Author: David Eppstein
Date: 22:11:07 09/30/99
Go up one level in this thread
On September 30, 1999 at 16:50:56, José de Jesús García Ruvalcaba wrote: >faster hardware benefits more the programs with better search policy, >regardless of how fast or slow they are. > By search policy I mean the extensions, pruning and other techniques used to >unbalance the search tree so the engine looks deeper at the moves that are more >"interesting". It is not important to outsearch your opponent in every branch of >the tree, only at those that are relevant. I like very much this answer, it sounds right to me. But I am curious what people think makes for a good strategy for unbalancing the tree. My own opinion: a line should be looked at longer when - it is important (in PV or not very far from it, e.g. conspiracy number search). Maybe this is why singular extensions might be useful? - it contains delaying moves that might be a symptom of the horizon effect (e.g. check and recapture extensions). - it ends in an difficult to evaluate position. Quiescence search sort of covers this, but I think not very well. What I mean is, for instance, a position with various latent tactical possibilities, forks, batteries, etc., should be searched a little deeper. Positions that look very placid should maybe be cut off early, because one or two ply will not make much difference to the positional qualities that should be included in the evaluation function. I think the third point, having an evaluation function that knows when it does or doesn't understand the position, may be the least well covered in current programs. I have something like it in my program: it will search up to a max of 1.5 plies deeper on lines that would otherwise end in a position where there is a threat (which I define as a capture that the player not-on-move could make if the player on-move passed) but it's hard to tell how well it works or whether my implementation technique is the most appropriate one.
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.