Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Stuffing the PV

Author: Tony Werten

Date: 05:28:33 12/07/00

Go up one level in this thread


On December 06, 2000 at 16:12:46, Robert Hyatt wrote:

>On December 06, 2000 at 15:52:47, David Rasmussen wrote:
>
>>On December 06, 2000 at 14:02:55, Scott Gasch wrote:
>>
>>>>Even more important... as you stomp down the pv path, don't forget to
>>>>increment the depth just as you would for a real search down that path.
>>>>IE checks, recaptures, etc...  and as you go a ply deeper, decrement
>>>>depth accordingly.
>>>>
>>>>Otherwise you will have _lots_ of fun debugging.
>>>
>>>Actually right now in StuffPV I am saving PV nodes in the hash with an EXACT
>>>hash entry and the score returned by Search but with a depth of zero IF the node
>>>already in the hash at that position is not the PV node (i.e. it got clobbered).
>>>
>>>I know that someday I need to correct this and make the depth into search_depth
>>>- ply + extensions or whatever... but for now the goal for right now is just to
>>>get these moves ordered first and back in the hash.
>>>
>>>Thanks,
>>>Scott
>>
>>If Bob means the same thing as you, then I'm confused. Because he doesn't do
>>that in Crafty. Correct me if I'm wrong.
>
>
>I store 0 for the draft in crafty for _all_ PV nodes.  I was answering the
>question based on how I would store a _real_ draft for PV nodes.  I don't
>see any benefit to doing this...

That's an idea worth trying. It would solve a complete different problem I have.
When my opponent plays the expected move, but I didn't search long enough, I
restart my search. Then I get 10 ply instantly (hashtable) , but die in ply 11
because I'm basicly starting a search without any knowledge. ( no killers,
history etc ) If I store the pv with draft=0 I force the engine to search this
pv and at least have some information.

Tony



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.