Author: Uri Blass
Date: 13:14:19 12/11/02
Go up one level in this thread
On December 11, 2002 at 14:04:00, Johan Hutting wrote: >On December 11, 2002 at 10:46:08, Sune Fischer wrote: > >>On December 11, 2002 at 10:00:38, Johan Hutting wrote: >> >>>On December 11, 2002 at 07:44:43, Sune Fischer wrote: >>> >>>>Chess is primarily tactics, >>> >>>Vincent does have a point, to get into a position where you can use those >>>tactics you require positional knowledge. Any 1600+ rated player can tell you >>>that :) >> >>Vincent was arguing a different point. I agree with him that it would be hard to >>write a top level program with only 2000 lines of code, but I don't believe it >>is impossible. > >Not impossible, but it would require perfect tuning. Problem is the computers >speeding up every year would make the engines with more knowledge catch up in >search depth fast. I think that better search means that the gap in search is going to become bigger at long time control and if you are 2 times faster in 5 minutes per game you may be 10 times faster at 120 minutes per game. better evaluation can help you to do better at longer time control but the same for better search. >> >>I'm pretty sure there is still room for improvement in the area of search. > >hopefully there is :) > >>And probably evaluation is secondary to search in 95% of the cases. > >I don't agree. 95% is a _lot_. See Gerbil for an example of stable search, few >eval. Movei of today(not public movei) has better evaluation and better search than gerbil except using hash tables.(in the last match against Gerbil it beated it 32-5 with 3 draws in 5 minutes per game (movei always changes it's mind and used no book). It is possible that movei was slightly lucky but still I expect movei to score more than 75% against gerbil in every time control in these conditions(no book and always change the first move) except very fast bullet of few seconds per game when movei can scores only 60-70%. The search of Gerbil is probably more stable than the search of movei but the quality of search is not measured by it's stability. Gerbil is not a good example because I do not consider it as a program with superior search. 2000 lines of codes are clearly enough to evaluate mobility,king safety,pawn structure and some other things in a productive way(mobility is already evaluated in a clearly productive way by movei but the evaluation can be improved). The rest of the evaluation is usually for special cases that do not happen in most of the games or for things that may be discovered by searching few plies deeper(it does not mean always tactics, and searching some plies deeper may be enough to translate advanatge that the program does not understand to advantage that the program understands). 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.