Author: Ralf Elvsén
Date: 05:35:24 10/06/00
Go up one level in this thread
On October 06, 2000 at 05:10:40, Severi Salminen wrote: > >> >>The third alternative is that the score of the best move is between alpha and >>beta. This is a potential PV. If this happens all the way back to the root >>this will indeed be part of a PV. > >This was interesting. So I can make use of this to store the PV and search those >moves first at the next iteration. So actually we _need_ fail-low situation in >storing PV. We store the PV if and only if we can got a<result<b case the whole >way from top to root? Can Alpha equal result or should result be bigger? I'm not sure what you mean by "_need_ fail-low situation in storing PV" . If the value of the best move is between alpha and beta there is neither fail-high nor fail-low (aside from the "equal-issue). Can the best value equal alpha or beta in a PV-node? If you start with +/-infinity at the root I think not, i.e. you have a potential PV-node if and only if alpha < bestvalue < beta . However, at least in my program this condition got more complicated when I started with windowed searches in the root and when I used the PV from the previous search depth to start the next depth. But I humbly suggest that you get the PV-stuff to work with +/-infinity -values in the root first and then move on to the nontrivial cases. Ot maybe they also are trivial? Every chess program seems to behave slighly differently when it comes to hashtables, extensions etc. It might be true for this as well... I'm no guru by any standards. Work out a way you feel comfortable with. Ralf
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.