Author: Uri Blass
Date: 00:49:15 01/04/04
Go up one level in this thread
On January 03, 2004 at 23:52:06, James T. Walker wrote: >On January 01, 2004 at 14:28:57, Robert Hyatt wrote: > >>On January 01, 2004 at 09:39:14, James T. Walker wrote: >> >>>On January 01, 2004 at 09:35:59, Uri Blass wrote: >>> >>>>On January 01, 2004 at 08:29:21, Martin Andersen wrote: >>>> >>>>>On January 01, 2004 at 08:07:49, Uri Blass wrote: >>>>> >>>>> >>>>>> >>>>>>Most of my tests are done without pondering. >>>>>> >>>>>>I have only a single processor to test so I do not trust matches with ponder on >>>>>>when programs do not ponder 100% of the time. >>>>>> >>>>>>Uri >>>>> >>>>>If you have ponder on the programs will use approx. 50% of CPU time each, right >>>>>? >>>> >>>>This assumption is wrong and program may ponder only in part of the moves when >>>>in part of the moves(for example when they have no pv) they may not ponder. >>>> >>>>Uri >>> >>>Why is this different from "normal" operation when one program is running on one >>>computer? >>>Jim >> >> >>Think about it a minute. This is on one computer, so each program gets 1/2 >>the time. Using two computers, when I don't have a move to ponder, you don't >>get part of _my_ machine to help you out... On a single computer, when one does >>not ponder, the other gets the entire machine. For that move, the computer is >>2x as fast as normal. >> >>note that this never happens with Crafty, since it _always_ finds a move to >>ponder, even if it has to do a search for the opponent to find such a move. >> >>Of course, in endgame table positions, Crafty might not need to ponder, which >>would let the opponent have all the CPU if it doesn't use tables. > >What you say makes sense but pondering is an integral part of the program. The >program that is more accurate at pondering the correct move may obtain a slight >advantage (deservedly so). If you ponder the wrong move you lose nothing. You >can only gain if you ponder the correct move. So programs should ponder >something even if it's stupid. I think that there are more important things then telling a program not to ponder on capture when they expect a capture that leads to known tablebases position. The opponent may decide not to capture but the gain from this possibility is too small for me even to consider implementing it in the near future(the main problem of chess program is searching too many stupid lines and I think that better pruning is clearly more important. 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.