Author: Miguel A. Ballicora
Date: 08:33:27 05/15/01
Go up one level in this thread
On May 14, 2001 at 18:53:34, Dieter Buerssner wrote:
>On May 14, 2001 at 12:47:32, Miguel A. Ballicora wrote:
>
>>Besides the wording, Giancarlo's point was clear. Do the programs "understand"
>>what's going on? Sometimes you avoid a6 because there so many good choices
>>that a program took a anyone. Still does not prove that that the program
>>understand the position.
>
>I agree. Also, I think (after a small experiment) in the discussed positions
>even stalemate detection in qsearch cannot help too much.
I guess that it would help shortening 1 ply the path required to see the draw
>>This is an attempt to make it harder for the programs
>>
>>2k2r2/2P3R1/pP6/1P6/8/8/1K6/8 w - - 0 56
>>
>>White should avoid bxa6. With enough time (and maybe table bases) programs
>>will avoid it because they will see a promotion in other line (Kb3 or Kc3) but I
>>guess that it will take many plies to find that, unless the program "understand"
>>the risk of playing bxa6.
>
>For Yace, this is indeed much more difficult than the position of GCP. This is
>what I get on my AMD k6-2 475, 20M hash, KRPKR and all 4-men TBs:
>
> nodes time score depth
[snip]
> 63260777 9:01.4 9.24 15. 56.Rg6 axb5 57.Rd6 b4 58.Kb3 Rg8 59.Kxb4 Kb7
> 60.Rd8 Rg4+ 61.Kc5 Rg5+ 62.Kd6 Rg6+ 63.Kd7 Rg7+
> 64.Ke8 Rxc7 65.bxc7 {HT} {580}
My program also needed 15 plies and chose Rg6. Actually, Rg6-Rd6 is an excellent
plan that I did not think about when I set the position! Very direct and freezes
black defenses, King can't move (Kb7 is answered by Rd8) and Rook can't leave
the first rank. The beauty is that it looks like a human plan! more logical
than "tactical". Funny how a logical plan comes from a pure tactical one!
In Giancarlo's 2nd position you got:
>I get:
[...]
> 11233200 1:00.3 0.00 19t 56...Rf3+ 57.Kc4 Rf4+ 58.Kb5 Rf5+ 59.Kc6 Rf6+
> 60.Kc5 Rf5+ 61.Kd4 Rf4+ 62.Ke5 Rf5+ 63.Kd6 Rf6+
> 64.Ke7 Re6+ 65.Kxe6H {-740}
In fact, 1 minute is very good to see the draw in my opinion!
>I think, more aggressive extensions for check (in rook endgames, Yace will not
>do too many extenions for this) could help to "understand" the positions better.
That will certainly help, but it is difficult to control it in other positions.
Maybe it is required a specific algorithm for this situations (it will be
expensive), but it is triggered only when there are no legal king moves && all
the pawns are blocked && all the legal moves belong to the same piece && that
piece is a rook.
One more thing in my to-do list :-).
Miguel
>
>Regards,
>Dieter
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.