Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Pawn Transposition Table

Author: Robert Hyatt

Date: 20:32:44 01/12/98

Go up one level in this thread


On January 12, 1998 at 17:49:04, Don Dailey wrote:

>On January 12, 1998 at 15:12:10, Robert Hyatt wrote:
>
>>On January 12, 1998 at 14:07:59, Don Dailey wrote:
>>
>>>You may not have a problem.  A assume given a longer search your
>>>numbers will climb to the 95-99% range.  Is your table pretty
>>>small?
>>>
>>>Now that you have ps hash tables there are lot's of wonderful tricks
>>>you can play.  I store lot's of info in mine that the evaluation
>>>can use further, like locations of passed pawns and holes etc.
>>>
>>>- Don
>>
>>So do/did I.  Only problem is that as I improve things, that which
>>depends only on the locations of pawns is getting smaller.  I flag
>>passers, protected passers, weak pawns, and so forth.  But the savings
>>is not really all that great since I have to use these in evaluating
>>pieces.
>>
>>But the "structure" can certainly grow.  And everything is basically
>>free
>>if you can stick it in there, and if you have enough memory for a decent
>>table (even 32,000 entries is enough for most positions if you look at
>>how many ways the pawns can realistically be moved around)...
>
>
>Yes,  mine keeps getting bigger as I add more features.
>
>A long time ago, purely by accident while implementing the pawn
>structure
>hashing I hashed the king into the pawn structure.  I knew my numbers
>were wrong and found the problem quickly.  But it gave me the idea to
>do this on purpose and get some king safety too.   I never did but
>thought it might be worth a try later.   I now see it's been done before
>which shouldn't surprise me.
>
>I like your king safety idea as seen in another post.  My king safety
>is good but expensive.  I'm considering using this idea but it won't
>cover all the king safety stuff.  I think it might cover the most
>expensive stuff though which is what I need.   Thanks for the idea.
>
>- Don


In Cray Blitz, and early versions of crafty, I had a separate hash table
for king safety.  But it wasn't quite as efficient due to the kings
being
thrown into the hash signature.

The current approach works well for the pawn part.  But the rest of my
king safety is highly dynamic and unhashable, because I distribute king
safety over all the opponent's pieces, since they are what makes my king
safety important.



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.