Author: Marc Bourzutschky
Date: 08:08:30 01/16/06
Go up one level in this thread
On January 15, 2006 at 18:03:12, GuyHaworth wrote: > >Re the various metrics, the best initial position to take is a 'metric-neutral' >one: understand that > > - different metrics provide different information > - no metric has yet delivered 'obviously sensible' moves all the time > - different metrics take different times to compute > - different metrics lead to different-sized EGTs > - no strategy based on metrics alone optimises theoretical value > one has to finesse re the opponent's fallibility with a k-move rule > >However, in attempting to produce better and better strategies based on >metric-based information from EGTs, [q.v. a recent paper by Bourzutschky, >Tamplin and Haworth in the J of Theoretical Computer Science], I find that I >always need the DTZ information. This is because, whatever else happens, one >must not give the opponent the opportunity to claim a k-move draw in the current >phase of play. So one can only choose from the moves that do not allow this, >provided there are some. > >It happens that the DTZ information gives the most compact EGTs and is quickest >to compute too. That is why it is the metric of choice for Yakov Konoval and >Mark Bourzutschky. > >g Another benefit of DTZ is that all moves within the solution are reversible, making a clear division of the game into distinct phases. This makes counting of records much more unique. For example, for a DTM endgame with a maximum depth of N, it is trivial to "extend" the record by prepending a series of captures or pawn promotions. In fact, a certain individual specializes in creating such meaningless records. -Marc
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.