Author: pete
Date: 11:38:53 07/08/99
Go up one level in this thread
On July 07, 1999 at 23:18:35, Robert Hyatt wrote: >On July 07, 1999 at 16:02:52, pete wrote: > >>On July 06, 1999 at 17:15:44, blass uri wrote: >> >>> >>>On July 06, 1999 at 13:33:01, Robert Hyatt wrote: >>> >>>>On July 04, 1999 at 06:19:17, Frank Schneider wrote: >>>> >>>>>On July 03, 1999 at 19:16:59, Gerrit Reubold wrote: >>>>> >>>>>>Hi all, >>>>>> >>>>>>please test your programs with the following position >>>>>> >>>>>>5rk1/1r3pp1/pp2pq2/3p4/3Q4/1PR5/P4PPP/4R1K1 w - - >>>>>> >>>>>>it is from a game which my program (Bringer) lost with white against The Crazy >>>>>>Bishop. The draw is very easy to see (for humans): Qxf6 gxf6, and then rook >>>>>>checks at h3, g3, f3... How long does your program take to find Qxf6 *with a >>>>>>draw score*. How many plies / seconds? Question to the programmers: What do you >>>>>>do to solve such positions fast? Extending on checks is not enough, my program >>>>>>needs a 12 ply search (8 minutes on a PII-300) to find the draw (Qxf6 is found >>>>>>earlier). >>>>>Gromit shows a drawscore after iteration 5 (1sec). >>>>> >>>>>Frank >>>>>> >>>>>>Greetings, >>>>>>Gerrit Reubold >>>> >>>> >>>>This is an evaluation issue. If your program thinks white is better, then it >>>>will see a draw. If your program likes black, then it will find that black >>>>doesn't _have_ to take the draw as the repetition is certainly not forced if >>>>black or white doesn't want to repeat. >>> >>>The repetition is forced. >>>It is a perpetual check. >>>I agree that a program can see the draw for the wrong reasons but it does not >>>change the fact that white has perpetual check after Qxf6 gxf6. >>> >>>Uri >> >>here I am disappointed ; this is a forced perpetual ; some progs see it , some >>don't , and although crafty usually is a good solver it has no clue about this >>position and goes for strange king manouevres like Kf1, Ke2 , etc even at long >>time controls ; crafty won't find this one at ply 15. >> >>some day there will be a programmer who admits something like this without >>strange eexcuses :); everybody here should know that not being able to >>understand a certain position doesn't say anything about the prog's overall >>strength. >> >>Pete > > >I believe I gave _both_ correct explanations. Crafty doesn't find it because >it can't see deep enough. It sees the king walking forward on one color of >squares, then walking backward on the other, which means that the rep is very >deep. > >However, the original question was interpreted by me as something different... >white can force the perp, black can not. If white thinks it is better, it >can obviously avoid it if it wants... So it is all about evaluation. If one >side has a really big king safety term (IE perhaps CSTal) it might like white >better because of the king safety issue. And it might well turn the draw down >even if it can see it. > >In _my_ case, I do _not_ worry about solving problems. They help little in >solving games positions, because you can spend way too much time following >checking lines that are futile... IE for every forced draw that Crafty walks >into (with this theme) there are dozens of positions where it plays _better_ >by not wasting the time following the checks.. > > >In this game, crafty plays the "odd" Kf1 because it does _not_ want to drive >the black king to where it wants to get, namely the center of the board. Hi, sorry for the impolite part of my last post , i had a _very_ bad day :=) I wanted to point out something different; the position is bad for white because of the strong free pawn at d ; that's why I can't understand the part about white being better ( crafty should know better ? , well in fact it does :) ); and I think king's safety or a doubled pawn at f are just of _very_ minor meaning in this position, and a prog who judges these factors high in this position is simply wrong. But I see your points . It is a question of the point of view ( one seeing it as chessprog programmer who wants to point out some general idea , other as chess game spectator ). Pete
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.