Author: Dieter Buerssner
Date: 08:30:41 02/07/03
Go up one level in this thread
On February 07, 2003 at 08:28:23, Guido wrote:
[Much snipped]
I agree with all your reasoning. I think the specific discussed case was kpkp,
where the problem should not exist?
>Perhaps a better solution to this problem shall exist, but I didn't find it.
The method you described is more or less, what a normal engine does during
search. A very easy to implement alternative in the search process would be: let
the probing code just return "not found". When you call the probing code,
instead of:
if (last move was promote or capture && pieces on board < max-table-men)
probe the tb
use
if ((last move was promote or capture
|| second last move was promote or capture)
&& pieces on board < max-table-men)
probe the tb
It is clear, that this will need a depth more in about half of the time
(depending on the position), to resolve the same position in the search tree on
average. OTOH, I think nobody really knows, which is the best maximum depth at
which one should probe the TBs at a certain search-depth (to get a good
compromize between speed and accuracy). As a pro, typically less different disk
accesses will be needed, and the caches (OS or internal TB-cache) should work
with a better hitrate. The additional branches needed, I consider will not have
any effect.
The promotion problem you mentioned would not happen in 5-men TBs (which I
consider most important at the moment, having both sides for 4-men will not
waste a lot of space), when we have the convention, that we have only the side
with more pieces to move. After a capture (say from 6-men) this could delay the
solution for another depth or 2. I think, this is also of rather little
importance.
I used this method, shortly after I first used TBs. With a slow internet
connection, I used kppkp.nbw as only 5-men TB for testing. It seemed to work
well in test positions and in games.
Regards,
Dieter
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.