Going back and reading one of the classic structure papers from the end of the last century, I found:
"where T is the test set (one reflection in each of at least 2,400 thin resolution shells[41] of the data)".
It seems to me that this could describe even a randomly selected free set.
Are the authors poking fun at the "thin shells" proponents, or finding an
expedient way to satisfy a reviewer's requirement? Or is there a legitimate
selection strategy that could be described like this?
The ref[41] is Kleywegt and Brunger 1996, which finds the results
are largely independent of the test set selection method.
eab
|