Arimaa Forum - Print Page


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

        Arimaa >> General Discussion >> Deadlocked game. Why not draw?
        
(Message started by: Boo on May 27^th, 2011, 7:07am)

Title: Deadlocked game. Why not draw?
Post by Boo on May 27^th, 2011, 7:07am

http://arimaa.com/arimaa/gameroom/opengamewin.cgi?client=1&gameid=185511&role=v&side=b

See the game starting from move 60.

What if one player cant improve the pieces, and the opponent doesnt want to, but continues to make stupid moves in his side of the board? This game could have continued to my death hadnt the opponent submitted illegal move (that repeated the position for the 3rd time).

Imagine such a situation in the real life tournament for the world championship final, where after several hundred of moves players will forget if this position repeats for 2nd time or 3rd time. Will such a game continue until somebody dies?

Title: Re: Deadlocked game. Why not draw?
Post by Fritzlein on May 27^th, 2011, 12:23pm

on 05/27/11 at 07:07:06, Boo wrote:

What if one player cant improve the pieces, and the opponent doesnt want to, but continues to make stupid moves in his side of the board?

Good question. Fortunately this doesn't happen in games involving two humans. The player who can make progress doesn't torture his opponent; instead he tries to make progress toward the win.

It would be different if the passive play were not a mistake on either side, i.e. if neither player could make progress. In my opinion, such a position "should" be a draw. At present, however, this type of position doesn't arise in games between two humans, so the debate about what to do when it happens is moot. We are all waiting around for Arimaa's first-ever natural draw. When we see it happen, there will be vigorous discussion about how it should have been handled.

Quote:

Imagine such a situation in the real life tournament for the world championship final, where after several hundred of moves players will forget if this position repeats for 2nd time or 3rd time. Will such a game continue until somebody dies?

Omar has foreseen the possibility, so he has instituted tournament rules to decide the outcome of the game. Note that if it ever happens, nobody will be happy about it, and everyone will have ideas about how to change the tournament rules. I am not saying that the situation has been handled in a good way, just that tournament rules already don't allow a game to drag on forever.

Every game has a parameter in its time control that specifies a maximum length. If that maximum is reached, the game stops immediately. For example, Postal Mixer games end after 300 days. The standard fast game (30s/move) has a two-hour cutoff. The World Championship finals had a six-hour cutoff. Etc. In practice, nobody ever hits the cutoff, but it is there just in case.

If the cutoff were reached, the player with more pieces at that time would win. If no pieces had been exchanged, Silver would win. If an equal number of pieces had been exchanged, then whoever captured last would lose. For example, in the game you link, you and Bomb each had 15 pieces left, but you would have lost because Bomb reduced you to 15 pieces before you reduced Bomb to 15 pieces.

That is an arbitrary and unsatisfactory way to resolve games, but you can rest assured that (A) games are guaranteed to end, and (B) if games do start ending that way, we'll figure out a better way to deal with it.

To answer the subject-line question, "Why not draw?", consider that there are no natural draws, and that Omar doesn't want "unnatural" draws. If it isn't drawn on the board (and it never has been so far) the players should fight it out. There is no fundamental need for agreed draws, so Omar wants to encourage players to play on to a decision.

It is unfortunate that one can't reason with a bot, so bots can't take Omar's lesson to heart. However, allowing for agreed draws when a bot is too stupid to press an advantage would be letting the tail wag the dog. In "real" games of Arimaa, the human vs. human kind, we don't want or need to allow agreed draws, so we avoid allowing them it bot games either. Omar hasn't watered down the rules of Arimaa in a concession to bot idiocy.

Fortunately, modern bots aren't nearly as passive as Bomb was in 2005. The trend in playing style, both for humans and for bots, has been towards greater aggression. Thus the case of the passive bot that won't take the fight to the other side of the board when it has an advantage seems mostly like a historical artifact.

Title: Re: Deadlocked game. Why not draw?
Post by Fritzlein on May 27^th, 2011, 12:43pm

I see your game didn't end after two hours, and the time control parameter was for eight hours. This is silly. Omar, could you change the cutoff for Bomb2005Fast (and other old fast bots) to two hours?

Title: Re: Deadlocked game. Why not draw?
Post by ocmiente on May 27^th, 2011, 12:45pm

I expect that there are other winning moves, but this one seemed like the most obvious to me:
82g Mg3n hh3w Rh2n Hg2e

Unfortunately, the bot would have to make that move to end the game. So, yeah, that would be pretty lame to be in that position with a bot.

Title: Re: Deadlocked game. Why not draw?
Post by Boo on May 28^th, 2011, 3:41am

Quote:

Yeah, 100% agree :)

Quote:

Every game has a parameter in its time control that specifies a maximum length.

Oh, this is fine. I didnt know that.

However, I dont understand the logic regarding the resolution of the game result if the cutoff is reached.

Quote:

If no pieces had been exchanged, Silver would win.

OK, this one would indicate, that the player who has the upper hand should lose, if he fails to make some progress. This is good - the game rules encourage the player who has the advantage to go forward and make some action going.

Quote:

the player with more pieces at that time would win.

Now this one is the direct opposite of the first one. A player who has the upper hand (e.g. captured 1 rabbit) is encouraged to turtle up. And the opponent with the material disadvantage should attack. ???

I think the game designer should write down some guidelines like "the player who has the advantage should generate action" and push them through the rules, and not make some weird switches.

Quote:

I expect that there are other winning moves, but this one seemed like the most obvious to me:
82g Mg3n hh3w Rh2n Hg2e

Yeah it is winning of course, but that would have given some freedom for my horse and camel :) So the bomb decided to search for alternative ways to improve.

Title: Re: Deadlocked game. Why not draw?
Post by Boo on May 28^th, 2011, 2:21pm

BTW shouldnt bots with bugs be removed from ladder?

http://arimaa.com/arimaa/gameroom/opengamewin.cgi?client=1&gameid=185669&role=v&side=b

Title: Re: Deadlocked game. Why not draw?
Post by Eltripas on May 28^th, 2011, 7:53pm

on 05/28/11 at 14:21:17, Boo wrote:

BTW shouldnt bots with bugs be removed from ladder?

http://arimaa.com/arimaa/gameroom/opengamewin.cgi?client=1&gameid=185669&role=v&side=b

You can't blame bomb for failing when you were playing that way, bomb is not going to be removed from the ladder, it is a great benchmark.

Title: Re: Deadlocked game. Why not draw?
Post by Belteshazzar on May 30^th, 2011, 7:58am

I would support removing bot_Bomb2005Blitz and bot_Bomb2005Fast from the ladder. Those were the last two I had to beat, and in both cases I ended up winning because the bot made an illegal move, which was quite an anticlimactic way to finish the ladder. The normally timed version of Bomb2005 should stay on the ladder, as it doesn't seem to submit illegal moves very often.

Title: Re: Deadlocked game. Why not draw?
Post by ocmiente on May 30^th, 2011, 11:56am

on 05/30/11 at 07:58:19, Belteshazzar wrote:

I would support removing bot_Bomb2005Blitz and bot_Bomb2005Fast from the ladder... because the bot made an illegal move...

If the bot did that on a regular basis, I would support that. However, bot_Bomb2005Fast doesn't lose that way very often. You might consider the bot's entire loss record (http://arimaa.com/arimaa/gameroom/pastrecord.cgi?id=2320&r=l), rather than basing it on one loss.

bot_Bomb2005Blitz loses this way more often (http://arimaa.com/arimaa/gameroom/pastrecord.cgi?id=2649&r=l), but even then, it's probably worth looking at the bot's entire win/loss record before claiming that it should be removed. That bot wins more than 60% of the time, and most of the time it loses for reasons other than submitting an illegal move.

Title: Re: Deadlocked game. Why not draw?
Post by mistre on May 30^th, 2011, 1:16pm

It would be interesting to see how many times (and percent of the time) each bot has lost by trying to submit illegal moves.

I think there are versions of clueless that are even worse than bomb in this department (whereas I have never seen bots like Opfor and Marwin try to submit illegal moves).

Title: Re: Deadlocked game. Why not draw?
Post by Janzert on May 30^th, 2011, 2:21pm

One thing to note, at the time Bomb was written the third time repetition moves were not illegal but merely losing moves.

Janzert

Title: Re: Deadlocked game. Why not draw?
Post by Fritzlein on May 30^th, 2011, 3:44pm

on 05/30/11 at 07:58:19, Belteshazzar wrote:

As I proponent of keeping all versions of Bomb2005 on the ladder, let me provide some data to support my case. Here are the results of all the games that bot_Bomb2005Blitz has ever played through April 2011:

score term count
----- ---- -----
0 i 15
0 p 31
0 t 53
0 e 116
0 m 531
0 g 1845
0.5 a . 2
1 e . 4
1 m 13
1 p 31
1 r 1404
1 g 1539
1 t 1856

Thus out of 7440 games, Bomb2005Blitz has lost 46 by repetition of position (or another kind of illegal move, if that ever happens). That's 0.6%.

I know it is a frustrating way to win, but I submit that it rarely happens naturally. Either the human is trying to provoke Bomb into repetition, or the human is just marking time, not trying to accomplish anything positive. This might happen more often in a blitz game, but it is still quite rare. The game that started this thread might be one of the few cases where the human had nothing better to do than shuffle pieces around; usually there is a way to improve position.

The reason I would like Bomb2005 to remain on the ladder is that it is one of our few direct measures of rating inflation. We can compare the rating these bots have now to the ratings they had then. In particular this is useful information for the P1 and P2 bots which play exactly the same, but also the Fast and Blitz versions give some long-term insight into rating inflation and the value of faster hardware.

I would be more inclined to support removing these bots from the ladder if illegal moves happened all the time, but since it is only a one-in-162-games occurrence, I feel the benefits of having them in the ladder outweigh the drawbacks.

Title: Re: Deadlocked game. Why not draw?
Post by Fritzlein on May 30^th, 2011, 3:54pm

on 05/28/11 at 14:21:17, Boo wrote:

BTW shouldnt bots with bugs be removed from ladder?

http://arimaa.com/arimaa/gameroom/opengamewin.cgi?client=1&gameid=185669&role=v&side=b

Bugs that trigger when you are trying to win naturally are much more of a problem in my mind than bugs that trigger when you play suboptimal moves in an attempt to trigger them.