Author: Robert Hyatt
Date: 07:07:07 12/10/03
Go up one level in this thread
On December 10, 2003 at 09:41:18, Bob Durrett wrote: >On December 09, 2003 at 23:22:12, Robert Hyatt wrote: > >>On December 09, 2003 at 19:36:05, Bob Durrett wrote: >> >>>On December 09, 2003 at 13:21:35, Robert Hyatt wrote: >>> >>>>On December 09, 2003 at 13:02:56, Bob Durrett wrote: >>>> >>>>>On December 09, 2003 at 11:13:56, martin fierz wrote: >>>>> >>>>>>On December 09, 2003 at 10:50:23, Sune Fischer wrote: >>>>>> >>>>>>[snip] >>>>>> >>>>>>>If the bare engine had been playing he would have had to add a few things the >>>>>>>GUI normally takes care of. >>>>>>>For UCI engines it is expected that the GUI handles certain (trivial) things. >>>>>> >>>>>>claiming a draw on 3-fold repetition is *not* a trivial thing. there are >>>>>>different possible cases: >>>>>> >>>>>>1) if your opponent avoids it, he loses >>>>>>2) if your opponent avoids it, he wins >>>>>> >>>>>>in case 2) you should of course claim the draw, because perhaps he will notice >>>>>>he could avoid it. in case 1) however, you can safely repeat the moves, and not >>>>>>claim the draw. it is *not* mandatory to claim a draw on the 3rd repetition. so >>>>>>you should basically not claim it if you might win if your opponent avoids the >>>>>>draw. >>>>>> >>>>>>how do you expect a GUI to make the right decision? imagine the following >>>>>>absurdity: jonny is running without GUI and happily repeats moves against >>>>>>shredder, and does not claim the draw because the engine doesn't know about it. >>>>>>shredder has a bug and allows a 3-fold repetition but will deviate before the >>>>>>fourth repetition. now shredders GUI stops shredder from moving, and says "i >>>>>>claim a draw with my move XY because of 3fold repetition" - this would have been >>>>>>hilarious for everybody except SMK :-) >>>>>> >>>>>>since 3fold repetition is something you claim or don't claim based on the >>>>>>current position, it is clearly something the GUI shouldn't be doing! >>>>>> >>>>>>cheers >>>>>> martin >>>>> >>>>>THIS suggests the obvious changes which should be made to engines and GUIs ASAP >>>>>by all chess programmers. >>>>> >>>>>Before an engine &/or GUI claims a draw, it should evaluate the position and >>>>>determine whether or not it has a strong advantage. >>>> >>>>The engine already _does_ this. It searches every root move individually >>>>and chooses the one that produces the best score. If you get a draw >>>>score back, you can safely assume that no other move will give you a >>>>"strong advantage" since the score of 0.00 was better than any other move. >>>> >>>>QED. It chose the drawing move, thinking a draw was the best outcome >>>>possible in this particular position. >>>> >>>> >>>>> If it does have a strong >>>>>advantage, then claiming a draw would be precluded by the programmer. In other >>>>>words, the software would be programmed in advance to make the sensible choice. >>>> >>>>Which it already does, as I explained. >>>> >>>> >>>>> >>>>>Similarly, the engine/GUI should be programmed to claim a draw by repetition in >>>>>cases where repetition can be played and when also the engine evaluates the >>>>>position as being a strong disadvantage. >>>>> >>>> >>>>The engine will evaluate the position as 0.00 in the above case. >>>> >>>> >>>> >>>>>Obviously, humans avoid claiming repetition draws whenever they are winning and >>>>>claim repetition draws whenever they can if they are losing otherwise. In >>>>>positions perceived to be equal, humans may or may not claim the draw by >>>>>repetition depending on other factors such as tournament standing. >>>> >>>>If you play a move that repeats for the third time. You can claim the >>>>draw. Or your opponent can claim the draw immediately when it is his >>>>move. Playing a repetition for the 3rd time and wishing your opponent >>>>would not notice reminds me of "Grumpy old men". "You can wish in one >>>>hand and crap in the other, and see which one fills up first." >>>> >>>>:) >>>> >>>>> >>>>>Chess computers should be programmed similarly. >>>>> >>>>>Bob D. >>> >>>Bob Hyatt: I don't mean to be unkind, but perhaps you should read my bulletin >>>again. >>> >>>Bob D. >> >>I also don't mean to be unkind, but must respond "why". Do you think I >>missed something or misunderstood something? >> >>This nonsense about playing a 3-repeat move and hoping the opponent won't >>see it is totally ridiculous in the context of alpha/beta searching that we >>are all using. > >Another relevant thought: > >I have seen engines play a move [and display 0.00] which is obviously intended >to give the opponent an opportunity to repeat the position a third time. Upon >further extensive analysis, I have sometimes found that the engine's doing so >was a terrible mistake. Two types of "terrible mistakes" can occur: > >(a) the engine could have played a much stronger move but didn't, or >(b) the opponent engine does not repeat the position a third time but instead >plays a much stronger move which could have been prevented. > >In my view, engines which make these mistakes have "bugs" in them, or simple >programming errors. > >Bob D. Let's define the context precisely here. I have seen what you describe. But it is not a bug. Any more than an engine grabbing a pawn that later loses the game is a bug. The context is "the engine search space". If this search space says "repeating for a 3rd time to get a draw score is the best I can do" then that is the best the engine can do, within its search horizon and search space. That isn't a "bug". It is a "shortcoming" of insufficient depth or knowledge. As a human have you ever played a move that you later discovered was bad? Was that a "bug" in your neurons, or just a lack of search or understanding?
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.