Zitat von
Mythbuster
Hallo,
was an der SSDF Liste zu Verzerrungen führt, sind die Paarungen: Häufig findet man völlig unsinnige Paarungen von Geräten, wo es zu eindeutigen Ergebnissen kommt ... wie 20:0 oder 19:1 ...
Jeder, der sich mit der Materie beschäftigt hat, wird feststellen, daß bestimmte Geräte regelrechte "Angstgegner" haben, wie Menschen auch. Wenn ich jetzt ein Gerät "A" in erster Linie viele Partien gegen seine Angstgegner spielen lasse und nur wenige Partien gegen Geräte, die ihm gut liegen, wird die ermittelte Elo Bewertung auf jeden Fall schlechter ausfallen, als bei einer gleichmäßigen Anzahl von Spielen gegen alle Gegner.
Sorry to post in English, I understand German very well, but it is difficult to write correctly. The SSDF was totally independant. The principles for testing were heavily discussed. Some of the computers played very few opening lines and that was an advantage when the computers were released because they outbooked older computers. When next years models came the computer with few opening lines were again easily outbooked by the new competitors. SSDF took the stand that a narrow opening library was a weakness and therefore allowed testing where a computer could loose 18-2 to another computer of approx. equal strength just because of the library. What is important is that these results did not have very much impact on the rating. In an article they showed that removing the "booked" result just differed a few points because of the great nr. of played games. A weakness was however that it took long time for a computer to stabilize itself on the list and it could initially get a higher rating than it deserved because of the library.
Regards from Norway (not Sweden
)
Erik