Author: Robert Hyatt
Date: 16:49:35 06/01/98
Go up one level in this thread
On June 01, 1998 at 09:58:07, Komputer Korner wrote: >On June 01, 1998 at 08:56:58, Robert Hyatt wrote: > >>On June 01, 1998 at 03:07:21, Komputer Korner wrote: >> >>>On June 01, 1998 at 02:42:40, Komputer Korner wrote: >>> >>>>I have done about 80 games of Crafty 15 vs Comet A90 >>>> and Comet is stronger at 10 2 on a Pentium 166Mhz. If I had a faster >>>>computer then Crafty would do better, and I am still having problems >>>>with uninterrupted matches because if Crafty resigns or if there is a >>>>perpetual the match will hang. Also Sometimes Comet will not start a >>>>match. It refuses to make the first move of a match sometimes. I have to >>>>click the match icon shortcut again. I have now included the beep >>>>initstring to see if this makes a difference with the match hangs. It >>>>hasn't solved the occaisonal Comet refusal to start a match however. >>>>-- >>>>Komputer Korner >>>> >>>>The following was done by Tracy Miller and posted in the r.g.c.c. >>>>"I have now completed five 5 0 matches with various programs against >>>>Comet using Tim Mann's wonderful new Winboard 3.6.9beta. Each match was >>>>30 games. Here are the results: >>>>EXChess 4.0 Comet 26.0 >>>>GnuChess 7.5 Comet 22.5 >>>>Tristram 1.5 Comet 28.5 >>>>Arasan 1.5 Comet 28.5 >>>>Crazy Bishop 21.5 Comet 8.5 >>>> >>>>I finally found a program that seems to be better than Comet at blitz. >>>>Check out Crazy Bishop's wonderful performance! I tried hard to make >>>>conditions equal for these programs. I made sure the hash tables were >>>>equivalent, wherever possible. The only thing I couldn't figure out how >>>>to do was tell which programs were pondering (thinking on opponents' >>>>time). " >>> >>>My Comet- Crafty matches are still crashing with pondering on. No one >>>knows why. >>>I will go back to taking the pondering off. P.S. I just went back to >>>pondering off. it sure speeds up the opening play. No crashes yet but >>>too early to tell. Another question to Bob Hyatt? will Crafty learn >>>actual new moves in it's opening book? Will it add new moves based on >>>previous good results? If so are these limited to the 65,000 position >>>learning limit that you implemented and if so why a limit at all? >> >>it doesn't "add" to its book. I don't trust such, because adding a >>blitz >>line can be a way to add a losing line. The "65536 limit" is on >>position >>learning, which has *nothing* to do with the book. These are simply >>positions >>where the eval dropped significantly, and prevent someone from taking >>crafty >>out of book on move 1 or 2 or so, where it can't learn anything about >>the >>opening, yet it still won't repeat losing moves over and over. >> >>Also, you reported that comet was doing better than Crafty at blitz >>games. >>Could you send me some of the log files? IE I'd like to check to be >>sure >>that it is losing games that are getting recorded as losses, that CPU >>usage >>is balanced, etc. On ICC, we have several comets and crafty's running, >>and >>I haven't seen any overwhelming losses yet. Comet plays well, but seems >>to >>get into dire difficulties in the endgame. IE on ICC last night we had >>a >>5 game match at 3 0, crafty won 4 drew 1. The day before crafty won 2, >>lost >>1. So really bad results might say something is wrong in your >>configuration, >>as in comet is pondering while crafty does, or whatever, or the games >>are >>getting recorded as a wrong result due to a bug in either program >>telling >>winboard the wrong game outcome... > >I am doing a 40 game match at home at time control of 10 2 and so far >no crashes overnight. The pondering is now off for both sides. When I >come home from work, I will send you everything including the result. It >looks like the crashes were because of pondering. Remember this is on a >Pentium 166. >-- >Komputer Korner version 15.11 will be out later today. It fixes one bug in learning that could crash things badly... if the last move played in the game (at the point where learning occurs) is legal for *both* sides, things could get out of whack and it would break. This afternoon, after spending the morning debugging this, I played 335 games with the time control set at 999 moves in 1 minute, which made each game take about 5 seconds *total* time. Zero crashes. I used two processors for each crafty, pondering enabled, learning enabled, tablebases enabled, etc. zero problems... and *zero* flagged games...
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.