Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hiarcs7.32 determines a Draw 39 moves before Crafty16.13 sees the Draw!!

Author: Robert Hyatt

Date: 12:38:44 09/01/99

Go up one level in this thread


On September 01, 1999 at 14:27:37, blass uri wrote:

>On September 01, 1999 at 09:39:58, Robert Hyatt wrote:
>
>>On September 01, 1999 at 08:49:23, blass uri wrote:
>>
>>>On September 01, 1999 at 07:56:29, Robert Hyatt wrote:
>>>>What is happening here is that Hiarcs is using some unsafe evaluation (IMHO)
>>>>to say that two minors vs one minor is a draw.  Rebel lost a game making this
>>>>assumption in the WCCC as Uri pointed out.
>>>
>>>Rebel evaluated KBB vs KN as 0.xx pawns advantage for the BB and not as a draw.
>>>
>>>>I don't call KBN vs KB a draw.  Most are.  But _all_ are _not_.  It makes you
>>>>look like a genius when it is right.  And like an idiot when it is wrong.  I
>>>>prefer the database approach which is 100% right.
>>>
>>>The problem even with tablebases is with KBN vs KBP
>>>In this case you can go by search for KBN vs KB by sacrificing the pawn but if
>>>you cannot go by search to KBN VS KB your evaluation will be clearly wrong.
>>>
>>>I do not say that the static evaluation should evaluate these positions as draws
>>>but it should evaluate positions of KBN vs KBP as close to a draw and crafty has
>>>not 0.xx evaluations for them.
>>>
>>>Uri
>>
>>It does for safe cases... and it used to for other cases.  IE KNN. many programs
>>say the KNN side can't win no matter what.  But if the other side has a pawn,
>>it can be a win and they screw up.  I let the tablebases handle that one now.
>>There are other cases.  As far as rebel, does it matter whether it says +.5 or
>>0.0?  It was wrong (seriously wrong) in either case as the two pieces won.  And
>>that is the 'unsafeness' I try to avoid.
>
>It is important if Rebel says 0 or 0.5 because if you say 0 for KBB vs KN then
>you may miss an easy draw that you evaluate as 0.00 and if you say 0.5 you can
>not miss an easy draw.
>
>KBB vs KN you can deal with tablebases but you have not tablebases for KBN vs
>KBP so you cannot avoid unsafeness in this case.
>
>Uri


correct,although in a tournament game I'll bet the stronger side can win
the pawn with a 3-4 minute search and force a proper evaluation... at least
I did when I tested this with Crafty.

as far as 0.00 or 0.5, why not -.5 or +5?  ie if the number is wrong, it
doesn't matter much as you are getting a wrong score and making a game-
critical decision based on this score.  IE the WCCC game.



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.