Have you ever tried to use Cython??? It’s terrible. Try to Cython a Python file and then let me know if you think it is that easy. The packages that you are referring to weren’t cross-compiled into C with Python.h and Cython, they were written in C and then called from Python.

Have you ever written a machine-learning module in C? Well I can tell you one thing, it will take a long time. Here are some benchmarks between Julia and other languages, so you can understand Julia’s speed compared to Python.

The Julia Language

If you never need to use anything but packages like Sklearn and Tensorflow, then I’m sure Python is fine. Massive data pulls and anything that is custom is where you’ll hit the significant performance issues.