Author: Robert Hyatt
Date: 07:11:36 07/08/01
Go up one level in this thread
On July 08, 2001 at 10:09:04, Robert Hyatt wrote: >On July 07, 2001 at 10:07:40, Brian Richardson wrote: > >>Hi >>Tinker scores book moves (w/l/d formula) and also uses a preferences file (? >>move flags) to avoid certain book moves, but only when in book. Sometimes the >>last possible book position has no book moves that Tinker would play with these >>constraints, so it falls into the normal search. Unfortunately, sometimes the >>normal search picks one of these book moves it already tried to avoid. Book >>moves flagged ? can simply be removed from the root move list (when not leading >>to check/mate issues), but how do others avoid book moves merely scored as poor? >>Thanks >>Brian > > >I don't do anything about this at the present. I tried a few things but >all were worse. IE you _could_ generate your ply-1 move list, and delete >the moves that are in your book file. For me, this was horrible. Imagine >a case where your opponent plays BxN, the only recapture is QxB, but you >lose that game and flag the move as bad. Now you just end up a piece down >as that move can't be played. It is better to propogate that "bad score" >back up the tree a bit. > >One thing I should do that I am not (yet) is that if all moves at this ply >are not playable, then I should back up and make the move two plies back >not playable for the same reason. And this should be done recursively to >back up to a point where there are options... > >Bob I should add that with my current book learning, if _all_ moves at a ply are unplayable, then the move two plies back _should_ also be unplayable due to the way I back up scores in the learning. But this is not (yet) guaranteed to happen.
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.