Author: Ian Osgood
Date: 12:14:05 04/01/99
Go up one level in this thread
On April 01, 1999 at 12:25:15, Stuart Cracraft wrote: > >Maybe it's just my imagination but having been >working on a program that tests on FICS, I've had >a devil of a time getting the rating up signficantly >over 2000 on the FICS server. Usually it will get >up to 2010-2040 and then evoke some interesrt in >some stronger types and get pounded down to 1970-2000 >before it creeps up enough to get the stronger ones >interested again (re-pounding, etc.) > >What's special about 2000 on FICS? Maybe just a typical >barrier for a program? What is the handle of your program? Typically, computers modify their formula to roughly select their opposition. Specifically, a) assesswin > k to avoid dropping points to low rated sandbaggers and in general play higher rated players. b) !computer to avoid getting crushed by 400MHz crafties c) inc > k if your computer behaves badly at small search depths In addition, many computers will refuse playing more than n games in a row with the same person to avoid repeated losses. When I researched formulas a while ago, I found that programs with !computer and assesswin>0 were on average 100 points higher for the same processor speed. >If there is (or isn't) anything special about it, >then I'm wondering what I could do short of >increasing my hardware. It's an old Pentium 133mhz >but I don't want to spend the money to upgrade it >and would rather improve the program. What did you expect? For comparison, Iodine is gnuchess on a P100, with assesswin > 1, and it maxes out at 2156. PoorGnu on a P120 with assesswin > 0 maxed out at 2242 but is currently under 2000. Note that neither have !computer. An account I run is SapphireII. It is a high quality program, but runs at only 32 MHz. It is routinely crushed by everything running on P166 or better except for a few rank amateur programs. (This is actually pretty good! Kittinger wrote some mean assembly.) Search depth is everything in comp-comp games. Faster processors prevail in the long run. Get an upgrade. K6's are good, cheap chess processors. (IMHO, G3's are the best chess processors.) >The usual: PVS, null-move, hashing. Evaluation consists >of about 50 terms. Evaluation is done at the leafs >rather than pc/sq style for the most part. If you >need further details, feel free to ask. Code available. > >--Stuart
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.