Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Clearing the Transposition Table

Author: Edward Screven

Date: 00:55:53 03/04/98

Go up one level in this thread


On March 03, 1998 at 20:18:34, Brian McKinley wrote:

>I have seen more than one reference to clearing the transposition table.
>Is this necessary, and if not, why is it a common practice?  The only
>benifit I can think of is a negligible increase in lookup speed until
>the  table fills up again.

i never completely clear the transposition table.  i always try to keep
as much information as i can.

i do distinguish between old and new entries, where new means from the
current search, and old means from a previous search. (i use this
information
to evict old entries faster than new entries, since it's less likely
that they will be relevant in the future.)  maintaining the distinction
is very cheap -- it doesn't require updating the transposition table
per-move.  instead, each entry includes a small half-move number field.
when an entry is created or updated its half-move number field is set
to the current game half-move number.  if i see an entry with a
non-current
half-move number, then i know it's old.

my program includes a simple notion of search goals, which are really
just parameters to the tip evaluator.  if these goals change, then
the current scores in the transposition table obviously become invalid.
but instead of simply erasing the entries, i replace the score
information
of the entry with (-inf,+inf), leaving the suggested move information
intact.  this is a win because the best move given the old goals is
usually
still a good move given the new goals.

i don't know if clearing the transposition table every move is common
or not, but, given the above, i can't think of any reason why i would
ever want to do so.

	- edward screven



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.