Author: Robert Hyatt
Date: 08:44:01 01/10/04
Go up one level in this thread
On January 10, 2004 at 01:58:21, scott farrell wrote:
>One thing I realised when working on all the qnodes vs SEE vs even captures etc
>etc. Is that anything you put in the qsearch, is less likely to be horizoned
>obviously. It also has an effect of not having to appear in the full width
>search PV, and therefore leaves more room for other moves in the PV, allowing
>for better play (is the assumption anyway).
>
>Obviously getting to the next ply faster with less qnodes also gets a better PV
>and better play.
>
>Take the example of an even capture, SEE says even , so it is pruned in the
>qsearch ( fairly common, and is what happens in crafty).
Crafty doesn't prune "equal" captures. It prunes only _losing_ captures
(losing according to SEE). If the expected material gain/loss is zero,
the capture is included in the q-search. That is why the lines
tmp=Swap();
if (tmp >= 0) {
are used.
>
>If it is actually winning either material or position, then the main full width
>will have to find it , and use 2 slots in the PV, if you recapExtend it, it only
>takes one. So I think this has the effect of extending those PVs that dont have
>the capture in it, but horizining it to the qsearch.
>
>With ideas from Tord, Sune, Robert, Uri, I think I'll modify the SEE for so that
>it returns a complexity level more than a score, or it can return a score or say
>_invalid_ if its over complex. But this is a whole nother thread below. So I
>will probably make it different again for even captures.
>
>Anyway, I am after comments/ideas on just qsearching all even captures. Lets
>leave alone the strict performance consideration for the minute. I am more
>intersted in people's opinion on changes to the PV/strength/play, etc etc. Given
>you work out code that does real fast on these even captures in the qsearch, you
>can minimize the performance loss somewhat.
>
>Scott
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.