circular iteration

[Alex Martelli]

Simon Brunning <simon.brunning at gmail.com> wrote:
   ...
> > is there a faster way to build a circular iterator in python that by
doing this:
> > 
> > c=['r','g','b','c','m','y','k']
> > 
> > for i in range(30):
> >     print c[i%len(c)]
> 
> I don''t know if it's faster, but:
> 
> >>> import itertools
> >>> c=['r','g','b','c','m','y','k']
> >>> for i in itertools.islice(itertools.cycle(c), 30):
> ...   print i

Whenever you're using itertools, the smart money's on "yes, it's
faster";-).

E.g., on a slow, old iBook...:

kallisti:~ alex$ python -mtimeit -s'c="rgbcmyk"' 'for i in range(30):
c[i%len(c)]'
10000 loops, best of 3: 47 usec per loop

kallisti:~ alex$ python -mtimeit -s'c="rgbcmyk"; import itertools as it'
'for i in it.islice(it.cycle(c),30): i'
10000 loops, best of 3: 26.4 usec per loop

Of course, if you do add back the print statements they'll take orders
of magnitude more time than the cyclic access, so /F's point on
premature optimization may well be appropriate.  But, if you're doing
something VERY speedy with each item you access, maybe roughly halving
the overhead for the cyclic access itself MIGHT be measurable (maybe
not; it IS but a few microseconds, after all).

I like itertools' approach because it's higher-abstraction and more
direct.  Its blazing speed is just a trick to sell it to conservative
curmudgeons who don't see abstraction as an intrinsic good -- some of
those are swayed by microseconds;-)


Alex