cpython/Doc/library/timeit.rst

353 lines
12 KiB
ReStructuredText
Raw Normal View History

2007-08-15 11:28:22 -03:00
:mod:`timeit` --- Measure execution time of small code snippets
===============================================================
.. module:: timeit
:synopsis: Measure the execution time of small code snippets.
.. index::
single: Benchmarking
single: Performance
2011-01-26 21:20:32 -04:00
**Source code:** :source:`Lib/timeit.py`
--------------
2007-08-15 11:28:22 -03:00
This module provides a simple way to time small bits of Python code. It has both
2012-10-01 23:35:39 -03:00
a :ref:`command-line-interface` as well as a :ref:`callable <python-interface>`
one. It avoids a number of common traps for measuring execution times.
See also Tim Peters' introduction to the "Algorithms" chapter in the *Python
Cookbook*, published by O'Reilly.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Basic Examples
--------------
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
The following example shows how the :ref:`command-line-interface`
can be used to compare three different expressions:
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. code-block:: sh
2007-08-15 11:28:22 -03:00
$ python3 -m timeit '"-".join(str(n) for n in range(100))'
10000 loops, best of 3: 30.2 usec per loop
$ python3 -m timeit '"-".join([str(n) for n in range(100)])'
10000 loops, best of 3: 27.5 usec per loop
$ python3 -m timeit '"-".join(map(str, range(100)))'
10000 loops, best of 3: 23.2 usec per loop
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
This can be achieved from the :ref:`python-interface` with::
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
>>> import timeit
>>> timeit.timeit('"-".join(str(n) for n in range(100))', number=10000)
0.3018611848820001
2012-10-01 23:35:39 -03:00
>>> timeit.timeit('"-".join([str(n) for n in range(100)])', number=10000)
0.2727368790656328
2012-10-01 23:35:39 -03:00
>>> timeit.timeit('"-".join(map(str, range(100)))', number=10000)
0.23702679807320237
2012-10-01 23:35:39 -03:00
Note however that :mod:`timeit` will automatically determine the number of
repetitions only when the command-line interface is used. In the
:ref:`timeit-examples` section you can find more advanced examples.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. _python-interface:
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Python Interface
----------------
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
The module defines three convenience functions and a public class:
2007-08-15 11:28:22 -03:00
.. function:: timeit(stmt='pass', setup='pass', timer=<default timer>, number=1000000, globals=None)
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Create a :class:`Timer` instance with the given statement, *setup* code and
*timer* function and run its :meth:`.timeit` method with *number* executions.
The optional *globals* argument specifies a namespace in which to execute the
code.
.. versionchanged:: 3.5
The optional *globals* parameter was added.
2007-08-15 11:28:22 -03:00
.. note::
Because :meth:`.timeit` is executing *stmt*, placing a return statement
in *stmt* will prevent :meth:`.timeit` from returning execution time.
It will instead return the data specified by your return statement.
2007-08-15 11:28:22 -03:00
.. function:: repeat(stmt='pass', setup='pass', timer=<default timer>, repeat=3, number=1000000, globals=None)
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Create a :class:`Timer` instance with the given statement, *setup* code and
*timer* function and run its :meth:`.repeat` method with the given *repeat*
count and *number* executions. The optional *globals* argument specifies a
namespace in which to execute the code.
2007-08-15 11:28:22 -03:00
.. versionchanged:: 3.5
The optional *globals* parameter was added.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. function:: default_timer()
2007-08-15 11:28:22 -03:00
2012-10-02 00:01:16 -03:00
The default timer, which is always :func:`time.perf_counter`.
2007-08-15 11:28:22 -03:00
2012-10-02 00:01:16 -03:00
.. versionchanged:: 3.3
:func:`time.perf_counter` is now the default timer.
2007-08-15 11:28:22 -03:00
.. class:: Timer(stmt='pass', setup='pass', timer=<timer function>, globals=None)
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Class for timing execution speed of small code snippets.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
The constructor takes a statement to be timed, an additional statement used
for setup, and a timer function. Both statements default to ``'pass'``;
the timer function is platform-dependent (see the module doc string).
*stmt* and *setup* may also contain multiple statements separated by ``;``
or newlines, as long as they don't contain multi-line string literals. The
statement will by default be executed within timeit's namespace; this behavior
can be controlled by passing a namespace to *globals*.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
To measure the execution time of the first statement, use the :meth:`.timeit`
method. The :meth:`.repeat` method is a convenience to call :meth:`.timeit`
multiple times and return a list of results.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
The *stmt* and *setup* parameters can also take objects that are callable
without arguments. This will embed calls to them in a timer function that
will then be executed by :meth:`.timeit`. Note that the timing overhead is a
little larger in this case because of the extra function calls.
2007-08-15 11:28:22 -03:00
.. versionchanged:: 3.5
The optional *globals* parameter was added.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. method:: Timer.timeit(number=1000000)
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Time *number* executions of the main statement. This executes the setup
statement once, and then returns the time it takes to execute the main
statement a number of times, measured in seconds as a float.
The argument is the number of times through the loop, defaulting to one
million. The main statement, the setup statement and the timer function
to be used are passed to the constructor.
2012-10-01 23:35:39 -03:00
.. note::
2012-10-01 23:35:39 -03:00
By default, :meth:`.timeit` temporarily turns off :term:`garbage
collection` during the timing. The advantage of this approach is that
it makes independent timings more comparable. This disadvantage is
that GC may be an important component of the performance of the
function being measured. If so, GC can be re-enabled as the first
statement in the *setup* string. For example::
2012-10-01 23:35:39 -03:00
timeit.Timer('for i in range(10): oct(i)', 'gc.enable()').timeit()
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. method:: Timer.repeat(repeat=3, number=1000000)
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
Call :meth:`.timeit` a few times.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
This is a convenience function that calls the :meth:`.timeit` repeatedly,
returning a list of results. The first argument specifies how many times
to call :meth:`.timeit`. The second argument specifies the *number*
argument for :meth:`.timeit`.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. note::
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
It's tempting to calculate mean and standard deviation from the result
vector and report these. However, this is not very useful.
In a typical case, the lowest value gives a lower bound for how fast
your machine can run the given code snippet; higher values in the
result vector are typically not caused by variability in Python's
speed, but by other processes interfering with your timing accuracy.
So the :func:`min` of the result is probably the only number you
should be interested in. After that, you should look at the entire
vector and apply common sense rather than statistics.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. method:: Timer.print_exc(file=None)
Helper to print a traceback from the timed code.
Typical use::
t = Timer(...) # outside the try/except
try:
t.timeit(...) # or t.repeat(...)
2012-11-02 17:07:26 -03:00
except Exception:
2012-10-01 23:35:39 -03:00
t.print_exc()
The advantage over the standard traceback is that source lines in the
compiled template will be displayed. The optional *file* argument directs
where the traceback is sent; it defaults to :data:`sys.stderr`.
.. _command-line-interface:
Command-Line Interface
2007-08-15 11:28:22 -03:00
----------------------
When called as a program from the command line, the following form is used::
python -m timeit [-n N] [-r N] [-s S] [-t] [-c] [-h] [statement ...]
Where the following options are understood:
.. program:: timeit
.. cmdoption:: -n N, --number=N
2007-08-15 11:28:22 -03:00
how many times to execute 'statement'
.. cmdoption:: -r N, --repeat=N
2007-08-15 11:28:22 -03:00
how many times to repeat the timer (default 3)
.. cmdoption:: -s S, --setup=S
statement to be executed once initially (default ``pass``)
.. cmdoption:: -p, --process
measure process time, not wallclock time, using :func:`time.process_time`
instead of :func:`time.perf_counter`, which is the default
.. versionadded:: 3.3
.. cmdoption:: -t, --time
2007-08-15 11:28:22 -03:00
use :func:`time.time` (deprecated)
2007-08-15 11:28:22 -03:00
.. cmdoption:: -c, --clock
use :func:`time.clock` (deprecated)
2007-08-15 11:28:22 -03:00
.. cmdoption:: -v, --verbose
2007-08-15 11:28:22 -03:00
print raw timing results; repeat for more digits precision
.. cmdoption:: -h, --help
2007-08-15 11:28:22 -03:00
print a short usage message and exit
A multi-line statement may be given by specifying each line as a separate
statement argument; indented lines are possible by enclosing an argument in
quotes and using leading spaces. Multiple :option:`-s` options are treated
similarly.
If :option:`-n` is not given, a suitable number of loops is calculated by trying
successive powers of 10 until the total time is at least 0.2 seconds.
:func:`default_timer` measurements can be affected by other programs running on
the same machine, so the best thing to do when accurate timing is necessary is
to repeat the timing a few times and use the best time. The :option:`-r`
option is good for this; the default of 3 repetitions is probably enough in
most cases. You can use :func:`time.process_time` to measure CPU time.
2007-08-15 11:28:22 -03:00
.. note::
There is a certain baseline overhead associated with executing a pass statement.
The code here doesn't try to hide it, but you should be aware of it. The
2012-10-01 23:35:39 -03:00
baseline overhead can be measured by invoking the program without arguments,
and it might differ between Python versions.
2007-08-15 11:28:22 -03:00
2012-10-01 23:35:39 -03:00
.. _timeit-examples:
2007-08-15 11:28:22 -03:00
Examples
--------
2012-10-01 23:35:39 -03:00
It is possible to provide a setup statement that is executed only once at the beginning:
.. code-block:: sh
$ python -m timeit -s 'text = "sample string"; char = "g"' 'char in text'
10000000 loops, best of 3: 0.0877 usec per loop
$ python -m timeit -s 'text = "sample string"; char = "g"' 'text.find(char)'
1000000 loops, best of 3: 0.342 usec per loop
::
>>> import timeit
>>> timeit.timeit('char in text', setup='text = "sample string"; char = "g"')
0.41440500499993504
>>> timeit.timeit('text.find(char)', setup='text = "sample string"; char = "g"')
1.7246671520006203
The same can be done using the :class:`Timer` class and its methods::
>>> import timeit
>>> t = timeit.Timer('char in text', setup='text = "sample string"; char = "g"')
>>> t.timeit()
0.3955516149999312
>>> t.repeat()
[0.40193588800002544, 0.3960157959998014, 0.39594301399984033]
The following examples show how to time expressions that contain multiple lines.
Here we compare the cost of using :func:`hasattr` vs. :keyword:`try`/:keyword:`except`
to test for missing and present object attributes:
.. code-block:: sh
2007-08-15 11:28:22 -03:00
$ python -m timeit 'try:' ' str.__bool__' 'except AttributeError:' ' pass'
2007-08-15 11:28:22 -03:00
100000 loops, best of 3: 15.7 usec per loop
$ python -m timeit 'if hasattr(str, "__bool__"): pass'
2007-08-15 11:28:22 -03:00
100000 loops, best of 3: 4.26 usec per loop
2012-10-01 23:35:39 -03:00
$ python -m timeit 'try:' ' int.__bool__' 'except AttributeError:' ' pass'
2007-08-15 11:28:22 -03:00
1000000 loops, best of 3: 1.43 usec per loop
$ python -m timeit 'if hasattr(int, "__bool__"): pass'
2007-08-15 11:28:22 -03:00
100000 loops, best of 3: 2.23 usec per loop
::
>>> import timeit
2012-10-01 23:35:39 -03:00
>>> # attribute is missing
2007-08-15 11:28:22 -03:00
>>> s = """\
... try:
... str.__bool__
... except AttributeError:
... pass
... """
2012-10-01 23:35:39 -03:00
>>> timeit.timeit(stmt=s, number=100000)
0.9138244460009446
>>> s = "if hasattr(str, '__bool__'): pass"
>>> timeit.timeit(stmt=s, number=100000)
0.5829014980008651
>>>
>>> # attribute is present
2007-08-15 11:28:22 -03:00
>>> s = """\
... try:
... int.__bool__
... except AttributeError:
... pass
... """
2012-10-01 23:35:39 -03:00
>>> timeit.timeit(stmt=s, number=100000)
0.04215312199994514
>>> s = "if hasattr(int, '__bool__'): pass"
>>> timeit.timeit(stmt=s, number=100000)
0.08588060699912603
2007-08-15 11:28:22 -03:00
To give the :mod:`timeit` module access to functions you define, you can pass a
2012-09-20 00:13:38 -03:00
*setup* parameter which contains an import statement::
2007-08-15 11:28:22 -03:00
def test():
"""Stupid test function"""
L = [i for i in range(100)]
2007-08-15 11:28:22 -03:00
if __name__ == '__main__':
2012-10-01 23:35:39 -03:00
import timeit
print(timeit.timeit("test()", setup="from __main__ import test"))
Another option is to pass :func:`globals` to the *globals* parameter, which will cause the code
to be executed within your current global namespace. This can be more convenient
than individually specifying imports::
def f(x):
return x**2
def g(x):
return x**4
def h(x):
return x**8
import timeit
print(timeit.timeit('[func(42) for func in (f,g,h)]', globals=globals()))