Author: Christophe Theron
Date: 12:12:07 03/18/98
Go up one level in this thread
On March 18, 1998 at 06:31:37, Ernst A. Heinz wrote:
>On March 17, 1998 at 21:42:52, David Blackman wrote:
>
>>Actually about positions like KNkp and how thay should be evaluated.
>>
>>It looks like Junior, Crafty and maybe others are evaluating such
>>positions
>>as not winnable as soon as they appear in the tree. Surely it would be
>>better
>>to allow the tree to grow out to its normal depth, and just detect these
>>(probably drawn) positions in the normal end-node evaluator. That way
>>your program "knows" it is a draw, unless the win is shallow enough to
>>see in the search, in which case it finds it.
>>
>>Of course that makes the search tree a bit bigger for a given depth.
>>Has anyone tried both approaches to see how big the cost is?
>
>"DarkThought" uses fast rules of thumb to disable the draw detection
>heuristic in such cases where it might fail. The costs are negligible --
>both as for runtime overhead and size of search tree.
>
>"DarkThought" solves all the critical positions posted in this thread.
>
>W.r.t. Christophe Theron's analysis of the Nolot-Position with key move
>Ng4 I like to add that for "DarkThought" the depth where it finds the
>solution not only depends on the null move search but also quite
>strongly
>on the aggressiveness of pawn-related extensions.
>
>=Ernst=
You are right. My point was to show that null move or related techniques
had a significant impact on this position.
Unfortunately, others parameters biased the result. So here is another
position given by Chrilly Donninger in his famous paper from ICCA
Journal, Sep 93:
White:
Pawn b4
Rook d4
King f2
Black:
Pawn a6 b5 h2
Rook g6
King h1
White to move. The key is 1.Rd1+, which is mate in 7. After 1...Rg1,
black is in zugzwang. White plays for example 2.Rf1, and wins.
Genius3/5 finds the key in 0.5 second or so at "ply" 5, which means IMO
ply 8 or 9 (K5-100, 384Kb hash).
Rebel9 finds that 1.Rd1+ has a positive score at ply 9 in 1 second, but
score is only +4.80, and finds the mate at ply 10 in 2 seconds (K5-100,
10Mb hash).
ChessMaster 4000 finds that 1.Rd1+ has a positive score at ply 7 in 2
seconds (+1.23), and finds the mate at ply 8 in 3 seconds.
Fritz2 didn't find the solution in 10 minutes.
I would like to hear about results of others programs in this position.
I suspect that check and pawns extensions would have an impact on the
time needed to find the mate, but my only point is: does program X finds
the mate?
Christophe
This page took 0.01 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.