Author: stuart taylor
Date: 16:29:31 04/29/02
Go up one level in this thread
On April 29, 2002 at 16:09:55, Roy Eassa wrote: >On April 29, 2002 at 15:58:01, Chris Taylor wrote: > >>On April 29, 2002 at 15:37:41, Roy Eassa wrote: >> >>>On April 29, 2002 at 15:26:57, Chris Taylor wrote: >>> >>>>Both programs had 196Mb hash and own opening books. Both had access to all >>>>5.4.3 piece endgame bases. One hour games, 8 played.... >>>> >>>>J7 v H8 4-4 >>>> >>>>H8 +3 -3 =2 >>>> >>>>Only a small number of games yet again....... >>>> >>>>Gambit Tiger is playing at the moment >>>> >>>>Chris. >>> >>> >>>Maybe Hiarcs 8 will join the crowd of programs that are all very close to the >>>same strength (IMO) and all somewhat weaker than Fritz 7 but somewhat stronger >>>than Hiarcs 7.32. That crowd includes Junior 7, Chess Tiger 14, Gambit Tiger 2, >>>Shredder 6.02, and probably Deep Fritz 6. (Chessmaster 8's strength is still a >>>mystery to me.) >> >>I will be playing the same format, for more of the top engines. By this look at >>SSDF lists or Chessfuns list! I will still not have enough games to form an >>opinion as to the strenth of these engines. But I think it is fun! >> >>I want to do longer time settings, but this means less games, or leave the >>computers to play longer. >> >>How does 30 moves in 30 mins x 2, then 1 hour to finish the games sound? >>2 hours per game, sort of.......... >> >>Chris. > > >My general feeling: if enjoying the QUALITY of the games is the most important >factor, then obviously the slower the better -- but of course you must sacrifice >a lot of statistical significance of the results. If, on the other hand, you >care most about statistical significance of the results, you can play MORE games >at faster time controls and NOT dramatically affect the rankings among programs. > I'm not talking about 1 second a move here, of course, which may be too short >to be of much significance. But I seriously doubt that a many-game match played >between two modern programs on fast PCs at 20 seconds a move (average) will >result in a dramatically different result than the same-length match played at >an average of 200 seconds a move. (The accuracy of play will of course be >affected, but not the ranking IMO.) > >Such a change in average time per move WOULD dramatically affect the results of >a match between any strong program and a human GM, but not the results between >two programs. I used to post some great ideas I had about the most valuable, as well as economic testing, which I still am sure would be the most effective in economy (of number of games) as well as value (for being able to see many observations or even conclusions). But no one took up my suggestions nor praised them all that greatly. So I don't have the energy to work it out all over again etc. S.Taylor
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.