4 comments

[ 2.9 ms ] story [ 20.5 ms ] thread
Question: How do you control false positive rate mining correlations?

Answer: We use false positives as a negative feedback mechanism which refines our models. We have dials for precision and recall depending on how much we want to risk looking stupid.

I would be highly interested in seeing how strongly this feeds into their system. A bit too much of the false positivity taken into account could make them look way too cautious and hinder any sort of adventurous recommendations. I'm sure they have taken this into account, though.

The linked page breaks the back button. Why would they do that?
Some of their answers seem a little nonsensical. Python, for example, is not blazingly fast, but it is good for some things like I/O bound systems.
Interesting response to "What do you think about TDD?" "I'm assuming you mean test driven development...taken to an extreme I think it doesn't make much sense for a consumer web startup. Probably makes more sense if I was building the space shuttle."

Space shuttles are traditionally built with formal methods and SEI CMM Level 5, so it's pretty hard to see them going the agile route. Conversely if done well, TDD's great for a web startup because it makes it so easy to evolve.