Author: Omid David Tabibi
Date: 13:48:56 06/15/04
Go up one level in this thread
On June 15, 2004 at 16:44:47, Tony Werten wrote: >On June 15, 2004 at 11:41:30, Omid David Tabibi wrote: > >>On June 15, 2004 at 07:48:48, Tony Werten wrote: >> >>>On June 15, 2004 at 04:52:53, Omid David Tabibi wrote: >>> >>>>On June 14, 2004 at 21:45:44, Anthony Cozzie wrote: >>>> >>>>>On June 14, 2004 at 20:12:09, Omid David Tabibi wrote: >>>>> >>>>>>On June 14, 2004 at 19:20:57, Anthony Cozzie wrote: >>>>>> >>>>>>>On June 14, 2004 at 16:31:17, Jaime Benito de Valle Ruiz wrote: >>>>>>> >>>>>>>>[d]8/8/1p4p1/p5kp/Pp1QP1Pq/1P6/6K1/8 w - - 0 62 >>>>>>>> >>>>>>>>White moves and wins! >>>>>>>>Can your engine find the right answer and give the proper "crushing" score >>>>>>>>(around 9 pawns!) along with a proper PV? How long does it take? >>>>>>>>Mine doesn't find anything after a long long time (over an hour), but with >>>>>>>>null-move disabled, it takes only a few seconds to show the full PV. >>>>>>>>(Some engines find the right answer, but with a small score, an absurd PV, and >>>>>>>>they change their mind some plies later) >>>>>>>> >>>>>>>>In this position the black king is in a very delicate position: After a couple >>>>>>>>of moves, if the black queen moves, white checkmates inmediately, so both kings >>>>>>>>and queens must remain still while pawns play on their own. >>>>>>>> >>>>>>>>Any good suggestion to avoid this problem with null-move? >>>>>>>> >>>>>>>>Regards, >>>>>>>> >>>>>>>> Jaime >>>>>>> >>>>>>>Zappa doesn't get it either (gxh5 @ 15 ply after 1 minute). Zappa relies on >>>>>>>double nullmove for zugzwang avoidance. This works find for positions with 1 >>>>>>>zugzwang, but in this position black gets hit with about 10 of them :) >>>>>> >>>>>>Ehem, ehem, heard about verified null-move pruning? :) >>>>>> >>>>>> >>>>>>> >>>>>>>anthony >>>>> >>>>> >>>>>It's a possibility. If I understand it correctly verifications performed >>>>>nonrecursively. >>>> >>>>The original formulation had problems with detecting several zugzwangs in a >>>>line, and I think it might have failed to find the correct move here. One of the >>>>things I changed in the new formulation is to make the algorithm find as many >>>>zugzwangs there might be along a line. >>> >>>Did you split black and white zugzwangs now ? >> >>I'm not quite sure what you mean, but the algorithm works the same for both >>white and black :) > >In your original paper, you said that after a fail high, you reduce depth, and >during this reduced search, you accepted new fail high on nullmoves as ok. > >IMO this was wrong because you could be doing a verification search for white, >but accepting a nullmove fail high for black. Yes, I solved that problem in the new formulation. > >Tony > >> >> >>> >>>Tony >>> >>>> >>>> >>>>> >>>>>Right now, my goals are "Deep Zappa" aka "ParZappa" aka "Zappa-NUMA" aka "Zappa" >>>>>and the autotuning stuff; when I get some free time I'll try to give it a shot. >>>>> >>>>>Oh, and I have to rewrite my eval again. >>>>> >>>>>anthony
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.