Author: Ralf Elvsén
Date: 16:32:50 02/26/01
Go up one level in this thread
On February 26, 2001 at 16:29:06, Severi Salminen wrote: >>I not so sure about my figure 8.5 anymore. In fact I think the >>probability that it is correct is pretty low... I locked on >>that solution immediately and then it's hard to change >>ones mind. So concerning the theoretical number derived >>from a branching factor, forget what I said (I'll maybe >>come back). My advice is still to use measured numbers. >>This is what I do myself. It has the additional benefit >>of avoiding my sloppy calculations :) > >I think the point of the original message was to make a decent test that >measures the speed of movegen+makemove. So, if modern chess programs have >effective branching factor of about 3 (maybe 2-4) it is wise to generate all >moves and make and unmake three of them. This has nothing to do with critical AB >tree but normal searching conditions. I'd guess 3 would be good approximation. > >Severi And that is exactly the question I tried to address. Given an (effective) branching factor, what is the proportion of makemove/movegen ? Have I been so unclear? I'm not sure if this ratio equals the branching factor. I'll have a look at it when I have the time. I guess someone around here knows but is too lazy to answer. As I understood the original question he wanted to run one movegen and then a number of make/unmake in the same proportion as will be found in a real search. He used a branching factor (however it was measured) to guesstimate the proportion. I doubted, and still do, that that is correct (even for a "normal search"). If it is the obvious thing to do I will be happy to be informed why. But all those considerations can be skipped if the proportion is measured. When I do a test like that I don't care about the branching factor, I don't see the relevance. Ralf
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.