Author: Robert Hyatt
Date: 05:56:43 08/14/00
Go up one level in this thread
On August 14, 2000 at 03:01:08, pavel wrote: >On August 14, 2000 at 02:03:03, Uri Blass wrote: > >>On August 14, 2000 at 00:10:22, Robert Hyatt wrote: >> >>>On August 13, 2000 at 23:41:03, Uri Blass wrote: >>> >>>>On August 13, 2000 at 20:51:38, Robert Hyatt wrote: >>>> >>>>>On August 13, 2000 at 18:47:42, Alvaro Rodriguez wrote: >>>>> >>>>>>On August 13, 2000 at 18:20:33, Robert Hyatt wrote: >>>>>> >>>>>>>On August 13, 2000 at 16:21:34, Uri Blass wrote: >>>>>>> >>>>>>>>On August 13, 2000 at 15:12:25, Christophe Theron wrote: >>>>>>>> >>>>>>>>>On August 13, 2000 at 11:42:57, Mike S. wrote: >>>>>>>>> >>>>>>>>>>On August 13, 2000 at 10:59:22, pete wrote: >>>>>>>>>> >>>>>>>>>>>(...) >>>>>>>>>>>[D]3r1rk1/2p1Rppp/p4n2/1p1b4/3P4/3B3P/PPPN2P1/4R1K1 b - - 0 1 >>>>>>>>>> >>>>>>>>>>It seems to me that Tiger, when playing 21...Bxa2?, cannot have expected 22.b3. >>>>>>>>>>Maybe he expected something like 22.Rxc7 Rxd4 23.Ra1 Bd5 24.Rxa6 or similar. I >>>>>>>>>>would be interested if Tiger "knows" this standard motif of locking up a bishop >>>>>>>>>>after it captured a border pawn on the 2nd (7th) row. I think, in such cases the >>>>>>>>>>lines beginnig with b3 etc. should be examined more closely than usual (?). >>>>>>>>>> >>>>>>>>>>Regards, >>>>>>>>>>M.Scheidl >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>>Chess Tiger 12.0e has a partial knowledge of this "standard motif". It knows >>>>>>>>>that the bishop is in trouble if it cannot leave a2, but the evaluation penalty >>>>>>>>>I give in this case does not prevent it to take the pawn. >>>>>>>>> >>>>>>>>>That means that if another move could lead to a positional advantage, Tiger >>>>>>>>>would play the other move. If there is no such move, Tiger will take the pawn >>>>>>>>>with the bishop. >>>>>>>>> >>>>>>>>>I know it sounds a little bit strange, but I have been thinking about this >>>>>>>>>problem for quite a while, and I have not found a good solution. For every >>>>>>>>>example of a trapped bishop that gets lost I have seen the opposite example >>>>>>>>>where the trapped bishop eventually escapes or completely shreds the side it has >>>>>>>>>been trapped in, which leads to a big pawn majority and a winning endgame. >>>>>>>> >>>>>>>>The question is what happens in cases when you cannot find by a search of few >>>>>>>>minutes that the bishop can escape and cannot find by a search of few minutes >>>>>>>>that the bishop is trapped. >>>>>>>> >>>>>>>>I believe that in most of these cases moves like Bxa2 are wrong but I may be >>>>>>>>wrong because I did not see a lot of examples when search cannot solve the >>>>>>>>problem. >>>>>>>> >>>>>>>>Uri >>>>>>> >>>>>>> >>>>>>>I prefer to be conservative here. Rather than trying (a) if the bishop isn't >>>>>>>lost, then take the pawn, I prefer (b) if the bishop can't get off of a2 by >>>>>>>the time the evaluation is called, then it is trapped. >>>>>>> >>>>>>>Works well for me, very inexpensive to test for. >>>>>> >>>>>>Crafty takes the pawn if the analysis shows that it can get out? So crafty takes >>>>>>no risk.. Interesting to see what the other program does in this positions.. >>>>>> >>>>>>Regards, >>>>>>Alvaro >>>>> >>>>> >>>>>That is correct. It has to see taking the pawn, _and_ the bishop getting off >>>>>of a2, within the search. Otherwise it assumes that the bishop is trapped and >>>>>gives it a huge penalty. >>>>> >>>>>I haven't seen it fail very often, and when it did fail, the position was >>>>>complex enough that it wasn't possible to understand it with a simple static >>>>>eval trick anyway. >>>>> >>>>>The amazing thing is that I _still_ see it happening on ICC... I got tired >>>>>of seeing crafty do that pretty quickly. I decided that sitting in a game >>>>>and worrying about whether it will play a move that even a 1600 player would >>>>>avoid was simply something I didn't want to do. As a result, I don't. :) >>>>> >>>>>A 2500 (GM-level) program simply can _not_ play such a move. If it does, and >>>>>a GM sees it, it will lose the next N games because he will set that trap over >>>>>and over... and the program will bite over and over. >>>> >>>>I disagree. >>>> >>>>I do not think it is so easy to set the trap again and again. >>>>Remember that you have to set it in positions that the program cannot detect the >>>>loss of the bishop by search. >>> >>>A GM/IM can do this pretty easily in half of the games. >>> >>>why do you think I fixed it? one particular IM was very good at setting this >>>up, over and over and over.. >>> >>> >>> >>> >>>> >>>>I guess that your opinion is based on your experience on ICC but the hardware >>>>today is faster and programs can see more things by search. >>>> >>>>Uri >>> >>> >>>Most of those positions can not be solved by search. All it takes is for >>>black to have a missing a pawn, and black's rook holds the bishop for a long >>>time. But with the bishop stuck at a2, all of black's energy goes into trying >>>to extricate the bishop. All of white's energy goes into trying to snare >>>black's king. White succeeds more than black. >> >>I am interested to see the games of this IM against crafty when the IM won. >>It will be interested to see if Fritz or Junior fall in the trap in most of the >>cases. >> >>I never saw the problem in tournament time control games of Fritz or Junior(both >>programs take a2 in the game crafty-tiger) and the players could prepare against >>the commercial programs at home. >> >>This is the reason that I guess that it is not easy to set the trap. >> >>Uri > >also note that if you try to "solve the problem" by simply doing "long and deep >search", IMO its not a smart thing to do. >would you do a long search on a "tournament time control" game if you fall on >such position? >or would you rather let the code (or probably penalty for such moves) take care >of it? >maybe you can say that "modern hardwares" are fast and strong enough to give the >program a lift in the search. >in this case. > >* we dont know how much time it needs for most programs to find the "mistake" >even in the best hardwares available..... > >* and if we just add the code the program (in this case crafty or rebel) it will >find the "mistake", by giving a penalty, as soon as it searches the bad move, >"taking the pawn with the rook". > >thanks >pavel The problem with relying on a deep search is that you might avoid most of these captures at the root of the tree, but you may well blow your position because you see at the _end_ of the variation you can "win" a pawn. And you merrily follow that path, wrecking your position and winning a pawn, until you suddenly see that all you have done is wreck your position...
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.