Author: Dave Gomboc
Date: 08:46:47 05/16/98
Go up one level in this thread
On May 16, 1998 at 10:27:28, Don Dailey wrote: >>>>>I dunno why Rebel+Fritz+genius and some others >>>>>are so horrible in these exchanges, but my first guess was piece square >>>>>tables. Now i say: perhaps mobility is a reason too. >>>>>The common thing of all these programs compared to Crafty, Zarkov, Diep >>>>>and some others is that they lack mobility terms. For the record, the above was written by Vincent Diepeeven, not myself. >I know this is not correct for Genius. One of the few things Richard >Lang ever "gave up" about his program was some details on the mobility. >He even does queen mobility. Of course this was a few years back, >it's possible he stopped using mobility. > >I am curious about different implementations of mobility. Here is a >quick >rundown of ways I've done BISHOP mobility through the years: > >1) Simple square count method (legal moves) >2) Simple square count where I pretend only pawns on board. >3) Combinations of 1 and 2 folded together. >4) Method 2 with table. >5) Method 1 or 2 but pawn attacked sqaures count less or not at all. >6) Not at all. > >Method 4 is something I found useful. I count the squares but use the >count to index a table. The table is designed to emphasize "minimal >mobility", I consider it more important to go from zero to 5 sqaures >that to go from 5 to 10. This helps a lot in positions where you choose >to "improve" a well developed bishop when you should be doing more >constructive things. I don't have proof this is any better, it's a seat >of the pants thing from observing lot's of games. > >I think method 5 is the very best, but my programs found it quite >expensive to compute, others might do better. I may eventually >implement this in Cilkchess because with bit boards it will probably >be fairly cheap. > >Method 3 seems like an improvement over either 1 or 2, but I believe >method 1 is the least good. But in some positions it makes more sense. >Nothing is completely ideal for every situation. So I generally give >more weight to pawn considerations. I also had a program that didn't >count squares where pieces blocked it but skipped over them until it >ran into a pawn. > >As far as rook mobility is concerned, I believe it's more important >to judge open and half open files. I think rank mobility is useful >if the weight is kept low, the same probably applied to file mobility >but again, prime consideration is openness. One consideration is >attacking enemy pawns. Is it mobility restricting, or is it a good >thing? The answer is both! I try to focus on what I'm measuring, >and I'm not measuring rook bonuses for attacking pawns. Since I'm >measuring mobility, I oount pawns that can be captured as mobile >squares but don't count any farther. I have a separate term for >various attacks on undefended pawns. > >Method 6, I don't know about. Right now cilkchess using this method >and I feel like I have problems too often with bishops. I have some >static" terms that judge a bishop's worth like pawn counts on certain >color squares and a term that attempts to measure a bad bishops >which I have mixed feelings about. Also a table with weak weights >to get the bishops out to the center a little. Sometimes though, >the bishops get confused about where they belong. > >- Don
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.