Author: Robert Hyatt
Date: 20:26:59 08/15/01
Go up one level in this thread
On August 15, 2001 at 23:23:53, Scott Gasch wrote: >Recently I've been messing around with singular extension (s.ex). As I see it, >the problem with s.ex is that with alpha beta you do not have the precise value >of a move -- you only have an upper or lower bound on the score. I am doing >fail hard alpha beta and have been playing with this: > >If all moves in a position are terrible (less than alpha - delta1) except one... >and that one is within delta2 of alpha, search that one move over again one ply >deeper. I started doing this with delta1 = MATE-300 as I was hoping to find >positions where there is only one move that will save you from mate. I've been >gradually reducing delta1, though, and dumping the positions where it would >extend. I also thought about changing the second condition: only extending if >the one move was > alpha (that would have to mean it's a PV node because if it >was a FH it would have returned immediately and never get to this code). > >What is the typical way to implement s.ex? What do you think about the ideas >above? How do you do it? > >Thanks, >Scott A good starting point is the paper on SE in the ICCA journal. I would guess 1988-1992 (I am not at the office so I don't have the journals handy to look it up). It is pretty complicated to implement if you do the full implementation as described by Hsu/Campbell. There are lots of pitfalls to overcome...
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.