Author: Dave Gomboc
Date: 14:28:11 09/28/99
Go up one level in this thread
On September 28, 1999 at 16:26:53, Dann Corbit wrote: >On September 28, 1999 at 16:22:11, blass uri wrote: > >>On September 28, 1999 at 15:53:35, Dann Corbit wrote: >><snipped> >>>1. I support the SSDF >> >>I think the main problems that I see with the ssdf games are: >> >>1)Big part of the games are not public so it is impossible to check if there are >>mistakes in the games(I found a mistake in some games when Junior was slowed >>down by a significant factor because the tester used another application in the >>same time and the tester repeated the games but unfortunately it is impossible >>to check if there are mistakes in secret games. >This is *mostly* an old problem. They seem to publish new games here. Do you >still see this as an issue? I'm not sure what percentage of their games are published here, but I think it's pretty low. Ideally, 100% of the new games would be available at an ftp site, but this may be too much to ask from some volunteers. >>2)The "learning" problem. >> >>Some programs have learning function and some programs(mainly old programs) have >>not learning functions. >>I found that Fritz5(p200) won rebel8(p90) the same game 5 times. >> >>I am sure that Fritz5 could earn a lot of rating if it played 400 games against >>Rebel8 instead of 40 games. >> >>The rating of a program is dependent on the question how many games every >>program play against others and is dependent on the opponents. >This is an interesting artifact. I think (in general) it shows that programs >that learn are vastly superior to those that don't -- don't you? > >In any case, the SSDF results are valid under exactly the conditions stated: >Machines configured exactly as described and run using the Chessbase autoplayer. > >If anything changes -- hardware -- mode of play -- whatever... >Then we don't have nearly so good an idea of how well the programs would do. > >People complain about the autoplayer -- what are the alternatives? The SSDF has >tens of thousands of games. I think they have a hard enough time running the >games right now without making all the moves manually (which is *Far* more prone >to human error -- what if they take too long making the moves?) > >If another autoplayer were used would it be any better? I, for one, would prefer the use of an open-source autoplayer. Perhaps the major benefit would be that there would be less (I dare not hope for none :-) accusations of cheating flying around. Dave
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.