Testing performance of Python code

27 Feb 2021 - fubar - Sreekar Guddeti

Python standard library consists timeit module that gives execution time for Python code.

Context

Code that computes what it is intended to compute is a first requirement of writing code. After that is accomplished, we are interested in optimizing the code in terms of readability, faster execution time.

Python focuses a lot more on readability than other languages. However, it also provides datastructures to improve performance. This is roughly measured in the execution time. To measure this time, Python standard library has the timeit module.

Let us compare the execution time to create an array of natural numbers. A simple but slower implementation is to use list datastructure.

import timeit

N = 100 # number of naturals

def list_array():
    L = [each for each in range(N)]
    return L

A better alternative is to create numpy arrays

import numpy as np

def numpy_array():
    A = np.arange(N)
    return A

To compare the execution time, pass these functions as argument to timeit() method. This method executes the code for default N=100000 i.e. million times. Invoke this method to time the two ways.

if __name__ == '__main__':

    # timeit.timeit() method repeats the code for default
    # N=100000 i.e. million times

    # to time a function
    time_list = timeit.timeit("list_array()", setup="from __main__ import list_array")
    print(f'time for list comprehension is {time_list}')

    time_array = timeit.timeit('numpy_array()', globals=globals())
    print(f'time for numpy array is {time_array}')    

The console output clearly shows better performance for the latter method.

time for list comprehension is 5.16344269999999
time for numpy array is 1.093457199999989