Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Parallel search article RBF

Author: Robert Hyatt

Date: 14:53:41 09/11/02

Go up one level in this thread


On September 11, 2002 at 16:02:02, Jay Scott wrote:

>On September 11, 2002 at 11:41:13, Robert Hyatt wrote:
>
>>it also takes a huge amount of memory since best-first
>>has to store the whole tree as it is traversed.
>
>That is not strictly true, since the fringes of the tree are (or can be)
>searched serially by single processors. The bulk of nodes do not need to be
>stored.
>
>Certainly the traditional hash table, which pretty much can store all nodes if
>there's enough memory and degrades gracefully if there's not, has the best
>tradeoffs.
>
>  Jay


How are you not going to store the tree when it is, by definition, "best
first"?  IE how can you figure out how to permute the minimax scores without
having the entire tree handy?  Or at _least_ the bottom layer, which is the
majority of the tree...

But in any case, without the entire tree, you can't figure out the PV nor do I
see how you can adjust the minimax scores, so I think the entire tree is in
memory or it isn't best-first.

That is _the_ advantage of depth-first, in fact, that this memory requirement
is not present.



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.