Author: Gian-Carlo Pascutto
Date: 05:15:13 04/23/02
Go up one level in this thread
On April 23, 2002 at 02:46:29, José Carlos wrote: > The original post was not about algorithmic deficiencies nor perfect > programs. I felt that it was a general implication. I agree they may exist, but they are 'edge' cases and not representative of normal behaviour. >I still believe even "perfect programs" (read top commercials) have this >behaviour, but I don't test them, so I don't have data. I believe this behaviour is minimal for top programs. >>ICC ratings are pretty meaningless. Too dependent on formula, programmer >>testing, and subject to large random fluctuations. > >Asking for numbers and then arguing they're meaningless doesn't make much >sense if you don't explain clearly why do you say they're meaningless. >Testing and formula don't usually make a difference among time controls. Formula and timecontrols are directly related. Most programs only allow a limited amount of timecontrols to be played. Diep only allows bullet against humans. Obviously, that will boost the rating up. Hossa used to allows computers to play, but now is human only. The rating went up by several hundred points in blitz, but less in standard.(*) Testing also makes a very large difference. There is a lot less standard played, so this rating is much less resistant to sudden strength changes. >Random fluctuacions exist, but they're not so big to make my examples >meaningless. I >showed a couple of programs whose bullet rating was 200 points higher than the >blitz rating, and another two whose bullet was 200 lower. That's not a random >fluctuation. 2 1 is bullet, and 3 0 is blitz. The games last about as long. 10 5 is also blitz. The division is completely arbitrary Random fluctuations on ICC are often > 200 points per day. -- GCP
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.