Author: blass uri
Date: 01:00:43 02/01/00
Go up one level in this thread
On February 01, 2000 at 03:14:32, Bertil Eklund wrote: >>1)The main difference is that his tournament is not tournament of long matches >>so he does not count cases when program win the same game again and again. > >Hi! > >What is this for nonsense, he plays 10-12 games with each program, Enrique plays >100-150 and we play 300-1000 games, and you don´t see the difference. Thorstens >tournaments is interesting but shows almost nothing and his comments shouldn´t >we talk about. He also use computers with 64 mb Ram vs SSDF 128mb and Enrique >256mb. You must be blind or a virgin if you can´t see the differences. Why do >you think his results of Fritz always is worse than CSTal. Of course you know it >but instead you mumbles of the major difference, etc. Hi! You are right that you play more game but when I said the main difference I wanted to explain the difference that thorsten meant. Fritz is not always worse than CStal in his tournaments. I remember one tournament when fritz was not worse than Cstal. The point is not about the question if Fritz is better than CStal but the fact that programs can take advantage by playing the same opening again and again and that they do it. I think that if you test program by not playing the same opponent again and again but playing the same games in a different order you will get a different rating. I do not think that the problem is the 64Mbytes of thorsten. Maybe fritz could score slightly better with 128 or 256 Mbytes but I know that enrique used the upgrade of Fritz6 in his last tournament and in the past Fritz was not the best program in enrique's games. Hiarcs7 won and Junior5 was second. The ssdf did not test Fritz6 and I believe that the reason is that chessbase knew before the upgrade that it is no improvement relative to Fritz5.32 >>Thorsten does not measure the learning of programs in matches. >> >>2)There is a simple way to increase the "playing strength of chess >>programs, measured in the way SSDF does". > >It´s not that easy. Typically we plays around 20 games with F3 P90 and F3 has a >learner, Genius5 and McP6 also have learners. The fact that fritz3 has a learner does not help much because it may change the move in a losing position. If you play enough games the learning will not help. because the book can considers also what to do after learning. I know nobody who does it but the problem with testing old programs when you count only results is that it is easy to do it. Uri
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.