Page 1 of 1

Quark 2.35 draw claim bug

PostPosted: 28 Mar 2005, 05:37
by Igor Korshunov
Hi there!

Quark claim a draw just one move before being mated !!!



[Event "Computer chess game"]
[Site "WILDCAT-FC58790"]
[Date "2005.03.28"]
[Round "-"]
[White "WildCat"]
[Black "Quark v2.35Paderborn"]
[Result "1/2-1/2"]
[TimeControl "60+2"]
[FEN "kqQ5/8/NKn5/8/8/8/8/8 w - - 0 1"]
[SetUp "1"]

{--------------
k q Q . . . . .
. . . . . . . .
N K n . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
white to play
--------------}
1. Qxb8+ Nxb8
{Not enough material to mate !} 1/2-1/2



2. Nc7# mate!!!

Best regards,
Igor

Re: Quark 2.35 draw claim bug

PostPosted: 28 Mar 2005, 14:56
by Sven Schüle
Hi Andrew,

Andrew Fan wrote:I guess Quark is a bit lazy with it's "not enough material" draws (KNvKN). Should not consider material draws if the king(s) are in any of the corners.

in the "Programming" subforum there is a recent thread about insufficient material and claiming draw. I'm sure you know that KNKN does not fall into this category, an engine may never claim draw here according to the rules. What you can do for KNKN if the enemy king is not in a corner is for example:
a) to offer a draw,
b) to accept a draw offer,
c) (at depth > 0) to return a zero value instead of searching the subtree,
d) (at depth = 0) to play a random move without any searching.

Besides checking for "enemy king not in corner", one could define more criteria which tell us if a mate-in-1 is possible (position of knights, own king), but this would go too far for such a rare case. One could even omit checking for corner square, just search one ply and return 0 if no mate is found. (A value >= 0 is guaranteed for the side to move in KNKN.)

More details belong into "Programming" indeed; sorry for bringing this up here. I just had the impression that "claiming draw" and "evaluating as draw" are mixed up sometimes.

Sven

Same bug in Crafty...

PostPosted: 29 Mar 2005, 08:02
by George Lyapko
Code: Select all
[Event "Computer chess game"]
[Site "LYAPKO"]
[Date "2005.03.29"]
[Round "-"]
[White "Crafty-19.19"]
[Black "Crafty-19.19"]
[Result "1/2-1/2"]
[TimeControl "120"]
[FEN "kqQ5/8/NKn5/8/8/8/8/8 w - - 0 1"]
[SetUp "1"]

{--------------
k q Q . . . . .
. . . . . . . .
N K n . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
white to play
--------------}

1. Qxb8+
{Insufficient material} 1/2-1/2


Best regards,
George

Re: Quark 2.35 draw claim bug

PostPosted: 29 Mar 2005, 10:55
by Thomas Mayer
Hi Igor,

first of all: thanks for pointing that out...

it's definitely new for me, because Quark should only claim draw when just one piece is on the board and not in the case you showed... I will check that...

Greets, Thomas