Publications

I think the lesson in the last two paragraphs in Gary Weiss email (GPS: see previous item) is very important. Researchers have done experiments modifying the class distribution, and have not recalculated the decision threshold (calling into question the results). I imagine that practitioners also might find this useful.