Measuring very short execution times has some pitfalls that can be avoided
with the help of the timeit
Python module.
Sometimes one might be interested in measuring the performance of a very small bit of Python code, such as a single statement or a function call. Measuring reliably very short execution times can be challenging, and one needs typically to gather some statistics for error estimates.
In order to avoid common traps in such a measurement, Python standard library
contains the timeit
module. The module can be used from the command line,
within a program, or within an interactive interpreter (especially convenient
with IPython) or in Jupyter notebooks. When a code snippet is measured with
timeit
, the module takes care of running the snippet several times and
calculating statistics.
IPython interpreter provides a special magic command %timeit
, which will
measure the statement following it using the timeit
module. In the
following example, we compare the execution time of sin
and cos
functions
with the same input:
IPython 6.1.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: from math import sin, cos
In [2]: %timeit sin(0.2)
68.8 ns ± 0.181 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
In [3]: %timeit cos(0.2)
71.1 ns ± 1.57 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each)
As can be seen, with timeit
each measurement consist of a huge number of
evaluations (10000000 loops), and the measurent is even repeated multiple
times (7 runs) for better statistics.
The timeit
module can be used also from command line. The expression to be
measured is given as string, and possible setup (e.g. module import) is
provided with -s argument:
python3 -m timeit -s "from math import sin" "sin(0.2)"
10000000 loops, best of 3: 0.052 usec per loop
The setup statement (from math import sin
) above is executed once initially,
and the actual statement then multiple times.
Whole Python scripts cannot be directly measured, but if the script provides
e.g. a main
function it can be measured. For larger entities one might want
to specify also how many times the statement is executed with -n argument:
python3 -m timeit -n 3 -s "from mandel_main import main" "main()"
3 loops, best of 3: 569 msec per loop
In above, a single measurement consists of three evaluations of main
and the
measurement is then repeated three times.
The timeit
module can be used also directly inside scripts, more information
can be found in the Documentation of timeit.