Author: Uri Blass
Date: 15:50:05 11/20/02
Go up one level in this thread
On November 20, 2002 at 17:49:43, Tony Werten wrote: >On November 20, 2002 at 17:31:57, Omid David Tabibi wrote: > >>On November 20, 2002 at 16:55:41, Gian-Carlo Pascutto wrote: >> >>>Nullmove in Deep Sjeng uses an algorithm of my own, but I can >>>switch it back to other systems easily. I did so for running >>>a few tests. >>> >>>I made a version which uses Heinz Adaptive Nullmove Pruning >>>and a version which uses your verification nullmove. >>> >>>I did a run over ECM-GCP, at 10 seconds per position. >>> >>> >>>So, you can see here that verification nullmove searches 20% more >>>nodes, took 12% more time, searched 1/3 of a ply less deep and >>>missed three more positions. >>> >>>That's pretty much a 'total loss' :) >>> >> >>That is a typical result when the quiescence search is too big. With the help of >>its extensions, even the standard version of your algorithm manages to detect >>most of the problems that verified null-move is meant to detect. So the >>verification search is just a waste of time in this case (and since your test >>was conducted in a fixed time per position, this waste of time didn't enable the >>verified version to reach deeper and detect what the other version did). >> >>Solution: >> >>By shifting to a far simpler quiescence search, you will get about the same >>tactical strength as you get currently (since this time, verified null-move will >>detect the problems, instead of your immense quiescence), but will end up with a >>significantly smaller search tree. > >And positionally a weaker program :( > >Tony positionally? I thought that big qsearch is about tactics. I plan to do my qsearch bigger in the future because I want to detect forks or cases that a piece is in trouble and have not good squares to go. I found that extending in the qsearch when the queen is pinned helped me to solve at least one more problem in the GCP test suite. detecting pins to the king is cheap for me because I calculate them for my move generator so it seems to be a good deal. 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.