Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: pawn hash tables

Author: Gerd Isenberg

Date: 14:18:34 08/01/03

Go up one level in this thread


On August 01, 2003 at 17:07:07, Omid David Tabibi wrote:

>On August 01, 2003 at 16:49:10, Gerd Isenberg wrote:
>
>>On August 01, 2003 at 16:01:51, Joshua Lee wrote:
>>
>>>Would someone please explain pawn hash and how much of the whole hash tables it
>>>is when you set them to lets say 256MB?
>>
>>Hi Joshua,
>>
>>as Rick already mentioned, pawn hash tables are for pure pawn related eval
>>stuff. It requires a separate pawn hashkey, incrementally updated if pawn
>>formation changes, similar to main hash table.
>>
>>The size depends on the size of your entries and the number of entries of
>>course. The number of entries should be much more less than in main hash table,
>>64K entries are quite enough IMHO. But the entry size may be bigger than in main
>>hash, depending on what you store (some aggregated eval score or several
>>bitbooards with some important properties).
>>
>>If i think about my eval with a lot of pawn-king or piece interactions...
>>May be i throw out pawn hash table in future IsiChess. A lot of pawn patterns
>>like, isolated, backward, double, triple, candidates, passers, outside passers,
>>duos, rams etc. may be generated on the fly with bitboards and stored a while.
>>
>>On the other hand, if you hash pawn eval, you may do a lot of really expensive
>>things with pawns -
>
>>the hit rate is about 99%.
>
>That's about the hit rate I get in closed positions, but for example searching
>from the start position, I only get a hit rate of of slightly over 90%.

You are right, Omid.
Starting position seems really be a worst case for pawn hash hit rate.
After 1.e4 e5 it becomes better, but even less 99%.

Gerd

>
>
>>
>>Regards,
>>Gerd
>>
>>>
>>>Thankyou



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.