Velocity Reviews (http://www.velocityreviews.com/forums/index.php)
-   Python (http://www.velocityreviews.com/forums/f43-python.html)
-   -   Re: looping versus comprehension (http://www.velocityreviews.com/forums/t957090-re-looping-versus-comprehension.html)

 Chris Angelico 01-30-2013 03:49 PM

Re: looping versus comprehension

On Thu, Jan 31, 2013 at 1:58 AM, Robin Becker <robin@reportlab.com> wrote:
> however, when I tried an experiment in python 2.7 using the script below I
> find that the looping algorithms perform better. A naive loop using list +=
> list would appear to be an O(n**2) operation, but python seems to be doing
> better than that. Also why does the append version fail so dismally. Is my
> test coded wrongly or is pre-allocation of the list making this better than
> expected?

First off, are you aware that your first three blocks of code and your
last four produce different results? The first ones flatten the list,
the others simply convert tuples to lists. With n = 3:

>>> points = []
>>> for xy in row:

points += [xy[0],xy[1]]
>>> points

[0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6]
>>> map(list,row)

[[0, 1], [1, 2], [2, 3], [3, 4], [4, 5], [5, 6]]

Once that's sorted out, then timings can be played with. But it's
worth noting that list appending is not going to be O(N*N), because
it's going to allow room for expansion.

ChrisA

 Steven D'Aprano 01-31-2013 05:17 AM

Re: looping versus comprehension

On Thu, 31 Jan 2013 02:49:31 +1100, Chris Angelico wrote:

> it's worth
> noting that list appending is not going to be O(N*N), because it's going
> to allow room for expansion.

This is true for list.append, which is amortized constant time. But it is
not true for list addition, alist + blist, which is O(N**2) and hence
gets really, really slow:

steve@runes:~\$ python -m timeit "L = []" "for i in xrange(1000): L = L + [1]"
100 loops, best of 3: 3.08 msec per loop
steve@runes:~\$ python -m timeit "L = []" "for i in xrange(5000): L = L + [1]"
10 loops, best of 3: 71 msec per loop
steve@runes:~\$ python -m timeit "L = []" "for i in xrange(25000): L = L + [1]"
10 loops, best of 3: 2.06 sec per loop

Notice that as the number of list additions goes up by a factor of 5,
the time taken goes up by a factor of 25.

--
Steven

 Roy Smith 01-31-2013 08:28 AM

Re: looping versus comprehension

In article <5109fe6b\$0\$11104\$c3e8da3@news.astraweb.com>,
Steven D'Aprano <steve+comp.lang.python@pearwood.info> wrote:

> On Thu, 31 Jan 2013 02:49:31 +1100, Chris Angelico wrote:
>
> > it's worth
> > noting that list appending is not going to be O(N*N), because it's going
> > to allow room for expansion.

>
> This is true for list.append, which is amortized constant time. But it is
> not true for list addition, alist + blist, which is O(N**2) and hence
> gets really, really slow:
>
> steve@runes:~\$ python -m timeit "L = []" "for i in xrange(1000): L = L + [1]"
> 100 loops, best of 3: 3.08 msec per loop
> steve@runes:~\$ python -m timeit "L = []" "for i in xrange(5000): L = L + [1]"
> 10 loops, best of 3: 71 msec per loop
> steve@runes:~\$ python -m timeit "L = []" "for i in xrange(25000): L = L + [1]"
> 10 loops, best of 3: 2.06 sec per loop
>
>
> Notice that as the number of list additions goes up by a factor of 5,
> the time taken goes up by a factor of 25.

It's not the addition, per-se, that's the problem. It's the creation of
a new list each time. If you use +=, it's back to O(n):

~\$ python -m timeit "L = []" "for i in xrange(1000): L += [1]"
1000 loops, best of 3: 275 usec per loop

~\$ python -m timeit "L = []" "for i in xrange(5000): L += [1]"
1000 loops, best of 3: 1.34 msec per loop

~\$ python -m timeit "L = []" "for i in xrange(25000): L += [1]"
100 loops, best of 3: 6.91 msec per loop

 All times are GMT. The time now is 04:13 PM.