View Single Post
Old 05-22-2013, 09:19 PM
doneit doneit is offline
Join Date: Jan 2013
Posts: 15
Default Re: All things considered...

Elroch, thanks for the reply and clarification. I think my post may seem a little pessimistic in retrospect. Maybe I should have said that for the unwary the devil seems very much in the detail (or in this case maybe that's "in the data"?). Lecture 17 highlights some of the subtle dangers and there are some almost surreal aspects to the theory throughout the course (I suppose starting with the fact that you can get any sort of handle on out-of-sample performance at all). My comments were really alluding to theory-vs-practice and relative importance there-of, especially the latter where, as you say, a disciplined approach to handling the data is needed and that's where experience counts.

For example I recently came across Grok and while I don't have the capability to assess it's claims I'm betting the human-element isn't fully excluded from the process for some of the reasons raised in lec.17. IOW's the topic may be *machine* learning but that doesn't mean the whole process can be mechanised (can data know it's biased?).

Reply With Quote