Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Positions to store in the hash table

Author: Matthew McKnight

Date: 11:14:47 12/17/03

Go up one level in this thread


On December 17, 2003 at 13:19:55, Alvaro Jose Povoa Cardoso wrote:

>On December 17, 2003 at 12:01:19, Matthew McKnight wrote:
>
>>On December 17, 2003 at 11:52:33, Alvaro Jose Povoa Cardoso wrote:
>>
>>>Hi, all.
>>>I saw several discussions in the past on hash table replacing schemes, optimal
>>>hash table sizes, wich information to store there, etc.
>>>But I don't remember any discussion about what positions are the best candidates
>>>to store in the HT.
>>>What most of us do is to store positions in wich the best move is a fail high
>>>or, an exact score, wich is correct for move ordering purposes.
>>>I was wondering if someone uses a different criteria for choosing positions to
>>>store in the HT.
>>>When in some middle game positions I see a drop in the hash table hits (below
>>>15%) of my engine I sometimes think if there is a better way to improve the
>>>transposition effect, if such a thing can be done.
>>>
>>>My best,
>>>Alvaro Cardoso
>>
>>I'm not sure if this answers your question, but I use two tables.  One of them
>>is the normal, "depth-preferred" table.  The other one is an "always store"
>>table.  If I am adding a position to the table, but the entry in the DP table
>>comes from a deeper search I leave the deeper information and put this new
>>information in the "always store table," replacing without prejudice.  Then in
>>the readPosition() function I first check the depth preferred, and if there is
>>nothing there I look in the always store table.  This makes a noticeable
>>difference with HT performance.
>>
>>Matt
>
>Thanks Matt,
>but my question is more in the line of how do we choose positions to try to
>store in the HT. I don't remeber seeing a discussion about that here.
>Currently we choose a position with a best move that caused a fail high or an
>exact score (above alpha). I wonder if that is all we can do.
>
>Regards,
>Alvaro

You can also store a lower bound for positions that fail low, so that if you
come to that position again, you can adjust your window if need be.




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.