Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Measuring closeness to a minimal tree

Author: Vincent Diepeveen

Date: 06:43:20 04/06/03

Go up one level in this thread


On April 06, 2003 at 09:01:43, Sune Fischer wrote:

>On April 06, 2003 at 08:45:30, Dan Andersson wrote:
>
>> A precise formula? Not practical AFAIK with regards to transpositions and the
>>possibility of choice of lower branching, higher transpositional subtrees.
>>
>>MvH Dan Andersson
>
>It might be possible, in the same way it is possible to use a hash for counting
>perft.

for small search depths that might be true to some extend. But the major
problems you get when you assume that the program uses a hashtable of its own.

So influences like overwriting hash entries and loading factor of it get of
overwhelming importance.

Note that if there would be a technique to determine a much smaller search tree
in an easy way for big search depths (so not too much overhead to count things)
then i would consider doing the experiment myself too.

It's very itneresting to know for me whether at a search depth of say 12 ply,
whether a good move ordering could get it from say 12 million nodes to 1.2
million nodes or whether 8 million nodes is more realistic. It tells basically
what you could achieve by a better move ordering.

>You have to add a few counters to your hash entry, each entry must know the
>average depth to which it searched. I suppose you might even need an array (like
>int [maxsearchdepth]), so when hitting this entry you add all the counters here
>to the current branch.
>You need of course some kind of triangle array for counters, just like when you
>assemble the PV.
>
>It's bound to be a mess, but probably doable :)
>
>-S.



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.