Numba - JIT compilation & optimisations for Python

[Will]

Numba

Something I stumbled upon recently that appears to give decent gains for ordinary Python code (i.e. without the ML frameworks).

It has some additional decorators to drive it & interesting features like vectorisation for parallel/matrix operations etc.

https://numba.pydata.org/

Not as fast as, but at least approaching Cython performance:

JIT fast! Supercharge tensor processing in Python with JIT compilation

Numba examples (Jupyter notebook)

[Ed_S]

Interesting - that’s new to me. I have had some success with Pypy, which at one point was improving in leaps and bounds, but I think it has calmed down now. (It’s not specifically for numeric work - there might be cases which suit one of these but not the other.)