[RogerL]

Let's see, I'm learning about unscented Kalman filters, moving object tracking in computer vision, PCA, and more. I'm learning about the problem domain that I'm working in. I have no interest in learning yet another API unless I absolutely have to, because it isn't really interesting knowledge that I have to call the fibble() function before calling xacxtyor() in this poorly documented API. That's not learning, it's the accumulation of facts.

Of course, sometimes learning new API/languages enables meaningful learning, and I'm all for it. Want to quickly get a handle on Kalman Filters? Probably a lot faster to put something together with numpy (say) and experiment on the REPL than to program it in a performant language like C or C++, even if that is where it will eventually end up running in your project. So, yes, there's a great reason to learn those languages (or Julia, or whatever).

But the endless march of APIs is pretty tiring to me, and I do everything I can to isolate myself from that. I'm pretty happy writing my C++, and Python (2.7, btw), while trying to solve rather hard problems.

Another way to put it - what I value is intellectual work, not memorization/puzzling things out. Learn a new math technique? Awesome! Learn a byzantine set of calls I need to make to make a widget widge on the screen? Not so much.

[JackMorgan]

Well it sounds like you have no qualm learning the right tool for the job. I would consider numpy to be an API, but the difference here is it's one that helps you with your domain. If widging widges was your domain I'd expect the professional thing to do would be for you to learn that API. I'm talking about people who in your job wouldn't bother with Julia or numpy and would just trundle along with C.

[noelwelsh]

This is great, but it's not addressing the point, which is "I am, however, very excited to try things that might reduce amount of bugs and potential security issues in my code, make it cheaper to run, cheaper to support."

Also, if you like math go learn some functional programming. The ideas of abstract algebra, which you will encounter in FP, are increasingly relevant to machine learning as well. E.g. http://jmlr.org/proceedings/papers/v28/izbicki13.pdf

[auvrw]

> But the endless march of APIs is pretty tiring to me

i hear you on that one

> and I do everything I can to isolate myself from that.

but still, you can't stop thinking about design entirely. when a new ideally-pure-C-but-C++-if-you-really-need-stl library is ready to actually be put to use you'll have to put an api of some sort on it. hopefully that api will be informed by your use of other people's apis, both good and bad.

don't forget that the whole point of computers is that they're immediately useful. you can still do most of the math you want with a pencil, paper, and stack of books.