Fast Python

Appendix B. Using Numba to generate efficient low-level code

Numba is a framework to auto-magically convert Python code to native code—CPU or GPU. It is, on the CPU side, an alternative to Cython. The reason we have an entire chapter on Cython and not Numba is because in this book we are interested in understanding how things work and not only in making them work. Numba, which is great otherwise, is not great from a pedagogical approach—due to being “magical.”

To solve real-world problems, Numba is as good as, if not better than, Cython as it requires less work on your part and produces similar results. From a usability perspective, I recommend you consider Numba as an alternative to Cython. Actually, it’s probably more pragmatic to consider ...