Author: David Blackman
Date: 01:31:42 06/26/99
Go up one level in this thread
On June 26, 1999 at 01:20:13, Peter Kappler wrote: >On June 26, 1999 at 00:23:51, David Blackman wrote: [ in reply to James Robertson ] > >>Maybe you are comparing apples and oranges. The way i count quiescence nodes >>it's hard to see how less than 50% of total nodes could be in the quiescence >>search. Anyone getting less consistently is probably either counting >>differently or doing a radically different type of search. 2/3 is probably ok, >>but don't give up on trying to improve it :-) > > >I think it depends on your pruning strategies. For me, the q-nodes are only >about 20% of the total nodes. > >I don't think I'm doing anything exotic in my counting approach. I increment my >node counter every time I call make_move, and if this happens to be in the >q-search, I count it as a q-node. > >--Peter This is exactly what i mean by a different way of counting qsearch nodes. The way i count (and i hope James does the same) is like this: If i am in the normal search with a draft of 1, i generate a move, decide i don't want to extend on that move, so after domove and the recursive call, the draft will be 0. Because of this i call the quiescence search instead of the normal search. Now the first thing the quiescence search does, is to increment the quiescence node counter. After that it tries a material+position evaluation, and if that fails high it returns without generating or making any moves, but still counting as one quiescence node. (If it doesn't fail high it generates captures and does recursive calls etc.) So with this way of counting, it is very hard to get quiescence nodes under 50% of total. Peter's method of counting is just as legitimate, but with exactly the same pruning and move ordering, it gets much lower percentages. If James is using my counting method and getting 2/3, no worries. If he's using Peter's counting method and getting 2/3, then he should look for improvements.
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.