need notes on bottom up vs top down data mining

both are incomplete, misguided
	- top down so focused and slow on one case study that technology rushes past
		- also, generality blind

	- bottom up can answer questions no one cares about. 

so combine both. 
	- top down for goals looking for questions
	- bottom up looking for answers
	- the missing link are the answers that address current questions
	- understand the open issues in the field/business so you can produce answers that matter (to someone)

but bottom up is not necessarily a slave to top down
	- bottom up can add main questions
		- e.g. stability
	- and some of these questions may be answerable before the top down ones
		- e.g. does CMM work? vs "stability"