Author: Peter Fendrich
Date: 15:58:08 11/23/00
Go up one level in this thread
On November 23, 2000 at 15:26:44, Amir Ban wrote: >On November 23, 2000 at 12:33:12, Peter Fendrich wrote: > >>On November 22, 2000 at 12:10:18, Uri Blass wrote: >> >>>I found from my experience that Gandalf does not do preprocessing. >>> -- snip -- >>There are always problems to find clear and simple definitions. >>I can see a problem even with this definition. >> >>Suppose that the program stores a table saying that knights are generally doing >>worse in the corners than in the center of the board and this table is not >>changed during the game. This is not preprocessing by your definition. >> >>Suppose that the program changes that table once in the root when we reach the >>endgame. This is preprocessing by your definition. Let's say that you ignore >>this case because the table is changed only once. Then you have the case that >>all pieces have a similar table each that changes only once when endgame is >>reached. Is that preprocessing? If not, suppose that you change these tables >>whenever the game reaches different stages in the game (opening, middle game, >>endgame etc). >>Is this preprocessing? >> > >It is preprocessing, and it is not minor at all. It will mix up your program's >behavior in stage transitions. It will swap pieces to reach a stage, or avoid >doing that, all for the wrong reasons. And when it sleepwalks into the next >stage, the program will wake up with a surprising new eval. > >Amir This behaviour is possible to reach even without rootnode information. Like this: if (material this and that) "let the king walk to the centre" else "let the king hide in the corner" The problem is to seamlessly transit into different stages. I'm more afraid of that judgements in the rootnode is based on conditions not present later in the evaluation. The Rootnode information can't be used without a critical look. I think it is possible to let the rootnode gradually affect the evaluation. In the evaluation it is still possible to override it all, avoiding the preprocessed information. Then the evaluation is no preprocessed. If the information is used however, it is based on complex algorithm's not applicable to do in the end-node evaluation. //Peter
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.