Author: Robert Hyatt
Date: 19:49:26 01/14/99
Go up one level in this thread
On January 14, 1999 at 13:51:26, John Coffey wrote: >On January 13, 1999 at 18:07:52, KarinsDad wrote: > >>On January 13, 1999 at 17:48:18, John Coffey wrote: >> >>> >>>The trouble with control of squares is that you have to take into account >>>the values of the pieces that attack a square. i.e. a pawn could control >>>a square. I had figured out that I could use bits for each type of piece >>>that attacked a square, where pawns use high bits and kings use low bits. >>>Add the bits for each side and the side with the highest number (more or less) >>>controls the square. >> >>Thanks John, >> >>I was toying with doing something in the back of my head about control of >>squares, but it was low priority. I had gotten as far as determining that the >>side with the least valuable resources controlling the square was in fact >>actually controlling it, but didn't have the time to really think about it. >> >>This saves me a lot of thought. >> >>KarinsDad :) > >I believe that determining who controls what squares - especially if weights >are given for controlling squares around the enemy king, could be a useful >addition to the evalution function. > >Now this issue for me is how to determine which pieces attack a particular >square. There are two choices.... > >1. Derive this from bitboards. Even with the advantages of bitboards this > seems to me to involve a great many steps. > >or > >2. My plan all along has been to update the information each time a piece >moves only on the squares affected. My evaluation of control of squares >would be incrementally evaluated as the tree is traversed. This could be slow, >but it has some advantages since a by-product of this method is that I would >have a current move list and information about captures etc, mobility etc. This >is likely to help with move ordering. > >I am posting this because I would like to get the opinion of others on this >method, especially Dr. Hyatt. > >john coffey Probably several others that can respond just as well as I can, but in any case, the king-side attack idea is a good one... something that Deep Blue does, for example. But the issue here is that it is not cheap, whether you compute it on- the-fly as I would, or incrementally as I tried a couple of years ago. The issue always turns into "am I better with this than without" and my conclusion (so far) is without. Because you can do other things to compensate to an extent. IE I don't get attacked fiercely very often. On occasion, yes, but in general, no. I watched about 8 blitz games against GM Dlugy this afternoon, and in 5 5 blitz games, Crafty did 10 plies in every search I watched, or better. And 10 plies in blitz sees a whole lot of stuff and covers up for an inability to notice a bishop and queen on an open diagonal, for example. (BTW, the blitz match went 8-0 for silicon. :) But, my advice is _always_ to try things you haven't tried before. Or try things others said didn't work for them. Programs are a summation of a great number of different pieces of stuff. And another 'piece' will affect different programs in different ways.
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.