Author: Robert Hyatt
Date: 14:25:23 08/21/02
Go up one level in this thread
On August 21, 2002 at 14:55:48, Gian-Carlo Pascutto wrote: >On August 21, 2002 at 14:48:26, Robert Hyatt wrote: > >>I just computed the branching factor for every iteration that was completed, >>leaving off the first one since I had nothing to compare it to (first one for >>each search). I added them all upp and divided by N, and got something just >>under 4.0... >> >>Anybody can do that again, rather than producing disinformation... > >This doesn't work. All that is printed are search times, not node >counts. You don't need node counts to compute an effective branching factor. All you need are a set of times. IE it takes 1 second to complete plies 1-6. It takes 4 seconds to finish ply 7, or 8 seconds total. It takes 16 more seconds to compute ply 8, or 24 seconds total. The output would look like this: depth time <etc> 6 1 ... 7 5 ... 8 21 ... If you take those numbers, you compute that 7 plies took 4 seconds, and 8 plies took 16 seconds. The branching factor for that test is 4 (16/4). You can do this for every search in the log. And it is just as accurate as trying to count nodes. IE you can do the same thing for Crafty even though it never produces node counts iteration by iteration... > >The paper literally says that at small depths not all cpus are used. That is very possible... But in Crafty at small depths the same thing happens.. > >Hence, when comparing the plies and only looking at times, you are >missing the variable of more cpus kickin in. You can't use this to >determine the branching factor of their search. > >-- >GCP Sure you can, particularly by keying on the last few plies of search which are taking appreciable time...
This page took 0.01 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.