Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SEE functions

Author: Andreas Guettinger

Date: 14:01:30 11/25/02

Go up one level in this thread


On November 25, 2002 at 16:31:01, David Rasmussen wrote:

>On November 25, 2002 at 15:16:35, Andreas Guettinger wrote:
>
>>On November 25, 2002 at 14:17:38, Robert Hyatt wrote:
>>
>>>
>>>Crafty's is not bad.  This is a classic trade-off issue.  It could easily be
>>>made more
>>>accurate.  IE not using absolute-pinned pieces and so forth.  But the question
>>>becomes,
>>>does the cost of the extra accuracy result in tree sizes that are small enough
>>>that the
>>>savings offsets the added computational cost...
>>>
>>>It's easy to make it more accurate.  But the question is, "is it worth it?"
>>
>
>We're not just talking tree sizes here, are we? See below.
>
>>
>>I would say no. The differences between MVV/LVA are already not that big, so
>>even if the SEE is not the most accuratest, if it is only used for move
>>ordering, that doesn't really. More important is that it is fast and only
>>executed when necessary.
>>
>
>For move ordering, Crafty's is probably fine. That is, I don't think that there
>is a lot to gain in move ordering and therefore tree size, by a more accurate
>SEE. But SEE is also, in Crafty at least, and in Chezzz (which uses a
>SEE-function similar to Crafty's), used to throw out apparently futile captures.
>If the SEE isn't good enough, this will lead to tactical problems, by throwing
>out what seems to be a futile capture, but isn't. I asked my original question
>with that aspect in mind. How safe is it to throw a capture out, using Crafty's
>SEE?
>
>/David


SEE and pruning is not exactly the same. I use the same SEE in quiescence search
and in normal move ordering, but pruning i do obviously only in quiescence,
because I don't want to miss any moves in the normal search. (As far as I know,
it's similar in Crafty, as I read many Crafty CCC postings which helped me
implement my SEE).
Pruning in quiescence can be done in account of the score you get from SEE. The
scores you get from SEE are accurate. You can then decide below which threshold
you want to prune the moves in quiesece. I currently prune all moves that give
me a negative score in SEE.
Pruning in quiescence can be optimised by experimentation. But quiescence is
very leaky by itself, since only captures are followed, so accuracy is not
needed for pruning.


Regards
Andreas




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.