Author: Vincent Diepeveen
Date: 14:39:17 09/17/01
Go up one level in this thread
On September 17, 2001 at 13:06:09, Bruce Moreland wrote: >On September 17, 2001 at 03:45:14, Vincent Diepeveen wrote: > >>DB didn't play a superb move here that is the bottom line. Some idiot >>who can't play chess looked at the game and concluded that f3 is a superb >>move because after a while several programs with bad king safety also >>get it at the end. > >My program switched to f3 eventually. That was the point of the post that I >made yesterday, after running the position for a long time, as *requested* by >you. > >So now you tell me that my program has bad king safety? No. your programs score for f3 and the other move was very similar. What you see is that f3 is nearly -5.0, now for gxf4 you also get the same. I wanted the -5.0 for f3 confirmed. That's all i needed, because how easily it technical is i already know myself. If you are also not happy with gxf4, which leads to a very complex position, then that's not interesting (unless it's like -mate in 10 or so). >I'm assuming that I'm misunderstanding something here, because I don't think you >would be saying this. I would venture a guess that if you were to check out >everyone's king safety, Ferret's would be pretty good. What i wanted to show and i think i'm succeeding here is that f3 is DEFINITELY not the way to go! I also realize that ferret just like diep just like shredder and all other strong programs will have problems to see that the remaining pawns after f3 Bh5 and some exchanging into endgame will be taking quite some time to get off the board. in the meantime they count as +1.0. So if you get -5.0 for a move like f3 knowing that the remainder of the pawns will be taken away very quick, then that says QUITE a lot to me! >>Ferret comes very far for f3 Bh5 because it sees it tactical very far. >> >>Nimzo and genius have a too weak endgame + king safety (preprocessor >>problem here) to see that after f3 either e3 or Bh5 is very bad for >>white and therefore play it. >> >>My assumption is that Deep Blue didn't saw it tactical and went for the >>f3 e3 endgame and evaluated this just like nimzo does: not so bad for >>white initially. Only -1.xx or so. Of course later in the game that >>went down when kasparov blindfolded won the endgame with a piece more. > >I conclude that you didn't read my post. After something like 18 hours, mine >wants to play f3, too. > >The reason is that gxf4 LOSES. So does f3. My program gets -5 in both cases >eventually. > >You know what move it eventually chose? Kg1, which failed high at -4.7. What >do you make of *this*? > >If you see all the way to mate or massive loss, and all moves lead to it, >programs as designed today will just pick one of the losses, not necessarily the >easiest one for a human to see. > >There is a famous game, Duchess vs Kaissa, 1977, where the following position >was reached: > >[D]Q5k1/4rp1p/1p1q1bp1/1B1n4/3Pp1P1/4B2P/PP3P2/2R3K1 b - - > >In this position, black played Re8, hanging a rook, rather than Kg7, and >initially this was regarded as a bug. Eventually they figured out that Kg7 >loses to mate in 5. > >Was it wrong to play Re8? If you do a short search with no extensions, it seems >really bad, so you might conclude that the program was bad. > >Same deal here, in my opinion. You can't criticize a program because it sees >that it is screwed before you do. > >If you'll recall, that's what caused the altercation I had with that Chessbase >beta-tester "author" who was operating Hiarcs in 1999. He declared my program's >moves "mistakes" when it was losing. It wasn't making mistakes, it was just >trying to avoid particularly bad terminal positions, and if it needed to give up >material at the root in order to do this, it did. > >bruce
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.