Author: Kurt Utzinger
Date: 03:16:03 03/01/04
Go up one level in this thread
On March 01, 2004 at 06:00:37, Claude Le Page wrote:
>Hi Kurt!
>I agree that statistical method is too heavy to get sure conclusions about
>settings;
>Now , there is an alternative method : one could name it "soft debugging"
>it is shorter in the sense that it needs much less games , but a deeper work:
>Some explanations
>I found , quite by chance , that many engines choose opening variations that
>theory considers as bad ; once it was sure that theory is right , this choice
>may be considered as a bug and a "better" setting wold induce a choice
>consistent with opening theory ;
>here are 2 examples:
>1
>Nimzoindian defence , Leningrad variation
>1 d4 Nf6 2 c4 e6 3 Nc3 Bb4 4 Bg5 h6 5 Bh4 c5 6 d5
>most engines choose 6...Nxd5 7 Bxd8 Nxc3 8 Qb3 Ne4+ etc only the shredder and
>junior families , and the newest crafty make a more regular choice
>IMHO , a "good" setting must repair this bug
>2
>2 knights defence
>1 e4 e5 2 Nf3 Nc6 3 Bc4 Nf6 4 Ng5 d5 5 exd5
>most engines choose 5...Nxd5
>there is a polemic about 6 Nxf7 , but after 6 d4 , Black is lost;
>often ,in correspondence games there is an attempt of rehabilitation
>but it is unsuccessful
>here too , a better setting must lead to choose Na5 , Nd4 , or b5
>Without doubt , one should find plenty of examples of this sort
>Each time a bug of this kind is corrected , it leads to a more general
>improvement
>Of course , it happens that settings don't suffice : in this case ,more drastic
>changes are needed
>Friendly Yours
>Claude Le Page
Hi Claude
This method of "soft debugging" is surely a good one, but as you
mentioned yourself: a time consuming (but most interesting) matter.
Greetings
Kurt
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.