Author: Harald Faber
Date: 05:16:13 05/25/99
Go up one level in this thread
On May 25, 1999 at 08:00:44, blass uri wrote: >>>The evaluation function should be changed and my Junior5.4 has a different >>>evaluation function. >> >>Do you believe that changing and not adding would make an improvement? > >I am sure about it. >The evaluation of Junior is based on some numbers in the program and changing >these numbers can improve Junior. It simply reminds me on tuning Chessmaster... >>>Junior clears the hash tables after every move so it cannot learn from previous >>>search and it is not because of the fact that it is a fast searcher. >> >>Whoops. Any reason for that? > >I do not know. >Amir Ban said in the past that it is not important and cannot help much but I >think that he was wrong. >I told him ideas to carry information from search to the next search and I think >that it can help significantly in tactical positions. That is my opinion too. >>Fast searching=little knowledge > >1)I think that a little knowledge that is correct is better than a lot of >knowledge that is not correct. I don't expect ANY of the current programs containing incorrect knowledge. >2) I am not sure if fast searching =little knowledge >It can be also some tricks to do the same things faster. Sure. Think of the (alleged?) 40% speed gain Fritz4->Fritz5. But I am sure that a program doing 20-30knps has MUCH more knowledge inside than a program at 200knps. >>Where do you see improvements? By fine-tuning that little knowledge? > >Yes and I am not sure if it is really a little knowledge. Depends on the definition of LITTLE of course. >I cannot compare it with other programs because I do not know what is the >knowledge of other programs. I see a problem in "counting" knowledge. >I also think that the fact that you need less time to represent the knowledge of >Junior (assuming that it is the case and Junior is not faster because of program >tricks) does not imply that it is a little knowledge. You are counting beans. ;-) Do you agree with "much less knowledge than Hiarcs, MCP, King and CSTal? ;-) >It is possible that Amir Ban is using knowledge that other programmers did not >think about and this knowledge is more important. Clear, think of Rebel-Tiger. >It is possible that by fine-tuning the "little knowledge" of Junior you can get >better program then by fine-tuning the Knowledge of Hiarcs7 even if both >programs look at the same number of nodes. But they don't. And personally I prefer a program that knows much about chess, e.g. about the wrong bishop. And I was VERY surprised to find that REBEL doesn't know! One of 2 things that irritate me about Rebel, the other one is king safety. I admit I find Junior's eval concerning king safety more accurate. >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.