Author: Robert Hyatt
Date: 08:42:50 09/17/99
Go up one level in this thread
On September 17, 1999 at 07:06:21, Georg v. Zimmermann wrote: >> >>Something to experiment with is folding the score in toward zero based upon the >>50-move counter. If pieces are coming off and pawns are moving, it is likely >>that progress is being made. Once they stop coming off and pawns stop moving, >>it's very possible that you're headed for a 50-move draw. I'm not talking about >>5 or 10 moves, that seems to happen in middlegames, but maybe around 15-20 it >>should start kicking in slowly. > >I like that idea better than my original idea (reducing overall score towards 0 >if all part-evaluation stay the same). >I would like to see it (as experiment) implemented in Crafty to see the effect. >I don't think thats hard to do. I already have lots of draw folding going on. Chess 4.x and Nuchess were the first to do what was suggested above... Slate called this a "weariness factor" in that if the score wasn't slowly going up, getting closer to the 50-move rule would pull it slowly down. I don't do this now, but did for a long time way back (in Crafty). You might check the comments in main.c to see when it was added and when it was removed, I don't recall the circumstances when/why now... But it is definitely a _very_ old idea dating back to the middle 70's at least... > > >> >>It's not necessarily true that the draw is going to happen if you just let the >>program play. When you get near 50-moves, the program will hit a brick wall -- >>it will see 50-move draw scores. At this point it might not have enough horizon >>to try something new, it might allow a draw, or try to make some progress out of >>panic. It would be nice to have the program program panic more gradually, so >>perhaps it can do something before it's had to put its pieces on dumb squares in >>order to avoid repetitions. > >Exactly, there are 2 drawbacks to how it works now: > >1.) We humans have to wait 42 moves :-( > >2.) Sometimes, in many endgames, the remaining 8 moves arn't enough to try >something new.
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.