Author: Dann Corbit
Date: 14:35:47 03/29/04
Go up one level in this thread
On March 29, 2004 at 17:28:37, Dieter Buerssner wrote: >On March 29, 2004 at 16:35:29, Dann Corbit wrote: > >>On March 29, 2004 at 16:22:21, martin fierz wrote: >> >>>On March 29, 2004 at 15:50:10, Dann Corbit wrote: >>> >>>>I have not done the math, but I am guessing no matter what king of move ordering >>>>you have (purely randome or the pv move every time) you will get something like >>>>this: >>>> >>>>nodes = some_constant * sqrt(mini_max_nodes) > >>>i disagree with your formula. it is definitely not some_constant. it is a >>>constant between 1...sqrt(N) taken to the power of D/2. else there would be no >>>point in improving move ordering, or at least, not as much as there is :-) >> >>What if the constant is one trillion? > >But, do you mean, that some_constant is a function of moveordering only, or a >function of depth (and moves), too. I think, the later will be the case. If yes, >your formula will of course be correct, but will not show any insight. Many >other formulas will be correct, too, then. > >nodes = some_constant * exp(mini_max_nodes) > >You will always be able to adjust some_constant, to get the correct result. Of >course, it will be very tiny. Perhaps I missed something. > >Regards, >Dieter Would you agree that if we always picked the second best node first rather than the best node, the search will still be proportional to the square root of the minimax nodes? If yes, then the conclusion is inescapable, I think.
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.