Author: Dann Corbit
Date: 15:39:28 12/28/05
Go up one level in this thread
On December 28, 2005 at 17:11:30, David Dahlem wrote: >On December 28, 2005 at 16:24:53, Dann Corbit wrote: > >>On December 28, 2005 at 16:03:47, David Dahlem wrote: >> >>><snip> >>> >>>>If you have a bitboard program, there is a way to get really cheap mobility >>>>scores. >>> >>>Could you elaborate on this method? >> >>Precalculate for every ray, based on the bit pattern on that ray. >> >>Move generation is also helped by this idea (both for capture/promotion and >>non-capture). >> >>consider this epd string: >>[D]4k3/1n6/8/pR5b/1P6/1n6/1N6/4K3 w - - >>For the rook's row, we have 1x000001 ( the x is because we ignore the square he >>sits on, he can never attack or defend that.) >>For this simple case, the non-captures are the zero squares and the captures are >>the non-zero squares anded with the opponent chessmen mask. >>Regardless of what the chessmen are, there are 5 non-captures and 0,1,or 2 >>captures depending on whether the 1 bits are friend or foe. >> >>For the column we have 010x1110. >>There is one non-capture and 0,1, or 2 captures depending on whether the nearest >>one bits are friend or foe. >> >>You get the idea. > >Thanks Dann. Mobility should only be considered for bishops, knights, rooks, and >queens, is this correct? I would test it in your eval for benefit. Tord does not calculate queen mobility, for instance. If you use bitboards, you can calculate the total pawn mobility in a couple operations. I think it is a lot more important in the endgame (e.g. down a piece with +3 pawns and your pawns are rammed sounds like trouble). >And I'm thinking the bonus should be scaled according >to piece values? I do a fitted curve using parabolic least squares fit from actual data. At least it gives you a good starting point. YMMV.
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.