Author: Robert Hyatt
Date: 08:44:11 06/05/01
Go up one level in this thread
On June 05, 2001 at 05:40:22, Daniel Clausen wrote: >Hi > >On June 04, 2001 at 22:45:44, Robert Hyatt wrote: >> It is so close to perfectly ordered that it can be called "perfectly ordered" >> with no danger of being wrong enough that it can be measured. > >Let's hope that no mathematician ever quotes you on this. ;) > >Still it's amazing to me to reach 30 plies that fast. (or in PM's case even >faster) 30 plies w/ perfect alpha-beta should be 15 plies w/o alpha-beta, right? very loosely, yes. But _very_. The point with MTD(f) is that the search is carried out with a null-window (alpha,alpha+1 for the bounds). That means that _every_ branch searched either fails high or fails low. If all you have is material, from the initial position the score _must_ be zero or else we would already know the game is won or lost by white. That is going to be a _very_ efficient search. And from the opening position, there are so many transpositions, and _every_ one will essentially be a perfect score (because in MTD(f) there are _no_ "EXACT" hash entries) it is just horribly efficient. Give it an unbalanced tactical position and this won't be true of course... > >I don't know about your engine but a 15-ply search w/o alpha-beta from the >initial position is.. welp.. out of the question. :) Maybe null-move helps with >this a lot? or I am missing something obvious? Or both? :) What approx. depths >would a non-MTD(f) engine reach in the same experiment? I can reach (say) 15 plies in 24 hours in any position. From the opening position I could certainly go to 20 or a bit beyond. But my tree is far different in shape since I don't do the null-window search _everywhere_. This is why my PV search takes 10X as long as searching the rest of the ply-1 moves added to gether. MTD doesn't behave like that. The first move flies by too. > >Regards, > >Sargon
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.