Author: Yen Art Tham
Date: 12:35:27 12/03/02
Go up one level in this thread
On December 03, 2002 at 09:26:42, Eduard Nemeth wrote:
>Very interesting post from SMK in CSS Forum (only german).
>
>Please read it, i thing that a translation is interesting for You!
>
>Read here:
>
>http://f23.parsimony.net/forum50826/messages/54995.htm
Which you accuse of here me, with position tests too schummeln is
the absolute irony. Thus that clearly becomes the following history:
On a boring, rained day I sat down here to the computer and a nice
Tool into Shredder built. By Hotkey I can "learn" a position or in
addition, a complete test series (e.g. the WM test) in Shredder in a
few milliseconds. What is called "learn"? Now, Shredder finds the
solutions and/or the courses which can be played then clearly
rather. This procedure is absolutely perfectly camouflaged, i.e. with
as awkward cheat as mirrors or tricks or something similar can one
it naturally not out-cheat. Learning is also controllable. I can adjust
it in such a way that it notices each Depp, since all courses are
out-played always immediately or however it to camouflage in such
a way that it is noticeable absolutely reliably to nobody. With
absolutely surely I mean absolutely by the way safe. If I so which
into Shredder build, then one can assume thus it cannot be
discovered if I it does not want. The "normal" play strength is not by
the way impaired thereby. With the completion of Shredder7 then
the question was acute whether I was to then build the "Tool" into
the commercial version or not and whether I already times few
well-known positions a Shredder "to teach" am. The reason, for it
spoke was to be shown, how unreasonable it is from position tests
any practice or analysis-relevant Elo numbers except a certain "test
number" to derive. A further reason is that that will beschissen
wants will usually also beschissen. Speak if someone it finds mad,
if a program fares well in a test, then I can supply with also that to
it. * Hero mode * I did not build it on then after careful consideration
nevertheless into Shredder7. Reason was that I the people, which
with Shredder and a test are occupied for hours and do not observe,
as Shredder in certain positions behaves and wants with other
programs to compare not verarschen want. So after the slogan:
Mach you the trouble and notes all solution times, those is anyway
all gefaked. * Hero mode out * If I already times thereby am:
Position tests are in my opinion suitable for comparing program in
this certain test with one another. They are not suitable for on my
account also closing any large conclusions on more general play
strength or however "analysis strength". The this like that is not
makes for me my life by the way many heavier. As beautiful would
be it nevertheless, if I had to hunt a new test version of Shredder
only once by a test and on the next morning had I then to the
breakfast the result whether the new version is now better or worse.
Unfortunately is many heavier this and lasts around a multiple
longer. Tests are sometimes suitable for recognizing weaknesses of
programs in certain positions or on my account also position
samples. That is however expresses with difficulty and it is not
thereby done simply only the positions to be taken, which are not
solved from a program. One must understand to try, why a program
a course not out easily and whether this is then a generally
accepted, important reason. It is not possible general statements to
meet how: X solves 60% in the final game, Y solves 70%, therefore
Y is better in the final game. Can position test show, where one
could set possibly with an improvement. Candidates are then the
positions, which solve other programs very fast, the own however
not. By far all such positions is not worthwhile itself it then however
to look for improvements. Is not been to me to destroy here any
life's works or like but I think, one should have done the
conclusions, which one draws from a test correctly arrange can.
Stefan
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.