Author: Don Dailey
Date: 04:42:09 11/28/97
Go up one level in this thread
I have to say that the trick in computer chess is to put your program
in a position where it has a CHANCE to win and then see what happens.
Unless your program is a HUGE favorite in each individual game, the
odds of surving till the end are against it. But still, in a widely
varied field of contestants, only the best have a reasonable chance of
making it.
Even in Hong Kong Deep Blue finished 3rd and its team commented before
the match that their "chances" of winning were probably less than 50%
due to
the limited number of rounds and the "luck" factor. For Cilkchess,
right
at this momement I believe Cilkchess's chances are significantly less
that Nimzo's because Cilkchess still has to play the King which is
percieved
as being the toughest match left for Cilkchess. The feeling is that
Nimzo has no difficult opponents left. Of course anything can happen.
Cilkchess is a 64 bit program running on a small Alpha SMP of 4
processors.
Each processor is 466 Mhz. The program is brand new and not the same
program that competed last year. We have only been working on the
program
for a couple of months and we are still heavily debugging the thing.
There have also been many flashes of UNbrilliance by Cilkchess so far
but
it has had some nice moments too. I still cannot believe we pulled out
a
draw in the Nimzo game.
Here are a few of the internals of Cilkchess:
1. Bitboards similar to Crafty
2. MTD(f) search
3. No pre-proccesor for evaluation
4. More mid game evaulation than our other MIT programs.
5. We are only doing about 500,000 nodes per second due to huge
knowledge
overhead. Some of the micro's here are not too far behind us!
6. No databases except king and pawn vs king.
7. Almost no endgame knowledge (see game 4) We do have a table of
about
50 simple endings that give us either an alternative material score
or a pointer to a function to handle scoring more specifically for
that
ending.
8. Lot's of bugs!
It is not clear if the program is any stronger than Cilkchess 1 or the
top
programs here but we think it may be due to the huge hardware advantage
we
enjoy. After the program matures, is debugged and fine tuned, it will
be
very much stronger than it is now.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.