Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: path history effect

Author: Robert Hyatt

Date: 19:43:41 11/17/02

Go up one level in this thread


On November 17, 2002 at 06:10:03, Georg v. Zimmermann wrote:

>Hi,
>
>The reason why this happens is black seems to be getting worse scores with
>increasing depths, so it uses the repetition to push the horizon away.
>
>I dont see it as problem of the hash table. When not using hash tables, couldnt
>the same thing happen ?

NO.  Because repetition detection would kick in and propogate a draw score
back up the tree.

But with hashing, the path from the root to the actual stored position doesn't
have to be the same as the path from the root to the position where the hash
hit occurs.  And that can lead to errors and there is nothing that can be
done other than to encode the complete path in the hash table.  And that
will wreck its utility totally...


>
>Or do you want to use the hash table to _detect_ such a symptom ? Interesting
>idea..
>
>Georg
>
>On November 16, 2002 at 17:28:46, Gerd Isenberg wrote:
>
>>Hi all,
>>
>>in this position from Virtual - IsiChess, WMCC99 Paderborn, i noticed probably
>>an path history effect with IsiChessMMX:
>>
>>[D] r3k2r/1q3ppp/p3p3/6P1/1n2pP2/2P5/1P4QP/1R2KB1R b Kkq -
>>
>>At depth 11 following PV occured:
>>
>>1.. Rd8 2.cxb4 Qxb4+ 3.Kf2 Qd2+ 4.Kg3 Qe3+ 5.Kh4 h6 6.g6 Qxf4+ 7.Qg4 Qxg4+
>>8.Kxg4 f5+ 9.Kf4
>>
>>with this final position:
>>
>>[D] 3rk2r/6p1/p3p1Pp/5p2/4pK2/8/1P5P/1R3B1R b k -
>>
>>At depth 13 a longer PV came up with the same score and the same final position.
>>I guess it was a true score hit from the hashtable from the iteration two plies
>>before. Instead of 7...Qxg4+ there were two delaying checks before the queen
>>exchange took place.
>>
>>Is it worth to consider and store the distance to root into an hashtable entry?
>>But what to do in case of true scores with appropriate drafts, if the stored
>>root distance is less the current?
>>
>>Comments welcome.
>>
>>Thanks,
>>Gerd



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.