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 second
19edition of *Python Cookbook*, 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
47A callable can also be passed from the :ref:`python-interface`::
48
49   >>> timeit.timeit(lambda: "-".join(map(str, range(100))), number=10000)
50   0.19665591977536678
51
52Note however that :func:`.timeit` will automatically determine the number of
53repetitions only when the command-line interface is used.  In the
54:ref:`timeit-examples` section you can find more advanced examples.
55
56
57.. _python-interface:
58
59Python Interface
60----------------
61
62The module defines three convenience functions and a public class:
63
64
65.. function:: timeit(stmt='pass', setup='pass', timer=<default timer>, number=1000000, globals=None)
66
67   Create a :class:`Timer` instance with the given statement, *setup* code and
68   *timer* function and run its :meth:`.timeit` method with *number* executions.
69   The optional *globals* argument specifies a namespace in which to execute the
70   code.
71
72   .. versionchanged:: 3.5
73      The optional *globals* parameter was added.
74
75
76.. function:: repeat(stmt='pass', setup='pass', timer=<default timer>, repeat=5, number=1000000, globals=None)
77
78   Create a :class:`Timer` instance with the given statement, *setup* code and
79   *timer* function and run its :meth:`.repeat` method with the given *repeat*
80   count and *number* executions.  The optional *globals* argument specifies a
81   namespace in which to execute the code.
82
83   .. versionchanged:: 3.5
84      The optional *globals* parameter was added.
85
86   .. versionchanged:: 3.7
87      Default value of *repeat* changed from 3 to 5.
88
89.. function:: default_timer()
90
91   The default timer, which is always :func:`time.perf_counter`.
92
93   .. versionchanged:: 3.3
94      :func:`time.perf_counter` is now the default timer.
95
96
97.. class:: Timer(stmt='pass', setup='pass', timer=<timer function>, globals=None)
98
99   Class for timing execution speed of small code snippets.
100
101   The constructor takes a statement to be timed, an additional statement used
102   for setup, and a timer function.  Both statements default to ``'pass'``;
103   the timer function is platform-dependent (see the module doc string).
104   *stmt* and *setup* may also contain multiple statements separated by ``;``
105   or newlines, as long as they don't contain multi-line string literals.  The
106   statement will by default be executed within timeit's namespace; this behavior
107   can be controlled by passing a namespace to *globals*.
108
109   To measure the execution time of the first statement, use the :meth:`.timeit`
110   method.  The :meth:`.repeat` and :meth:`.autorange` methods are convenience
111   methods to call :meth:`.timeit` multiple times.
112
113   The execution time of *setup* is excluded from the overall timed execution run.
114
115   The *stmt* and *setup* parameters can also take objects that are callable
116   without arguments.  This will embed calls to them in a timer function that
117   will then be executed by :meth:`.timeit`.  Note that the timing overhead is a
118   little larger in this case because of the extra function calls.
119
120   .. versionchanged:: 3.5
121      The optional *globals* parameter was added.
122
123   .. method:: Timer.timeit(number=1000000)
124
125      Time *number* executions of the main statement.  This executes the setup
126      statement once, and then returns the time it takes to execute the main
127      statement a number of times, measured in seconds as a float.
128      The argument is the number of times through the loop, defaulting to one
129      million.  The main statement, the setup statement and the timer function
130      to be used are passed to the constructor.
131
132      .. note::
133
134         By default, :meth:`.timeit` temporarily turns off :term:`garbage
135         collection` during the timing.  The advantage of this approach is that
136         it makes independent timings more comparable.  The disadvantage is
137         that GC may be an important component of the performance of the
138         function being measured.  If so, GC can be re-enabled as the first
139         statement in the *setup* string.  For example::
140
141            timeit.Timer('for i in range(10): oct(i)', 'gc.enable()').timeit()
142
143
144   .. method:: Timer.autorange(callback=None)
145
146      Automatically determine how many times to call :meth:`.timeit`.
147
148      This is a convenience function that calls :meth:`.timeit` repeatedly
149      so that the total time >= 0.2 second, returning the eventual
150      (number of loops, time taken for that number of loops). It calls
151      :meth:`.timeit` with increasing numbers from the sequence 1, 2, 5,
152      10, 20, 50, ... until the time taken is at least 0.2 second.
153
154      If *callback* is given and is not ``None``, it will be called after
155      each trial with two arguments: ``callback(number, time_taken)``.
156
157      .. versionadded:: 3.6
158
159
160   .. method:: Timer.repeat(repeat=5, number=1000000)
161
162      Call :meth:`.timeit` a few times.
163
164      This is a convenience function that calls the :meth:`.timeit` repeatedly,
165      returning a list of results.  The first argument specifies how many times
166      to call :meth:`.timeit`.  The second argument specifies the *number*
167      argument for :meth:`.timeit`.
168
169      .. note::
170
171         It's tempting to calculate mean and standard deviation from the result
172         vector and report these.  However, this is not very useful.
173         In a typical case, the lowest value gives a lower bound for how fast
174         your machine can run the given code snippet; higher values in the
175         result vector are typically not caused by variability in Python's
176         speed, but by other processes interfering with your timing accuracy.
177         So the :func:`min` of the result is probably the only number you
178         should be interested in.  After that, you should look at the entire
179         vector and apply common sense rather than statistics.
180
181      .. versionchanged:: 3.7
182         Default value of *repeat* changed from 3 to 5.
183
184
185   .. method:: Timer.print_exc(file=None)
186
187      Helper to print a traceback from the timed code.
188
189      Typical use::
190
191         t = Timer(...)       # outside the try/except
192         try:
193             t.timeit(...)    # or t.repeat(...)
194         except Exception:
195             t.print_exc()
196
197      The advantage over the standard traceback is that source lines in the
198      compiled template will be displayed.  The optional *file* argument directs
199      where the traceback is sent; it defaults to :data:`sys.stderr`.
200
201
202.. _timeit-command-line-interface:
203
204Command-Line Interface
205----------------------
206
207When called as a program from the command line, the following form is used::
208
209   python -m timeit [-n N] [-r N] [-u U] [-s S] [-h] [statement ...]
210
211Where the following options are understood:
212
213.. program:: timeit
214
215.. cmdoption:: -n N, --number=N
216
217   how many times to execute 'statement'
218
219.. cmdoption:: -r N, --repeat=N
220
221   how many times to repeat the timer (default 5)
222
223.. cmdoption:: -s S, --setup=S
224
225   statement to be executed once initially (default ``pass``)
226
227.. cmdoption:: -p, --process
228
229   measure process time, not wallclock time, using :func:`time.process_time`
230   instead of :func:`time.perf_counter`, which is the default
231
232   .. versionadded:: 3.3
233
234.. cmdoption:: -u, --unit=U
235
236   specify a time unit for timer output; can select ``nsec``, ``usec``, ``msec``, or ``sec``
237
238   .. versionadded:: 3.5
239
240.. cmdoption:: -v, --verbose
241
242   print raw timing results; repeat for more digits precision
243
244.. cmdoption:: -h, --help
245
246   print a short usage message and exit
247
248A multi-line statement may be given by specifying each line as a separate
249statement argument; indented lines are possible by enclosing an argument in
250quotes and using leading spaces.  Multiple :option:`-s` options are treated
251similarly.
252
253If :option:`-n` is not given, a suitable number of loops is calculated by trying
254increasing numbers from the sequence 1, 2, 5, 10, 20, 50, ... until the total
255time is at least 0.2 seconds.
256
257:func:`default_timer` measurements can be affected by other programs running on
258the same machine, so the best thing to do when accurate timing is necessary is
259to repeat the timing a few times and use the best time.  The :option:`-r`
260option is good for this; the default of 5 repetitions is probably enough in
261most cases.  You can use :func:`time.process_time` to measure CPU time.
262
263.. note::
264
265   There is a certain baseline overhead associated with executing a pass statement.
266   The code here doesn't try to hide it, but you should be aware of it.  The
267   baseline overhead can be measured by invoking the program without arguments,
268   and it might differ between Python versions.
269
270
271.. _timeit-examples:
272
273Examples
274--------
275
276It is possible to provide a setup statement that is executed only once at the beginning:
277
278.. code-block:: shell-session
279
280   $ python -m timeit -s 'text = "sample string"; char = "g"'  'char in text'
281   5000000 loops, best of 5: 0.0877 usec per loop
282   $ python -m timeit -s 'text = "sample string"; char = "g"'  'text.find(char)'
283   1000000 loops, best of 5: 0.342 usec per loop
284
285In the output, there are three fields. The loop count, which tells you how many
286times the statement body was run per timing loop repetition. The repetition
287count ('best of 5') which tells you how many times the timing loop was
288repeated, and finally the time the statement body took on average within the
289best repetition of the timing loop. That is, the time the fastest repetition
290took divided by the loop count.
291
292::
293
294   >>> import timeit
295   >>> timeit.timeit('char in text', setup='text = "sample string"; char = "g"')
296   0.41440500499993504
297   >>> timeit.timeit('text.find(char)', setup='text = "sample string"; char = "g"')
298   1.7246671520006203
299
300The same can be done using the :class:`Timer` class and its methods::
301
302   >>> import timeit
303   >>> t = timeit.Timer('char in text', setup='text = "sample string"; char = "g"')
304   >>> t.timeit()
305   0.3955516149999312
306   >>> t.repeat()
307   [0.40183617287970225, 0.37027556854118704, 0.38344867356679524, 0.3712595970846668, 0.37866875250654886]
308
309
310The following examples show how to time expressions that contain multiple lines.
311Here we compare the cost of using :func:`hasattr` vs. :keyword:`try`/:keyword:`except`
312to test for missing and present object attributes:
313
314.. code-block:: shell-session
315
316   $ python -m timeit 'try:' '  str.__bool__' 'except AttributeError:' '  pass'
317   20000 loops, best of 5: 15.7 usec per loop
318   $ python -m timeit 'if hasattr(str, "__bool__"): pass'
319   50000 loops, best of 5: 4.26 usec per loop
320
321   $ python -m timeit 'try:' '  int.__bool__' 'except AttributeError:' '  pass'
322   200000 loops, best of 5: 1.43 usec per loop
323   $ python -m timeit 'if hasattr(int, "__bool__"): pass'
324   100000 loops, best of 5: 2.23 usec per loop
325
326::
327
328   >>> import timeit
329   >>> # attribute is missing
330   >>> s = """\
331   ... try:
332   ...     str.__bool__
333   ... except AttributeError:
334   ...     pass
335   ... """
336   >>> timeit.timeit(stmt=s, number=100000)
337   0.9138244460009446
338   >>> s = "if hasattr(str, '__bool__'): pass"
339   >>> timeit.timeit(stmt=s, number=100000)
340   0.5829014980008651
341   >>>
342   >>> # attribute is present
343   >>> s = """\
344   ... try:
345   ...     int.__bool__
346   ... except AttributeError:
347   ...     pass
348   ... """
349   >>> timeit.timeit(stmt=s, number=100000)
350   0.04215312199994514
351   >>> s = "if hasattr(int, '__bool__'): pass"
352   >>> timeit.timeit(stmt=s, number=100000)
353   0.08588060699912603
354
355
356To give the :mod:`timeit` module access to functions you define, you can pass a
357*setup* parameter which contains an import statement::
358
359   def test():
360       """Stupid test function"""
361       L = [i for i in range(100)]
362
363   if __name__ == '__main__':
364       import timeit
365       print(timeit.timeit("test()", setup="from __main__ import test"))
366
367Another option is to pass :func:`globals` to the  *globals* parameter, which will cause the code
368to be executed within your current global namespace.  This can be more convenient
369than individually specifying imports::
370
371   def f(x):
372       return x**2
373   def g(x):
374       return x**4
375   def h(x):
376       return x**8
377
378   import timeit
379   print(timeit.timeit('[func(42) for func in (f,g,h)]', globals=globals()))
380