Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: nodecount == 1 move right?

Author: Vincent Diepeveen

Date: 16:35:31 02/15/06

Go up one level in this thread


On February 15, 2006 at 19:19:07, Joshua Shriver wrote:

>I thought nodecount had a strong definition already. 1 node == 1 move
>possibility in the search tree. Not sure how this number could be made different
>other than what the actual number is.
>
>Josh

Well even in the case of Diep it's difficult to count like for example Fruit is
doing it.

In case of Diep, should i count the number of generated moves, or should i count
the number of visited positions, or should i count the number of visited nodes
including the number of avoided visits?

You know there is a huge difference there.

Let me give you an example. At a simple outdated K7 the number of generated
moves a second in search is 2.2 million a second. That doesn't include a fast
searcher of mine which generates effectively at 6 million a second.

But let's show a huge difference between the fast searcher and diep already in
terms of counting nps.

The fast searcher is detecting a repetition in top of the search. So that counts
as a node. In diep it isn't counted as i detect the transposition
already directly after making the move, so before doing another iteration (diep
is not recursive). In most programs after making an illegal move, the king gets
captured. In diep this is highly unlikely to happen. It will never search an
illegal position. Usually it detects illegal moves directly after making a move
as my incremental attacktables give it already.

Of course in the fast searcher, just keeping track of incremental attacktables
is already too expensive, as it is searching at 3 million nps (K7 2.1Ghz) and
quite a bit more at A64.

This is a huge difference in nps practical.

The old diep had a very stable nps. If i count like the fast searcher is
counting and this is how most engines count, in some middlegames i get 2.5 times
more nps than diep reports now. In openings position of course repetitions and
illegal positions happen less, so obviously the increase in nps there is pretty
tiny but still significant in absolute numbers. It's more like 30% there. Not
250%.

Yet on average the increase is 50-60% in a game in nps on average.

That's quite significant.

Vincent



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.