Author: Robert Hyatt
Date: 09:44:46 12/13/00
Go up one level in this thread
On December 13, 2000 at 12:17:20, Ed Schröder wrote: >On December 13, 2000 at 11:31:01, Robert Hyatt wrote: > >>On December 13, 2000 at 06:16:54, Uri Blass wrote: >> >>>On December 13, 2000 at 05:41:34, Günther Simon wrote: >>> >>>>On December 13, 2000 at 02:17:16, Uri Blass wrote: >>>> >>>>>On December 13, 2000 at 01:33:25, Jouni Uski wrote: >>>>> >>>>>>Oh not again!! This trojanishe Esel (g5 sacrifice). >>>>>> >>>>>>Jouni >>>>> >>>>>I think that avoiding hxg5 is a good test position for chess programs. >>>>> >>>>>I am surprised that shredder falled into this trap in tournament time control >>>>>because I know that one of the things that was improved in shredder5 relatively >>>>>to shredder4 is tactics. >>>>> >>>>>I believe that other chess programs may do better. >>>>> >>>>>Junior6 probably can avoid hxg5 on the same hardware at tournament time >>>>>control(I say probably because I have not Junior6 and I only checked that >>>>>Junior5.9 can see 0.00 evaluation for hxg5 at depth 17 and later changes its >>>>>mind to dxc4) >>>>> >>>>>Uri >>>> >>>>As far as I know some progs now have implemented an Anti-Trojan function. >>>>One of them must be Crafty. I am just wondering why it is not implemented >>>>in all newer chess software?! >>>> >>>>Günther >>> >>>I know that the anti trojan function is not used by crafty against copmputers. >>> >>>The antitrojan function can cause problems if the program will be afraid to take >>>g5 when it is good to do it so I dislike it. >>> >>>I prefer to see anti-trojan extensions when the computer discovers trojan >>>situation(positions when hxg5 hxg5 is possible at the root) >>> >>>It means that if hxg5 hxg5 is possible(I assume the side who sacrifice is white >>>without loss of generalization) then the program will consider the typical >>>trojan attack moves as 0 plies in the search. >>> >>>Usually taking the piece at g5 is the tactical mistake so doing the extensions >>>only when hxg5 hxg5 is possible at the root can solve the practical problem >>>without doing the program slower in most of the cases. >>> >>>Uri >> >> >>The trojan horse detection is now automatic in crafty. If it sees the key >>elements of the trap, at the root position, then it will enable the anti- >>trojan code and in the game given, it will _never_ consider hxg. If one or >>more key elements are missing, then it will rip it given the chance, because >>it will be safe. > >As you see Uri "sometimes" pre-processing can be very useful. > >Ed :) Bob PS. The reason I did this is that the trojan detection code was fairly simple in nature, noticing that the h-file was open, and that the opponent had a queen somewhere and a rook on the file. This was enough to prevent it from playing hxg and opening the file, but then later it would screw up badly if the file got opened in a different way. Or if the file was threatened to be opened. Now the code is only triggered when the piece is sitting on g4/g5, and it discourages Crafty from taking it. Once the piece is taken, the code is turned off as it is then too late. Of course, Crafty will never take the piece, but the nice thing is that if the file gets opened in a different way, the anti-trojan stuff isn't turned on and the huge scores don't confuse things. It already knows that the open file is bad... but the score wasn't big enough to prevent taking the piece. Now I get the best of both worlds. Even if the solution is a bit "ugly"... But at least, it is now on duty 100% of the time, without requiring any manual intervention when someone starts to use it.
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.