Author: Robert Hyatt
Date: 12:07:10 06/24/01
Go up one level in this thread
On June 24, 2001 at 10:43:06, Rafael Andrist wrote: >On June 24, 2001 at 09:54:11, Robert Hyatt wrote: > >>On June 24, 2001 at 04:09:29, Rafael Andrist wrote: >> >>>On June 23, 2001 at 22:24:58, Robert Hyatt wrote: >>> >>>>On June 23, 2001 at 20:02:53, Vincent Diepeveen wrote: >>>> >>>>> >>>>>I do not know much details on schach, but if you give it an entire >>>>>night, then it still will show the same bad branching factor as >>>>>we were used to in 1997. >>>> >>>> >>>>What in the name of heaven are you talking about? "bad branching factor >>>>we were used to in 1997?" In 1996 I was using R=2 null-move, as was Bruce >>>>and others. _we_ were not used to any huge branching factor. My branching >>>>factor today is almost _exactly_ what it was in 1996... >>>> >>>> >>>>My branching factor also remains fairly constant regardless of the depth. >>>>It makes no sense to me that it would get smaller and smaller as you go >>>>deeper. Because eventually it would have to go below 1 and the search >>>>would actually get _faster_ as you go deeper. And that is fully nonsense. >>>> >>> >>>But the BF get's clearly smaller if the search goes deeper. At least with a >>>full-window-search. It's also clear that this process isn't linear. So your >>>second argument that it would ge below zero is nonsense itself. We live in a >>>completly non-linear world! >>> >>>Rafael B. Andrist >> >> >>I don't have any data for searches beyond 15 plies in the middlegame, but for >>the 350 or so searches I have, I don't see the BF as better from 14->15 than it >>is for 12->13... I don't see why it would be so either. unless you assume >>infinite everything like hash table size, etc. > >In my experiments, I didn't go so deep. But the it seems to me, that the BF is >an exponential function of the search depth (different for each side). At this >depths the BF changes only 0.1 or less. It's clear that in some situations the >BF can go up too, e.g. if you queen a pawn in a pawnendgame. > >Can you give me a typical position where you mesured the values? Then it would >be possible to compare. > >Rafael B. Andrist I was using the "plytest" data from the "crafty goes deep" ICCA paper. The positions are on my ftp machine (350+ of them). For positions where you don't change your mind at all, branching factor should be optimal. For positions where you change several times, branching factor will be horrible as the trees are _much_ bigger when that happens.
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.