Author: Vincent Diepeveen
Date: 15:28:37 10/16/02
Go up one level in this thread
On October 16, 2002 at 17:44:11, Jeremiah Penery wrote: If you search for me postings a few years ago you will see already that i quote there that DIEP and Schach and older The King versions and all programs that use singular extensions, that they completely die when getting at depths 10-12 ply where you get near the tactical barrier, which i defined to be around 12 ply for most game positions back then. It still happens. At 10 ply singular extensions are just 60% of the plies possible in DIEP. at 12 ply it's already possible at 70% of the plies to get extended and extended. >On October 16, 2002 at 11:05:50, Uri Blass wrote: > >>On October 16, 2002 at 09:40:21, Jeremiah Penery wrote: >> >>>On October 16, 2002 at 07:31:01, Vincent Diepeveen wrote: >>> >>>>On October 15, 2002 at 13:31:07, James Swafford wrote: >>>> >>>>There is 2 things >>>> >>>> a) theoretic branching factor without nullmove >>>> >>>>that's true for the last so many plies of deep blue and >>>>the theoretic truth. >>>> >>>> b) branching factor with hashtable. >>>> >>>>You will see that deep blue is somewhere, like all software programs >>>>in the middle of this. >>>> >>>>126 mln nodes a second. And it got like 12 ply with a bit more >>>>than 3 minutes. >>>> >>>>that's like total branching factor = 12th root from >>>>126 mln * 180 seconds = 6.0 >>> >>>If you mean (180*126m)^(1/12), that equals 7.294. >>> >>>>So the real overall b.f. which was achieved was 6.0 >>>> >>>>that's lower than the theoretic branching factor of sqrt(40^18) >>> >>>If a decent program didn't have a lower number than the 'theoretic branching >>>factor...' I'd be very surprised. >> >>I believe that it is possible to get >>a good program with branching factor of 6 >>and a lot of extensions. > >I didn't say it was impossible, I said I'd be surprised by it. However, now >that I think about it, Chessmaster might have that kind of branching factor. > >>I see no contradiction >>plies do not have to be eqvivalent. > >Ah! This is the important statement. Vincent seems not to understand it. > >>>BUT, if you insist on this number, and you think they only got 12 plies, they >>>need to search 4096000000 nodes for 12 plies. They can accomplish that in 32.5 >>>seconds with 126M n/s. 12.2 plies takes 47 seconds. In 180 seconds, they can >>>do 12.93 plies, in fact. sqrt(40) is a branching factor of 6.32. With a >>>branching factor of 6, they can reach 13.3 plies in 180 seconds based upon your >>>formula. >> >>No >> >>You forget that there is also a constant. >>If the number of nodes to get depth n is >>a*6.32^n I see no problem. > >Problem with what? > >The real problem here is trying to calculate their branching factor with this >formula. The formula assumes the search tree is perfectly uniform, and theirs >was highly NON-uniform. Therefore this particular calculation is meaningless. >I only used it because Vincent seemed insistent upon using it.
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.