Author: Robert Hyatt
Date: 14:19:52 09/12/02
Go up one level in this thread
On September 12, 2002 at 15:53:39, Jay Scott wrote: >On September 11, 2002 at 17:53:41, Robert Hyatt wrote: >>How are you not going to store the tree when it is, by definition, "best >>first"? > >Heh. By not being best-first everywhere, as you know perfectly well. :-) In this >paper, they used a short depth-first search as an "evaluation function" for the >best-first search, so the stored best-first tree is only a tiny part of the >entire tree that is searched. That's what I meant. It isn't that "tiny". And as you use more time to go deeper, all those new nodes are "best first" and have to be stored... > >Other methods are possible too. For example, you may be able to discover >(especially taking your time limit into account) that a certain subtree that has >already been searched is unlikely to be (or to be discovered to be) any good. >"That move looked OK at first, but now I'm almost certain it's awful!" That >storage can be recovered immediately, and the stump marked "do not search >again". A full-up "rational search" would take into account memory costs as well >as time costs, and trade them off continuously against move quality. I can see that failing when you suddenly discover that all _other_ moves drop drastically due to a tactical threat that move actually prevents... :)
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.