Author: Dann Corbit
Date: 13:20:05 06/04/02
Go up one level in this thread
On June 04, 2002 at 16:14:07, Robert Henry Durrett wrote: [snip] >I was thinking about the use [somehow] of chess knowledge to evaluate the root >node, but without using any search to accomplish this evaluation. If you have no search at all, this approach will fail. Guaranteed. >This appears >to have been incorrect. It now looks like you are still relying on examination >of possible lines emminating from the position, in addition to anything else you >may be doing, to evaluate the root node BEFORE selecting the "children." > >Is this closer to the truth? Here is what happens: The root node's children get examined. The very best looking one becomes the pm (predicted move) which is the first thing in the pv. That node gets examined very carefully. All the other nodes get a "zero window search" (which is actually one unit wide). These searches happen very quickly most of the time. If your evaluation is pretty good, the pm guess will usually be right (90% or better). Some of the worst moves will get the fat trimmed off with null-move pruning. They won't be searched as deeply. You don't have any choice about selecting the children. The children are each and every legal move from each and every node in the tree. They don't all get searched at the same depth. Extensions will lengthen some paths and pruning will shorten some others. Have you seen Bruce Morland's or Colin Frayn's tutorials? They are very good at explaining how computer chess searching works.
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.