Author: Andrew Dados
Date: 12:54:02 01/30/02
Go up one level in this thread
You may see entire thread at: http://groups.google.com/groups?hl=en&frame=right&th=ce0be94fc8e7f93&seekm=332208AC.797E%40nwlink.com#link11 It is dated 1997/03. Indeed seems like Campbell meant 11-12 plys. And seems like nothing changes...:) --------------------------------------- Robert Hyatt (hyatt@crafty.cis.uab.edu) wrote: : Vincent Diepeveen (vdiepeve@cs.ruu.nl) wrote: : : In <5fne5f$i9o$1@hpax.cup.hp.com> fotland@cup.hp.com (Dave Fotland) writes: : : Thanks for you mail, this clears up things about the DB mystery! : : So they will try to beat Kasparov with a 11 ply fullwidth searching program. : : That is hardly 6 moves. that'll give them a lot of strategically insights. No! Dr Campbell said that deep blue (and Deep Blue jr), could search full width two ply deeper if didn't do their search extensions. He also said that the search extensions could go to 30 or 35 ply, since Kasparov can search that deeply. So they are NOT just doing an 11 ply full width search. Also, the 11 ply full width is not equivalent to traditional full width since it has 7 ply full, variable ply selective, followed by 4 ply full width. So they 4 ply full width of tactics at the endpoints of a selective branch. If they can do 2 more ply without the selective search, then 24/25 of the nodes in the search are selective, and only 1/25 are in the full width search. So don't compare it your basic full width searcher. : Noo... you didn't read carefully... 9 plies for a "game in 15 minutes" : game. That's *far* different from a 40/2hr time control... I still : suspect they are searching in the 14-15 ply range on DB hardware. I : suspect the 2 ply gain is a misinterpretation by someone, because the : RS6000 version should be much worse than only 1/25th the speed of the : real DB... No. I asked Dr Campbell specifically about this, and he said 11 or 12 ply against Kasparov. The SP2 has 32x the CPU's at 2x the speed each, so the basic search is 64x faster, but they only have about 16x the number of chess chips. And against Kasparov of course they have a higher time limit. If I remember right, he said something like 10M NPS for Jr, 200 NPS for Deep Blue, so only about 2 more ply. He said that if they did just a full width search that they could get 13 or 14 ply. But they think the selective search is really important so they can see as deeply as Kasparov. Also, Deep Blue Jr didn't have an opening book or endgame databases. It does think on its oponent's time. : : My estimate (based on a 6 games match): : : Kasparov - Deep Blue 6 moves : 5.5 - 0.5 : I'm still at 4-2... 2 wins, 4 draws... and hoping they do better. : (that's Kasparov winning 2 games, drawing 4)... I'm a weak chess player (1500), but I watched some games that were played by players stronger than myself, and I was impressed by Deep Blue Jr. My uninformed opinion is not worth much, but I think Deep Blue can win a game or more. -- David Fotland fotland@cup.hp.com
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.