Arimaa Forum - Print Page


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

        Arimaa >> Off Topic Discussion >> The big difference between Chess and Arimaa?!
        
(Message started by: OLTI on Mar 14^th, 2006, 6:10am)

Title: The big difference between Chess and Arimaa?!
Post by OLTI on Mar 14^th, 2006, 6:10am

One Million US Dollars match!
Classical chess World Champion Vladimir Kramnik vs Deep Fritz one of the strongest chess program (if not the strongest) of the world.

Title: Re: The big difference between Chess and Arimaa?!
Post by OLTI on Mar 14^th, 2006, 6:12am

Here you can find more

http://www.chessbase.com/newsdetail.asp?newsid=2947

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Mar 14^th, 2006, 3:07pm

Go Kramnik! If he wins, I'll bet he'll get a shot at Hydra for two million.

Title: Re: The big difference between Chess and Arimaa?!
Post by OLTI on Mar 14^th, 2006, 4:30pm

I'm wondering how would increase the strength of bot_bomb if it run on the same hardware of Hydra. 64 computers connected and operating as if they are a single machine, with the processing power equivalent to more than 200 standard PCs.

Here for more information about his power.

http://tournament.hydrachess.com/ahydra.php

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Mar 14^th, 2006, 5:43pm

on 03/14/06 at 16:30:59, OLTI wrote:

I'm wondering how would increase the strength of bot_bomb if it run on the same hardware of Hydra.

Yes it is a very interesting question. I expect the increase in playing strength varies directly with increase in log of computing power. So if 2 times the speed results in 50 rating points more, then 1000 times the speed results in 500 rating points more. (1000 = 2^10 and 50 * 10 = 500.)

However, not everyone expects a linear relationship to hold. Perhaps if Bomb could run a million times faster it would still fall for the elephant-horse attack, and therefore lose anyway.

We do know that Bomb performed poorly in the 2005 Postal Championship where it was thinking for hours on every move, as compared to humans thinking only 10 minutes or so. It lost games to players rated 2171, 2164, 1933, 1918, 1687, and 1556. It defeated players rated 1851, 1810, 1605, and 1471. That's a performance rating of only 1716, despite the huge thinking time.

Title: Re: The big difference between Chess and Arimaa?!
Post by jdb on Mar 14^th, 2006, 7:01pm

Increased time per move, in general, helps more on games with a lower branching factor than games with a higher branching factor. With a lower branching factor, the extra time allows relatively more search depth.

Title: Re: The big difference between Chess and Arimaa?!
Post by omar on Mar 14^th, 2006, 11:09pm

I just sent an email to the Hydra team to ask if they might be interested to have Hydra play Arimaa and see how it does against the humans.

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Mar 15^th, 2006, 1:06am

The odds are they won't give you the time of day, but it doesn't hurt to ask. That would be so amazingly cool if the Hydra team were interested.

Title: Re: The big difference between Chess and Arimaa?!
Post by PMertens on Mar 15^th, 2006, 9:06am

That would be really cool ...
After all their cute machine must be bored from beating humans in chess

Title: Re: The big difference between Chess and Arimaa?!
Post by omar on Mar 15^th, 2006, 2:45pm

on 03/15/06 at 01:06:47, Fritzlein wrote:

The odds are they won't give you the time of day, but it doesn't hurt to ask. That would be so amazingly cool if the Hydra team were interested.

Yes, it never hurts to ask.

I did get a kind reply from Chrilly Donninger the lead programmer saying that Hydra is especially programmed for chess and its not trivial to change it for arimaa.

Title: Re: The big difference between Chess and Arimaa?!
Post by seanick on Sep 28^th, 2006, 9:42am

its too bad- I guess the FPGA part of the program they run is too tightly coupled with chess. You would think they would be interested in trying to adapt it, after all the FPGA's are meant to optimize the same thing for chess that take the most time in Arimaa as well. that added performance, only allows them to get to P4.5 (P18 for chess, /4 for arimaa...) in approx. 1 second of evaluation.
Someone here played bot_bombP4 right? I imagine Fritzlein but don't recall the specifics. The match was called early because it took a really long time per move, and wasn't noticably better, right? Would it have been more difficult if it was the same playing strength, but at blitz speed?

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Sep 28^th, 2006, 5:06pm

on 09/28/06 at 09:42:22, seanick wrote:

Someone here played bot_bombP4 right? I imagine Fritzlein but don't recall the specifics. The match was called early because it took a really long time per move, and wasn't noticably better, right? Would it have been more difficult if it was the same playing strength, but at blitz speed?

Yes, I played part of a game against BombP4. It wasn't enough moves for me to tell whether there was a difference in strength or not. BombP4 might actually be noticably stronger in tactical situations, which means that to win one would have to take advantage of Bomb's deep strategic misunderstandings.

Certainly a blitz bot becomes tougher if you can soup up its calculation speed. All of us humans are blunder-prone at blitz speed. I think beating Bomb at blitz relies in part on tactical weakness in Bomb, which would be eliminated with enough processor speed. If BombP5 (=Bomb on Hydra hardware?) could play at blitz speed, I'm not sure it could be beaten by today's humans, except by trial and error discovery of a loss the computer would repeat.

As long as I'm making s.w.a.g.s, though, I should say that I think BombP5 would still lose the Arimaa Challenge at two minutes per move, where the humans avoid most tactical blunders. I think Omar could open up the Arimaa Challenge to any type of supercomputer hardware, and humans would still be safe in the absence of software advances.

I was just noticing today in my game with chessandgo how often "slow" plans are critical in Arimaa. I had his camel hostage with my elephant and his horse hostage with my camel, both next to the same trap. I hit on the plan of herding the horse hostage over to my other home trap so his elephant wouldn't be able to defend both hostages. This was the decisive idea of the game, forcing chessandgo to let me capture a rabbit and the hostage camel for only a horse. Without that idea, I would have been losing, because I lacked total control of any trap. Yet it would have taken at least four moves to execute this plan, probably five moves to secure material advantage.

I'm trying to imagine a full-width alpha-beta searcher that can look ahead four moves on each side. Even with perfect move-ordering (which there never is), and thus optimal pruning, one would have to evaluate 15000^4 nodes, approximately 5 x 10^16. To move within an hour, it would have to search more than a quadrillion nodes per second. Any guesses on when such a computer will be available?

Title: Re: The big difference between Chess and Arimaa?!
Post by 99of9 on Sep 29^th, 2006, 6:59am

on 09/28/06 at 09:42:22, seanick wrote:

allows them to get to P4.5 (P18 for chess, /4 for arimaa...) in approx. 1 second of evaluation.
Someone here played bot_bombP4 right?

I thought the P18 for chess was using the definition of 1ply=1play for both players? If that's true, bombP4 would only be equivalent to P2.0 in chess.

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Sep 29^th, 2006, 9:11am

on 09/29/06 at 06:59:57, 99of9 wrote:

I thought the P18 for chess was using the definition of 1ply=1play for both players?

I'm pretty sure the standard usage of "ply" in chess is that one play for both players is two ply. Furthermore, I would be stupefied if any computer could full-width alpha-beta search 18 moves for both players (36 ply) at all, never mind in one second.

Title: Re: The big difference between Chess and Arimaa?!
Post by PMertens on Sep 29^th, 2006, 11:06am

Quote:

BombP4 might actually be noticably stronger in tactical situations, which means that to win one would have to take advantage of Bomb's deep strategic misunderstandings

True ... the trick is to never get involved in tactical situations.

I am quite confident that brute force without a really decent eval will not be sufficient in my lifetime.
(Meaning: a P6_Arimaa_Score is not very frigthening ... while a P20 would probably kick my butt)

Title: Re: The big difference between Chess and Arimaa?!
Post by PMertens on Sep 29^th, 2006, 2:20pm

and one more difference: Fritzl would never forfeit a game because of restroom problems ::)

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Sep 29^th, 2006, 11:22pm

on 09/29/06 at 14:20:17, PMertens wrote:

and one more difference: Fritzl would never forfeit a game because of restroom problems ::)

I confess, I have gone to the toilet in the middle of an on-line Arimaa game. Another reason to like accumulation of reserve...

Title: Re: The big difference between Chess and Arimaa?!
Post by OLTI on Nov 27^th, 2006, 10:05am

The match is underway.
Live games at http://www.rag.de/microsite_chess_com/interaktiv.html

Title: Re: The big difference between Chess and Arimaa?!
Post by RonWeasley on Nov 27^th, 2006, 11:09am

Excellent! I watched Kramnik blunder!

Title: Re: The big difference between Chess and Arimaa?!
Post by DorianGaray on Nov 28^th, 2006, 8:46am

Thanks for the link OLTI. I enjoyed watching those games. :)

Title: Re: The big difference between Chess and Arimaa?!
Post by DorianGaray on Nov 28^th, 2006, 8:50am

on 03/14/06 at 17:43:59, Fritzlein wrote:

I don't know if anyone can tell what would be the results at those speeds. It would be like bomb having several years (with the current computer) while the human player would only have a few minutes (at most) to choose his move.

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Nov 28^th, 2006, 11:06am

Wow, I (along with the whole chess world) am in shock. There are plenty of instances of chess grandmasters blundering, but I don't think the reigning World Champion of chess has ever overlooked a checkmate in one move.

I felt pretty dumb after I blundered a horse to Adanac in our game in the fifth round of the 2006 World Championship, but now I won't beat myself up no matter what mistakes I make this year. If Kramnik can be sporting and dignified after a mistake like his, I can only applaud him, and try to behave half as well myself when it is my turn to blunder horrifically.

I hope Kramnik fights back for at least one win in the remaining four games, if not a match victory. But if computers have come to dominate even the World Champion of chess by a substantial margin, there is one benefit for us: It makes Arimaa all the more intriguing.

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Nov 28^th, 2006, 11:31am

on 11/28/06 at 08:50:27, DorianGaray wrote:

I don't know if anyone can tell what would be the results at those speeds.

I agree, Dorian, that we have no way of telling what pure hardware speedup would do. If we look to the history of chess, we can plot an almost linear relationship between rating of chess computers and log of computing power. But chess theory was not advancing in the way that Arimaa theory is, which could have a drastic effect on plotting a similar trajectory for the ratings of Arimaa bots.

While Fotland was still actively developing, he kept up with Arimaa theory to a certain extent. For example, when everyone began to beat Bomb with camel hostages, he tuned Bomb to keep its camel safer. But now Arimaa theory marches on while Bomb is static. Hardware speedup can't compensate for that. For example, if it turns out that swarming is a good strategy, a super-fast Bomb would play defensively super-effectively, and actually cooperate in getting itself buried under a flood of pieces.

I expect a linear relationship to hold between the log of computing power and the playing strength only if theory isn't advancing, or if developers keep putting the latest theory into their bots. Given how far ahead of bots we are strategically, I expect that any significant improvement in bot play will come from software advances rather than hardware advances. In other words, we have exactly the situation Omar hoped to create. (which is pure luck on Omar's part, I'd say ;-))

Title: Re: The big difference between Chess and Arimaa?!
Post by nbarriga on Nov 28^th, 2006, 7:26pm

It's too bad that arimaa bots improvement has stalled a little. Bomb looked promising when actively developed, but now i think there isn't any bot near bomb, not to say near top human players. I think, to keep up the original goals behind arimaa, Omar has to come up with new ways to encourage development. the only thing i can imagine right now that would so that, is forcing WCC participants to opensource their bots.

Otherwise, all the effort put on bots that get discontinued is lost.

For example, one way of easily telling if raw computer power could improve significantly a bot performance, would be to paralelize the best bot available, to run on a cluster, or even as a BOINC(boinc.berkeley.edu) project.

This is coming from a disapointed bot developer(don't worry, disapointed at myself, not the commuty behind arimaa, you're great guys!). It is very difficult to start a bot, because you have to put all the accumulated arimaa strategy on it.

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Dec 13^th, 2006, 7:49am

The thread title "The big difference between Chess and Arimaa" invites another answer: Humans can no longer compete at chess. With Kramnik losing two and drawing four in this match, and never looking close to a win, it is hard to see a top human ever beating a top computer again.

One could speculate how long humans would hold out at Arimaa if the Challenge prize were $1,000,000, but that misses the point. If the human got half a million just for defending the prize, we would have humans studying the game so hard they would be rated 3000.

Title: Re: The big difference between Chess and Arimaa?!
Post by DorianGaray on Dec 13^th, 2006, 12:22pm

on 12/13/06 at 07:49:51, Fritzlein wrote:

...and never looking close to a win...

Maybe it's just me but I believe that he was about to win the one he blundered away, don't you agree Fritz?

Title: Re: The big difference between Chess and Arimaa?!
Post by Fritzlein on Dec 13^th, 2006, 3:01pm

on 12/13/06 at 12:22:46, DorianGaray wrote:

Maybe it's just me but I believe that he was about to win the one he blundered away, don't you agree Fritz?

Other people have said that the game was dead even just before Kramnik blundered, but I'm not good enough to tell by myself. My evaluation goes like this: number of pieces is even, king safety is about even, pawn structure is about even, I don't see any way for either player to win a piece, checkmate, or queen a pawn. Ergo, it must be a draw. ;-)

But seriously, when I say Kramnik never looked to be winning, I was relying on someone else's opinion. In one game Kramnik was up a pawn, so in fact I personally did think he might have winning chances, but people who are good at chess said Kramnik was actually behind, and as it turned out he later had to sweat to hold the draw. It's best if you don't trust anything I say about grandmaster chess positions!

Title: Re: The big difference between Chess and Arimaa?!
Post by chessandgo on Dec 13^th, 2006, 3:52pm

as far as I recall, they said that to avoid checkmate, Kramnik had to move his King, which would have allowed Fritz to get a draw by perpetual check ...

Title: Re: The big difference between Chess and Arimaa?!
Post by 722caasi on Sep 11^th, 2007, 6:08pm

In this position
5N1k/q5p1/7p/4P3/pp2Q3/8/1P4PP/2b4K b - - 0 1
After
1 ...... Kg8
2 Ng6 Bxb2
3 Qc8 Kh7
4 Nf8 Kg8
5 Ng6 Kh7
and so on with perpetual check