1:mod:`timeit` --- Measure execution time of small code snippets
2===============================================================
3
4.. module:: timeit
5   :synopsis: Measure the execution time of small code snippets.
6
7**Source code:** :source:`Lib/timeit.py`
8
9.. index::
10   single: Benchmarking
11   single: Performance
12
13--------------
14
15This module provides a simple way to time small bits of Python code. It has both
16a :ref:`timeit-command-line-interface` as well as a :ref:`callable <python-interface>`
17one.  It avoids a number of common traps for measuring execution times.
18See also Tim Peters' introduction to the "Algorithms" chapter in the *Python
19Cookbook*, published by O'Reilly.
20
21
22Basic Examples
23--------------
24
25The following example shows how the :ref:`timeit-command-line-interface`
26can be used to compare three different expressions:
27
28.. code-block:: shell-session
29
30   $ python3 -m timeit '"-".join(str(n) for n in range(100))'
31   10000 loops, best of 5: 30.2 usec per loop
32   $ python3 -m timeit '"-".join([str(n) for n in range(100)])'
33   10000 loops, best of 5: 27.5 usec per loop
34   $ python3 -m timeit '"-".join(map(str, range(100)))'
35   10000 loops, best of 5: 23.2 usec per loop
36
37This can be achieved from the :ref:`python-interface` with::
38
39   >>> import timeit
40   >>> timeit.timeit('"-".join(str(n) for n in range(100))', number=10000)
41   0.3018611848820001
42   >>> timeit.timeit('"-".join([str(n) for n in range(100)])', number=10000)
43   0.2727368790656328
44   >>> timeit.timeit('"-".join(map(str, range(100)))', number=10000)
45   0.23702679807320237
46
47
48Note however that :mod:`timeit` will automatically determine the number of
49repetitions only when the command-line interface is used.  In the
50:ref:`timeit-examples` section you can find more advanced examples.
51
52
53.. _python-interface:
54
55Python Interface
56----------------
57
58The module defines three convenience functions and a public class:
59
60
61.. function:: timeit(stmt='pass', setup='pass', timer=<default timer>, number=1000000, globals=None)
62
63   Create a :class:`Timer` instance with the given statement, *setup* code and
64   *timer* function and run its :meth:`.timeit` method with *number* executions.
65   The optional *globals* argument specifies a namespace in which to execute the
66   code.
67
68   .. versionchanged:: 3.5
69      The optional *globals* parameter was added.
70
71
72.. function:: repeat(stmt='pass', setup='pass', timer=<default timer>, repeat=5, number=1000000, globals=None)
73
74   Create a :class:`Timer` instance with the given statement, *setup* code and
75   *timer* function and run its :meth:`.repeat` method with the given *repeat*
76   count and *number* executions.  The optional *globals* argument specifies a
77   namespace in which to execute the code.
78
79   .. versionchanged:: 3.5
80      The optional *globals* parameter was added.
81
82   .. versionchanged:: 3.7
83      Default value of *repeat* changed from 3 to 5.
84
85.. function:: default_timer()
86
87   The default timer, which is always :func:`time.perf_counter`.
88
89   .. versionchanged:: 3.3
90      :func:`time.perf_counter` is now the default timer.
91
92
93.. class:: Timer(stmt='pass', setup='pass', timer=<timer function>, globals=None)
94
95   Class for timing execution speed of small code snippets.
96
97   The constructor takes a statement to be timed, an additional statement used
98   for setup, and a timer function.  Both statements default to ``'pass'``;
99   the timer function is platform-dependent (see the module doc string).
100   *stmt* and *setup* may also contain multiple statements separated by ``;``
101   or newlines, as long as they don't contain multi-line string literals.  The
102   statement will by default be executed within timeit's namespace; this behavior
103   can be controlled by passing a namespace to *globals*.
104
105   To measure the execution time of the first statement, use the :meth:`.timeit`
106   method.  The :meth:`.repeat` and :meth:`.autorange` methods are convenience
107   methods to call :meth:`.timeit` multiple times.
108
109   The execution time of *setup* is excluded from the overall timed execution run.
110
111   The *stmt* and *setup* parameters can also take objects that are callable
112   without arguments.  This will embed calls to them in a timer function that
113   will then be executed by :meth:`.timeit`.  Note that the timing overhead is a
114   little larger in this case because of the extra function calls.
115
116   .. versionchanged:: 3.5
117      The optional *globals* parameter was added.
118
119   .. method:: Timer.timeit(number=1000000)
120
121      Time *number* executions of the main statement.  This executes the setup
122      statement once, and then returns the time it takes to execute the main
123      statement a number of times, measured in seconds as a float.
124      The argument is the number of times through the loop, defaulting to one
125      million.  The main statement, the setup statement and the timer function
126      to be used are passed to the constructor.
127
128      .. note::
129
130         By default, :meth:`.timeit` temporarily turns off :term:`garbage
131         collection` during the timing.  The advantage of this approach is that
132         it makes independent timings more comparable.  The disadvantage is
133         that GC may be an important component of the performance of the
134         function being measured.  If so, GC can be re-enabled as the first
135         statement in the *setup* string.  For example::
136
137            timeit.Timer('for i in range(10): oct(i)', 'gc.enable()').timeit()
138
139
140   .. method:: Timer.autorange(callback=None)
141
142      Automatically determine how many times to call :meth:`.timeit`.
143
144      This is a convenience function that calls :meth:`.timeit` repeatedly
145      so that the total time >= 0.2 second, returning the eventual
146      (number of loops, time taken for that number of loops). It calls
147      :meth:`.timeit` with increasing numbers from the sequence 1, 2, 5,
148      10, 20, 50, ... until the time taken is at least 0.2 second.
149
150      If *callback* is given and is not ``None``, it will be called after
151      each trial with two arguments: ``callback(number, time_taken)``.
152
153      .. versionadded:: 3.6
154
155
156   .. method:: Timer.repeat(repeat=5, number=1000000)
157
158      Call :meth:`.timeit` a few times.
159
160      This is a convenience function that calls the :meth:`.timeit` repeatedly,
161      returning a list of results.  The first argument specifies how many times
162      to call :meth:`.timeit`.  The second argument specifies the *number*
163      argument for :meth:`.timeit`.
164
165      .. note::
166
167         It's tempting to calculate mean and standard deviation from the result
168         vector and report these.  However, this is not very useful.
169         In a typical case, the lowest value gives a lower bound for how fast
170         your machine can run the given code snippet; higher values in the
171         result vector are typically not caused by variability in Python's
172         speed, but by other processes interfering with your timing accuracy.
173         So the :func:`min` of the result is probably the only number you
174         should be interested in.  After that, you should look at the entire
175         vector and apply common sense rather than statistics.
176
177      .. versionchanged:: 3.7
178         Default value of *repeat* changed from 3 to 5.
179
180
181   .. method:: Timer.print_exc(file=None)
182
183      Helper to print a traceback from the timed code.
184
185      Typical use::
186
187         t = Timer(...)       # outside the try/except
188         try:
189             t.timeit(...)    # or t.repeat(...)
190         except Exception:
191             t.print_exc()
192
193      The advantage over the standard traceback is that source lines in the
194      compiled template will be displayed.  The optional *file* argument directs
195      where the traceback is sent; it defaults to :data:`sys.stderr`.
196
197
198.. _timeit-command-line-interface:
199
200Command-Line Interface
201----------------------
202
203When called as a program from the command line, the following form is used::
204
205   python -m timeit [-n N] [-r N] [-u U] [-s S] [-h] [statement ...]
206
207Where the following options are understood:
208
209.. program:: timeit
210
211.. cmdoption:: -n N, --number=N
212
213   how many times to execute 'statement'
214
215.. cmdoption:: -r N, --repeat=N
216
217   how many times to repeat the timer (default 5)
218
219.. cmdoption:: -s S, --setup=S
220
221   statement to be executed once initially (default ``pass``)
222
223.. cmdoption:: -p, --process
224
225   measure process time, not wallclock time, using :func:`time.process_time`
226   instead of :func:`time.perf_counter`, which is the default
227
228   .. versionadded:: 3.3
229
230.. cmdoption:: -u, --unit=U
231
232    specify a time unit for timer output; can select nsec, usec, msec, or sec
233
234   .. versionadded:: 3.5
235
236.. cmdoption:: -v, --verbose
237
238   print raw timing results; repeat for more digits precision
239
240.. cmdoption:: -h, --help
241
242   print a short usage message and exit
243
244A multi-line statement may be given by specifying each line as a separate
245statement argument; indented lines are possible by enclosing an argument in
246quotes and using leading spaces.  Multiple :option:`-s` options are treated
247similarly.
248
249If :option:`-n` is not given, a suitable number of loops is calculated by trying
250successive powers of 10 until the total time is at least 0.2 seconds.
251
252:func:`default_timer` measurements can be affected by other programs running on
253the same machine, so the best thing to do when accurate timing is necessary is
254to repeat the timing a few times and use the best time.  The :option:`-r`
255option is good for this; the default of 5 repetitions is probably enough in
256most cases.  You can use :func:`time.process_time` to measure CPU time.
257
258.. note::
259
260   There is a certain baseline overhead associated with executing a pass statement.
261   The code here doesn't try to hide it, but you should be aware of it.  The
262   baseline overhead can be measured by invoking the program without arguments,
263   and it might differ between Python versions.
264
265
266.. _timeit-examples:
267
268Examples
269--------
270
271It is possible to provide a setup statement that is executed only once at the beginning:
272
273.. code-block:: shell-session
274
275   $ python -m timeit -s 'text = "sample string"; char = "g"'  'char in text'
276   5000000 loops, best of 5: 0.0877 usec per loop
277   $ python -m timeit -s 'text = "sample string"; char = "g"'  'text.find(char)'
278   1000000 loops, best of 5: 0.342 usec per loop
279
280::
281
282   >>> import timeit
283   >>> timeit.timeit('char in text', setup='text = "sample string"; char = "g"')
284   0.41440500499993504
285   >>> timeit.timeit('text.find(char)', setup='text = "sample string"; char = "g"')
286   1.7246671520006203
287
288The same can be done using the :class:`Timer` class and its methods::
289
290   >>> import timeit
291   >>> t = timeit.Timer('char in text', setup='text = "sample string"; char = "g"')
292   >>> t.timeit()
293   0.3955516149999312
294   >>> t.repeat()
295   [0.40183617287970225, 0.37027556854118704, 0.38344867356679524, 0.3712595970846668, 0.37866875250654886]
296
297
298The following examples show how to time expressions that contain multiple lines.
299Here we compare the cost of using :func:`hasattr` vs. :keyword:`try`/:keyword:`except`
300to test for missing and present object attributes:
301
302.. code-block:: shell-session
303
304   $ python -m timeit 'try:' '  str.__bool__' 'except AttributeError:' '  pass'
305   20000 loops, best of 5: 15.7 usec per loop
306   $ python -m timeit 'if hasattr(str, "__bool__"): pass'
307   50000 loops, best of 5: 4.26 usec per loop
308
309   $ python -m timeit 'try:' '  int.__bool__' 'except AttributeError:' '  pass'
310   200000 loops, best of 5: 1.43 usec per loop
311   $ python -m timeit 'if hasattr(int, "__bool__"): pass'
312   100000 loops, best of 5: 2.23 usec per loop
313
314::
315
316   >>> import timeit
317   >>> # attribute is missing
318   >>> s = """\
319   ... try:
320   ...     str.__bool__
321   ... except AttributeError:
322   ...     pass
323   ... """
324   >>> timeit.timeit(stmt=s, number=100000)
325   0.9138244460009446
326   >>> s = "if hasattr(str, '__bool__'): pass"
327   >>> timeit.timeit(stmt=s, number=100000)
328   0.5829014980008651
329   >>>
330   >>> # attribute is present
331   >>> s = """\
332   ... try:
333   ...     int.__bool__
334   ... except AttributeError:
335   ...     pass
336   ... """
337   >>> timeit.timeit(stmt=s, number=100000)
338   0.04215312199994514
339   >>> s = "if hasattr(int, '__bool__'): pass"
340   >>> timeit.timeit(stmt=s, number=100000)
341   0.08588060699912603
342
343
344To give the :mod:`timeit` module access to functions you define, you can pass a
345*setup* parameter which contains an import statement::
346
347   def test():
348       """Stupid test function"""
349       L = [i for i in range(100)]
350
351   if __name__ == '__main__':
352       import timeit
353       print(timeit.timeit("test()", setup="from __main__ import test"))
354
355Another option is to pass :func:`globals` to the  *globals* parameter, which will cause the code
356to be executed within your current global namespace.  This can be more convenient
357than individually specifying imports::
358
359   def f(x):
360       return x**2
361   def g(x):
362       return x**4
363   def h(x):
364       return x**8
365
366   import timeit
367   print(timeit.timeit('[func(42) for func in (f,g,h)]', globals=globals()))
368