Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Tiger evaluation function

Author: James T. Walker

Date: 08:56:42 04/11/01

Go up one level in this thread


On April 11, 2001 at 11:27:09, Robert Hyatt wrote:

>On April 11, 2001 at 03:01:27, Ernst Walet wrote:
>
>>On April 10, 2001 at 23:23:21, Derek wrote:
>>
>>>This position is a study that I saw on the following website:
>>>http://www.chesscafe.com/endgame/endgame.htm
>>>
>>>White: Kc5,Nd5,Ng6,Bh7
>>>Black: Ka5,Ba2,a3
>>>
>>>White to move
>>>
>>>Rebel-Tiger 14.0, using all 3,4, and 5-man tablebases
>>>
>>>After the moves:
>>>
>>>1.Nc3 Bb1 2.Nxb1 a2 3.Nd2
>
>Bb1 is probably not the best move.  It leads to an instant mate in 30
>according to crafty, while Be6 prolongs it.
>
>Nxb1 a2 Nd2 leaves things at a mate in 29 for white.
>
>a1=B changes it to a mate in 12.
>
>There is a bug somewhere.  I would suspect that the program is recognizing that
>KBNN vs KB is somehow a draw, when it is not.  It ends up as a KBN vs K ending
>or a KBNN vs K ending, either of which is easily won.
>
>
>>>
>>>Tiger analyzes 3...a1=Q as lost for Black, but shows 3...a1=B as a draw for
>>>Black. This is clearly incorrect, as after 4 Nb3+ Ka4 5 Nxa1, it's evaluation
>>>changes to show that Black is mated in 10 moves. Is this a serious evaluation
>>>error in the program?
>>
>>
>>What happens if you disable the tablebases?
>>
>>Ernst.
>
>
>That would be a good test...

There appears to be a serious problem here with Tiger.  With tablebases off and
after 2...a2  Tiger plays Nd2 expecting a1=N which I guess is OK.  But if I play
a1=B Tiger instantly plays Nc4+ with score of 0.00.  Tiger expects to see Ka4
Nb6 Ka3 Nc4 Ka4 to repeat and draw. ??

With tablebases on, after 2....a2 Tiger plays Nd2 expecting Ka6.  If I play a1=B
again Tiger again plays Nc4+?? scored 0.00.  Definitely a bug somewhere.



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.