Doc/lib/libtimeit.tex

   1 \section{\module{timeit} ---
   2          Measure execution time of small code snippets}
   3
   4 \declaremodule{standard}{timeit}
   5 \modulesynopsis{Measure the execution time of small code snippets.}
   6
   7 \versionadded{2.3}
   8 \index{Benchmarking}
   9 \index{Performance}
  10
  11 This module provides a simple way to time small bits of Python code.
  12 It has both command line as well as callable interfaces.  It avoids a
  13 number of common traps for measuring execution times.  See also Tim
  14 Peters' introduction to the ``Algorithms'' chapter in the
  15 \citetitle{Python Cookbook}, published by O'Reilly.
  16
  17 The module defines the following public class:
  18
  19 \begin{classdesc}{Timer}{\optional{stmt=\code{'pass'}
  20                          \optional{, setup=\code{'pass'}
  21                          \optional{, timer=<timer function>}}}}
  22 Class for timing execution speed of small code snippets.
  23
  24 The constructor takes a statement to be timed, an additional statement
  25 used for setup, and a timer function.  Both statements default to
  26 \code{'pass'}; the timer function is platform-dependent (see the
  27 module doc string).  The statements may contain newlines, as long as
  28 they don't contain multi-line string literals.
  29
  30 To measure the execution time of the first statement, use the
  31 \method{timeit()} method.  The \method{repeat()} method is a
  32 convenience to call \method{timeit()} multiple times and return a list
  33 of results.
  34 \end{classdesc}
  35
  36 \begin{methoddesc}{print_exc}{\optional{file=\constant{None}}}
  37 Helper to print a traceback from the timed code.
  38
  39 Typical use:
  40
  41 \begin{verbatim}
  42     t = Timer(...)       # outside the try/except
  43     try:
  44         t.timeit(...)    # or t.repeat(...)
  45     except:
  46         t.print_exc()
  47 \end{verbatim}
  48
  49 The advantage over the standard traceback is that source lines in the
  50 compiled template will be displayed.
  51 The optional \var{file} argument directs where the traceback is sent;
  52 it defaults to \code{sys.stderr}.
  53 \end{methoddesc}
  54
  55 \begin{methoddesc}{repeat}{\optional{repeat\code{=3} \optional{,
  56                            number\code{=1000000}}}}
  57 Call \method{timeit()} a few times.
  58
  59 This is a convenience function that calls the \method{timeit()}
  60 repeatedly, returning a list of results.  The first argument specifies
  61 how many times to call \method{timeit()}.  The second argument
  62 specifies the \var{number} argument for \function{timeit()}.
  63
  64 \begin{notice}
  65 It's tempting to calculate mean and standard deviation from the result
  66 vector and report these.  However, this is not very useful.  In a typical
  67 case, the lowest value gives a lower bound for how fast your machine can run
  68 the given code snippet; higher values in the result vector are typically not
  69 caused by variability in Python's speed, but by other processes interfering
  70 with your timing accuracy.  So the \function{min()} of the result is
  71 probably the only number you should be interested in.  After that, you
  72 should look at the entire vector and apply common sense rather than
  73 statistics.
  74 \end{notice}
  75 \end{methoddesc}
  76
  77 \begin{methoddesc}{timeit}{\optional{number\code{=1000000}}}
  78 Time \var{number} executions of the main statement.
  79 This executes the setup statement once, and then
  80 returns the time it takes to execute the main statement a number of
  81 times, measured in seconds as a float.  The argument is the number of
  82 times through the loop, defaulting to one million.  The main
  83 statement, the setup statement and the timer function to be used are
  84 passed to the constructor.
  85 \end{methoddesc}
  86
  87
  88 \subsection{Command Line Interface}
  89
  90 When called as a program from the command line, the following form is used:
  91
  92 \begin{verbatim}
  93 python timeit.py [-n N] [-r N] [-s S] [-t] [-c] [-h] [statement ...]
  94 \end{verbatim}
  95
  96 where the following options are understood:
  97
  98 \begin{description}
  99 \item[-n N/\longprogramopt{number=N}] how many times to execute 'statement'
 100 \item[-r N/\longprogramopt{repeat=N}] how many times to repeat the timer (default 3)
 101 \item[-s S/\longprogramopt{setup=S}] statement to be executed once initially (default
 102 \code{'pass'})
 103 \item[-t/\longprogramopt{time}] use \function{time.time()}
 104 (default on all platforms but Windows)
 105 \item[-c/\longprogramopt{clock}] use \function{time.clock()} (default on Windows)
 106 \item[-v/\longprogramopt{verbose}] print raw timing results; repeat for more digits
 107 precision
 108 \item[-h/\longprogramopt{help}] print a short usage message and exit
 109 \end{description}
 110
 111 A multi-line statement may be given by specifying each line as a
 112 separate statement argument; indented lines are possible by enclosing
 113 an argument in quotes and using leading spaces.  Multiple
 114 \programopt{-s} options are treated similarly.
 115
 116 If \programopt{-n} is not given, a suitable number of loops is
 117 calculated by trying successive powers of 10 until the total time is
 118 at least 0.2 seconds.
 119
 120 The default timer function is platform dependent.  On Windows,
 121 \function{time.clock()} has microsecond granularity but
 122 \function{time.time()}'s granularity is 1/60th of a second; on \UNIX,
 123 \function{time.clock()} has 1/100th of a second granularity and
 124 \function{time.time()} is much more precise.  On either platform, the
 125 default timer functions measure wall clock time, not the CPU time.
 126 This means that other processes running on the same computer may
 127 interfere with the timing.  The best thing to do when accurate timing
 128 is necessary is to repeat the timing a few times and use the best
 129 time.  The \programopt{-r} option is good for this; the default of 3
 130 repetitions is probably enough in most cases.  On \UNIX, you can use
 131 \function{time.clock()} to measure CPU time.
 132
 133 \begin{notice}
 134   There is a certain baseline overhead associated with executing a
 135   pass statement.  The code here doesn't try to hide it, but you
 136   should be aware of it.  The baseline overhead can be measured by
 137   invoking the program without arguments.
 138 \end{notice}
 139
 140 The baseline overhead differs between Python versions!  Also, to
 141 fairly compare older Python versions to Python 2.3, you may want to
 142 use Python's \programopt{-O} option for the older versions to avoid
 143 timing \code{SET_LINENO} instructions.
 144
 145 \subsection{Examples}
 146
 147 Here are two example sessions (one using the command line, one using
 148 the module interface) that compare the cost of using
 149 \function{hasattr()} vs. \keyword{try}/\keyword{except} to test for
 150 missing and present object attributes.
 151
 152 \begin{verbatim}
 153 % timeit.py 'try:' '  str.__nonzero__' 'except AttributeError:' '  pass'
 154 100000 loops, best of 3: 15.7 usec per loop
 155 % timeit.py 'if hasattr(str, "__nonzero__"): pass'
 156 100000 loops, best of 3: 4.26 usec per loop
 157 % timeit.py 'try:' '  int.__nonzero__' 'except AttributeError:' '  pass'
 158 1000000 loops, best of 3: 1.43 usec per loop
 159 % timeit.py 'if hasattr(int, "__nonzero__"): pass'
 160 100000 loops, best of 3: 2.23 usec per loop
 161 \end{verbatim}
 162
 163 \begin{verbatim}
 164 >>> import timeit
 165 >>> s = """\
 166 ... try:
 167 ...     str.__nonzero__
 168 ... except AttributeError:
 169 ...     pass
 170 ... """
 171 >>> t = timeit.Timer(stmt=s)
 172 >>> print "%.2f usec/pass" % (1000000 * t.timeit(number=100000)/100000)
 173 17.09 usec/pass
 174 >>> s = """\
 175 ... if hasattr(str, '__nonzero__'): pass
 176 ... """
 177 >>> t = timeit.Timer(stmt=s)
 178 >>> print "%.2f usec/pass" % (1000000 * t.timeit(number=100000)/100000)
 179 4.85 usec/pass
 180 >>> s = """\
 181 ... try:
 182 ...     int.__nonzero__
 183 ... except AttributeError:
 184 ...     pass
 185 ... """
 186 >>> t = timeit.Timer(stmt=s)
 187 >>> print "%.2f usec/pass" % (1000000 * t.timeit(number=100000)/100000)
 188 1.97 usec/pass
 189 >>> s = """\
 190 ... if hasattr(int, '__nonzero__'): pass
 191 ... """
 192 >>> t = timeit.Timer(stmt=s)
 193 >>> print "%.2f usec/pass" % (1000000 * t.timeit(number=100000)/100000)
 194 3.15 usec/pass
 195 \end{verbatim}
 196
 197 To give the \module{timeit} module access to functions you
 198 define, you can pass a \code{setup} parameter which contains an import
 199 statement:
 200
 201 \begin{verbatim}
 202 def test():
 203     "Stupid test function"
 204     L = []
 205     for i in range(100):
 206         L.append(i)
 207
 208 if __name__=='__main__':
 209     from timeit import Timer
 210     t = Timer("test()", "from __main__ import test")
 211     print t.timeit()
 212 \end{verbatim}