Author: Uri Blass
Date: 00:50:44 10/18/02
Go up one level in this thread
On October 18, 2002 at 03:32:00, Uri Blass wrote: >On October 17, 2002 at 23:48:27, Robert Hyatt wrote: > >>On October 17, 2002 at 19:57:41, Uri Blass wrote: >> >>>On October 17, 2002 at 18:12:02, Robert Hyatt wrote: >>> >>>>On October 17, 2002 at 16:34:08, Murray wrote: >>>> >>>>>On October 17, 2002 at 10:07:41, ujecrh wrote: >>>>> >>>>>>(snip) >>>>>>We can track opponent's thinking time and, without trying to match it, add some >>>>>>time or search extensions when an unusual delay has occured. >>>>>> >>>>> >>>>>But humans also ponder when the computer is thinking. It could be argued that if >>>>>the human is having to spend a long time thinking in a difficult position, the >>>>>computer should play just as quickly or quicker than normal, to reduce the >>>>>human's chance of seeing through the complications. >>>>> >>>>>Murray Cash >>>> >>>> >>>>If the computer had any _idea_ about what makes up a complex/hard-to-analyze >>>>position, >>>>this would be a good plan. But it doesn't have a clue about whether the >>>>opponent has an >>>>easy or difficult position to play, and trying to play games with time usage >>>>will more often >>>>than not blow up in your face.... >>> >>>The computer may ponder about all the possible moves but use different time for >>>different moves(I remember that I read that this idea is used by aristarch). >>> >>>If based on the scores it can see that there is a forced move it can give it >>>almost all of the time but if it see 5 moves with almost the same score it can >>>continue to analyze them and use almost 1/5 of the time for everyone of them. >> >>This idea is simply no good. I have explained why a dozen times or two, over >>the past 10 years... It doesn't work now. It didn't work 20 years ago. It >>won't >>work 20 years from now either. >> >>The _best_ way to ponder is to pick the best move and go with it, unless the >>program >>is so bad it can't predict right even 50% of the time, which is very low... > >I think that the best way to ponder is not so simple >I do not try today to use it and movei ponder only on the expected move >or wait in case that it finished to ponder. > >It is certainly better than using the same time for all moves >but if you ponder more time on the better moves than I believe that >pondering on all the moves may be better. > >You should start by pondering most of the time on the expected move but if you >fail high on the search after the expected move or if the difference in >evaluation relative to other moves is small then you can increase the time for >the other moves. > >Small difference means less than 0.1 pawn and >if the difference in evaluation is 0.3 pawns you still may >use 70-80% of your time for the expected move and if the difference >is more than 1 pawn you may use even more than 90% of your time for the expected >move. > >Uri I can add that I have no experience in threads but I guess that it is possible to have a thread for every legal move of the opponent and to update every second the percentage of time that is going to be used for every thread. Uri
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.