Author: Robert Hyatt
Date: 19:45:50 03/25/99
Go up one level in this thread
On March 25, 1999 at 17:09:24, Dave Gomboc wrote: >On March 24, 1999 at 22:42:20, Robert Hyatt wrote: > >>On March 23, 1999 at 12:18:29, Vincent Diepeveen wrote: >> >>>On March 22, 1999 at 07:57:06, Robert Hyatt wrote: >>> >>>>On March 21, 1999 at 20:05:08, Charles L. Williams wrote: >>>> >>>>>Is it advantageous to keep updating the PV after entering the quiescence search, >>>>>or should this search be kept very simple? I'm getting ready to add one, >>>>>captures only. Thanks in advance. >>>>> >>>>>Chuck >>>> >>>> >>>>If you don't you won't see the end of the PV when you display it at the root, >>>>and this can make testing/debugging difficult, since you won't have any idea >>>>what the complete path looks like from the root to the position where you did >>>>an Evaluate() and then accepted that score. In fact, your entire PV will have >>>>problems if you don't 'latch it' at the precise instant the score is 'ok'. >>> >>>Sure. an easy solution is there however: if you get your PV from the >>>hashtable and if you are using several probes at the hashtable, >>>then you can get the PV out of the hashtable and this includes then the >>>q-search too. >>> >>>Quick and easy! >>> >>>Greetings, >>>Vincent >> >> >>It's easy if you can be _sure_ you won't overwrite a PV hash entry. IE how many >>times do you get the first result back quickly, then do a bunch of searching on >>later moves because they have lots of checks that produce lots of extensions. >> >>What do you do if the PV gets overwritten before you need it? I tried this a >>long time ago. And didn't like the result. We back up the PV so infrequently >>I can't really measure the time it takes anyway... it is very minimal... > >I would appreciate it if someone described a way to ensure that hash positions >belonging to the PV do not get overwritten without adversely affecting general >performance. As things stand I am tracking the best move at the root >separately, because occasionally my root position gets trashed. > >I am using replacement scheme BIG-1 (q.v. Dennis Breuker's Ph.D. thesis). > >Dave Gomboc I don't see a reasonable way to do it. every time you 'back up a PV' you could flag every hash entry in that path as "PV" and _never_ replace them. But that would be costly. And without that, you will occasionally step on the PV entry and cut the PV off at that point. It wouldn't hurt, really, except it makes debugging/tsting harder as I want the path _and_ score, not just the score and the first few moves.
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.