Author: Amir Ban
Date: 05:26:37 04/22/98
Go up one level in this thread
On April 22, 1998 at 01:56:49, blass uri wrote: >for big depths Junior's evaluation at depth d should be equal to >Junior's evaluation at depth d-1 after the best move it found. >This can be done recursively for example for d>=13. > >This idea will help Junior to follow its main line. >did you try it amir? > I think you are saying that if you have an evaluation and main line for a certain depth, then if you play the best move and evaluate to one ply less, you should get the same evaluation and main line. I don't know what "big depth" has to do with it. Anyway you say that if this doesn't hold, then making it true would be an improvement. As an ideal, it should be true. But first, d-1 does not necessarily mean one ply less in Junior, so maybe that explains everything. Second, there are several valid reasons why it is not (always) true for Junior. Your statement that getting rid of this would be an improvment I'm not sure I agree with. Maybe you are taking a purist view and saying that this is what you expect from a program, so fixing it would be getting rid of a bug, but from the point of view of playing strong chess and winning games, it's not obvious to me that this is bad behavior. I agree that it is useful to have a search algorithm that is conceptually simple, with predictable results, so that its behavior can be easily followed, and bugs are immediately obvious. This is very important for a developer who is trying to understand and control a complex search algorithm. But life is tough, and the Junior search algorithm fits the opposite description :( As to following the main line, the opponent usually spoils it by playing something different. Amir >at small depths the main line can change often and it is more >important to find tactics than to follow the main line >but at big depths it is more important to follow the main line > >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.