Author: Robert Hyatt
Date: 13:58:47 10/24/02
Go up one level in this thread
On October 24, 2002 at 14:49:17, Omid David wrote: >On October 24, 2002 at 14:41:24, Robert Hyatt wrote: > >>On October 24, 2002 at 14:04:07, Fernando Villegas wrote: >> >>>I wonder if every or some commercial prorams has the same "panic mode" DB had. >>>Sometimes I gues they does have one. At leats after some move that does make >>>them happy, I have observed that time allotment change, depth of the search >>>deepens and so on. Chris, Ed? >>>Fernndo >> >> >>I think that since Bert and I first published the "using time wisely" paper in >>the JICCA >>many years ago, _everybody_ has been doing that in some form or another. For >>years, >>we watched people bouncing up and down in their chairs hoping their program >>would >>see that the move it was about to play was the wrong move because the score had >>dropped >>way down. The only question was could their machine find a better move before >>running >>out of time. >> >>We decided to stop doing that and added the "fail-low time extension" idea and >>then reported >>on it at the next ACM event. > >Do you have a copy of that paper? > Not an electronic copy. It was published as "Using time wisely" in a volume of the JICCA. There was a later addendum called "using time wisely revisited" published in the same journal a couple of years later... The basic ideas were 1. use more time on fail lows, less time on "obvious" moves. 2. use more time right out of book. 3. Use more time early in the game, knowing that pondering will save time as the game wears on and make up for using the time early. > > >>The next year everybody was doing something that >>seemed very >>logical to me a couple of years earlier... >> >>The only question is how you define "panic mode". Deep Blue had something more >>refined >>that just "the score dropped" but I never followed Hsu's explanation since it >>was very brief and >>unclear (to me). Something like "if the size of the tree can't be expressed in >>a canonical form, >>then we consider the tree to be unstable." >> >>His idea worked, even if I didn't understand how it worked. But it would use >>more time in >>positions where it had not yet failed low, which was the important point.
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.