Author: Robert Hyatt
Date: 09:10:12 01/27/05
Go up one level in this thread
On January 27, 2005 at 05:44:21, Tord Romstad wrote: >I have played around with various variants of singular extensions from time >to time, but never had much success, even in tactical test positions. The >problem seems to be that the general search depth is reduced too much because >of the many stupid variations which get extended. When I look at my log files, >it seems that most of the singular extensions made are for recaptures after >obviously losing captures, like the line 1. e4 e5 2. Qh5 Nc6 3. Qxh7 Rxh7. >The move Rxh7 is singular, but making an extension in a line like this >does not seem like a good idea. > >Do those of you who successfully use singular extensions employ some technique >to avoid triggering the extension in the type of situation described above? > >Tord Hsu addressed this, you have to avoid calling a recapture "singular" for that reason. There are other similar cases. Also in the Cray Blitz implementation, I found that until I copied Hsu's "sticky transposition table" idea, the search was unstable, in that on one iteration a move would be singular, on the next iteration it would not be. That made the score bounce around. The sticky TTable flagged SE moves so that they would either remain singular forever, or sometimes would remain singular for the next iteration even if the SE test would fail there. In the implementations I have played with, I did my normal extension stuff first, and if I already planned on extending (say a recapture extension, or getting out of check) then I didn't worry about the SE tests at all... Clearly if I try QxN, then I don't want pxQ to be singular just because it is the only legitimate way to recapture. I don't even trigger the recapture extension for that case, so why would I bother with a singular extension? There is probably a lot more of that sort of idea that I have not tried, but I will at some point in the future when I readdress this for the N+1th time. :)
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.