How novel would it be to link vector entries or variables' importance to the correlation between the function's derivatives and frequency components?

Isn't the importance of variables determined typically through things like coefficient analysis in regression or other?

No idea but it sounds fascinating

@antigermanist No idea either. AI spit it out in a semi-related discussion, which is why I asked.

I just wanted to confirm if it was somewhat plausible, very plausible, or total bullshit.

The hyperdimensional search engine strikes again!

Please have my usual upvote after leaving me curious, fascinated and very confused again.

I'm gonna give my very uniformed opinion.

@Wisecrack as usual, has a very valid standpoint... In principle.

I guess it's rational to measure variable use. That's a solved problem.

Thing is, it's not mathematical. Derivative of a variable doesn't make sense in the programming aspect.

You can make it statistical, which is what many code analyzers already do.

But I don't think there's any value to be extracted from that.

And of course, you get the upvote anyway, because even if I don't agree, it takes knowledge and guts to post it, and who knows, maybe in the end tell me I'm a fucking idiot.

@CoreFusionX When I speak of variables, I'm not talking about internal variables to a function, but its parameters, especially in relation to the vector inputs of a neural net.

But I appreciate the hell out of anyone that has even an inkling of what I'm talking about, posting their opinion or insight.

@Wisecrack that's how you do investment portfolios

find a common space between those vectors, rewrite everyone in that space. Build a covariance matrix between everybody and solve it for covariance 0. Pretend that common space is called time and do it with stocks and BAM! You've got a risk free portifolio

Not everybody will make the cut though, so you can happily discard anyone unfit for duty

@kobenz "that's how you do investment portfolios"

ooh, now thats something new. Tell us more!