Author: Robert Hyatt
Date: 10:45:56 08/25/04
Go up one level in this thread
On August 24, 2004 at 15:43:00, Scott Gasch wrote: >On August 24, 2004 at 11:46:28, Robert Hyatt wrote: > >>On August 24, 2004 at 11:34:31, Alessandro Scotti wrote: >> >>>On August 24, 2004 at 11:24:28, Anthony Cozzie wrote: >>> >>>>On August 24, 2004 at 10:40:15, Alessandro Scotti wrote: >>>> >>>>>Ok but then you don't propagate the mate score out of the quiesce, right? >>>> >>>>And why not? >>> >>>Well if I knew why I wouldn't be asking! (*) :-) >>> >>>For now my reason is: there is no proof that a score is mate out of quiesce, >>>because quiesce doesn't examine all possible moves but only part of them. After >>>reading several replies this doesn't seem to be a popular opinion, though, but >>>that's how I see things at present... >>> >>>(*) Hope English tenses are correct here! >> >> >>You miss the point. I reach a node X, I can play a move which lets my opponent >>mate me, or I can stand pat. The way around this is to notice that you are in >>check, and force the q-search to look at _all_ legal replies, and eliminate the >>"stand pat" option. >> >>Either way will produce a valid search result. The latter will find forced >>mates more accurately of course... > >To expand on what Bob said: also remember that if you have a position in the >qsearch where a side is in check BUT that side could have stood pat in the >previous qsearch positions then it does not really help to return a MATED score >now. You can probably find that exact description in the Crafty main.c from back when I had copied the Cray Blitz q-search. Just prior to the 1996 WMCCC event in Jakarta I removed the checks/check evasions in the q-search. Until then I did as you are explaining as otherwise backing up mate scores will fail where the side getting mated can "stand pat" earlier in the q-search. > You know that because they chose not to stand pat before that they are >"down". But because you have to assume that they had some other move option >than to continue trading / capturing when they chose not to stand pat, this >position is not forced and therefore not a true mate. > >In my qsearch I do something like this: > > Are we in check AND has the side on move never had the chance to stand pat > up to now? > > YES? Then do not allow a stand pat here and search all legal replies and > possibly return a MATED-IN-N score here. This position is forced. > > NO? Then allow a stand pat here and search a subset of the moves. Do not > return a MATED-IN-N score here even if you detect a mate because it is > not forced. > I didn't like that as it caused a few hash table mate oddities. IE your q-search finds a mate, but obviously not the shortest mate possible. That can lead to quirks in PV output and so forth if you carry hash scores across searches... >This makes sense to me but I'd welcome any comments ;) > >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.