Author: GuyHaworth
Date: 10:22:52 03/08/04
Go up one level in this thread
[ The 'DTZ' metric is so-called because a P-push and/or conversion is not only an irreversible move but also zeroes the move-count re the FIDE 50-move rule. However, in other game-domains, the concept of an irreversible move persists while there is no move-count to zero. These games include checkers, (?) International Draughts, and variants of the standard 08x08 chess. So I'm going to use 'DTI' for 'Depth To Irreversible move' rather than 'DTZ' in the future. ] The reason why it is easier, at the moment, to compute DTC EGTs rather than DTZ EGTs is this: 1) In a 'DTC pass' of an endgame, all positions which are given a depth are given the same depth. In noting that they are to be given a depth in that pass, it is therefore only necessary to set a bit in a bit-vector. In a later activity, the relevant index-positions in the database are given the relevant depth. 2) In a 'DTZ pass' of an endgame, the current practice is to survey the whole endgame rather than 'P-slices' of it which each have a fixed set of P-positions. It follows that in a single 'DTZ pass', different positions can be given different depths. One therefore needs more than a bit-vector in RAM to remember how to incrementally update the EGT itself. I think this ups the RAM-requirement by a factor of 4, or 8 if 16-bit index-entries are used to accomodate depths greater than 126 (as in several 6-man endgames). The remedy to this - and one which Dieter Buerssner has implemented I think - is to slice up an endgame into 'P slices' and create sub-EGTs for these, in the right order of course. This requires a canonical numbering of the 'Pawn position profiles' which takes into account the a-h symmetry. If this is done, the creation of DTZ EGTs requires somewhat more 'passes' than the DTC EGT creation, but this is not a major consideration. A further benefit is that, if P-ful endgame EGTs are stored by 'P-slice', clustering of positions is better, and one might consider bringing down (at runtime during a game) only the relevant subset of the EGT as the Ps will be somewhat advanced. g
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.