Author: Dan Honeycutt
Date: 09:10:02 03/12/04
Go up one level in this thread
On March 11, 2004 at 12:56:37, Fabien Letouzey wrote: > >On March 11, 2004 at 12:40:40, Dan Honeycutt wrote: > >>Can you explain your statement in an earlier thread "I never "hash cut" at PV >>nodes (I only read the best move) so that my PV is never truncated." > >>At the start of each iteration i save the pv from the previous iteration and set >>a flag. while the flag lasts i use it to order moves and avoid hash cuts. >>however, once the old pv plays out and the flag is cleared any new pv is >>(likely) truncated. do you have a way around that? > >So you flag the "leftmost branch" only right? If one of your PV move is wrong >(another one is found to be better), you "lose" the flag, is that it? This >reminds me of a non well-known algorithm called P-alpha-beta which is a hybrid >between alpha-beta and PVS. It is interesting in itself, but I disgress ... > >If you use a PVS variant, a common way of knowing that the current node type is >"PV node" is to test if (beta!=alpha+1) (non-null window, this assumes you are >using PVS). If you never apply hash cuts at those nodes, your PV should always >be complete. > >Note that this may be affected by "window reductions" (aspiration, "shrunk" >window after a re-search failed high, ...). > >But the short answer is (beta!=alpha+1) :) If you don't use PVS, then you can't >predict in advance if the current node is going to be PV or not of course :) > >Maybe I did not understand your question correctly, don't hesitate asking for >more precision. > >Fabien. You all are a bit over my head but thanks Fabian and others. I lose the flag right away. I simply follow the pv from the previous iteration (avoiding hash cuts) till it runs out then the search goes it's merry way. the pv looks like iteration pv moves : 2 a b 3 a b c 3 d e f 4 d e f g 4 h everything looks like i think it should till that 1 move pv pops out. if i read Bob's reply right this is normal, don't worry about it. i am using PVS so the beta!=alpha+1 gives me something to tinker with. thanks again Dan H.
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.