Page 1 of 1

Standpat and check

Posted: Sat Feb 06, 2010 6:20 pm
by vladstamate
Hi there,

How should I deal with stand pat score calculation in quiesce search, in the situation where we are in check? Evaluations while in check are not reliable. So I cannot raise alpha or return beta (I have a fail-hard algorithm) based on the standpat score.

So what do I do with it?

Regards,
Vlad.

Re: Standpat and check

Posted: Sat Feb 06, 2010 7:09 pm
by jwes
vladstamate wrote:Hi there,

How should I deal with stand pat score calculation in quiesce search, in the situation where we are in check? Evaluations while in check are not reliable. So I cannot raise alpha or return beta (I have a fail-hard algorithm) based on the standpat score.

So what do I do with it?
If you know you are in check, just extend and ignore (or not calculate) the stand-pat score. Otherwise, you can do an expensive incheck() or an unreliable evaluate(). You can test to see which works better for you.

Re: Standpat and check

Posted: Sun Feb 07, 2010 5:53 pm
by vladstamate
Right,

That was good advice. I just did that in quiesce search: if in check, do not return beta or change alpha based on stand pat score and try to also generate the moves that takes us out of check, not only captures.

That improved the engine noticeably. Plisk went from 400 to 427 in the 9 STS tests.

Thanks,
Vlad.