Author: Sune Fischer
Date: 04:04:01 05/28/02
Go up one level in this thread
On May 28, 2002 at 06:47:48, Uri Blass wrote: >On May 28, 2002 at 04:05:47, Sune Fischer wrote: > >>On May 27, 2002 at 23:09:49, Robert Hyatt wrote: >> >>>I put the 1/2 ply limit in to _prevent_ unlimited depth searches. That was >>>my point. My limit seems to control tree explosion pretty well. But I _still_ >>>can search beyond depth=64 if I don't put a limit in the recursive search call >>>as well, because I can search beyond 20 plies in real positions, which would >>>take me well beyond 60 plies with lots of extensions. And it only takes _one_ >>>such line to blow the subscript bounds and cause bad things to happen. >> >>I increased the limit from 60 to 200 at the cost of an additional ~150 kB. >>But that's not important since its just a global array that will never be used. >>I believe it is then safe to stop the search at 150 plies, that's 50 plies left >>for extensions(!), it will simply have to do, even if I get a crash in 1 of >>10000000 games ;) >> >>-S. > >I have a limit of 100 plies and also a limit of 32 to the nominal depth. >If the program gets depth 32 it simply plays. > >It never happened to me that the limit of 100 plies was not enough because of >extensions. > >In theory it may happen but I do not know of a single practical position when it >happens and I do not know if it is possible to compose a position when it >happens because if the position is so simple that the program can search 32 >plies there are not going to be a lot of extensions. I think of a KR-KR endgame, there could be a lot of checks extending, but I really don't go _very_ deep there, also with tablebases I wouldn't ever search such a position. So to be realistic it has to be a position not found in tablebases, and you need to go _very_ deep which probably means blocked pawns positions, and they won't give a whole lot of extensions in my program, not enough to exceed the safty buffer anyway. I can't prove it, but I'm sure it's perfectly safe. Actually I plan to put a limit on the extensions, so that it will never go more than 2x as deep as the prescribed search depth. Seems overkill to go that much deeper in just one line anyway. >I extend often slightly more than one ply for checks and in theory there may be >a sequence of a lot of checks when every side reply a check by check so I cannot >be sure that extensions for the 100 plies cannot happen in a small depth but >practically it seems that it never happened in games because my program never >crushed in games including some games at long time control. > >Uri
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.