need to reduce pf rate
	shepperd's association rules?

is there some trheshold effect between distefano04 (5 data sets)
and here (12 data sets)
	is there something about too much?

we don't explore other learners. much work to select current combinations.
we are aware of other researchers trying to do better. that is work in progress
but, informal sources tell suggest that our preferred methods aren't statisticaly
significantly worse than our preferred method.

this "mega" learning could be quite common in the future. once we get a data
source, it becomes a data flood.