Author: Tord Romstad
Date: 06:32:46 08/09/04
Go up one level in this thread
On August 09, 2004 at 07:39:21, David B Weller wrote: >On August 08, 2004 at 04:35:54, Tord Romstad wrote: > >>3. Mate threats. These can be detected statically, or by using the return >>value of a null move search. Extend if the null move search returns >>-MATE_VALUE+Ply+2. > >Question/Comment: *Some* experiments have indicated better results from ><-MATE_VALUE+300 [ie, any mate score]. Doesnt this make sense? Because, after >all, mated is mated [an absolute score]. Then again, I suppose the distance to >mate IS somewhat indicative of the actuall threat ... comments... I don't know. I've never tried to extend for mates in more than one move. My intuition tells me that extending for deep mates is probably not a good idea. Defending against a mate in several moves is usually much easier than defending against a mate in one. Of course, my intuition is very often wrong regarding computer chess. Perhaps extending for deep mates is a good idea. The only way to find out is to try. >>4. Attacking moves. If you evaluate all internal nodes (as I do), compare >>the value of the king safety component of the evaluation function before >>and after a move is made. If a move dramatically reduces the opponent's >>king safety, you extend. >> >>In order to avoid search explosions, you may want to use fractional >>extensions. I extend by a full ply only for checks. >> >Tord, does this mean you extend less for the mate threat extension and the >1reply to check? Yes. Checks are always extended by a full ply (this could probably be improved). Mate threats and 1 reply to check are extended by something between 1/3 and 3/4 of a ply, depending on the position. >It seems backward to me IMHO [I say this intuitively, not >because of experimental data] It seems there are trillions of worthless checks >but far fewer 1reply and mate-threats ... yes/no/maybe ?? Most of the experimental data I have seen indicate that your intuition is incorrect. Check extensions are very important. If you remove them from a chess engine, the playing strength will usually drop noticably. 1reply and mate threats, on the other hand, tend to help a lot in test suites, but usually don't have a big effect on the playing strength in real games. Tord
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.