Author: Robert Hyatt
Date: 18:24:34 08/18/01
Go up one level in this thread
On August 18, 2001 at 16:33:28, Heiner Marxen wrote: >On August 18, 2001 at 14:31:17, Koundinya Veluri wrote: > >>On August 17, 2001 at 23:28:50, Robert Hyatt wrote: >> >>>On August 17, 2001 at 15:16:54, Alvaro Jose Povoa Cardoso wrote: >>> >>>>Only now I've implemented the trick of avoiding generating any moves by >>>>searching the hash move first (if there is one) and bailing out early if this >>>>move causes a fail high. >>>>Now, I just don't get the same node counts. I get tree size reductions. >>>>This is a bug right? >>>>I should get exactly the same node count, just a litle faster right? >>>> >>>>Help appreciated, >>>>Alvaro Cardoso >>> >>> >>>Node count should not change. You did take care of the issue that once >>>you search the hash move and get no cutoff, you then generate moves and >>>you do _not_ search the hash move a second time? If you do, that can >>>change the node counts... >> >>What about extensions for forced moves? When there's only one check escape, I >>extend the search, but this extension is not stored in the hash table and I >>cannot confirm that the hash move is a forced move without generating any moves. >>So in this case, if the hash move is tried before generating check escapes, it >>would not be searched to the same depth (hence changing the node counts). Or >>should I always generate check escapes before trying any moves to avoid this >>situation? >> >>Koundinya > >You could just put this info (1 bit) into the hash entry, along with the move. >When storing the move first, it should be available... so just remember it. > >Regards, >Heiner I am a bit lazier. If I am "in check" then I generate only legal moves before I try the hash move. This handles that case with ease, since if I am not in check, I can't trigger a one-rep extension and using the hash move doesn't cause confusion. But if I am in check, I generate em all first which makes things also correct for onerep...
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.