Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Origin of Test Suites/Positions?

Author: Mike S.

Date: 15:50:12 06/22/02

Go up one level in this thread


On June 22, 2002 at 10:42:46, Robert Henry Durrett wrote:

>On June 22, 2002 at 06:45:54, Mike S. wrote:
>(...)
>>which were selected for human training or reading pleasure so to speak, nasty
>>things like in-between moves (which may be unimportant for the human when they
>>are obviously waste of time), can ruin the test positions.
>
>Interested in better understanding of "nastiness" of in-between moves, as it
>applies to computer test suites.  What "nasty" problems are caused by this?

Some engines will play the intended solution immediatly, some others might see
it too, but will play the in-between move before (i.e. a strong but simply
refuteable threat) which delays the solution move. In these cases, you can't
measure the solution time of the program which plays the in-between move, in the
original test position. Afterwards it's a different position. Also there will be
engines which play the in-between move but would not have found the solution
move previously (which you can never really tell because of that). IOW, useless
for a test suite.

In practise, demands are even more strict. For example, when I released my
Quicktest, I was not aware that many people use test suites (seemingly only)
with *automatic* testsuite processing functions. Some of those functions can't
handle avoid move tests. So these program "solved" the am positions by "finding"
the wrong move immediatly... :o)

But I keep the am positions, and ask people to test manually. It takes only ~30
mins total. It's also much more informative IMO, to watch the thinking process
of the various engines "live".

>One idea which intrigues me that it may be possible to write a program which
>would search a large database of games to find positions which meet the criteria
>you suggest.  Ever heard of anything like that?  Think it is a good idea?  I
>realize that the only large databases would be of human games, but maybe that's
>unavoidable?

(see posting above)
http://www.talkchess.com/forums/1/message.html?237027

That describes the "manual" method. It should be possible to write a program
which finds such evaluation jumps (especially if both engines show one), if the
annotations/search info follow a standard format of some kind, i.e. from a PGN
file. Maybe such a program even exists, I don't know. But such a program could
only guess, IOW. offer a "rough" selection of positions which *may* be useful.

Another idea is to quickly browse through games in the Fritz or ChessBase GUI,
with the *material* info line on (or in another program which shows the material
balance in a similar way). Usually, the (tactical) interesting positions will be
somewhere near the changes of the material balance. For example, if the winning
program gets behind in material. Which most often means that it has played a
successful sacrifice.

I think I found a lot of test positions that way.

Regards,
M.Scheidl



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.