Author: Pat King
Date: 17:29:17 01/27/00
Go up one level in this thread
On January 26, 2000 at 10:48:41, Tijs van Dam wrote several interesting questions, none of which he will get a consistent answer on. For what it's worth, here are my answers. >Hello, > >Very often I read about "my branch factor is ...", or "my program calculates ... >nodes per second". It is not always clear to me what is meant here, especially >on the topic of fail-high percentages. > >Speed: >* Do you count quiescence nodes in the number of nodes per second? > >Fail high: >* Do you count quiescence nodes that fail high because one of the moves is very >good? >* Do you count quiescence nodes that you skip because Eval()>beta? >* Do you count nodes that you don't search further because >SearchNullMove()>beta? >* Do you count those nodes as "fail high on the first move"? Every call to SearchXXX() increments the node count, because I've done a Move() and will do an Unmove() >* Do you count nodes that you skip because HashProbe()>beta? No, because no Move/Unmove > > >Branch factor: >* Is this > - the average quotient of the number of nodes searched in a ply and the >previous, or > - the average quotient of the number of nodes searched in the last completed >ply in a search and the one before? Closer to the latter: EBF = ln(N)/D, where EBF = "Effective Branching Factor", N = nodes searched as defined above, and D is the nominal depth of the search, not accounting for quiesence or extensions. > >Quiescence: >* Do you count leaf nodes of the normal search, thus the "root nodes" of >quiescence search, as quiescence nodes? Not currently count q nodes seperately. > > >Greetings, >Tijs van Dam One only moderately successful (as judged by my own low standards) approach. I have found my "EBF" useful in evaluating how effective changes to my search were, and it seems to compare well with the figures one sees thrown around here from time to time, but as you've noticed, there isn't a lot of consistency in how people measure things. Pat King
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.