[Tutor] could somebody please explain...

[Steven D'Aprano]

On Wed, Oct 01, 2014 at 09:43:29AM -0700, Clayton Kirkwood wrote:

> # program to test time and count options
> 
> import datetime,operator, sys
> from datetime import time, date, datetime
> date = datetime.now()
> dayofweek = date.strftime("%a, %b")
> print("Today is", dayofweek, date.day, "at ", date.time())
> 
> start = 0
> count_max=int(input("give me a number"))
> start_time = datetime.now()
> 
> print( start_time )
> while start  < count_max:
>     start=start + 1
> 
> end_time = datetime.now()
> print( "s=s+1 time difference is:", (end_time - start_time) )


The first problem you have here is that you are not 
actually timing how long it takes to add "start + 1". 
You're actually timing eight things:

- lookup the value of start;
- lookup the value of count_max;
- check whether the first is less than the second;
- decide whether to loop, or exit the loop;
- if we're still inside the loop, lookup start again;
- add 1 to it;
- store the result in start; 
- jump back to the top of the loop.


So the results you get don't tell you much about the speed of start+1.

Analogy: you want to know how long it takes you to drive to work in the 
morning. So you wake up, eat breakfast, brush your teeth, start the 
stopwatch, have a shower, get dressed, get in the car, drive to the gas 
station, fill up, buy a newspaper, and drive the rest of the way to 
work, and finally stop the stopwatch. The time you get is neither 
accurate as "driving time", nor "total time it takes to get to work" 
time.

Ideally, we want to do as little extra work as possible inside the 
timing loop, so we can get a figure as close as possible to the time 
actually taken by + as we can.

The second problem is that you are using datetime.now() as your clock. 
That's not a high-precision clock. It might be only be accurate to a 
second, or a millisecond. It certainly isn't accurate enough to measure 
a single addition:

py> from datetime import datetime
py> x = 1
py> t = datetime.now(); x + 1; datetime.now() - t
2
datetime.timedelta(0, 0, 85)


This tells me that it supposedly took 85 microseconds to add two 
numbers, but as I showed before with timeit, the real figure is closer 
to 0.09 microseconds. That's a lot of noise! About 85000% noise!

Unfortunately, it is tricky to know which clock to use. On Windows, 
time.clock() used to be the best one; on Linux, time.time() was the 
best. Starting in Python 3.3, there are a bunch more accurate clocks in 
the time module. But if you use the timeit module, it already picks the 
best clock for the job. But if in doubt, time.time() will normally be 
acceptable.

https://docs.python.org/3/library/time.html

https://docs.python.org/3/library/timeit.html



-- 
Steven