Author: h.g.muller
Date: 10:57:39 03/04/06
Go up one level in this thread
On March 03, 2006 at 20:20:30, Walter Faxon wrote: >Building a TB on the fly is a neat trick but requiring so much time makes its >use problematic. Well, I would not say that. I wonder for which time controls you would not be able to spend 3-5 minutes once you get in this situation. You could even do it while pondering. Once you have the TB you can play the rest of the game 'a tempo', and will gain the invested time back in 10 moves or so... What is problematic is to spend those 3-5 minutes in normal search for a few moves, the first of which returning a blunder that will cost you the game! I don't think with current technology levels it is too moch to ask from a computer to recognize situations where retrograde search is far more effective than forward search. >Perhaps another good trick would be to find a way to index an >existing TB (or set of TBs) so that only a very small number of physical disk >read operations would load the relevant TB subset, at the cost of a fraction of >one second. Of course the TB would be optimized for this rather than to >minimize the total space used. > >-W This is trivial: simply use the Pawn positions as most-significant components of the index. Then all positions with the current Pawn structure are adjacent, and easily loaded: there are less than 4000 of those due to the Kings. Problem is that even to retrieve the Queen's wing structure from a TB after the King's wing is cleared (which will happen at small search depth) you still need a 9-men TB KppppKppp. If you want to have that TB in advance (i.e. before knowing that you had those particular Pawns on the Queen's wing), you should pre-empt every possible Pawn structure. This requires on the order of 64^2*48^7/(2*4!*3!) = 8350 GB (uncompressed). It seems infinitely easier to spend the 5 minutes...
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.