Author: David Rasmussen
Date: 03:58:28 01/31/02
Go up one level in this thread
On January 30, 2002 at 18:37:21, Dann Corbit wrote: >What does "fullwidth" mean in this context? Obviously, it cannot possibly mean >that every branch of the tree is examined exhaustively to 18 ply (which would be >absurd). > >Was 18 ply claimed for the opening? For the middle game? For the endgame? As >a broad average? > >If (for instance) I take crafty or chess tiger or pepito and let it run >overnight, and 18 ply search is not unusual in the least bit. > But Crafty does null move pruning and futility pruning in q-search etc. Deep Blue didn't, I'm assuming. Of course with a "modern" search policy (especially one like Tiger's, which seems extreme and very effective), a 200Mn/s machine can reach 18 ply, but DB didn't do that, and it can hardly be called full width. In this context (DB), I consider "full width" to mean not the entire minimax tree to 18 ply in every line, but the "entire" alpha-beta tree for whatever move ordering they might have had (90-95% first fail highs on avg. I assume), without any forward pruning. That is a pretty big tree. Everything we are used to measure everything against today, is forward pruned in some way. DB weren't, I assume (again). >How does that compare with deep blue at 10 minutes? > >I don't think it is certain (either for or against) and I don't think that the >terms used are certain. > >Just like "ply" means 20 different things to 20 different programmers. Even >"nodes" does not always mean the same thing. > Agree with you on nodes, but ply? Ply is pretty well defined, I think. >In computer chess, there really are no precise definitions, because nobody has >done a standardization. Therefore, the terms are very general. I am not sure >(either way) what the real search throughput was. > >At any rate, it was good enough to beat Kasparov. Sure. I am not disagreeing with that :) /David
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.