How can you profile a Python script?

2009-02-25 python performance profiling time-complexity

Project Euler and other coding contests often have a maximum time to run or people boast of how fast their particular solution runs. With Python, sometimes the approaches are somewhat kludgey - i.e., adding timing code to __main__.

What is a good way to profile how long a Python program takes to run?


Python includes a profiler called cProfile. It not only gives the total running time, but also times each function separately, and tells you how many times each function was called, making it easy to determine where you should make optimizations.

You can call it from within your code, or from the interpreter, like this:

import cProfile'foo()')

Even more usefully, you can invoke the cProfile when running a script:

python -m cProfile

To make it even easier, I made a little batch file called 'profile.bat':

python -m cProfile %1

So all I have to do is run:


And I get this:

1007 function calls in 0.061 CPU seconds

Ordered by: standard name
ncalls  tottime  percall  cumtime  percall filename:lineno(function)
    1    0.000    0.000    0.061    0.061 <string>:1(<module>)
 1000    0.051    0.000    0.051    0.000<lambda>)
    1    0.005    0.005    0.061    0.061<module>)
    1    0.000    0.000    0.061    0.061 {execfile}
    1    0.002    0.002    0.053    0.053 {map}
    1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler objects}
    1    0.000    0.000    0.000    0.000 {range}
    1    0.003    0.003    0.003    0.003 {sum}

EDIT: Updated link to a good video resource from PyCon 2013 titled Python Profiling
Also via YouTube.

In Virtaal's source there's a very useful class and decorator that can make profiling (even for specific methods/functions) very easy. The output can then be viewed very comfortably in KCacheGrind.

It's worth pointing out that using the profiler only works (by default) on the main thread, and you won't get any information from other threads if you use them. This can be a bit of a gotcha as it is completely unmentioned in the profiler documentation.

If you also want to profile threads, you'll want to look at the threading.setprofile() function in the docs.

You could also create your own threading.Thread subclass to do it:

class ProfiledThread(threading.Thread):
    # Overrides
    def run(self):
        profiler = cProfile.Profile()
            return profiler.runcall(, self)
            profiler.dump_stats('myprofile-%d.profile' % (self.ident,))

and use that ProfiledThread class instead of the standard one. It might give you more flexibility, but I'm not sure it's worth it, especially if you are using third-party code which wouldn't use your class.

The python wiki is a great page for profiling resources:

as is the python docs:

as shown by Chris Lawlor cProfile is a great tool and can easily be used to print to the screen:

python -m cProfile -s time <args>

or to file:

python -m cProfile -o output.file <args>

PS> If you are using Ubuntu, make sure to install python-profile

apt-get install python-profiler 

If you output to file you can get nice visualizations using the following tools

PyCallGraph : a tool to create call graph images

 pip install pycallgraph


 pycallgraph args


 gimp pycallgraph.png

You can use whatever you like to view the png file, I used gimp
Unfortunately I often get

dot: graph is too large for cairo-renderer bitmaps. Scaling by 0.257079 to fit

which makes my images unusably small. So I generally create svg files:

pycallgraph -f svg -o pycallgraph.svg <args>

PS> make sure to install graphviz (which provides the dot program):

pip install graphviz

Alternative Graphing using gprof2dot via @maxy / @quodlibetor :

pip install gprof2dot
python -m cProfile -o profile.pstats
gprof2dot -f pstats profile.pstats | dot -Tsvg -o mine.svg

A nice profiling module is the line_profiler (called using the script It can be downloaded here.

My understanding is that cProfile only gives information about total time spent in each function. So individual lines of code are not timed. This is an issue in scientific computing since often one single line can take a lot of time. Also, as I remember, cProfile didn't catch the time I was spending in say

Following Joe Shaw's answer about multi-threaded code not to work as expected, I figured that the runcall method in cProfile is merely doing self.enable() and self.disable() calls around the profiled function call, so you can simply do that yourself and have whatever code you want in-between with minimal interference with existing code.

A while ago I made pycallgraph which generates a visualisation from your Python code. Edit: I've updated the example to work with 3.3, the latest release as of this writing.

After a pip install pycallgraph and installing GraphViz you can run it from the command line:

pycallgraph graphviz -- ./

Or, you can profile particular parts of your code:

from pycallgraph import PyCallGraph
from pycallgraph.output import GraphvizOutput

with PyCallGraph(output=GraphvizOutput()):

Either of these will generate a pycallgraph.png file similar to the image below:

enter image description here

Ever want to know what the hell that python script is doing? Enter the Inspect Shell. Inspect Shell lets you print/alter globals and run functions without interrupting the running script. Now with auto-complete and command history (only on linux).

Inspect Shell is not a pdb-style debugger.

You could use that (and your wristwatch).

@Maxy's comment on this answer helped me out enough that I think it deserves its own answer: I already had cProfile-generated .pstats files and I didn't want to re-run things with pycallgraph, so I used gprof2dot, and got pretty svgs:

$ sudo apt-get install graphviz
$ git clone
$ ln -s "$PWD"/gprof2dot/ ~/bin
$ -f pstats profile.pstats | dot -Tsvg -o callgraph.svg

and BLAM!

It uses dot (the same thing that pycallgraph uses) so output looks similar. I get the impression that gprof2dot loses less information though:

gprof2dot example output

My way is to use yappi ( It's especially useful combined with an RPC server where (even just for debugging) you register method to start, stop and print profiling information, e.g. in this way:

def startProfiler():

def stopProfiler():

def printProfiler():
    stats = yappi.get_stats(yappi.SORTTYPE_TTOT, yappi.SORTORDER_DESC, 20)
    statPrint = '\n'
    namesArr = [len(str(stat[0])) for stat in stats.func_stats]
    log.debug("namesArr %s", str(namesArr))
    maxNameLen = max(namesArr)
    log.debug("maxNameLen: %s", maxNameLen)

    for stat in stats.func_stats:
        nameAppendSpaces = [' ' for i in range(maxNameLen - len(stat[0]))]
        log.debug('nameAppendSpaces: %s', nameAppendSpaces)
        blankSpace = ''
        for space in nameAppendSpaces:
            blankSpace += space

        log.debug("adding spaces: %s", len(nameAppendSpaces))
        statPrint = statPrint + str(stat[0]) + blankSpace + " " + str(stat[1]).ljust(8) + "\t" + str(
            round(stat[2], 2)).ljust(8 - len(str(stat[2]))) + "\t" + str(round(stat[3], 2)) + "\n"

    log.log(1000, "\nname" + ''.ljust(maxNameLen - 4) + " ncall \tttot \ttsub")
    log.log(1000, statPrint)

Then when your program work you can start profiler at any time by calling the startProfiler RPC method and dump profiling information to a log file by calling printProfiler (or modify the rpc method to return it to the caller) and get such output:

2014-02-19 16:32:24,128-|SVR-MAIN  |-(Thread-3   )-Level 1000: 
name                                                                                                                                      ncall     ttot    tsub
2014-02-19 16:32:24,128-|SVR-MAIN  |-(Thread-3   )-Level 1000: 
C:\Python27\lib\                                                                                                           22        0.11    0.05
M:\02_documents\_repos\09_aheadRepos\apps\ahdModbusSrv\pyAheadRpcSrv\                                                22        0.11    0.0
M:\02_documents\_repos\09_aheadRepos\apps\ahdModbusSrv\                                                    22        0.11    0.0
M:\02_documents\_repos\09_aheadRepos\apps\ahdModbusSrv\pyAheadRpcSrv\                                       1         0.0     0.0
C:\Python27\lib\                                                                                    1         0.0     0.0
c:\users\zasiec~1\appdata\local\temp\easy_install-hwcsr1\psutil-1.1.2-py2.7-win32.egg.tmp\psutil\     4         0.0     0.0
C:\Python27\lib\                                                                          1         0.0     0.0
c:\users\zasiec~1\appdata\local\temp\easy_install-hwcsr1\psutil-1.1.2-py2.7-win32.egg.tmp\psutil\ 4         0.0     0.0
<string>.__new__:8                                                                                                                        220       0.0     0.0
C:\Python27\lib\                                                                                                       4         0.0     0.0
C:\Python27\lib\                                                                                                 1         0.0     0.0
<string>.__new__:8                                                                                                                        4         0.0     0.0
C:\Python27\lib\                                                                                                   1         0.0     0.0
C:\Python27\lib\                                                                                                   4         0.0     0.0
C:\Python27\lib\                                                                                  1         0.0     0.0
C:\Python27\lib\                                                                                                      3         0.0     0.0
C:\Python27\lib\                                                                                         1         0.0     0.0
C:\Python27\lib\                                                                               1         0.0     0.0
C:\Python27\lib\                                                                                         1         0.0     0.0
c:\users\zasiec~1\appdata\local\temp\easy_install-hwcsr1\psutil-1.1.2-py2.7-win32.egg.tmp\psutil\               4         0.0     0.0 

It may not be very useful for short scripts but helps to optimize server-type processes especially given the printProfiler method can be called multiple times over time to profile and compare e.g. different program usage scenarios.

In newer versions of yappi, the following code will work:

def printProfile():

Also worth mentioning is the GUI cProfile dump viewer RunSnakeRun. It allows you to sort and select, thereby zooming in on the relevant parts of the program. The sizes of the rectangles in the picture is proportional to the time taken. If you mouse over a rectangle it highlights that call in the table and everywhere on the map. When you double-click on a rectangle it zooms in on that portion. It will show you who calls that portion and what that portion calls.

The descriptive information is very helpful. It shows you the code for that bit which can be helpful when you are dealing with built-in library calls. It tells you what file and what line to find the code.

Also want to point at that the OP said 'profiling' but it appears he meant 'timing'. Keep in mind programs will run slower when profiled.

enter image description here


line_profiler (already presented here) also inspired pprofile, which is described as:

Line-granularity, thread-aware deterministic and statistic pure-python profiler

It provides line-granularity as line_profiler, is pure Python, can be used as a standalone command or a module, and can even generate callgrind-format files that can be easily analyzed with [k|q]cachegrind.


There is also vprof, a Python package described as:

[...] providing rich and interactive visualizations for various Python program characteristics such as running time and memory usage.


To add on to,

I wrote this module that allows you to use cProfile and view its output easily. More here:

$ python -m cprofilev /your/python/program
# Go to http://localhost:4000 to view collected statistics.

Also see: on how to make sense of the collected statistics.

cProfile is great for quick profiling but most of the time it was ending for me with the errors. Function runctx solves this problem by initializing correctly the environment and variables, hope it can be useful for someone:

import cProfile
cProfile.runctx('foo()', None, locals())

A new tool to handle profiling in Python is PyVmMonitor:

It has some unique features such as

  • Attach profiler to a running (CPython) program
  • On demand profiling with Yappi integration
  • Profile on a different machine
  • Multiple processes support (multiprocessing, django...)
  • Live sampling/CPU view (with time range selection)
  • Deterministic profiling through cProfile/profile integration
  • Analyze existing PStats results
  • Open DOT files
  • Programatic API access
  • Group samples by method or line
  • PyDev integration
  • PyCharm integration

Note: it's commercial, but free for open source.

There's a lot of great answers but they either use command line or some external program for profiling and/or sorting the results.

I really missed some way I could use in my IDE (eclipse-PyDev) without touching the command line or installing anything. So here it is.

Profiling without command line

def count():
    from math import sqrt
    for x in range(10**5):

if __name__ == '__main__':
    import cProfile, pstats"count()", "{}.profile".format(__file__))
    s = pstats.Stats("{}.profile".format(__file__))

See docs or other answers for more info.

There's also a statistical profiler called statprof. It's a sampling profiler, so it adds minimal overhead to your code and gives line-based (not just function-based) timings. It's more suited to soft real-time applications like games, but may be have less precision than cProfile.

The version in pypi is a bit old, so can install it with pip by specifying the git repository:

pip install git+git://[email protected]

You can run it like this:

import statprof

with statprof.profile():

See also

I think that cProfile is great for profiling, while kcachegrind is great for visualizing the results. The pyprof2calltree in between handles the file conversion.

python -m cProfile -o script.profile
pyprof2calltree -i script.profile -o script.calltree
kcachegrind script.calltree

To install the required tools (on Ubuntu, at least):

apt-get install kcachegrind
pip install pyprof2calltree

The result:

Screenshot of the result

I ran into a handy tool called SnakeViz when researching this topic. SnakeViz is a web-based profiling visualization tool. It is very easy to install and use. The usual way I use it is to generate a stat file with %prun and then do analysis in SnakeViz.

The main viz technique used is Sunburst chart as shown below, in which the hierarchy of function calls is arranged as layers of arcs and time info encoded in their angular widths.

The best thing is you can interact with the chart. For example, to zoom in one can click on an arc, and the arc and its descendants will be enlarged as a new sunburst to display more details.

enter image description here

When i'm not root on the server, I use and run my program like this:

python -o callgrind.1

Then I can open the report with any callgrind-compatible software, like qcachegrind

It would depend on what you want to see out of profiling. Simple time metrics can be given by (bash).

time python

Even '/usr/bin/time' can output detailed metrics by using '--verbose' flag.

To check time metrics given by each function and to better understand how much time is spent on functions, you can use the inbuilt cProfile in python.

Going into more detailed metrics like performance, time is not the only metric. You can worry about memory, threads etc.
Profiling options:
1. line_profiler is another profiler used commonly to find out timing metrics line-by-line.
2. memory_profiler is a tool to profile memory usage.
3. heapy (from project Guppy) Profile how objects in the heap are used.

These are some of the common ones I tend to use. But if you want to find out more, try reading this book It is a pretty good book on starting out with performance in mind. You can move onto advanced topics on using Cython and JIT(Just-in-time) compiled python.

Simplest and quickest way to find where all the time is going.

1. pip install snakeviz

2. python -m cProfile -o temp.dat <PROGRAM>.py

3. snakeviz temp.dat

Draws a pie chart in a browser. Biggest piece is the problem function. Very simple.

I recently created tuna for visualizing Python runtime and import profiles; this may be helpful here.

enter image description here

Install with

pip3 install tuna

Create a runtime profile

python -m cProfile -o

or an import profile (Python 3.7+ required)

python -X importprofile 2> import.log

Then just run tuna on the file



Magic function for gprof2dot to profile any Python statement as a DOT graph in JupyterLab or Jupyter Notebook.

enter image description here

GitHub repo:


Make sure you've the Python package gprof2dot_magic.

pip install gprof2dot_magic

Its dependencies gprof2dot and graphviz will be installed as well


To enable the magic function, first load the gprof2dot_magic module

%load_ext gprof2dot_magic

and then profile any line statement as a DOT graph as such:

%gprof2dot print('hello world')

enter image description here

The terminal-only (and simplest) solution, in case all those fancy UI's fail to install or to run:
ignore cProfile completely and replace it with pyinstrument, that will collect and display the tree of calls right after execution.


$ pip install pyinstrument

Profile and display result:

$ python -m pyinstrument ./

Works with python2 and 3.

If you want to make a cumulative profiler, meaning to run the function several times in a row and watch the sum of the results.

you can use this cumulative_profiler decorator:

it's python >= 3.6 specific, but you can remove nonlocal for it work on older versions.

import cProfile, pstats

class _ProfileFunc:
    def __init__(self, func, sort_stats_by):
        self.func =  func
        self.profile_runs = []
        self.sort_stats_by = sort_stats_by

    def __call__(self, *args, **kwargs):
        pr = cProfile.Profile()
        pr.enable()  # this is the profiling section
        retval = self.func(*args, **kwargs)

        ps = pstats.Stats(*self.profile_runs).sort_stats(self.sort_stats_by)
        return retval, ps

def cumulative_profiler(amount_of_times, sort_stats_by='time'):
    def real_decorator(function):
        def wrapper(*args, **kwargs):
            nonlocal function, amount_of_times, sort_stats_by  # for python 2.x remove this row

            profiled_func = _ProfileFunc(function, sort_stats_by)
            for i in range(amount_of_times):
                retval, ps = profiled_func(*args, **kwargs)
            return retval  # returns the results of the function
        return wrapper

    if callable(amount_of_times):  # incase you don't want to specify the amount of times
        func = amount_of_times  # amount_of_times is the function in here
        amount_of_times = 5  # the default amount
        return real_decorator(func)
    return real_decorator


profiling the function baz

import time

def baz():
    return 1


baz ran 5 times and printed this:

         20 function calls in 15.003 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
       10   15.003    1.500   15.003    1.500 {built-in method time.sleep}
        5    0.000    0.000   15.003    3.001 <ipython-input-9-c89afe010372>:3(baz)
        5    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

specifying the amount of times

def baz():

I just developed my own profiler inspired from pypref_time:

By adding a decorator it will show a tree of time-consuming functions

@Profiler(depth=4, on_disable=show)

Install by: pip install auto_profiler


import time # line number 1
import random

from auto_profiler import Profiler, Tree

def f1():

def mysleep(t):

def fact(i):
        return 1
    return i*fact(i-1)

def show(p):
    print('Time   [Hits * PerHit] Function name [Called from] [Function Location]\n'+\
    print(Tree(p.root, threshold=0.5))

@Profiler(depth=4, on_disable=show)
def main():
    for i in range(5):


if __name__ == '__main__':

Example Output

Time   [Hits * PerHit] Function name [Called from] [function location]
8.974s [1 * 8.974]  main  [auto-profiler/]  [/test/]
├── 5.954s [5 * 1.191]  f1  [/test/]  [/test/]
│   └── 5.954s [5 * 1.191]  mysleep  [/test/]  [/test/]
│       └── 5.954s [5 * 1.191]  <time.sleep>
|   # The rest is for the example recursive function call fact
└── 3.020s [1 * 3.020]  fact  [/test/]  [/test/]
    ├── 0.849s [1 * 0.849]  f1  [/test/]  [/test/]
    │   └── 0.849s [1 * 0.849]  mysleep  [/test/]  [/test/]
    │       └── 0.849s [1 * 0.849]  <time.sleep>
    └── 2.171s [1 * 2.171]  fact  [/test/]  [/test/]
        ├── 1.552s [1 * 1.552]  f1  [/test/]  [/test/]
        │   └── 1.552s [1 * 1.552]  mysleep  [/test/]  [/test/]
        └── 0.619s [1 * 0.619]  fact  [/test/]  [/test/]
            └── 0.619s [1 * 0.619]  f1  [/test/]  [/test/]

With a statistical profiler like austin, no instrumentation is required, meaning that you can get profiling data out of a Python application simply with

austin python3

The raw output isn't very useful, but you can pipe that to to get a flame graph representation of that data that gives you a breakdown of where the time (measured in microseconds of real time) is being spent.

austin python3 | > my_script_profile.svg