Author: Robert Hyatt
Date: 07:26:12 10/23/99
Go up one level in this thread
On October 23, 1999 at 01:37:47, Christophe Theron wrote: >On October 22, 1999 at 13:08:44, Robert Hyatt wrote: > >>On October 21, 1999 at 15:55:21, Jeremiah Penery wrote: >> >>>On October 21, 1999 at 12:54:47, Christophe Theron wrote: >>> >>>> >>>>But no program of today do brute force. All the good programs are highly >>>>selective. >>>> >>>>If you are not talking about "true brute force" (simple-alpha beta + QSearch) I >>>>don't know what you mean. Everything else is selective. >>> >>>We must be using different definitions. I don't consider the use of null-move, >>>for example, a selective search. >> >> >> >>A word of caution. "selective search" has a precise definition in the CS >>literature, dating all the way back to Shannon's "How to program a computer >>to play chess" in the 1950's. >> >>Selective search means, very explicitly, to generate all moves at a node in >>the tree, and then to dismiss some of those a priori without any searching of >>any kind. This is also called 'forward pruning'. >> >>The idea of selective extensions was mentioned by Shannon, in the context of >>what he called a "variable depth search" which is exactly what all of our >>extensions and null-move reductions actually accomplish. >> >>I personally don't call a "null-move search" a "selective search" because it >>is an incorrect term of a previously established term. Any more than I would >>self-define the term "ampere" to mean resistance, rather than using the more >>accepted "Ohm". I don't know of anyone doing what I would call purely >>selective search, although our q-search is a perfect example, since we toss >>out some moves with no searching of any kind, while keeping others and searching >>them deeper, all in a pretty arbitrary way. >> >>It makes more sense to keep a common vocabulary when we discuss things so that >>every post doesn't require a personalized "glossary of terms" so that we can >>communicate. :) > > >You are right. It is indeed better that we all use the same vocabulary (and the >idea of a glossary in the "Computer Chess Resource Center" pops up again). > >But in this case, how can we call the kind of "selectivity" we get from, for >example, a null move search? > >Are we going to call this "variable depth search"? It's pretty misleading I >think. I agree totally. We need a 'term'. Because we have several algorithms that fit this 'thing'. Null-move. razoring. Both reduce the effective depth. Then there are all the extensions. Perhaps "selective extensions" although that doesn't clearly handle null-move type anti-extensions... > >It is true that we would need some discussion about these terms. The terms >"selectivity" and "full width" are often misunderstood. I agree. full-width is _also_ often misused. IE I would claim that Crafty is a classic 'full-width' or 'brute-force' approach, since it does match the approach Shannon defined. But it is also 'more' because of its selective extensions. Some think full-width means 'fixed depth' too. Very confusing. > >A suggestion, the null move selection could be put in the category of "dynamic >selectivity" (or dynamic pruning) instead of simply "selectivity", to emphasize >on the fact that selection is decided by a search? not bad. Because traditional 'selective search' is based on 'static analysis' to discard moves before they are searched. > >What do all interested readers think? > > > Christophe
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.