Author: Alessandro Damiani
Date: 13:58:03 09/26/99
Go up one level in this thread
On September 26, 1999 at 16:22:22, Robert Hyatt wrote: >On September 26, 1999 at 14:08:00, Alessandro Damiani wrote: > >> >>Yes, as it is always the case one has to know all properties of the objects one >>handles with. >> >>I detect passed pawns incrementally, with a few ANDs. This could only be >>possible because of BitBoards. >> >>Alessandro >> > > >I don't do anything incrementally at present... that is an efficiency >issue. I have ignored incremental eval because it locks you into a box, since >the code is mixed between Evaluate() and the Make()/UnMake() code. I found >that in Cray Blitz, (which did a lot of incremental work) I found myself stuck >in a box that served as a limiting factor. At present I am far more interested >in what is good and what is bad in the evaluation code. Once I am satisfied >with something, I will go back and look at it from an efficiency point of view. >Right now I am looking at it with a "design with change in mind" sort of >software engineering approach.. make it easy to add things whether they are >efficient or not. If they are good, then make them efficient. IE it is very >possible that a lot of my pawn stuff could be done incrementally and would maybe >faster. But once I go to that much work, I would likely tend to keep using that >code even if something better came long idea-wise. I'm not ready to commit to >most of what I do, until I am sure it works as I want.. You are perfectly right. I do incremental work to get information the leaf evaluation needs. For instance, passed pawns are detected incrementally but evaluated at the leaf nodes. Since passed pawns are "standard" knowledge I have done it this way. Alessandro
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.