Python Profiling¶

With hotshot¶

It is possible to use hotshot python module for profiling and see the results in a graphical way using the linux tool KCachegring.

- First, create the profiling file with hotshot :

import hotshot
p=hotshot.Profile("fichier")
p.start()
# python code to profile
...
p.stop()

- Then, you have to convert the result file in a format readable by KCachegring. You can use the tool hotshot2calltree to do that:

hotshot2calltree fichier_hotshot > fichier_kcachegrind

- Finally, open the file with KCachegrind:

kcachegrind fichier_kcachegrind

- To get the 20 first functions ordered by time and nb calls in python:

import hotshot.stats
stats=hotshot.stats.load("fichier")
stats.strip_dirs()
stats.sort_stats('time', 'calls')
stats.print_stats(20)

With cProfile¶

It seems to be better to use cProfile as hotshot is now deprecated.

import cProfile, pstats

cProfile.runctx("function_to_profile()", globals(), locals(), "fichier.profile")

p=pstats.Stats("fichier.profile")
p.strip_dirs().sort_stats('time', 'calls').print_stats(20)

It is possible to see a graphical representation of the results using the tool RunSnakeRun : http://www.vrplumber.com/programming/runsnakerun/