Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How I Learned to Stop Hating 141

Author: Stuart Cracraft

Date: 13:34:30 09/05/04

Go up one level in this thread


On September 05, 2004 at 12:24:13, Uri Blass wrote:

>On September 05, 2004 at 12:15:44, Stuart Cracraft wrote:
>
>>On September 05, 2004 at 09:27:01, Henk Bossinade wrote:
>>
>>>On September 04, 2004 at 17:54:50, Stuart Cracraft wrote:
>>>
>>>>On September 03, 2004 at 09:47:26, Jan K. wrote:
>>>>
>>>>>You should look for bugs in your search....i find the move even with no
>>>>>extensions but the threat extension set to 1/2 ply and no checks in qsearch.
>>>>>Takes 60 seconds and almost the same number of nodes like your full search.
>>>>
>>>>I am not sure where else to look.
>>>>
>>>>With nothing other than hashing and null move, no extensions including
>>>>no checking extensions no check-evasion extensions, etc., how many
>>>>ply does it take your program to solve WAC 141?
>>>>
>>>>I *have* to have recapture extension enabled in order to solve it
>>>>currently in anything approaching real-time.
>>>>
>>>>Stuart
>>>
>>>I get ply 9 with only hashing. Maybe your move ordering isn't optimal for WAC
>>>141.
>>>
>>>(proc. VIA C3 700Mhz)
>>>  4    0.05    -70        3741   Kg2f1 Nf4e2 Qc1d1 Qc7a5
>>>  4.   0.06    -70        4250   Kg2f1 Nf4e2 Qc1d1 Qc7a5
>>>  5    0.31    -70       15588   Kg2f1 Nf4d3 Qc1d2 Nd3e1 Qd2d1
>>>  5.   0.34    -70       20946   Kg2f1 Nf4d3 Qc1d2 Nd3e1 Qd2d1
>>>  6    0.84    -70       92728   Kg2f1 Nf4d3 Qc1d2 Nd3e1 Qd2d1 Qc7a5
>>>  6.   1.09    -70      104292   Kg2f1 Nf4d3 Qc1d2 Nd3e1 Qd2d1 Qc7a5
>>>  7    6.07    -70      533246   Kg2f1 Re8e2 Qc1b1 Re2d2
>>>  7.   6.59    -70      658602   Kg2f1 Re8e2 Qc1b1 Re2d2
>>>  8   17.44    -70     2165687   Kg2f1 Re8e2 Qc1b1 Re2d2 Rh1g1 Qc7a5 Bb3d1 Qa5b4
>>>  8.  21.14    -70     2381705   Kg2f1 Re8e2 Qc1b1 Re2d2 Rh1g1 Qc7a5 Bb3d1 Qa5b4
>>>  9   99.10    -70     6629614   Kg2f1 Re8e2 Qc1b1 Re2d2
>>>  9  169.96     ++    12524259   Qc1xNf4 Bd6xQf4 Rh4xPh5
>>>  9  240.97    260    22211001   Qc1xNf4 Bd6xQf4 Rh4xPh5 Pg6xRh5
>>>  9. 240.98    260    22211008   Qc1xNf4 Bd6xQf4 Rh4xPh5 Pg6xRh5
>>>
>>>Cuts  : Null=0 Delta=0 SEE=0 Eval=0 Rzr=0 Xfut=0 Fut=0
>>>Ext.  : Check=0 OneReply=0 Threat=0 Recap=0 RevCheck=0 Pawn=0
>>>Misc. : Hashprobes=12907069(10% hits) Moveorder=95% Bf=6.53 Nps=92171(53% qui)
>>>        Eval=11501487
>>
>>What is your move ordering?
>>
>>Mine is:
>>
>>   hash move or PV move
>>   all captures
>>   history heuristic scores
>>   centrality terms
>>
>>There is considerable overlap in the last three above.
>>
>>I have experimented with forcing SEE<0 of all captures down to the bottom
>>but it did not speed up due to cost of SEE, over MVV/LVA, for me.
>>
>>I have experimented with killer moves (no improvement) and increasing
>>the value of history heuristic over captures or vice versa to no avail.
>>
>>I have printed out the sorted move list at various ply and it appears
>>reasonable to me. Nothing outlandishly wrong or even overtly.
>>
>>I don't handle a lot of special cases like promotions due to perceived
>>rarity in the tree.
>>
>>Stuart
>
>I think that it is a mistake and having queen promotion in the qsearch should
>help you.
>
>Uri

My code has this as conditional compilation flag. It is normally disabled
since it did not help test result. I will retest to be sure since there
were some big fixes in the search and big jump in test result since last
test of this flag.

Stuart




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.