Author: Bob Durrett
Date: 17:39:21 12/03/02
Go up one level in this thread
On December 02, 2002 at 22:27:07, Bob Durrett wrote: > >Intuitively, is seems that it should be possible to write a program to do the >following: > >(1) Examine a given position to generate good "fantasy positions" based, >perhaps, on use of a pattern recognition scheme. A library of relevant patterns >might be called up out of memory and then a "very smart" subprogram would >synthesize candidate fantasy positions. This synthesis might utilize a routine >to maximize the position evaluation score, where placement of the available >pieces is treated as a variable. > >(2) Analysis to find possible legal move sequences leading from the "given >position" to the currently considered "fantasy position." Legal move sequences >would be an output. > >(3) Conventional search engine analyses whose purpose would be to determine >whether or not any of the legal move sequences would be playable. The main >purpose would be to rule out legal move sequences which would not likely occur. >This would happen if the intermediate positions had "best moves" which were >different from the moves in that sequence. > >Maybe some of this could be done more or less concurrently and/or interactively. > >Admittedly, this is a "half-baked" idea. It seriously needs refinement. > >Could any of the programmers here at ICD CCC produce such a program if they >wanted to? > >Bob D. Since there has been no response, I assume that I failed to communicate. Fantasy positions considered in human chess are closely related to the real position under evaluation. The requirement, that useful fantasy positions must be reached by legal move sequences from the position being evaluated, forces people to consider modified forms of the real position, but with the available pieces re-arranged somewhat on the chess board. Generally, repositioning of ALL of the pieces and pawns would never be considered. It is much more usual to consider repositioning of only a few. The mental process involved here treats the position of the pieces [and sometimes pawns] as an independent variable. The first problem seems to be to identify good fantasy positions which satisfy the above requirements. There may be more than one way to do that, I don't know. To avoid wasting time on evaluation of bad fantasy positions, it might be best to use a position evaluation function to quickly look at the alternatives. Limited conventional searching to identify inherent tactical features might also make sense. This could weed out many unpromising fantasy positions. The use of a library of relevant position fragments may or may not be a good idea. The overall concept is to try to emulate the human process of imagining fantasy positions followed by evaluation of their merit and identification of good ways to reach those positions. Would a chess program doing that be competitive with conventional chess engines? Probably not. But the real question is whether or not a decent chessplaying program could be produced by using the idea. Bob D.
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.