Author: Christophe Theron
Date: 00:28:10 12/26/97
Go up one level in this thread
On December 25, 1997 at 23:01:02, Fernando Villegas wrote:
>Wait a second: why some people give so total importance to deep lines
>and to be capable of looking deeper lines? I understand that it is
>important as much as tactical, evident shots appears, but even so it
>seems to me that the core of the issue is if the engine -or human- knows
>what he is seeing. I can imagine a program made out in such a fashion as
>to always lose pieces even if he see 100 plies IF his evaluation does
>not score material. Or another capable of being mated because ONLY see
>material, but has not idea of what a king attack is. So Junior must have
>something else that a deeper look. Even more, I suspect that his deeper
>look has to do with a better evaluation strategy. Is not what GM do?
>They look just one or two moves in each ply, so in some ocasions they
>can see long lines, 15 or 20 or more plies deep. If you read the
>interview he gave to me in WCCR, you will get a hint, the way as he
>conceives the concept of ply, what is a ply. Take a look at it...
>Fernando
Wait a minute (a second is not enough).
Of course I was joking by saying that maybe Junior sees 4 or 5 plies
farther than everyone.
I have myself spent a lot of time making my program compute as deep as
possible with a classical approach: when I say "ply N", that means that
the program has seen (most) of the possible moves on the first N plies
from the root, and has seen only the capturing and promoting moves in
the following plies. This is for example Crafty's search strategy (with
some exceptions for extensions and check evasions).
Using this "classical" search, Tiger is able to compute to ply 12-14 in
the middle game on the PII-300 64Mb computer I used in Paris. In blitz,
it was ply 8-9 (middle game). From what I've seen in Paris, Tiger is
quite good in that field (brute depth), and is now even better. I
remember two games, against Fritz and DarkThought. Tiger was computing
quite the same depth as these two (fast) opponents, except in the
endgame against DarkThought (which was running on Alpha 767MHz with
256Mb RAM...).
One thing that I know for sure now: if you want to go deeper, really
deeper, you have to use another approach. It doesn't mean that you will
play better, of course. But seeing deeper lines seems to be a little
more "human like", and more interesting.
Maybe one way to do it is to have a different quiescence search. Because
I always felt stupid to look only at capturing moves when your queen is
under attack. But until now I didn't succeed to make something more
efficient that a classical quiescence search...
Seems that Junior has a good search strategy. Not necessary a really new
one... Mr Shannon said: "don't evaluate a position if it is not quiet".
Most of the programs (including mine) don't evaluate a position if the
side to move has a good capture. Right, but what if the other side has a
good capture, too?
Christophe
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.