Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: another hashing question

Author: martin fierz

Date: 09:28:05 11/22/00

Go up one level in this thread


On November 22, 2000 at 10:06:24, Robert Hyatt wrote:

>On November 22, 2000 at 05:38:32, martin fierz wrote:
>
>>hi,
>>
>>i was wondering about this PV-hashtable thing: in my checkers program, i do not
>>generate a separate PV. i have two hashtables, a 'shallow' and a 'deep' one,
>>positions close to the root of the tree are stored in shallow, the rest in deep.
>>in shallow, i take care not to fill the table and not to overwrite entries. i
>>took this idea from the crafty source code. in deep i overwrite everything
>>without looking twice. question: i have no PV, but i have a guarantee that for
>>the first 10 ply i have a hash move - because thats about what fits in shallow.
>>for the next 10 ply i won't have a hash move. do you think that this is a
>>serious problem? should i be generating a PV and stuffing hash moves back in the
>>table (that surely costs some speed...)? typical search depths are 15-19 ply in
>>a couple of seconds. do these last 5-9 ply really matter?
>>
>>cheers
>>  martin
>
>
>They are often important for debugging, more than anything else.  IE the goal
>is to play the right move.  Nothing says you have to have the right PV, when
>playing a game.  But if you can't see the PV when testing, you have a harder
>problem to understand what is going on and why.
>
>DB didn't have a separate PV because the hardware didn't support the idea.  They
>seemed to play just fine.  So yes, it can work...

thanks for the answer. of course i can get a PV from the hashtable this way, as
long as the position is still in the hashtable, which it is for the first 10
ply. but i see what you mean, i could get the whole PV to see which position
determines the final evaluation and also to see if the quiescense search is
performing properly. i suppose i should implement it :-)

cheers
  martin



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.