Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: My fail soft reduces quality of collected PV. Help needed.

Author: Volker Böhm

Date: 08:45:31 04/20/04

Go up one level in this thread


On April 20, 2004 at 11:33:38, Fabien Letouzey wrote:

>On April 20, 2004 at 11:27:08, Volker Böhm wrote:
>
>>(our example below) If you search with window [+8,+9] and get +10 as result then
>>you can store +10 to hash ("fail-soft-soloution") or +9 (== beta)
>>("fail-hard-soloution"). In my test suite this "fail-hard-soloution" was good
>>enough to get allmost everytime a good pv. But additional nodes! It is too
>>significant to live with.
>
>We agree that fail hard does not solve the problem all the time then.
>About additional nodes; do you have figures?
Yes we agree, never gets perfect.

In WAC Test it took about 7% more time and failed on two more positions (max.
time 10 sec on celeron 1,3 GHZ). But it´s a special case as we are currently
programming a new engine from scratch. The test is made with a material-only
eval and nearly all simple search features. (Nullmove, IID, PVS, Hash, Initial
Window, Attack-Table based SEE, Killers).

>
>Fabien.



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.