fast enough to keep the interest of the spectators and players.

But why allow the human player to select the time control and  not the program; or have the time control negotiated per game; per player or per whole match.

My reason for this choice is that
 * The computer program should be able to demonstrate that it can handle playing a variety of time controls. Just as the program should not be tuned to play against any one player, it should also not be turned to play only at a specific time control. Thus the program should not be given a choice in the exact time control but should be able to rely on the time controls being bound to reasonable limits.
 * Human players are more easily able to incorporate the match games into their busy life schedule if they are able to select the time control for the game without compromise. Thereby increasing the chance that the selected human players will be able to participate in the challenge match.
 * I realize that this makes the challenge more difficult for computers, particularly in the current years when the gap between humans and computers has already become so wide. However, I am sure that the time control factor will be quite insignificant in future years when hardware improvements will have balanced the scale between humans and computers.
 * This further ensures that when humans eventually lose an Arimaa challenge type match they will have no reason to deny that the computer program has surpassed their skill level.

As alway, before making any changes I would like to receive suggestions and feedback to make sure that I am not overlooking something.
Title: Re: Flexible time control for the challenge match
Post by omar on Oct 14th, 2006, 10:46am
A technical difficulty is deciding how exactly to compute the average time-per-move given the time control.

I am currently thinking of using something like this:

 ATPM = M/(2-P/100) + R/50
 if (ATPM > T*0.95) then
   ATPM = T*0.95

where M = time per move from the time control
     P = percent of unused time added to reserve
     R = starting reserve time
     L = max limit on reserve
     G = total game time
     T = max time per move

Please feel free to suggest a different equation.
Title: Re: Flexible time control for the challenge match
Post by unic on Oct 14th, 2006, 11:14am
One disadvantage of a flexible time control is that the program would be required to parse the game state file (the third file)... while currently, as Fairy knows the time control in advance, it just parse the position file.

Not that it would be that difficult to add parsing of the game state file...
Title: Re: Flexible time control for the challenge match
Post by jdb on Oct 14th, 2006, 1:14pm
Having a bot handle (semi intelligently) a wide range of time controls is a pain. Please dont allow a whole bunch of different time controls.
Title: Re: Flexible time control for the challenge match
Post by Fritzlein on Oct 16th, 2006, 10:45am
I agree with jdb about not allowing whole classes of different time controls.  Yes, it shows some sort of intelligence to be able to manage time under different constraints, but that's not the kind of intelligence we're after.  Bot developers should be allowed, as much as possible, to focus on making their bots play Arimaa well.  Their distractions and obstacles should be minimized where possible.

If you are going to offer a choice of time controls, then you should limit it in advance to a small number of similar possiblities.  Let there be one time control for two minutes per move, one time control for 90 seconds per move, etc.  Let all the time controls be the same "shape", i.e. let them all have proportionally the same time per move, initial reserve, and maximum game length.  Having equal proportions prevents players from manipulating the time control to win on a technicality.

Imagine if the humans were getting clobbered in the Challenge but in desperation they discovered that a certain time control made the bot time out, or lose on score.  Yes, that would be a display of human ingenuity and adaptability that the bots couldn't cope with, but it would be completely beside the point of the Challenge.  The Challenge is about humans being better at an abstract strategy game per se, not about being able to win on a technicality of the match conditions.

As a Challenge defender, I would never want to play at three minutes per move, not only because it would be boring, but also because I think it would make my chance of winning less.  Playing that slowly would make me tired by the end of the game, and I would be as likely to blunder at the end from fatigue as to blunder in the middle from lack of thinking time.

That said, I can understand if you want to allow one 3-minute time control as an option to defenders.  If someone believes they will do better at that speed, fine, it's their funeral.  However, I would caution against 3/3/100/15/8 as an option.  That only forces the game to go 80 moves before the score function kicks in.  If someone wants three minutes per move, they shouldn't be allowed to play a shorter delaying game with the intent to win on score when time runs out.  (This is particularly relevant given that Bomb is currently programmed to kick in aggressive play by move 50 if nothing has happened, leaving only 30 moves of delaying play for the human to get a techinical win.)  The proportions should be the same, i.e. 3/3/100/15/12 as an alternative to 2/2/100/10/8, so that either game goes 120 moves before score kicks in.

I truly expect that 2/12/75/0/8/5 would be a better time control than the 2/2/100/10/8 which is now part of the rules.  I think 2/12/75/0/8/5 would be better for the spectators and better for the players, for reasons I've laid out elsewhere.  That said, I also believe in standardization.  It would be better for you to just stick to 2/2/100/10/8 forever with no variations or choices allowed, than to allow too much wacky experimentation on the part of the challenge defenders.

Omar, I disagree with you all the time about tournament rules and conditions, but ultimately I trust your judgment.  I trust that whatever you decide will be reasonable, even if what I would have decided (or what unic would have decided) would have taken us down a different path.  For the present, Arimaa is better off with you as a benevolent dictator than with too much chaos.
Title: Re: Flexible time control for the challenge match
Post by omar on Oct 20th, 2006, 7:55am
Thanks for the feedback guys. I really appreciate the input.

Looks like there is sufficient reason to not use a wide range of time controls. A good middle ground seems to be to allow a set of predefined time controls from which the human player can select at the start of the game.

So what should be the set of allowed time controls. I am considering using the follow ones:

3:00/3:00/100/15/10
2:30/2:30/100/12/9
2:00/2:00/100/10/8
1:30/1:30/100/8/6
1:00/1:00/100/6/4

3:00/16:00/75/0/10/7
2:30/14:00/75/0/9/6
2:00/12:00/75/0/8/5
1:00/10:00/75/0/6/4
1:00/8:00/75/0/4/3

Though I originally thought a MTPM of 5 minutes should be the max, a proportional layout shows that it should be a little higher for the slowest speed games. I remember having used over 10 minutes on some of my challenge match moves.

Note that this will not apply to the 2007 challenge match, but is being discussed for future years.

Omar
Title: Re: Flexible time control for the challenge match
Post by Fritzlein on Oct 25th, 2006, 9:10pm

on 10/20/06 at 07:55:54, omar wrote:
3:00/3:00/100/15/10
2:30/2:30/100/12/9
2:00/2:00/100/10/8
1:30/1:30/100/8/6
1:00/1:00/100/6/4

3:00/16:00/75/0/10/7
2:30/14:00/75/0/9/6
2:00/12:00/75/0/8/5
1:30/10:00/75/0/6/4
1:00/8:00/75/0/4/3
I have a preference for the second set, although I realize that banking 75% of unused time is untested.  In theory, if you don't bank all of the unused time, it provides an incentive for the players to move at a steady pace rather than blitzing out some moves and dawdling on others.  A steadier pace is more congenial to the spectators than alternating slow and fast.

Also I have a preference for adding a 30 second option for the convenience of the players.  I would certainly use it if I were defending the challenge this year, and be happy about all the time I saved.  If I can beat the top bot 95% of the time at 15 seconds per move, playing at 30 doesn't seem all that risky.

But the most importantly, if the player is given a choice of time control at all, I think it is good to limit it to five or six similar choices, i.e. one of the above sets, not both.


Quote:
Though I originally thought a MTPM of 5 minutes should be the max, a proportional layout shows that it should be a little higher for the slowest speed games. I remember having used over 10 minutes on some of my challenge match moves.
I still have a slight preference for a maximum time per move of five minutes regardless of time control.  However, I might be bored in any case by a game at three minutes per move, so bumping it up slightly probably won't bother me much more.

I am thinking of this as a player too, by the way.  I would be rather annoyed if Bomb got into trouble and thought for half an hour on one move.  I suppose I could tolerate seven minutes.  :P


Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB © 2000-2003. All Rights Reserved.