Author: James T. Walker
Date: 08:29:15 02/19/03
Go up one level in this thread
On February 19, 2003 at 07:56:41, Uri Blass wrote: >On February 19, 2003 at 00:50:04, James T. Walker wrote: > >>On February 18, 2003 at 22:57:47, Charles Worthington wrote: >> >>>I am cosidering making Crafty my primary engine for 3+2 and under games on the >>>chessbase server. Does anyone have blitz stats (ie. w-l-d) for crafty vs various >>>other engines? >> >>In my opinion Crafty is not a good blitzer. > >I have a different opinion > >Crafty has 6 versions in the top 14 free programs in blitz > >see http://www.digichess.gr/infiniteloop/ratings/IL3b_rate.txt Yes but look at the version numbers. The later numbers are lower on the list. Also I'm curious as to how the games were played. Auto232? One computer with ponder off? I'm completely unimpressed with one computer ponder off results. They ignore an important part of chess engines. That is guessing the opponents move and using the time to compute. There are other questions about single computer testing which need to be answered. I believe auto232 is a realistic test of programs in their "real world" element. > >I do not see Crafty as a better tournament player. > >Worst Crafty is better than best Comet in blitz based on the blitz list when in >tournament time control things are different and there are tournaments when >Comet can get better result than Crafty inspite of the fact that a lot of games >are played What is a lot of games? If programs are close, several hundred will not tell which is best. The opponents played by each is important also. >see http://wbec-ridderkerk.nl/html/his2ndedition.html > But 76 games is nothing but a small test. In my blitz database after severla hundred games Chess Tiger 15 was about 15 points below Chess Tiger 14. After more than 1000 games it is now slightly ahead of Chess Tiger 14. I'm not saying I'm right and you are wrong. I'm saying that my opinion is based on my experience only. I'm aware of the conditions of my test and the accuracy of my database. Others I'm not sure about. I know there are many problems with auto232 matches such as both machines scoreing a game a draw when one actually one. There are problems with programs that lose blit games on time and are never penalized for it. There are games when both programs call the game a draw when it was not agreed upon but because one machine missed the others move the "master" timed out and since the score was close to zero they just call it a draw/loss/win arbitrarily. I'm not saying my database is perfect either since I'm aware that humans make mistakes. > > It appears to me that Crafty has >>become better at longer time controls and worse at short time controls in the >>last few years. The only info I have on Crafty is for Crafty 18.13 which scored >>101 out of 278 games at G/5min vs various top programs in my blitz database. I >>have Crafty 19.01 scoreing only 37.5 out of 100 games vs Deep Sjeng Beta on 2 >>XP2400+ machines using auto232 (G/5min). I don't own a dual. >>Jim > >I think that the reason for your result is simply the fact that Crafty is weaker >than the top programs and not that it is better at tournament time control. > >I have no proof that Crafty is relatively better than other programs at >tournament time control(comparing with Comet suggest the opposite). > >Uri Again Uri, I'm not saying I'm right and you are wrong. Only that we have different opinions based on different experiences. I do play mostly top name commercial programs. I do not play much with low level amateur programs. I also belive that when one program is a lot better than the other it exagerates the difference in a head to head match. Therefore relative ratings should be based on many opponents and not comparison to only one opponent. I've rambled on too long just to agree that we disagree. Regards, Jim
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.