Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Node count differences with Omids verification search

Author: Tony Werten

Date: 11:11:40 11/22/02

Go up one level in this thread


On November 22, 2002 at 12:58:42, Andreas Herrmann wrote:

>On November 22, 2002 at 11:35:38, Tony Werten wrote:
>
>>On November 22, 2002 at 10:39:16, Uri Blass wrote:
>>
>>>On November 22, 2002 at 10:30:34, Robert Hyatt wrote:
>>>
>>>>On November 22, 2002 at 07:21:13, Uri Blass wrote:
>>>>
>>>>>On November 22, 2002 at 05:50:09, Omid David Tabibi wrote:
>>>>>
>>>>>>Vincent,
>>>>>>
>>>>>>You haven't read the article yet. READ IT. I have described everything I have
>>>>>>done quite clearly, and have presented the results. All the node counts are also
>>>>>>available at my website.
>>>>>>
>>>>>>Regarding your checkmate question: except for the "Mate in 4" and "Mate in 5"
>>>>>>positions which I was interested to know only if the checkmating was found or
>>>>>>not, in all other test positions (ECM, WCS, Neishtadt) the search was conducted
>>>>>>to the mentioned depth even if a checkmate is found at eariler plies. This was
>>>>>>done in order to ensure that the fixed depth results are comparable in terms of
>>>>>>tree size.
>>>>>>
>>>>>>Omid.
>>>>>
>>>>>What do you search after finding a mate?
>>>>>Does genesis search for a shorter mate and in this case does it knows not to
>>>>>search when it get ply that means that no shorter mate can be found?
>>>>>
>>>>>Movei simply stop searching after finishing an iteration with mate score.
>>>>>
>>>>>Uri
>>>>
>>>>
>>>>I would suspect he does like most everyone else.  Search until you reach depth
>>>>2*N-1 where N is the mate depth as in mate in 9.  I need to search 17 plies to
>>>>verify that no shorter mate than mate in 9 exists...
>>>
>>>I do not know what most do but I do not care after finding a mate to find the
>>>shorter mate.
>>>
>>>The result is the same even if I do not find the shortest mate.
>>>I care only to finish the iteration.
>>
>>You'll care when you add tablebase support.
>>
>>When there are 6 pieces on the board and your program sacrifies his queen to
>>find a checkmate in 123, you'd wish it had searched a bit longer to find the
>>mate in 4.
>
>Hi Tony,
>
>but then you must be in depth 245 to cut the search by checkmate in 123 and
>search abort after 2*N-1.

No, if you can't find anything better then you take the checkmate in 123. But if
you allocated fe 2 minutes, you can spend that time to find a shorter one. I
started with this idea when I saw Fritz make exactly this mistake. It's not too
bad, it just looks silly.

>
>In previous versions i have also finished the search after 2*N-1. In my current
>version i finish my search after 2*N+2, because of Nullmove and other pruning
>algos. But it's mainly interresting to search a little bit deeper if there was
>found a mate for the opponent, perhaps in some rare cases a draw will be find 2
>or 3 plys deeper.
>
>Andreas
>
>PS: Thanks again for your DLL. Works fine :)

You're welcome. Please don't forget to drop Eugene an email. The dll gives acces
to his tables, so you should let him know/ask permission. (Theoreticly the
latter, practicly the first )

Tony




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.