An introduction to variable and feature selection [pdf]

[sonabinu]

http://www.jmlr.org/papers/volume3/guyon03a/guyon03a.pdf

[notroot]

Anybody wanna verify that the fnr in this sentence on the bottom of page 5 (1151) is an error and should be fpr?

"ROC curves that plot 'hit' rate (1-fpr) as a function of 'false alarm' rate fnr are instrumental..."

Was gonna email them but considering published in '03 figured we were more interested in this than they are at this point.

[Bill_Dimm → notroot]

The fpr and fnr should be swapped (note: I haven't read the rest of the paper yet). Specifically, it should be:

ROC curves plot (1-fnr) as a function of fpr.

[elcapitan]

"Optimum brain damage procedure" made my day. I'll try that at work.

[jmount → elcapitan]

It is neat- the need to look at the second derivative in "Optimal Brain Damage" (because we know the first derivatives will be zero due to the use of optimization) also underlies Fisher information https://en.wikipedia.org/wiki/Fisher_information .

[lowmagnet]

Can we get a year tag? This appears to be from 2003.