Author: Uri Blass
Date: 12:44:48 11/23/00
Go up one level in this thread
On November 23, 2000 at 14:36:42, Christophe Theron wrote: >On November 23, 2000 at 12:22:52, Fernando Villegas wrote: > >>On November 23, 2000 at 09:48:00, Harald Faber wrote: >> >>>Beta tests will never show up all flaws and bugs a program contains. >>>As well as software is never 100% bug-free. >>> >>>IMO, there are other things to test, more important, than to take a look if the >>>prog can mate with B&N. >> >>Harald: >>What can be more important than to see if a modern program can or cannot mate >>B+N? One more step and you say it does not matter if a chess program cannot >>mate in no way. IMHO, it is one of those basic features you cannot miss. Then, >>is not a matter of asking tester to discover ALL bugs or voids, but at least to >>check some esentials. >>Worse: this is not a bug, but a VOID. Not something running badly, but something >>that does not exist at all. >>I believe the silence of Theron about this issue until now is a kind of >>recognition this is something to be fixed very quick and not something to be >>dispelled just like that.. >>Fernando >>Fernando > > > >Chess Tiger 13.0 does not know how to mate with KBN against K. > >That's something I know since ages. > >If the beta testers had notified me this "problem" during the beta test stage, >it would not have been fixed anyway. Because anyway I already know that, and I >had much more important things to fix first. > >Last year, Chess Tiger 12.0 has topped the SSDF list without knowledge of how to >mate with KR or KQ against K. > >Surprising? > >Actually, given enough time, the search of the program is able to solve the >problem of mating with KR against K. But if you have Chess Tiger 12.0 and try to >set up a KR/K position and give the program very little time to play, you will >see that it will never win. > >In Chess Tiger 13.0 and Gambit Tiger 1.0, knowledge has been added to solve >instantly KQ/K and KR/K. > > >It's not because you, Fernando, have decided that mating with KBN against K is >important that it is really important. > >I have watched my program playing hundreds, thousands, of games, and it has >happened maybe 1 or two times in several years. > >So adding this knowledge would probably mean a 0.5 elo improvement. I can add that one of the reason that adding this knowledge will not give much rating is the fact that programs use tablebases and they will do everything to avoid KBN vs K endgames. They are going to prefer being queen and rook down and only not to to be in a position of K vs KBN. I believe that a better evaluation for the ssdf games is to evaluate K vs KBN as 3 pawns advantage for the side with the KBN if there is no short mate. I sssume that you cannot use the name of the opponent as part of the evaluation and if you can use the name of the opponent then it is better to evaluate it as a loss when the name of the opponent is Fritz6 or Rebel or Junior6 or other examples and to evaluate it as a draw or almost as a draw when the name of the opponent is Junior5 or one of the tigers or other cases when the opponent does not know to mate with bishop and knight(I assume that the tigers usually cannot solve it by search in the ssdf games). 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.