Author: Heiner Marxen
Date: 10:53:25 04/27/03
Go up one level in this thread
On April 27, 2003 at 10:23:17, Geoff wrote: >Hi > >I had a study of this problem and came to a different conclusion. That test >position is interesting as in the solution there is an actual board repeat at >ply 5. It is then black to move and a mate for white in 2. Note if the test >position had of been black to move first it is mate for white in 2 instead of a >mate in 5 (A kind of a zugwang position I guess) Hence, a repetition detection must include the "side to move" with the board. >I decided the fault in 1.73 was due to the lines in the actual search function > >/* if this isn't the root of the search tree (where we have to pick a move and >can't simply return 0) then check to see if the position is a repeat. if so, we >can assume that this line is a draw and return 0. */ > >if (ply && reps()) > return 0; > >To me this seems unsound as it will cause it to miss the mate in this test >position. I could be wrong but I dont think the problem was in the reps () >function at all. > >But because version 1.83 has those same 2 lines in but still finds the mate >correctly, to me this then implies that the new hash version of the reps() >function is flawed ? I.e it doesnt see the repeat at Ply 5 ? > >If anyone is still following me ;-) I would be interested in you agree or >disagree? I disagree insofar as the unchanged lines most probably are not the cause for differing behaviour. I rather suspect that the reps() function in V1.83 is more accurate with respect to the "side to move". Cheers, Heiner > Regards Geoff > > > >On April 26, 2003 at 17:10:32, Geoff wrote: > >>Hello Tom >> >>Thanks for putting me out of my misery, I still hadn't spotted the problem ;-) >> >>You are correct that was what was causing the problem, I commented out the lines >> if (ply && reps()) >> return 0; >> >>just to prove the point and it then finds the correct solution >> >>10: 9991 4034 30732760 g5c5 h5g5 c5f5 g5h5 f5g5 h5g5 a5g5 a6a5 g5a5 >> >>it makes sense that this test position showed up the problem as black has a move >>sequence g5h5 h5g5 and white has a sequence a5g5 g5a5 even though there isn't an >>actual board repeat. >> >>Looks like that clever algorithm for reps detection was to too smart for its own >>good. Pity as it looked speedy, guess I will have to put the hash code in now >>instead, or maybe spot the flaw in the original algorithm >> >>Thanks for the help. >> >> Geoff >> >> >>On April 26, 2003 at 15:02:47, Tom Kerrigan wrote: >> >>>I wrote you an e-mail about this, but might as well post as well. >>> >>>The change between 1.7 and 1.8 in the search code is the repetition detection >>>method. I know the 1.7 rep detector was buggy, so it must have thought there was >>>some sort of repetition in the mating line. >>> >>>-Tom
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.