Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Is Genius the only asymmetry program out there?

Author: Christophe Theron

Date: 15:47:00 10/08/03

Go up one level in this thread


On October 08, 2003 at 13:31:40, Dana Turnmire wrote:

>I found this post which is several years old and thought it was interesting.
>
>"Here the point:
>
>Indeed Genius is NOT playing accurate chess.
>It is betraying us with boring evaluations that shall convince us:
>everything is save. And suddenly, nothing is true.
>
>Please try out the following example:
>[d] rqr3k1/3Qbp2/p1n1p1p1/1pp5/2P2P2/2N3P1/PP3PB1/R3R1K1 w - -
>
>Position Karpov-Topalov:
>
>1. Rxe6!
>
>
>W: Kg1 Qd7 Ra1e1 Bg2 Nc3 a2b2c4f2f4g3
>B: Kg8 Qb8 Ra8c8 Be7 Nc6 a6b5c5e6f7g6
>
>White to move !
>
>Please put this position into Genius4!
>
>Put the level on infinite to join the evaluations and main-lines.
>Click on compute and wait...
>
>First, Genius4 wants to play cxb, then it changes in the 6th ply
>to Bxc6.
>
>Consider you are black and white is your mail-chess-opponent.
>You have bought Genius4 because some people have told you
>that it is a very useful program for analysis.
>You believe these people.
>You use Genius to show you which move your opponent could play next,
>so that you are prepared on this.
>Because you expect Genius move Bxc6 you could analyse Ra7 or other
>moves in the time, your opponent considers about his move.
>
>Now - after 6 minutes on my Pentium 120 Mhz,  my Genius4 is
>through the 7th ply and has still Bxc6 with +0,66 in the main-line!
>So - because I am black, I am a little in weaker position, but not
>much! Maybe I will draw this game, you consider... and feel save!
>
>Genius4 has computed 19 plys selective, amazing. It is known to be
>a strong chess program, so I can trust it.
>It is now 8 plys deep and I would expect that WITHIN THESE 8
>PLY's I CAN FEEL TOTALLY SAVE THAT MY OPPONENT CAN NOT
>KILL ME WITHIN A FEW HALF MOVES.
>
>Thats what the materialists will always try to tell me,
>it is for sure within a certain range, that this and that is the best
>or strongest move, and that there are no surprises.
>
>Genius is now running through the 8th ply, after 9'44 " it changes
>into a main line with cxb instead of Bxc6.
>
>Ok, I should expect that my opponent plays these 2 alternatives.
>If he plays Bxb6 I will play Ra7, if white plays cxb I will answer
>with the same move.
>
>After 23 minutes Genius looks for Rxe6 in the 34th branch of the 8th
>ply.
>Will it change into the RIGHT move.
>NOOOOOOOOOOOOO.
>It fails and still insists of cxb !
>
>After 26' Genius4 starts to compute into the 9th ply (21 selective!!).
>
>But it still has no idea of the Master move: Rxe6.
>
>Now Save the position. Do a cold-reset and edit the position again,
>but
>with a different Rook. Put the rook from c8 to f8.
>
>Let Genius4 consider now in infinite and look into the main
>line, especially the 2nd move of the main line.
>
>Genius4 will play Rc8 and we have the original position.
>Than it expects cxb.
>
>After 2'36 seconds Genius4 evaluates suddenly -1.24 and has the
>following main-line:
>Rc8 Rxe6 Ra7 ....
>
>Thats it !!!!
>
>So Genius4 sees the Rxe6 sacrifice although the plot is one ply
>deeper in the tree.
>But it sees Rxe6 now within the ply 2-4-6-8 instead of 1-3-5-7-9.
>Because Genius4 is asymetrical, it considers in ply
>1-3-5-7-9 only few best moves and is not fully computing all moves
>accurate.
>In ply 2-4-6-8 it computes all opponent moves, so it sees all THREADS
>against it.
>
>This is the reason why genius is so difficult to kill, although it has
>a weaker position. It sees almost everything what YOU can imagine,
>and concentrates on defending instead of DOING!
>
>The above example shows that this has nothing to do with Genius
>tactically strength!
>
>Genius is not tactically to weak to find Rxe6.
>It finds the whole stuff when the plot is half a move deeper in the
>tree.
>
>Since I have studied Richard programs, I have collected positions
>where Richards programs failed because of the asymetrical approach.
>
>This approach may be one reason why genius is that strong,
>but it could also be the MAIN reason why genius has not made
>much programming progress over the last versions."
>
>I tried the test position with several programs and Genius 7 is the only one
>that will not play 1.Rxe6!
>
>Is this article accurate?



I guess it was a post from Thorsten, right?

I have tried this position again with Genius2 (the Windows version) and to me it
does not look like asymetry in search but asymetry in evaluation.

Genius indeed uses asymetry in search but not the way Thorsten describes it. The
method Thorsten thinks about either would not work (the savings would be very
small) or if it worked then it would be even more efficient to use it for both
sides.

But anyway in the example above it looks more like asymetry in evaluation
(evaluating a King attack as more serious when its own King is attacked).

I remember that Bob told us that Crafty was doing similar. I think I'm doing
something similar in Chess Tiger but I would have to have a closer look to the
source (there have been so many changes in the code that I do not remember for
sure).



    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.