Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash replacement scheme

Author: Robert Hyatt

Date: 13:51:06 10/12/01

Go up one level in this thread


On October 12, 2001 at 16:48:32, Alvaro Jose Povoa Cardoso wrote:

>On October 12, 2001 at 14:05:50, Robert Hyatt wrote:
>
>>On October 12, 2001 at 13:55:36, Alvaro Jose Povoa Cardoso wrote:
>>
>>>I red some stuff somewhere of a hash replacement scheme that uses the node count
>>>of the subtree bellow the node we are considering.
>>>It came to my mind the following:
>>>At the top of the search we probe the hash table for a position and we would use
>>>this node count to determine if the hash position should be given credit or not.
>>>_But_ at the top of the search we simply don't have a node count of the subtree
>>>bellow this node because we didn't search anything yet.
>>>So my question is: How do we compare the current position (wich has no subtree
>>>node count) with the hash position (wich has a subtree node count)?
>>>Using the 'draft'  instead o f the node count we don't have this problem.
>>>
>>>Any comments, please?
>>>
>>>Regards,
>>>Alvaro Cardoso
>>
>>
>>You don't use the node count for matching.  you use it for _replacement_.  IE
>>if you go to store an entry in the table, you first ask "which has the largest
>>node count" and you keep the one that does.  That is similar (but not identical)
>>to the concept of "depth-preferred replacement" that most of us use...  we keep
>>the entry that represents the _deepest_ search...
>
>
>I understand the replacement part.
>What I meant was: when I read from the hash table at the top of the search and
>if the hash position match the current position I still want to check if the
>hash position is worth being considered. In the depth-preferred scheme I do this
>by checking if the hash draft is greater than the current draft. But in the
>subtree node count scheme I simply don't have the current subtree node count
>because I haven't searched anything yet.
>It is this part that puzzles me.
>
>Alvaro Cardoso


You _still_ have to check the depth before using the score.  The node count
is an _additional_ piece of information you use only to control the replacement
policy.  Not when you use the entry...




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.