Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Diep as a strong sparring opponent (longish)?

Author: Eelco de Groot

Date: 17:00:21 10/13/03

Go up one level in this thread


I think that all sounds logical, Russell. But just thinking about it right now
as just an interested person I also agree with Uri that chessprograms do not
approach perfect or solved play at long or very long timecontrols. My opinion:
chess is far from solved and tactics still play a role even if you have many
days for a move. Not in every game of course, but if thirtynine games ended in
perfect draws -in 2020?- it will now be that one extremely hard to find move in
game forty that decides who wins the match. So in a way tactics becomes even
more important when random errors and imperfect heuristics causing the
occasional brilliant moves but also blunders, become obsolete and are replaced
with almost perfect calculations.

To put it in extremes for Vincent: "Positional heuristics are not necessary if
you can calculate beans perfectly and very deep"

Going towards the other end of the scale, short timecontrols, I think you may be
very right about the relative weaknesses becoming skewed. At least it sounds
right. Every computation you want to do that does not scale with search depth
-or with the number of nodes searched-, will have a different influence in
shorter or longer timecontrols. I you do a calculation of a fixed iterval, say a
second, at every root, it will hurt in extremely fast blitz.

Another practical problem; at long timecontrols the search might become unstable
and you haven't tested that enough. Or some unforeseen overflow may occur, or
the maximum number of halfmoves is set at 64 to save memory or control
extensions but causes the search to stop too soon at very large plydepths
(theoretically).

And there are other things to consider too that make Christophe's theory more
complex: the "ideal" search tree may in theory look like a fractal at any size
(that's how I imagine it in my mind) but for a very large tree the endnodes will
be more often in a next stage of the game, if the root is in the middlegame the
leaves will be in the endgame etc. As a non-chessprogrammer I imagine that may
make a difference in the way you search the larger tree even if the general
pattern of the search stays the same.

 (Computer-)chess is complex!

  Regards,Eelco




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.