OSNews

[q]
However, for any heavy lifting no one seriously considers Python. For my work on neural networks, I always described the difference between using C++ and Python as this: Would you rather spend 3 hours writing code that will take 3 days to run? Or would you rather spend 3 days writing code that will take 3 hours to run? Keep in mind that well written code gets reused, and it is not uncommon for code to be rerun millions of times during the course of your research.
[q]

It really depends on the task/field/condition, you cannot make big bold statement in the blank. But in scientific communities, matlab, mathematica, etc... (depending on the field) are huge, and used more than C++ most of the time, more exactly anytime you can get away with it (which again depends on the field: in video processing, you will have a hard time doing much in pure python for sure).

I find your example unconvincing, because especially in the academic environment, you keep changing the requirements, you are in a constant prototyping loop. The 3 hours run / 3 days coding almost never makes sense: code reuse in academic science hardly happens once you get above a certain level of abstraction, and any non trivial project involves a lot of non processing stuff. You have all the problems around the processing itself, such as data IO, visualization, etc... which are a big PITA to do in low level languages (that's also a problem in environments like matlab BTW), so you start using a scripting language for that, and you realize that you can use it more and more. You don't want to parse XML in C++, you don't want to do write your visualization in C++. Sure, you will not implement your core sparse matrix handling in python, but this is already done in fortran anyway. Fortran libraries, for numerical things, are good enough.

The point is nobody wants to use C++ anymore: you still need C speed for some stuff, but you really want to get away from it as fast as possible. That's not the default choice anymore.