Author: Uri Blass
Date: 13:58:39 07/16/04
Go up one level in this thread
On July 16, 2004 at 16:37:44, Steve Coladonato wrote: >On July 16, 2004 at 12:47:09, Uri Blass wrote: > >>On July 16, 2004 at 07:08:52, Steve Coladonato wrote: >> >>>I tend to agree with you and have posted previously on this subject. The book >>>is human not engine. You will see comments about the recent WCCC match and see >>>where so and so had an inferior position out of book. >>> >>>I have suggested in the past that to truly test the engine, the books should be >>>the same and even suggested a book depth of between 12 and 20 moves (24-40 ply). >>> In the past, before Shredder was Chessbase, I felt that Stefan Meyer-Kahlen's >>>algorithms or heuristics were a bit better than the competition (I haven't >>>followed Shredder very much since it was acquired). >>> >>>Today, I believe the improvements in the performance of the engines is mainly >>>attributed to the hardware not anything the programmers have done. >> >>I think that it is dependent on the programmer and you cannot generalize. >> >>There are programmers who do almost no progress in software and there are >>programmers who get significant progress. >> >>programs that did significant progress in the last year are mainly weaker than >>the commercial programs(Thinker,Fruit,Gothmog) but it is possible that at least >>one of them may become better than the commercial programs in the near future. >> >>I am also not sure about progress of Shredder and Junior since the last version >>and we will know better when the ssdf tests Shredder9 and Junior9. >> >>Uri > > >Uri, > >I'm not suggesting that the programmers do not improve their code. What I am >suggesting is that the evaluation of a position at a given ply depth for a given >program is basically unchanged and that the improvement is coming from being >able to see deeper. No it is wrong. Fruit1.0 did not evaluate mobility Fruit1.5 evaluates mobility so it is clear that the evaluation is changed. Given a restriction on the amount of time a program could >analyze a position, with older hardware it may have only been able to get to 12 >ply. Today, with newer hardware, it could probably get to 18 ply. And I have >seen with every engine so far, that what it evaluates at 12 ply, say +1.7 at 18 >ply this could easily be evaluted at -2.2. Consequently, the engines are >picking better lines because it sees deeper. I did not claim that better hardware is not significant but the question is if the improvement in hardware is bigger than the improvement in software and it is different for different engines. The example of 12 plies and 18 plies is not realistic because there are engines that did a jump of more than 100 elo in some months(see fruit) and hardware even did not get twice faster in the same months. > >My earlier interest in Shredder was two fold. One, it was amateur. Two, SMK >seemed to be doing something different in how he was evaluating positions. Movei Gothmog Fruit Crafty and other engines also do something different in how they evaluate positions so I do not understand your point. It's >kind of hard for me to explain but I think that the evaluation algorithms and >heuristics are basically the same (with a couple of exceptions) and the weight >assigment is the biggest difference. > >Steve No I do not take evaluation of another program and change weights and I am sure the same is for many programs. I learned from other programs but the idea for part of my evaluation is not taken from other programs. 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.