Author: Thom Perry
Date: 05:24:55 01/28/99
Go up one level in this thread
On January 27, 1999 at 19:21:09, Robin Smith wrote: >On January 27, 1999 at 18:08:03, blass uri wrote: > >> >>On January 27, 1999 at 17:45:16, Robin Smith wrote: >> >>>I have been playing over some chess positions with my computer (Pentium 166) and >>>have found some very strange behavior in the program Junior 5 in some endgames. >>> >>>strange position #1 >>>white Kd5, Pb5, Pc6 >>>black Ka2, Rb1 >>> >>>White to move wins easily with Kc5; but Junior wants to play c7??, thinking that >>>this also wins (evaluation ~=+2.5 pawns). Only after a think of several minutes >>>does it switch to Kc5, but not because it sees that c7? Rxb5+ draws, it is >>>because it decides Kc5 is even better than +2.5 pawns. All of this wouldn't be >>>all that odd since the draw involves a perpetual check, except that once Junior >>>5 plays c7 it INSTANTLY (less than 1 second) correctly sees the position is >>>drawn! >>> >>>even stranger position #2 >>>white Kd3 Ra2 >>>black Kd1 Pe2 >>> >>>Black to move, Junior wants to play Ke1, thinking that white will be mated in 7 >>>after Rxe2+; but e1=N+ draws! No mater how long Junior thinks it doesn't see >>>the underpromotion. >> >>It is a known fact that Junior5 does not consider underpromotions. >>I think that this was done to save time by not considering underpromotions. >> >>I understood that it is not simple to change the code of Junior5 to give Junior5 >>the possibility of considering underpromotions. >> >>> >>>strangest position #3 >>>white Kf6, Be5 >>>black Kh8, Bg8, Pg6 >>> >>>White to move can mate on the move with Kxg6 mate, but Junior 5 shuns both the >>>mate and the win of the pawn, preferring to draw the game with some silly bishop >>>move. This occurs no matter how long you let Junior 5 think! This bug must >>>have to do with Junior thinking that it "knows" B vs. B is "insufficient >>>material" and thus it shuns taking the pawn in the hopes that it might find a >>>win by keeping the pawn on the board. Incredible! >> >>Junior5 can find the win if you change the engine parameters of insufficient >>material. >> >>Uri > >Thanks for the info. I will try your suggestion on position #3. It still seems >to me to be odd behavior for a program to miss mate in one using it's default >settings. But no doubt the insufficiaent material knowlege improves Juniors >play more often than it hurts it. > >Robin You hit it in a nutshell. That's exactly why I have shied away from programs like F5, J5, Rebel 10, etc. How do you know what settings to use for what situations? It's interesting that these programs often find the solutions "after the fact". In other words, someone or some other program finds the solution, then the response is "But if you set AAA to BBB, and CCC to DDD, and turn off anti-this, but activate anti-that, then reduce the hash size...." Wow! I just fire up MCP and run with it. It may not always come up with the strongest solution, but at least I don't have to guess what it's strongest setup mode is. In my opinion, there's such a thing as "too many bells and whistles" which can sometimes give self defeating results. I wouldn't want any chess program where I would constantly be wondering, "Is the setup I'm using giving me the best results?" Better to have less with faith, than more without.
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.