Author: Tony Werten
Date: 04:33:50 02/07/02
Go up one level in this thread
On February 07, 2002 at 06:29:47, Sune Fischer wrote: >On February 07, 2002 at 04:55:23, Tony Werten wrote: > >>On February 06, 2002 at 10:45:25, Sune Fischer wrote: >> >>>On February 06, 2002 at 10:30:15, Tony Werten wrote: >>>>>So it would seem, but the search is exponential and not linear. >>>>>I think you should not consider the "depth" but rather the number of nodes >>>>>searched. >>>> >>>>Doesn't make a difference. Depth and number of nodes are the "same". >>> >>>Not at all, nodes is an exponential function of depth. >> >>Yes, should have said highly related. > >I had a feeling that "same" was a bit fuzzy ;) > >>My point is that when you give a program 1M nodes more than the other, at low >>depths this might be a couple of ply, at higher depths, it's less than a ply. >> >>Calling this diminishing returns isn't correct IMO. It's just the way a >>searchtree works. > >Yes, and we wouldn't be doing that, we would be multiplying by a BF factor, e.g. >doubling the search time each time. >The question is, if we double up every time, do we see diminishing returns at >some point, or will there be a constant change in rating with each doubling? > >If you compare computers rating with humans, you will see that almost all >programs are a few hundred elo higher in blitz and bullet than in standard >tournament time control. >Of cause we don't know what kind of diminishing returns a human has, so it >doesn't tell all that much, but there is a clear tendency. > > >>I believe DR is the fact that 4-3 scores a bit better then 8-6 >>and 12-9 > >That would mean diminishing returns. That's what the DR stands for :) I believe there is DR but not as big as the ply vs ply+1 tests indicate. Even more if there was no DR (ie 12-9 scores same as or better than 4-3), ply vs ply+1 would still indicate there was. Tony > >>OK. Hmm, how about giving a limited amount of petrol to accelerate a car ? If >>the first car goes slow, you can go twice as fast and arrive a few hours befor >>him. Else it might only be a few percent and a few minutes. > >What you are thinking of is the function 1/x (because time=distance/velocity), >it shows similar characteristics as exp(-x), but is not the same. > >-S.
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.