Page - 124 - in Programming for Computations – Python - A Gentle Introduction to Numerical Simulations with Python 3.6, Volume Second Edition

124 5 SomeMorePythonEssentials y = np.zeros(1000) # ...use the function add t = timeit.Timer(’for i in range(len(x)): x[i] = add(i, i+1)’, \ setup=’from __main__ import add, x’) x_time = t.timeit(10000) # Time 10000 runs of the whole loop print(’Time, function call: {:g} seconds’.format(x_time)) # ...no use of function add t = timeit.Timer(’for i in range(len(y)): y[i] = i + (i+1)’, \ setup=’from __main__ import y’) y_time = t.timeit(10000) # Time 10000 runs of the whole loop print(’Time: {:g} seconds’.format(y_time)) What will happen here? Well, first of all, note that there are two calls to timeit.Timer, one for each of the two loops from above. If we look at the first call totimeit.Timer, i.e., t = timeit.Timer(’for i in range(len(x)): x[i] = add(i, i+1)’, \ setup=’from __main__ import add, x’) we notice that two arguments are provided. You may recognize the first argument, for i in range(len(x)): x[i] = add(i, i+1), asaone-lineversionof the first loopfromabove, i.e. the loopoverx (usually,weprefer towritesuch loopsnot onasingle line.However,whenusedasanargumentina functioncall likehere, the one-line version is handy). This first argument, given as a string, is what we want the timingof.Thesecondargument,setup=’from __main__ import add, x’, is required for initialization, i.e., what the timer needs to do prior to timing of the loop. If you lookcarefullyat the string-partof this second argument,younotice an importstatementforaddandx.Youmaywonderwhyyouhavetodothatwhenthey aredefined inyourcodeabove,but stay relaxedabout that, it is simply theway this timer function works. What is required for the timer function to execute the code given in the first argument, must be provided in the setup argument, even if it is defined in thecodeabove. The followingline, x_time = t.timeit(10000) # Time 10000 runs of the whole loop will cause the whole loop to actually be executed, not a single time, but 10000 times! There will be one recorded time, the time required to run the loop 10000 times. Thus, if an average time for a single run-through of the loop is desired, we must divide the recorded time by (in this case) 10000. Often, however, the total time is fine for comparison between alternatives. Theprint command brings the recordedtime to the screen,before thenext loop is timed inanequivalentway. Why is the loop run 10000 times? To get reliable timings, the execution times must be on the order of seconds, that is why. How many times the requested code snippet needs to be run, will of course depend on the code snippet in question. Sometimes, a single execution is enough. Other times, many more executions than 10000 are required. Some trial and error is usually required to find an appropriate number. Executing theprogramproduces the followingresult, Time, function call: 2.22121 seconds Time: 1.4738 seconds