NumCpp VS RxCpp

Yeah, as someone who writes C++ daily for their ML related job, I concur that the cost of executing a convolutions dwarves the overhead of calling from Python. So as much as I like C++ over Python (because static compilation to find little typos or type mismatches ahead of time is much nicer than exploding 5 minutes later into my batched vision recognition problem 😠), generally for small problems, Python is a nice quick and dirty approach. I do have my eye though on this little C++ numpy clone.

Despite being written in C itself, the primary external API is for Python, and though it is possible to call via C, it's quite ungainly (several ref-counted Py_* calls and structs). It's probably easier to just consume a library that targets C++ directly like xtensor (https://xtensor.readthedocs.io/en/latest/numpy.html) or NumCpp (https://github.com/dpilger26/NumCpp).

I am trying to feed 2 different 3 column 1 row arrays into a linspace function using the NumCPP package, but i'm getting errors such as:

And, prior to that https://github.com/ReactiveX/RxCpp

Yeah, I know this problem with operators =), original RxCpp implementation also has this problem with implementing all functions into base class and calling dependent functions internally.

Also you can check rxcpp with documentation about reactive approach there. It is functional version of observer/publisher-subscriber patterns with ability to be multithreaded. You can send events from one side, subscribe from another and modify events in meanwhile

I use a very large header-only library RxCpp. Simply adding #include "RxCpp/rx.hpp" to one .cpp file adds >1 second of compilation time. I'd like to use it as a module, but when I try to import "RxCpp/rx.hpp";, I get a bunch of errors.

+1 for Rx, specifically RxCpp. I've used it in concert with Qt with great results.