I doubt whether threat detection (using null move) can be used when arrived at depth <= R + 2 or so. At depth 1 or 2 heavy pruning takes place when using move count pruning or something similar. So threat detection can not be trusted there. If pruning decisions depends on whether there is a threat or not that may give errors.
If you don't do pruning during thread detection the search slows down, so what to do ?
Threat detection near the leaves
Moderators: hgm, Rebel, chrisw
-
Alexander Zacharias
- Posts: 5
- Joined: Thu May 31, 2012 5:59 pm
- Location: Germany
Re: Threat detection near the leaves
I'd say it's a trade off between speed and accuracy, as nearly always. So if you believe the accuracy loss of a certain algorithm to outweight the speed gain in terms of playing strength / elo the best thing you could do in my opinion (besides proving it analytically 
) is verifying it: measure the playing strength of the two program versions, with and without that feature.
) is verifying it: measure the playing strength of the two program versions, with and without that feature.