Author: Robert Hyatt
Date: 11:07:21 12/09/03
Go up one level in this thread
On December 09, 2003 at 13:50:14, martin fierz wrote: >On December 09, 2003 at 12:32:15, Robert Hyatt wrote: >[snip] > >>I don't see how this matters since the engine doesn't understand the concept >>of "my opponent might make a mistake." > >and this is exactly where the problem is. i understand it, and you understand >it, but 99% or 100% of all chess engines don't understand it. which doesn't >change matters: if you have a draw "in hand", you don't claim it. you let the >other guy decide whether he claims it or makes a mistake. computers will say the >position is 0.00. it is >= 0.00. of course it is 0.00 if the opponent plays >well. but you don't know that he will play well, so you have him prove it. >even if your engine doesn't understand the concept, your opponent can still make >a mistake :-) yes. But back to the real-world. Alpha/beta doesn't do that. It only notices that the draw is the best score it can produce. It has no way to pass back information like "this is a repetition, but the opponent can blunder and lose if he isn't careful" vs "this is a repetition, and if I don't claim it, my opponent might choose to deviate and win, because he screwed up to let me repeat the position for the second time." IE the Jonny case. I only know it is a draw, and I would claim it (using alpha/beta) because I can't tell what happens if I don't, as the search doesn't tell me. > >[snip] > >>>there are so many positions where both sides are forced to repeat, because they >>>lose if they don't. e.g. i sac something, get a perpetual and if you go on the >>>wrong square with your king i win. if i make a quite move you defend against the >>>perpetual and win. this is a very common situation. i will definitely not claim >>>the 3fold repetition in such a position, because you just *might* go on the >>>wrong square. i can't do anything about it, the game is a draw. but since you >>>can still go wrong i won't claim the draw just yet, i have no reason to. i can >>>always claim it later. >> >>This does not happen in computer vs computer games. They resolve this at the >>_second_ repetition, they don't have to wait for the third and then play through >>it to fix the problem. > >replace "this does not happen" with "this should not happen in c-c games". for >example, you can imagine the situation where it's hard to see that avoiding the >draw is a loss, in fact, it first looks like a win. you're searching that move >which avoids, then you fail low, allot more time, and the next time you search >that position you're in time trouble and play the move which looks like a win >but is a loss. this is possible if you clear your hashtables, for example.... >besides, there is always the possibility of a bug. e.g. shredder repeated the >winning position 3 times, instead of 2. according to your theory, "this does not >happen". of course it does, it was a bug. if it happens to SMK, it can happen to >anybody ;-) Shredder didn't do it intentionally. It had a bug. Jonny intentionally said "I want that draw, as any other move loses". It seems pretty clear to me how it should have been handled. > >cheers > martin
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.