Author: Vasik Rajlich
Date: 08:07:36 02/06/04
Go up one level in this thread
On February 06, 2004 at 10:49:11, Tord Romstad wrote: >The ongoing discussion about mobility evaluation has given me a new idea >which could perhaps be of interest to some: > >One of the many difficult problems in mobility evaluation is to decide, for >each square a piece can move to without getting captured, how big a bonus one >should give for being able to move there. One of the main criterions I use >is whether the piece would contribute to the attack of the enemy king or the >defense of the friendly king from the square. This helps my engine to >notice attacking potential some time before the attack materializes on the >board. I also give an extra mobility bonus for pieces which have the potential >to do more than one task, for instance if the piece can move to squares where >it attacks the enemy king *and* to squares where it defends the friendly king. > >My new idea is this: Why not compute two different mobility scores for each >side? It seems more flexible to compute a separate "offensive mobility" and >"defensive mobility" for every piece. One could then use other components >of the eval to decide upon the weightings for offensive vs. defensive mobility >for each side. > >Has anybody tried something like this before? How well does it work? > >Tord I do something slightly similar. I score each piece according to how close it is to the enemy king, and to its own king. These scores are fed into a function which also takes into account the material on the board, holes in the pawn structure, and the difference in "attacking force". (For example if there are two pieces vs rook and pawn, the two pieces are a greater attacking force.) Unfortunately, everything is in an embryonic stage and I can't really say if it's any good. The key to having a good evaluation is coming up with some way to test it, piece by piece. Self-play is not enough, you'll never play enough games to show a 10-point improvement. Vas
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.