Author: Christophe Theron
Date: 22:10:39 03/30/98
Go up one level in this thread
On March 30, 1998 at 20:34:13, Christophe Theron wrote:
>On March 30, 1998 at 13:09:34, Jon Dart wrote:
>
>>Recently my program "Arasan" played a game on FICS against
>>"noro", and reached the following position:
>>
>>8/8/5R2/r5P1/8/p5P1/K5kP/8 b - -
>>
>>Here the computer played Rxg5 and eventually lost. It appears
>>that Kxh2 draws (Crafty finds Kxh2 right away. The analysis
>>goes Kxh2 g6 Rg5 Kxa3 Kxg3 Kb4 Kg4 Kc4 Kh5 Kd4 Rxg6).
>>
>>This isn't really hard, but it helps if the program knows
>>the passed pawn is not dangerous even though the king is
>>behind it. It looks to me like Crafty has special case
>>code for just this ending (in EvaluatePassedPawns).
>>Genius 5.0 also likes Kxh2 after a short search. I'm
>>wondering if any other programs find this difficult.
>>
>>--Jon
>
>Chess Tiger settles for Kxh2 in about 74 seconds with a score of -0.12.
>
>Rxg5, the previous considered move, had a score of -1.25 at that time.
>
>Not very satisfying. This position is really interesting because it
>shows a common case that happens quite often in rook endgames. Being
>able to solve this quickly should give a real advantage in rook & pawns
>endgames.
>
>I keep it in my private collection. Thanks.
>
>
> Christophe
OK. Forget my previous message (see above). I just didn't catch the
right position. In fact, I have added FEN import/export in Chess Tiger
this afternoon, and reading the position above was the first test. But I
just realize something was wrong and I didn't import it correctly.
Now this is fixed, and Tiger reacts differently with the right position.
It finds Kxh2 in 16 seconds (K5-100MHz, 16Mb hash).
And my speech about knowledge is pure bullshit, too. It's a tactical
position. Interesting, but mainly tactical. I think I wouldn't take the
risk to try to add specific knowledge to solve this problem. The only
knowledge Tiger uses to solve the problem is: "when there are only 2
kings left on the board, the game is draw". I don't check if the kings
are of the same color or not, so there is room for some improvement
here. :)
Genius5 finds this in 11 seconds on my computer, and it seems that it
doesn't find it by knowledge, but just because it sees that all the
material can be exchanged.
Christophe
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.