Author: Robert Hyatt
Date: 15:13:10 06/23/99
Go up one level in this thread
On June 23, 1999 at 14:13:54, Alex Boby wrote: >On June 23, 1999 at 13:47:06, Robert Hyatt wrote: > >>On June 23, 1999 at 09:35:09, William Bryant wrote: >> >>>As I understand the use of Singular extensions, >>> when, after generating all the _legal_ moves, there is only one move to be >>> made by the side on move, extend 1 ply. >>> >> >> >> >>this isn't a singular extensions... this is a special case of singular >>extensions in general, and is most frequently called the "one reply to check >>extension." >> >>>Since you only extend a maximum of 1 ply at any node, if the King is in check, >>> or any other extensions are also triggered you would be extending anyway >>> and this only extends this branch of the tree when it is the only >>> condition extending the search. >> >> >>Depends. In Crafty, I extend 1 ply when I check the opponent. Then, if he has >>only one legal move, he extends 3/4 ply for that reason... still keeping the >>rule of "no more than one ply of extensions for every ply of search done." >> > > What exactly is the meaning of extending by 3/4 of a ply? Isn't 'depth' an >integer argument to the search? > >Alex Boby Suppose you count a ply as "100". to do a 2 ply search, you search to depth=200, but for each ply you subtract 100. So no difference, yet. Now suppose you are searching to 10 plies. This turns into a depth=1000 with this scheme, right? And suppose you extend 3/4 of a ply. Depth would be 1075, and you are right, that does nothing at all since for each ply we subtract 100, and when we get to a depth<100, we call that time for the q-search. But suppose _two_ moves in the same path extend by 3/4 ply. Now each added 75, and we end up with a depth=1150. And 1150 is one more ply deeper than 1000 or 1075, correct? IE the idea is that it takes two 3/4 ply extensions before you really extend anything. And when you think of it, what this means is that the first 3/4 ply extension is ineffective, but the next three of them extend the search by one ply. Then the 5th does nothing, but the next three extend by one ply: 1000 1075 [no change] 1150 [1 extra] 1225 [2 extra] 1300 [ 3 extra] 1375 [no change again] and so forth... This isn't new at all... We were doing this in the 1970's...
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.