Author: Bruce Moreland
Date: 14:12:16 09/11/01
Go up one level in this thread
On September 11, 2001 at 15:04:28, Robert Hyatt wrote: >On September 11, 2001 at 12:54:32, Bruce Moreland wrote: > >>On September 11, 2001 at 11:40:22, Rafael Andrist wrote: >> >>>On September 11, 2001 at 10:36:00, Robert Hyatt wrote: >>> >>>>On September 10, 2001 at 13:37:58, Rafael Andrist wrote: >>>> >>>>>On September 07, 2001 at 13:41:57, Robert Hyatt wrote: >>>>> >>>>>>I did this in Cray Blitz _many_ years ago (coordinated squares is the term I >>>>>>hear used most often). And I was amazed that it took longer to find the right >>>>>>move. After a mountain of debugging output, I discovered what I mentioned >>>>>>previously... "hash grafting" (the art of grafting parts of the tree from >>>>>>one zone to another by using the hash table) was helping the dumber version, >>>>>>but not the smarter one. >>>>>> >>>>> >>>>>You mix some things together! Knowledge about opposition is only one of the >>>>>tools you need to to construct a system of co-ordinated squares (german: >>>>>Gegenfeldsystem). If you implement this correctly, you should find the correct >>>>>move instantly i.e at ply 1 as my chess program Wilhelm does. >>>>> >>>>>Rafael B. Andrist >>>> >>>> >>>>You find the right move instantly... But you don't _know_ it is the right move >>>>until the score jumps. It _could_ just be a draw. And in the case of Cray >>>>Blitz, using coordinated squares, it took 25-26 plies to see the big score. >>>>It saw the right move normally at around ply=18 with the +2.5 score. With >>>>the coordinated squares stuff, it got the Kb1 move instantly, but the score >>>>didn't reach +2.5 until 7-8 plies longer than the simple version. >>>> >>>>That was the point. The better the move ordering, the less "grafting" helps >>>>a shallow search find a deep solution. >>> >>>I now by eval from ply 1 that i win a pawn. In Fine 70, White is already a pawn >>>up, so I get instantly an evaluation around 2. >>> >>>Rafael B. Andrist >> >>Nobody has written an article on how to use coordinate squares in a chess >>program. If you write this up, I bet the ICCAJ would publish it. >> >>bruce > > >What I did was pretty simple... and was very similar to what I do in pawn-only >endings in Crafty. IE detecting/maintaining opposition. Note that I didn't >keep this code in Cray Blitz. It was just something that Harry or someone >suggested that I try to see what I thought. I ran into problems in that in >some cases, staying on the right "square" is the grossly wrong plan, when you >do that rather than centralizing. I don't think that coordinate squares have anything to do with maintaining opposition. bruce
This page took 0.01 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.