Author: Steve Maughan
Date: 08:53:08 01/01/01
Go up one level in this thread
Bob, Thanks again for the help. >This statement is still simply incorrect. You don't _touch_ the PV. The issue >is that you are searching, and someone whispers in your ear "Hey, your alpha >value is -1.2... I actually have searched this position before and I proved >that the score is actually no worse than -1.0... so change your alpha and >keep searching." You don't affect the PV, or anything else, other than make >the search more efficient. If the new alpha value is >= your old beta value, >you should have _already_ bailed out with a fail high. If your new beta value >is lower than your current alpha value, you should have already bailed out with >a fail low. I don't understand the concept of "cutting off the PV"... The PV >is backed up from below, not built as you search downward... OK lets take this as an example. Alpha = -1.2 but the hash value (at a greater depth) is a lower bound of -1.0, say also that beta = +2.0 . Suppose that if the position is searched without changing the value of alpha it would come back as -1.1. This is possible as it is a shallower search that the one that gave the lower bound of -1.0. This means that if we raise alpha the search will fail low at -1.0. This will then be returned to the parent node as +1.0 and may form part of the PV as it is between the parents alpha (-2.0) and beta (+1.2). By "cutting off the PV" I mean ending it at that node with a signature, usually <ht>. I have seen PV from Crafty end in <ht> - how do these come about if it's not by "cutting off the PV". Clearly I am missing something here otherwise Crafty and DarkThought wouldn't work and that's not the case. The strange thing is that when I adjust my code to do as you say I get an illegal PV - maybe it's another bug. Thanks again, Steve
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.