Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Extensions and Futility Pruning

Author: Frank Schneider

Date: 23:42:18 05/05/99

Go up one level in this thread


Hi James!

On May 05, 1999 at 12:35:50, James Robertson wrote:

>On May 05, 1999 at 02:18:11, Frank Schneider wrote:
>
>>On May 04, 1999 at 20:51:00, James Robertson wrote:
>>
>>>Whenever my program makes an extension, say, at depth = 4 it finds it is in
>>>check, it will not do any futility pruning/razoring for the rest of the branch.
>>>I recently read that some programs will do pruning/razoring even if there is an
>>>extension, unless it is at the current ply.
>>Gromit is one of those programs. It uses part-ply-extensions and has lots of
>>them, therefore there are extensions in many paths of the tree and switching
>>off futilitypruning after an extension would result in a much larger tree.
>>
>>>Is this a better way to do things?
>>>It seems like it would nullify some of the effects of extending, but it could
>>>also save a tremendous amount of time too. What should I do?
>>Test what works better for you.
>>(I've never tried switching off pruning if an extension applied - I'll try
>>that sometime).
>>
>>>
>>>Thanks for any help,
>>>James
>
>I tried out both ways on the first 50 positions in WAC, and the version that did
>not switch off pruning was about 20% faster. I made a mistake and ran the new
>version at 5 seconds a move, while the old one at 10 seconds, and the new still
>solved exactly as many as the old!
So it seems it is better not to switch off pruning (I'll test switching it off
later).
One remark: using the first 50 WAC positions as a testsuite it is no
surprise that both versions solved the same positions. Doing 7-ply-searches
Gromit quickly solves WAC1 and WAC3-WAC50 but to find WAC2 Gromit really needs
some time. IMO there are only 20-40 interesting positions in WAC, therefore
I'd suggest using for example ECM which has more positions (>700) and more
difficult positions (Gromit solves about 520/770 doing 8-ply-searches). So
far I have only little experience with ECM (I used WAC before, too), but it
seems to be a good choice. The drawback is that it takes a long time to
test so many positions, therefore it has to done automatically (for example
at night).


Frank




This page took 0.06 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.