Author: Greg Lazarou
Date: 20:46:11 02/20/99
Go up one level in this thread
Isn't it dangerous to use the result of a half done search? High up at the tree you may have not yet looked at all the possible responses and you may have overlooked a killer move by the opponent... I do a full search of 1 ply, then if we have time 2 ply, then if we have time 3 ply etc. When I run out of time I don't use the incomplete search's results, but instead revert to the findings of the prior full search. Am I missing something and is there a safe way to use the results of the last incomplete search? Greg On February 20, 1999 at 19:17:14, Peter Kasinski wrote: >On February 20, 1999 at 16:34:57, Robert Hyatt wrote: > >>On February 20, 1999 at 11:08:30, Peter Kasinski wrote: >> >>>I am having problems interupting search at an arbitrary point. >>>Unless it completes the iteration I get strange results. Typically I get the PV >>>move and all is fine. Sometimes I get a PV move with a strange (usually very >>>low) score. And once in a while I get a moronic move with a low score. >>>I don't allow this before first iteration is completed. >>>Any idea what is going on? >>> >>>Thank you, >>>PK. >> >> >>this is easy and hard. When you decide to 'time out' all you have to do is >>start 'returning' to back your way out of the recursive calls. But you _must_ >>be certain that you know you are 'backing out'.. you do _not_ store any hash >>stuff on the way back, you do not test for cutoffs, you do not back up scores >>or PVs, you don't modify scores or the PV, ie you do _nothing_ but return. >> >>If you don't notice you are 'aborting' and update something you will get >>wildly bogus results... > > >That's what I get. > >> >>In crafty, I do this _everywhere_ I call Search: >> >> val=-Search(-beta,-alpha,etc).. >> >> if (abort_search) return(0); >> >>abort_search is a global variable that is set 'true' when the time check >>says 'stop searching. You _must_ put this after _every_ call to search(), >>quiesce() or any other recursive routine you might use to do selective >>stuff... >> >>that prevents any score/pv update after that variable has been set.. > > >Thanks Bob, as always. > >PK.
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.