Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Move Ordering and PV

Author: Pat King

Date: 15:52:47 05/07/00

Go up one level in this thread


On May 05, 2000 at 13:26:51, Tom Kerrigan wrote:

>On May 05, 2000 at 11:16:34, Pat King wrote:
>
>>It seems clear to me that you need pretty good move ordering for PV to pay off
>>vs. plain AB. But how good? One could use only the pvmove[][] structure to
>>implement it (although I use just the hash, and probe it to find the PV). Is
>>this enough to make it pay? Are history tables and killer moves more effective
>>w/ PV than AB?
>>
>>Thanks.
>>
>>Pat
>
>I think you're confusing PV and PVS, which are two totally different things.
>
>PV is the Principal Variation, i.e., your best line. This is stored in the
>pvmove[][] array that you mention, and can be used with any type of search,
>including plain old alpha-beta. You obviously want to order the PV high at the
>start of a search. Assuming that the PV will be stored in your hash table is
>just fine, as long as you have a reasonably large hash table and a good
>replacement scheme.
>
>PVS is Principal Variation Search, which (I believe) is also known as
>null-window search. It's a method used to decrease the size of the alpha-beta
>windows within the search. It really has nothing to do with Principal
>Variations, except at the conceptual level, because it requires good move
>ordering to be effective.
>
>-Tom
Only confusing the acronyms. Yes, I meant PVS, and as you note, it requires good
move ordering. But HOW good? Random ordering would just hurt performance. Does
just the PV structure or hash beat break even? Have to have history? Killer
move?

Pat



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.