Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Mate in 19 - Glaurung Seems Faster

Author: Uri Blass

Date: 10:13:13 12/01/05

Go up one level in this thread


On December 01, 2005 at 13:06:14, Uri Blass wrote:

>On December 01, 2005 at 12:46:53, Tord Romstad wrote:
>
>>On December 01, 2005 at 10:42:00, Steve Maughan wrote:
>>
>>>Centrino 2 GHz
>>>
>>>[D]3n3k/bp3pp1/8/8/R1Q2R2/6p1/B2p1PP1/q1rn2K1 w - -
>>>
>>>Engine: Glaurung Mainz (128 MB)
>>>by Tord Romstad
>>>
>>>10/45  0:11   0.00     1.Rh4+ Kg8 2.Rh8+ Kxh8 3.Qh4+ Kg8
>>>                       4.Qxd8+ Kh7 5.Rh4+ Kg6 6.Qd6+ Qf6
>>>                       7.Qxg3+ Qg5 8.Qd6+ Qf6 9.Qg3+ Qg5 (4.858.674) 431
>>>
>>>11/47  0:21   +M19     1.Rh4+ Kg8 2.Rh8+ Kxh8 3.Qh4+ Kg8
>>>                       4.Qxd8+ Kh7 5.Rh4+ Kg6 6.Qd3+ f5
>>>                       7.Qxg3+ Kf6 8.Qd6+ Kg5 9.Rh5+ Kxh5
>>>                       10.Bf7+ Kg5 11.Qg3+ Kf6 12.Qg6+ Ke5
>>>                       13.Qe6+ Kf4 14.g3+ (9.646.864) 453
>>
>>Hi Steve,
>>
>>The PV is incomplete, which makes me a bit worried.  Glaurung should never
>>announce
>>mate unless the PV ends in a mate.  Do you use a GUI or some other tool which
>>truncate
>>the PVs?  Otherwise, I have a new bug to track down.  :-(
>>
>>The current development version of Glaurung is much faster in this position, by
>>the way.
>>On my iMac G5 (which is about half as fast as your Centrino), it announces mate
>>in 21
>>after 6 seconds and 8 plies, and reduces this to mate in 19 at the next
>>iteration after 11
>>seconds.  The reason for the improved solution time is simple:  I am
>>experimenting with
>>unlimited checks in the qsearch (Glaurung Mainz searches checks only in the
>>first 3 plies
>>of the qsearch).
>>
>>Tord
>
>unlimited checks in the qsearch cannot work.
>You may not have a limit of plies but with no limit the search is going to
>explode.
>
>Uri

I can add that even if you allow only one check at every ply in the qsearch you
may have
check 3 replies,check 3 replies,....

This means that if you have 20 consecutive checks you may need to search 3^20
positions to finish your qsearch.

Of course you can limit number of nodes in the qsearch but qsearch with no
limits does not make sense to me.



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.