Chess... just another game...

The Mayor

Well-Known Member
#1
I am always watching out for Crafty... I used to run Crafty on my Sparc station in Ohio (it was a robot on one of the Internet Chess Servers) about 10 years ago.

(Dead link: http://www.ru.is/wccc05/default.asp?Page=Notepad&ID=3)

Not as good as the 2004 finish, but still world class.

There's a great article on Computer Chess in this week's New Yorker as well.

--Mayor
 
#2
rumor mill...

Rumor has it that the author is doing a major rewrite of the "smarts" of the program, planned debut at the next WCCC event. Might be a few surprises built in as well...
 

The Mayor

Well-Known Member
#3
I hope...

I hope the author writes a program that crushes "Hydra" -- shouldn't be too hard for that kind of genious programmer.

--Mayor
 
#4
How many openings do they teach chess programs?

As a dedicated novice I've been of the opinion that you need to learn only one opening as white (and learn it well), and a dozen or so defenses to be a good novice player.

So being some openings are clearly more powerful than others, why would you teach a program more than one, and for that matter, why would a pro use more than one?

(P-Q3 fan myself. Keeps them guessing.)
 

The Mayor

Well-Known Member
#5
When I ran it...

I ran an opening book size of several hundred megabytes, that was in 1995. I am sure these things now have books of gigabyte size. What is far larger, however, are the endgame databases. They are huge and exhaustive.

Crafty allowed the person running it to customize the opening repetoire -- I used to aim towards gambit and open lines that were unusual, to quickly get the human into a position that was open, complicated, and unknown. Using that strategy, the program I ran go over a 2500 rating. Look at the top ratings it achieved and their dates. This program lives on fics (free internet chess server)

Code:
fics% finger meru
Finger of meru:

Last disconnected: Sun Jun  1, 13:06 PDT 2003

          rating     RD      win    loss    draw   total   best
Blitz      2244    350.0   14865    1427     900   17192   2576 (09-Jul-1996)
Standard   2177    350.0     948     123      93    1164   2407 (03-Mar-1997)
Lightning  2293    350.0    1079     137      42    1258   2548 (26-Jun-1996)


 1: Crafty v16.13 (1 cpus)
 2: Meru runs using xboard and a zippy interface (Thanks Tim) on a random
\   machine.
 3: Message all complaints/problems/questions to Meru's operator "mayor"
 4: Sometimes manually operated or with human assistance
--Mayor
 
#6
computers...

are blessed with perfect memory, and near instantaneous access. So the opening book can be as big as you want. For example, how about taking 10 million grandmaster games, cataloging their opening moves, and then using that to play the opening?

Trivial to do with a computer.

BTW one for white is not quite enough. For example, 1. e4. You still have to deal with a bunch of choices by black. 1. ... e5, 1. ... e6, 1. ... c5, 1. ... c6, 1. ... d6, 1. ... Nf6, etc. Gets pretty hairy to be prepared for anything your opponent throws at you, regardless of your first move...
 
#7
file sizes.

a gigabyte opening book is pretty common. The file "enormous.pgn" many use for opening book preparation is well over a gigabyte of ASCII text.

The endgame tables boggle the mind. We have completed the 6-piece endings now, and we have about 1.2 terabytes of data as a result. The new 5-piece format we all use still takes 7 gigabytes for all the 3-4-5 piece endings possible. The 6's stretch that to 1.2 terabytes. 7's are a ways off for most...
 
#8
Ah so that's how chess programs work

They use a large number of canned sequences.

I was under the impression that a computer looks at a current board position, traces through the opponent's possible responses to each move and the computer's next move, all the way to a checkmate, and using logic determines which move is most likely to result in a checkmate of the opponent. Using this method it wouldn't need any archived games at all, just a lot of processing ability.

Perhaps the processing task could be relieved by playing it more the way a human plays chess in breaking your goals down to more proximate ones that by experience we know will lead to a checkmate. E.g., I know if I can successfully fork an opponent's queen the game is essentially over, therefore forcing the fork is as good as a checkmate. Or a two-step defensive process, I realize my rook is forked, but one of my defensive options leads to a horsetrap will will partially compensate the loss of the rook, as long as it does not lead to any bigger problem, and so on. Higher level players don't often get rooks forked or knights trapped (I guess) but you have to consider the possibility when making a move, so maybe this would be another way to look at writing a chess program.
 
#9
actually

the "book" is only used to get the game into a known playable position, rather than something dead lost. But after that, the programs play very much like humans, in that they have some sort of code that moves the pieces around, and another piece of code that says "this position is good for white with a score of +1.5, this position is better for black with a score of -.4, and so forth. It then picks the pathway through the minimax tree that maximizes things for both sides. We rarely see "checkmate" in the game, and in fact, those are the _easy_ parts of the game to deal with since checkmate is very precisely defined. It is the positions where there is no checkmate, there is no way to win material, where things get interesting. Now the computer has to decide "what to do when there is nothing obvious to do?" and that's an interesting problem to solve.

The computers can now beat just about every player on the planet, thanks to ridiculously fast hardware of today. So the interesting games become computer vs computer. :)
 
#10
Its absurd! ...

"The computers can now beat just about every player on the planet, thanks to ridiculously fast hardware of today. So the interesting games become computer vs computer. :)"
------------------

... how interesting will a .0001 second game be (3 years away)? zg
 
#12
not quite the way we do it. :)

I do lots of testing where I play "game in one second", and it is interesting to watch as the pieces are just a "blur". What I see is "blurrrr", game over, "blurrrr", game over. I only do this to get a quick feel for whether a change is good or bad. Serious chess is still played at 40 moves in 2 hours. Even computer vs computer. Give them more time, they search deeper, and find better moves. So while game in 1 second can be played, and the programs are incredibly good even at that speed, the longer games are more interesting. A good human will find lots of program mistakes in the game/1sec games. But he won't find very many at all, if any, in the 40 moves in 2 hours games...

Of course, humans play both fast and slow chess. I've played who-knows-how-many one minute games myself against other humans. And I make far more tactical or positional errors there than I do in long games, say 40/60 moves in one hour or more...
 
#13
Chess skill as a predictor of blackjack skill

Seems to me that when playing chess, there are some days when I can beat anybody you can put in front of me, and other days where I cannot win at all and play only on the lowest level.

So maybe playing a game of chess would be a good self-test before going out to play BJ. A day when I'm leaving pieces exposed might also be a day when I'm reversing the count, etc.
 
#14
different skills

The ability to play chess requires significant visualization ability, so that you can mentally play out a sequence of moves, and "look" at the resulting position to see whether you like it or not. A good memory is also important for remembering opening variations that are tricky or dangerous. And "generalization skills" help so that you can apply something learned in one game to a completely different game position, but with common elements.

BJ is a purely mechanical skill of recognizing cards, adding 'em up, and recalling BS and BS departure indices. There are lots of times when I don't feel like playing chess, as playing a 2200-level player (about where I am) turns into a mental strain that requires lots of energy (at least for me). I've played lots of late-night chess, and I can tell you without a doubt, the more tired I become, the worse I play. I've noticed that in lots of players. I can play BJ dead tired and not think about the counting part of the game at all. In fact, it is very hard for me to not count cards no matter what kind of card game I play. :)
 
Top