Author: Uri Blass
Date: 04:43:44 02/11/03
Go up one level in this thread
On February 11, 2003 at 06:40:31, Sune Fischer wrote: >On February 10, 2003 at 23:25:32, Robert Hyatt wrote: > >>On February 10, 2003 at 16:21:23, Sune Fischer wrote: >> >>>I'm fooling around with some root node aspiration search, >>>Looks like it is a tad unstable, I wonder if that is normal. >>> >>>It produces different preferred moves at different plies. >>>E.g. normally it prefers e4 at ply 11, but using a narrow window of 0.1 pawn it >>>prefers e4 already at ply 9. Sure the hashtable is going to make a difference >>>here, I realize that, but still a change on the root move is not an >>>insignificant effect. >>> >>>The narrower the window, the more instability. Is this to be expected? >>> >> >>I don't see why. With _one_ possible exception. "lazy evaluation". It >>might cause lazy eval errors if you are not careful. But if you disable >>lazy evaluation, and you get different best move/score at the root based >>on the aspiration window, you have a problem. > >I tried disabling lazy eval, didn't have any effect. But my margin is also >pretty high compared to the positional bonusses. > >>Just beware of things based on alpha/beta, such as lazy evaluation, or >>perhaps some sort of forward-pruning issue (such as null-move). > >Without nullmove and lazy eval it is alot more stable, I still do not get >exactly the same scores but the moves are now the same. > >>If you want to post a position, I can twiddle my aspiration to whatever you >>are testing with and run the same searches and post what happens to me at those >>window widths... > >That would be great actually, just to see if Crafty is completely stable. > >I'm looking for some kind of bug that shows up from time to time. Here in this >position it has a ~0 score, every other program I've tried shows between -1 and >-1.5! It's not that hard to see, even Gerbil gets it. > >[D]r2r2k1/p1p2pbp/nq4p1/1p1bp3/1P1P2P1/P3P2P/2BN1P2/R1BQ1RK1 w - e6 0 18 > >Nb3 is suggested by crafty, ruffy and many others, frenzee wants to play e4. Of >course I'm not surprised it can't find the right move if it can't give a correct >score (or is it the other way around?:). Something must be broken, I just tested >and it's not related to nullmove or lazy eval.. :( > >-S. Are you sure that Nb3 is the right move The score is clearly negative for white but Movei also likes e4(latest movei find it faster than the public Movei but both find e4 after enough search time). Uri
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.