Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Replacement hash table

Author: h.g.muller

Date: 11:35:26 03/05/06

Go up one level in this thread


I experimented a little bit with the replacement policy, and was surprised to
find how well the most simplistic strategies performed. Sometimes they clearly
outperformed my attempts to do it in an intelligent way... :-S

It might depend a little on the details of your normal search, my search was
very conservative, doing iteratively deepening the search in the smallest
possible steps from the retrieved hash depth (if any) to the requested depth.

Under those conditions simply overwriting anything there might have been in the
entry to which a position hashes on the first try, without any attempt to rehash
and without any regard to the depth of the result you overwrite worked quite
well. What worked absolutely disastrous was to hold on to deep results that were
not suitable for the current window limits (so bounds of the wrong type, rather
than exact values). That brougt the search almost to a crashing halt, because
the table would be completely polluted by deep results that were no longer of
any use (because of a readjustment of the value of the root position, and
without the benifit of being able to hash these positions at lower depth with
useful bounds the search would never get deep enough to correct the situation.



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.