Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Too hard for my engine, how about yours?

Author: Sune Fischer

Date: 16:03:21 11/04/02

Go up one level in this thread


On November 04, 2002 at 18:46:50, Gerd Isenberg wrote:

>On November 04, 2002 at 18:11:59, Sune Fischer wrote:
>
>>On November 04, 2002 at 16:55:10, Uri Blass wrote:
>>>
>>>Depth 11 was finished few minutes ago and the score is very convincing.
>>>It seems that depth 11 is good not only to find the fail low but also to find
>>>the fail high for Be6
>>>
>>>11 67 161361 228198851 c8e6
>>>11 319 201885 285759097 c8e6 f4c1 c3c2 b1b2 e6c4 b2a3 c7c5 a3b2 c2d1n g1d1 b4d3
>>>d1d3 c4d3 d4c5 d6c5 c1c5 d8b8 b2c3 d3e4
>>>
>>>depth=11 +3.19 c8e6 f4c1 c3c2 b1b2 e6c4 b2a3 c7c5 a3b2 c2d1n g1d1 b4d3 d1d3 c4d3
>>>d4c5 d6c5 c1c5 d8b8 b2c3 d3e4
>>>Nodes: 285759097 NPS: 141545
>>>Time: 00:33:38.85
>>>
>>>Uri
>>
>>Congrats, I knew Movei could do it though :)
>>Your PV is 19 plies, so a bit extended.
>>Looks like there is a knight promotion, I wonder how often these underpromotions
>>aren't fail lows, cause I'm thinking of removing them completely, better branch
>>factor and all you know...
>>Of couse I would keep them at the root, so as not to have a bug, what do you
>>think?
>>
>>-S.
>
>Carefull - at least the knight promotion should be considered. But if the
>queen-promotion is rejected by capture of the promoted queen, underpromotions
>make not much sense, because they may captured as well. If a queen promotion
>forces a stalemate-draw, also rook as well as bishop promotions should be
>considered.

Well yes there is a risk, but how often does it happen?
I think searching underpromotions is basicly a waste of time, if the queen
promotion isn't good enough, then most likely the others can be discarded as
well. Bishop promotions must be very rare, I don't recall ever having seen a PV
with a bishop promotion, 99 of 100 it's a queen.

And as long as it's deep enough in the search the noise will be very high
anyway. I mean, I do not even search my losing captures in qsearch (I sort with
SEE) so in that context I consider the search of underpromotions overkill/waste
of time (simply cull them like you cull losing captures!).

What happens to the branch factor if you don't?
Imagine you have a position with 2-3 pawns promoting, in some cases you might
have two pawns that both have three possible squares, that's 6 moves for those
two pawns, and 6 moves times 4 promotion types, that's 24 moves! Quite a BF
explotion :o)

-S.




>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.