Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Difficult endgame test position (but solvable for humans!)

Author: Oliver Roese

Date: 21:45:37 06/09/00

Go up one level in this thread


On June 09, 2000 at 23:36:17, Peter McKenzie wrote:

>On June 09, 2000 at 19:14:52, Oliver Roese wrote:
>
>>On June 09, 2000 at 15:11:00, blass uri wrote:
>>
>>>On June 09, 2000 at 13:59:44, blass uri wrote:
>>>
>>>>On June 09, 2000 at 12:11:16, Ernst A. Heinz wrote:
>>>>
>>>>>>[D]8/8/Pk5K/2p5/2P1pp2/8/5P1P/8 w - - 0 1
>>>>>>
>>>>>>Programs cannot see at evaluation time that black is winning because of
>>>>>>unstoppable pawn.
>>>>>>
>>>>>>If I give them to analyze at 1 ply depth they give a big advantage for white.
>>>>>
>>>>>What do you mean by "big advantage"?
>>>>>
>>>>>"DarkThought WCCC'99" knows about White's unstoppable passed Pawn
>>>>>on h2 and it also knows about the potential breakthrough of the
>>>>>two Black Pawns on e4 and f4. Both somehow cancel each other out
>>>>>in the static evaluation such that the overall score is just the
>>>>>+1 material advantage of White.
>>>>>
>>>>>=Ernst=
>>>>
>>>>Some other programs gave more than +2 for white and in one of the cases(I think
>>>>it was hiarcs7.32) they gave even about +7 pawns at 1 ply depth.
>>>>
>>>>Uri
>>>
>>>Here are some scores of chess programs at depth 1
>>>
>>>chessmaster6000(7.21)
>>>Hiarcs7.32(7.18)
>>>Fritz5.32(6.16)
>>>Genius3 depth 1 with selective search=0(5.39)
>>>Rebel century1.2a(4.36 and the score at depth 2 is even 7.76 and only at depth 3
>>>becomes negative)
>>>Junior5.9 (3.41)
>>>Crafty17.10(2.97)
>>>CometB20(2.80)
>>>CometB11(2.20)
>>>Junior4.6 (1.78)
>>>Exchess2.51(0.79)
>>>
>>>No program can see that white is losing at evlauation time or even at depth 1
>>>
>>>Uri
>>
>>How would a human look at the position?
>>It would notify that both white and black have unstoppable passers/candidates
>>and a directed search is required to see, who is the winner.
>>"Deep Thought" evaluates it this way, but the others fail. Apparently they do
>
>I doubt that Deep Though handles this position much differently than most
>programs, what makes you think otherwise?
>
You are right.
It should have read "DarkThought WCCC'99".
See above.


>>not recognize blacks strength. Despite the fact that bitboards were celebrated
>>for allowing that efficiently...
>>I find some high values astonishing. Is it really sound to estimate such high
>>evals?
>
>The evaluation function is full of similar heuristics, but of course some are
>more speculative than others.  The main thing is that the heuristics are right
>more often than not.  After that, everything becomes a speed/accuracy tradeoff
>where the balance of speed vs accuracy is according to the author's taste and
>ability.
>
>>
>>Oliver Roese



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.