Author: Tord Romstad
Date: 11:08:35 10/26/05
Go up one level in this thread
On October 26, 2005 at 12:27:13, Gerd Isenberg wrote: >On October 26, 2005 at 06:58:41, Tord Romstad wrote: > >>This condition is the >>reason for the name "history pruning", which in my opinion is very >>unfortunate. History is just one of several conditions which can be >>used, and we are not talking about pruning, but reductions. I prefer >>the term "late move reductions", but it seems I am quite alone. > >Yes, i don't like "history pruning" as well. Cool. I am not completely alone, then. :-) >What about "right wing reductions" ;-) Perhaps, but it has political connotations which may not be universally accepted. :-) >Do you apply "late move reductions" only at "expected" cut-nodes or on all-nodes >as well? I use them absolutely everywhere, except at the root and when the remaining depth is less than 2 plies. I use late move reductions even at PV nodes. In principle, I want to be more restrictive, but to my frustration this always makes my program weaker. :-( >>I have found the technique to work even better (especially in tactical >>positions) with the following enhancement: If, at the node directly >>following a reduction, the null move fails low, and the moving piece >>in the move that refuted the null move is the same as the moving piece >>in the reduced move, immediately cancel the reduced-depth search and >>re-search the move with full depth. The point is that in cases like >>this, the reduced depth move often contain some serious tactical >>threat, and deserves a deeper search. > >As always - thanks for sharing your ideas and improvements. >Sounds logical - have to think about the control flow of the search. Another detail is that this enhancement allowed me to do late move reductions a little bit closer to the leaves. Previously I couldn't use them at remaining depth less than 3 plies without introducing too many tactical mistakes. It now seems to be safe to reduce the limit to 2 plies. >BTW. what about your strange 10% speedup problem you mentioned some time ago in >WBF? For the benefit of readers who don't read the Winboard Forum, I'll recapitulate the story here: Some time ago, I tried to remove queen mobility evaluation from my program. I expected this to make my program a tiny bit faster, but to my surprise it made it 10% slower (which is a lot for such a small change). >Is it still present and did you find an explanation - or was it only a >temporary chaotical "phantom", which disappeared after some code changes? I never found an explanation, and it was apparently just a chaotical fantom which disappeared after changes in parts of the code which should have been entirely unrelated to this. Fortunately, it seems that the slow situation was the exceptional one. :-) Tord
This page took 0.02 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.