Author: Heiner Marxen
Date: 09:21:38 10/05/02
Go up one level in this thread
On October 04, 2002 at 19:32:30, Omid David wrote: >On October 04, 2002 at 19:09:22, Jeremiah Penery wrote: > >>On October 04, 2002 at 18:06:21, Omid David wrote: >> >>>On October 04, 2002 at 17:27:05, Jeremiah Penery wrote: >>> >>>>On October 04, 2002 at 16:51:38, Roy Eassa wrote: >>>> >>>>>On October 03, 2002 at 19:18:16, Jeremiah Penery wrote: >>>>> >>>>>>[D]8/7p/6pP/k4pP1/b1p1pP2/KpPpP3/1P1P4/7Q w - - 0 1 >> >>>No blockade-detecting algorithm should announce a draw in this position, since >>>the blockade can be "broken". And I think there might be a victory for white >>>hidden: >>> >>> >>>1.Qd1 >>> >>>A) 1... Kb5 >>> 2.Qh5 >>> i) 2... gxh5 >>> 3.g6 hxg6 4.h7 white winning >>> ii) 2... Ka5 >>> 3.Qxg6 hxg6 4.h7 white winning >>>B) 1...Bc6 (or any other bishop move) >>> 2. Qxb3 cxb3 3.Kxb3 unclear situation >> >>My reasoning is that it's way harder to detect a blockade by search than it is >>to detect a winning line in some "blockaded" positions. In this position, if >>Qd1 Bc6 Qxb3 etc. is winning for white, the search will find it pretty quickly. >>However, without detecting this as a blockade, the program will NEVER detect the >>draw, and it will shuffle the pieces around forever, while declaring itself >>winning by 6 pawns or more. Either way, the end result will be a draw. >> >>My 'algorithm' is a pretty simple thing. I just wanted it to be fast, and not >>miss any cases, though I didn't care too much for false cases (because of the >>above). Unless my thing is 100% correct, and never misses any potential >>blockade draws, there will be error anyway (without the algorithm, you will say >>+6 in the position above, and many similar positions, though it is a draw - with >>the incorrect algorithm, you may say draw in some winning position). >> >>If there are cases where my stuff is completely and totally wrong, I will try to >>fix it. Otherwise, I'll take some inaccuracy and be reasonably happy that it >>works most of the time. > >Your perspective is really interesting, which allows a certain amounts of >"unharmful" inaccuracies. > >My method on the other hand, refrains from declaring a draw, even if it's 99.99% >sure that it actually is. Anything less than 100.00% accuracy, and the returned >answer will be "I don't know!" Now, that is quite an interesting proporty, since it could be used in a mate finder/prover like Chest. I'm interested to see your source code! From your web site I see that you do this as your MSc. Any chance that your source will be available sometime? Cheers, Heiner
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.