Author: Vincent Diepeveen
Date: 18:27:49 01/25/02
Go up one level in this thread
On January 25, 2002 at 03:46:36, Steve Timson wrote: > >>obviously that is evaluation function. NOT singular extensions. >>your evaluation is too optimistic about 3 pawns vs piece. > >While seeing the solution early is driven by 'faulty' eval, singular extensions >are definitely helping to see it much faster than the non-SE version. > >SE version 1st switches to the solution at ply 8 after 2.2 seconds. By the 10th >ply, and 19 seconds it has a +3.03 score. > >Non SE version 1st switches to the solution at ply 10 after 10 seconds. It >takes it until 15 ply and 167 seconds to see the +3.03 score and pv (that is >seen at 10 ply with SE). >So the shallow solution is seen in 1/5 the time with SE and the deeper solution >is seen in 1/8 the time. Not that either of them are solutions for the right >reason. > >So SE is helping chester to apply its faulty eval faster... :) Yes but SE also limits your program to 10 ply. how do you ever get to 15 ply with SE ? For now i have them turned on in DIEP too, but if there is just one reason to kick them out i already have uncommented them :) They reduce branching factor bigtime here. at 10-11 ply i lose like 0.5 ply to them, above that it even gets worse of course. Because with R=3 reduction and not doing them in the root this means 10 ply search depth ==> only at 6 ply they get used. Or at 6/10 of the search = 60% At 12 ply that's already 8/12 = 75% and then extension over extension it's horror. But they DO work for shots like this.
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.