Author: Robert Hyatt
Date: 18:50:50 05/05/01
Go up one level in this thread
On May 05, 2001 at 21:10:06, Paul wrote: >On May 05, 2001 at 20:46:45, Robert Hyatt wrote: > >>On May 05, 2001 at 19:26:11, Paul wrote: >> >>>Ok .. I'm into testsuites now :) ... but what is the proper way to record the >>>solution times? Do you take the time at which the keymove is found (the final >>>fail high), or the one when the fail high research has finished and produced a >>>new pv & score? Because there can be a big difference, eg. when you're at ply 13 >>>the research can take quite some time. Or is this a silly question? >> >>It depends on the engine. The _right_ time is the time after which it would >>play the right move. IE for Crafty, if a move fails high at the root and >>you see the Nxe4!! move displayed, it would play it _right then_ if time >>ran out... so the fail-high time is the time to use. > >Well ... in my GUI, if a rootmove fails high, the move lights up in green, does >that count? ;) > >But seriously, what do you do when it fails high (after a 'null' window search) >and the research doesn't? Is the pv from the (previous) bestmove still there or >is it gone at the moment of the failhigh? My guess is that you only replace the >bestmove & pv when time runs out or when the failhigh is resolved. I know, I >should look this up in the archives, must have been asked before! > >>If a program won't play the fail-high move without resolving the score, then >>you have to use the time it needs to actually decide to play the move... > >Ok, so I'll have to change to your way to get better solution times. > >Thanks, >Paul I don't notice fail highs on the null-window search if the re-search fails low. It never happened so far as the search goes. The fail high I was talking about was the search failed high at the root (new best move) which means the null-window search failed high, and the re-search _also_ failed high.
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.