xrange question

[Paul Prescod]

John Flynn wrote:
> 
>...
> 
> Yep, I see your point about the clarity of for-loops.
> 
> Out of curiosity, at what point do you find that 'xrange' becomes most
> useful? Do you have a rule of thumb, eg. more than, say, 500 iterations uses
> an 'xrange' instead of a range? Or does xrange _replace_ range for you in
> for-loops?

No, xrange is much slower for small loops so I don't replace all ranges
with xranges. Range is also more idiomatic. I think on my machine xrange
starts to "win" around 2000.  But allocating even a million items in a
regular range is still quite fast:

>>> start = time.time() ; a=range(1000000) ; print time.time() - start
0.30999994278

So even at a million either one is fine on modern machines. Plus,
consider that if you make a range of a million items you are probably
actually doing something a million times and the construction of the
range list is hardly going to be your bottleneck.

I wouldn't usually think twice to spend a third of a second getting
ready to process a million lines of text or a million patient records or
whatever...

Here's a program you can use to get a feeling for the behavior of range
and xrange:

import time

def testfunc(func, size):
    # always do a million iterations, but do it in 100-element chunks,
or
    # 1000-element chunks etc.

    # I always want to do a million iterations to get a good sense for
    # the performance of the function I'm using.
    loopfactor = 1000000/float(size)
    start = time.time()
    for i in xrange(loopfactor): 
        for i in func(size):
            pass
    return time.time() - start
    
for i in [1, 100, 1000, 2000, 5000, 10000, 100000, 1000000]:
    print i, "iterations"
    print "Range time: ", testfunc(range, i) 
    print "XRange time: ", testfunc(xrange, i) 
    print
-- 
Take a recipe. Leave a recipe.  
Python Cookbook!  http://www.ActiveState.com/pythoncookbook