Basics [Cython]

---

• Author: Diego Inácio
• GitHub: github.com/diegoinacio
• Notebook: basics_Cython.ipynb

---

Basic functions and operations using Cython and Python.

Installation

---

Installation command for anaconda and pip:

$ conda install -c anaconda cython

or

$ pip install Cython

import cython

Compilation

---

A Cython source file has the name of the module followed by the extension .pyx. For example, given the source file examples_cy.pyx with a simple function which returns a string.

def hello_cython():
    return 'Hello, Cython!'

The following step consist of creating the setup.py, which will be responsible for the compilation process.

from setuptools import setup
from Cython.Build import cythonize

setup(
    name="Examples Cython",
    ext_modules=cythonize("examples_cy.pyx")
)

Given that, the compilation step is done by running the command:

$ python setup.py build_ext --inplace

from examples_cy import hello_cython

print(hello_cython())

Hello, Cython!

Performance

---

The following example, we will try to approximate the value $\large\pi$ with the idea of $\\tan^{-1}1=\frac{\pi}{4}$ using the power series of arctan, defined by:

$$\large 4 \sum_{n=0}^{N}\frac{(-1)^n}{2n+1} \approx \pi $$

where $N$ tends to the infinity.

def pi_py(N):
    pi = 0
    for n in range(N):
        pi += (-1)**n/(2*n + 1)
    return 4*pi

print(pi_py(1000000))

3.1415916535897743

In the same Cython source file examples_cy.pyx, lets include the function and adapt it to be compiled.

cdef double pi_cy(int N):
    cdef double pi = 0
    cdef int n
    for n in range(N):
        pi += (-1)**n/(2*n + 1)
    return 4*pi

p.s.: compile it again running the command:

$ python setup.py build_ext --inplace

from examples_cy import pi_cy

# Time measurement over the situations
print('[Python] pi_py |', end=' ')
%timeit -n 5 -r 5 pi_py(1000000)
print('[Cython] pi_cy |', end=' ')
%timeit -n 5 -r 5 pi_cy(1000000)

[Python] pi_py | 429 ms ± 2.16 ms per loop (mean ± std. dev. of 5 runs, 5 loops each)
[Cython] pi_cy | 21.1 ms ± 229 µs per loop (mean ± std. dev. of 5 runs, 5 loops each)

Cython and Jupyter Notebook

---

To enable support for Cython compilation in Jupyter Notebooks, we have to run firstly the command:

%load_ext Cython

It will allow the C functions declaration inside cells, using the magic function %%cython for multiple lines.

p.s.: the function call must be within the same cell

%%cython

cdef int factorial(int x):
    if x <= 1:
        return 1
    return x*factorial(x - 1)

print(factorial(10))

3628800