Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hash Collisions

Author: Robert Hyatt

Date: 11:21:31 02/23/99

Go up one level in this thread


On February 23, 1999 at 13:17:59, J. Wesley Cleveland wrote:

>On February 23, 1999 at 13:00:18, Don Dailey wrote:
>
>>On February 22, 1999 at 21:57:24, Robert Hyatt wrote:
>>
>>>On February 22, 1999 at 19:53:14, Peter McKenzie wrote:
>>>
>
><snip>
>>>>
>>>>OK, so the $64 question is: will using this table cause any measurable
>>>>improvement in your program?
>>>>
>>>>For example, will your node count drop if you do a fixed depth search on a test
>>>>suite?
>>>>
>>>>cheers,
>>>>Peter
>>>
>>>
>>>No..but you should get fewer undetected hash signature collisions...  Whether
>>>that will affect scores/moves at the root for any position is a good question.
>>>Without a good answer, at present..
>>
>>I would guess that using the "better" numbers will decrease the number of
>>collisions.  I don't know if the improvement is enough to actually
>>measure (Hey, the program is player stronger now!) but it is an improvement
>>that comes at no cost and as such should be hard coded into the program.
>>
>>A good program will have a large number of little improvements like this
>>one.  Get 10 or 20 things like this together and they start adding up.
>>You can easily live without any particular one, but not all of them
>>together.
>>
>
>I think that the big advantage here would be to avoid the occasional times that
>you get a *terrible* set of hash numbers. (the more important the event, the
>more likely this is to occur ;)


Note that I'd suspect that most are using 'psuedo-random' numbers,which means
you get the _same_ set each time you play a game.  If you don't do this, you
can't use the hash signature for the opening book, for example...



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.