Author: Robert Hyatt
Date: 07:13:59 03/05/06
Go up one level in this thread
On March 05, 2006 at 05:31:05, Chan Rasjid wrote: >On March 04, 2006 at 20:12:47, Robert Hyatt wrote: > >>On March 04, 2006 at 18:26:47, Chan Rasjid wrote: >> >>>This is for an earlier thread somewhere. >>> >>>It may be piece12-to may be better than from-to for history stuff. >>> >>>A side may attempt to retain a certain board configuration or a part if it >>>strong >>>especially near the palace. Then such spaces may be good or bad for certain >>>piece-type-color and the from sqaure has no relevance, only the type and color. >>> >>>Also in endgame, the significant pieces are pawns and kings and the above >>>analysis >>>may also apply. >>> >>>The above non-random goodness or badness of squares towards type-color will give >>>the history value a bias towards fh/fl and it is this non-random bias that makes >>>history useful. >>> >>>Rasjid >> >>I tried both and could not get piece-to to out-perform to-from. They were >>close, but to-from was always just a bit faster for me. I ran several test >>suites, including ecmgcp for testing, plus some non-tactical positions to make >>sure they didn't behave differently... >> >>This probably varies by engine, no doubt... > >history[from + piece12][to] - can this work? >This is only my rough idea, may be complete silliness...and I am busy with >my bitboard generation. I actually tried <piece><to><from> (a 15 bit value) just to be thorough, but it offered nothing over <to><from> and makes the array much larger... I have not tried the approach you suggested... > >There is "collision" for from1 + pc12_1 == arr2 + pc12_2 >when from1 != from2 and pc12_1 != pc12_2.But maybe things may tubalanced itself >out and stiil may be best of the 3 history > >With from-to, there is already collision of :- >a) B 2 color, Q 2 color, P 1 color >b) R 2 color, Q 2 color, P 1 color >c) N 2 color, else ok. > >Maybe a brute run for collision can help, ie using a brute loop of >[from+pc12][to] to flick out the collision characteristic, etc.. to see if they >are a problem. Or maybe adding a suitable constant K to >[from + pc12][to + K] may do the trick. > >Rasjid
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.