Question

Python is a high-level interpreted language, which greatly reduces the time taken to prototyte and develop useful statistical programs. However, the trade-off is that pure Python programs can be orders of magnitude slower than programs in compiled languages such as C/C++ or Forran. Hence most numerical and statistical programs often include interfaces to compiled code (e.g. numpy which is written in C) or more recently, are just-in-time compiled to native machine code (e.g. numba, pymc3). Fortunately, it is relatively easy to write custom modules that comple to native machine code and call them from Pytthon, an important factor in the popularity of Python as a langugae for scientific and statistical computing.

We will use the example of calculating the pairwsise Euclidean distance between all points to illustrate the various methods of interfacing with native code.

Adapted and extended from http://nbviewer.ipython.org/url/jakevdp.github.io/downloads/notebooks/NumbaCython.ipynb

A = np.array([[0.0,0.0],[3.0,4.0]])
n = 1000
p = 3
xs = np.random.random((n, p))