Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: more info and another question

Author: James Swafford

Date: 11:02:33 09/13/02

Go up one level in this thread


On September 13, 2002 at 13:48:54, Georg v. Zimmermann wrote:

>This is common. Search through the archives, you will find a lot of discussions
>about it. Especially if your extensions depend on the window.
>
>The simpler your search, the less this will happen. So if you disable all
>selectivity (null move, extensions, etc. ) and it still happens you might have a
>problem.

The first thing I did was disable the null search.  Extensions play
into this somehow: it is a mate in three, and the fail high occurred
in iteration 2.

That doesn't explain it away, though.  If it triggered extensions
the first time around, then it should trigger an extension the next
time around.  So the puzzle remains unanswered. :)

I will keep digging -- if there's a bug I want to find it.

--
James



>
>Georg
>
>
>On September 13, 2002 at 12:06:18, James Swafford wrote:
>
>>[D] r1bq2rk/pp3pbp/2p1p1pQ/7P/3P4/2PB1N2/PP3PPR/2KR4 w - -
>>
>>You can see from the position above and the search output that this
>>is a simple mate.  Somehow the program must see the mate during
>>iteration 2, hence the fail high at the root (it was searching with
>>a window of +/- 1 pawn of the previous score of 15).
>>
>>After the fail high, the search is restarted with a window of
>>15,+infinity, but Qxh7 doesn't do so well on the research (it gets
>>a -675), so it doesn't get backed to the pv, the rest of the moves
>>fail low, and I have to start a search in iteration three.
>>
>>I find it odd that the program could see Qxh7 was good enough to
>>fail high, but on the research failed low.  How frustrating. :)
>>
>>Is this common, or do I have a bug?
>>--
>>James
>>
>>
>>Problem 1:
>>r1bq2rk/pp3pbp/2p1p1pQ/7P/3P4/2PB1N2/PP3PPR/2KR4 w - - h6h7;
>>
>>parsing: h6h7...done.
>>correct move(s): h6h7
>>FW Dpth   Nd Cnt    CSecs  Score PV Line
>> 1& ->        15        0   -570 h6g7 g8g7
>> 1& ->       216        0     10 h6g5 g6h5
>> 1& ->       589        0     15 h6f4 d8e8
>> 1  ->      2299        6     15 h6f4 d8e8
>> 2& ->      3373        6     15 h6f4 d8f6 f4f6
>> 2& ->      6558       11    +++ h6h7
>> 2& ->      7149       11    ---
>> 2& ->      7150       11     15 h6f4 d8f6
>> 2  ->     27746       39     15 h6f4 d8f6
>> 3& ->     32509       44     20 h6f4 g6h5 f4f7 g7h6
>> 3& ->     74189       94    +++ h6h7
>> 3& ->     75366       99  59997 h6h7 h8h7 h5g6
>> 3  ->     77892       99  59997 h6h7 h8h7 h5g6



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.