Author: blass uri
Date: 08:07:07 05/02/98
Go up one level in this thread
On May 02, 1998 at 10:00:39, Moritz Berger wrote: >On May 02, 1998 at 09:57:17, Moritz Berger wrote: > >>The whole search is different with bigger hash tables ... Sometimes >>Fritz simply cuts of an extension much earlier with smaller hash tables, >>but this might well be a case of "playing the right move for the wrong >>reasons", i.e. within the limitations of the evaluation function I >>expect Fritz to find the "better" move more reliably with bigger hash >>tables, even if this means it takes longer because a cut-off due to hash >>table collision is not effective with the smaller tables. > >There was a typo here: The concept means that > >small hash tables > >-> hash collision > >-> Fritz doesn't follow a line it normally would extend due to certain >criteria >(because it gets wrong scores at some point from the hash table) > >-> it "stumbles" upon the right move earlier because it had taken that >"irregular" shortcut before ... > >Moritz This is not the case of playing the right move for wrong reasons because with small hash tables fritz5 calculated the losing move at depth 15 for a long time before it decided it is a losing move (more than 2 pawns for black) and with big hash tables fritz5 prefered to believe its memory and not calculate the losing move at depth 15 for a long time. I used the tactical choice of engine in the position. Of course in the initial position it is white to move and not black(I posted black and it was a mistake). It can be understood from what I posted after the position.
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.