Author: blass uri
Date: 14:09:39 07/17/00
Go up one level in this thread
On July 17, 2000 at 16:38:42, Jeroen Noomen wrote: >Hi all, > >The Dutch championship and Dortmund have made one thing very clear: Computers >still have no idea what to do in blocked positions and how to prevent or defend >against a king's attack. > >IMO the situation is as follows, what I will call 'the big chess program >paradox': > >1. I am programmer and wondering what to do about these > horrible king side attacks and blocked positions. > >2. I try a lot of extra knowledge and EUREKA, it works! > >3. Unfortunately there is a side effect: The NPS drops from > 400 KNodes to 200 KNodes. If the difference is only 400 Knodes to 200 Knodes then I think that in the worst case the difference will be less than 70 elo ssdf rating. 70 elo ssdf rating is the difference from being twice slower and the fact that it has knowledge that other programs do not have should help. When the hardware get better diminishing returns from search will help you in comp-comp games. > >This 'little' side effect makes me start thinking: > >4. If I make this program commercially available, it will > do bad against other programs. Why? Simple: I had 400 KNodes, > now I have only 200, so I am outsearched. Not always. A program with better evaluation can detect a threat that a program with bad evaluation does not see and avoid pruning the right line. > >5. If it scores bad against other programs, I get a very low > rating in Sweden and I score (probably) bad in tactical > positions. You can build an opening book that leads to closed position for the ssdf list in order to score better against programs. > >6. After this, all people reading magazines, visiting CCC and > playing computer programs will say: 'This is a bad program. > It scores bad, is slow and thus it is weak'. > >Conclusion: I, the programmer, will NOT do this. It's commercially >not a good thing to do! > >And then I ask myself: Do I need more NPS versus humans? Is 400 >Knodes enough? Or do I fancy 4.000? What is the point against strong >GM's to make it stronger in an aspect of the game (tactics) where it >is already superior? > >Dear programmers, Christophe, Amir, Bob, Ed, Stefan and all others: >I am very interested in your opinion about this matter! > >Jeroen > > >PS You want to know why I started loving Rebel and I still love it? >Because it is far slower than the top programs, but KNOWS more about >positional chess! It is not slower than hiarcs or chessmaster in nodes per second. 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.