Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: are you saying deep blue cleared hash with every move

Author: Robert Hyatt

Date: 06:51:42 09/09/01

Go up one level in this thread


On September 09, 2001 at 08:39:13, Vincent Diepeveen wrote:

>On September 08, 2001 at 23:18:09, Robert Hyatt wrote:
>
>>On September 08, 2001 at 14:02:41, Vincent Diepeveen wrote:
>>
>>>On September 08, 2001 at 13:53:03, K. Burcham wrote:
>>>
>>>Yes it had to, because it was a preprocessor program,
>>>perhaps it didn't clear 'best move' entries in the software.
>>
>>
>>This is not a given.  You can be a pre-processor without clearing the hash
>>between _every_ pair of moves.  You only need to clear the hash scores if you
>>change some of the evaluation weights.  Nothing says a pre-processor has to
>>do this after each move.  Early versions of crafty did some pre-processing, but
>>only at two distinct places in the game, namely on the transition between
>>opening to middlegame and middlegame to endgame.  That resulted in _exactly_
>>two hash table clear operations in a game.  I see no reason why your assumption
>>has to be valid at all.
>
>the important thing is that it didn't use hashtables at all in
>hardware and that it of course couldn't use transposition entries from
>a previous search. Obvoiusly you don't need to clea rit, but practically
>it comes down to the same thing.


Why?  If the root positions for move #X and move #X+1 are similar, then no
evaluation weights need to be changed.  And the hash table _and_ scores can
be re-used just fine.  This is what I did in early versions of Crafty.



>
>If you can't use search scores from a previous search, then in human
>eyes it is completely justified to say that this sucks bigtime!



But there is nothing to say they couldn't.  Crafty used to be a pre-processor
for lots of things, and it knew when to clear the hash and when it didn't need
to, with a simple check in the preprocessor phase.



>
>>
>>
>>>
>>>Note that it didn't use hashtables onboard of each hardware
>>>cpu either.
>>>
>>>>
>>>>
>>>>
>>>>
>>>>i assume deep blue had some type of memory storage for eval
>>>>    of previous lines searched. if it did are you saying  that
>>>>       deep blue cleared its memory (hash), with each move?
>>>>
>>>>if it did not clear its memory with each move then your seven
>>>>   second search limit does not make sense to me.
>>>
>>>>and can you tell me which game this move was taken from, so that i
>>>>  can get a better analysis of this position. i can find it, but you
>>>>    can save me a lot of time, if you will.
>>>>
>>>>i will post my results after the analysis of this position.
>>>>
>>>>kburcham



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.