I
Ivan Voras
Are there any performance/size differences between using tuples and using
lists?
lists?
C
Christos TZOTZIOY Georgiou
Assuming you know that tuples are immutable and lists are mutable (they
can grow, shrink, get sorted etc), tuples consume less memory (lists
allocate more memory than your list needs, to avoid reallocating space
on every append/insert). This accounts for size.
Performance-wise, I don't think there is a noticable difference between
indexing lists and tuples (but you can always use the timeit module).
IIRC Guido has stated that tuples are not intended to be immutable
lists, but a kind of handy unnamed structs (a la C) or records (a la
Pascal).
Keep in mind also that lists have useful methods (like count and index)
that tuples don't. This about functionality.
L
Lukasz Pankowski
Ivan Voras said:
I made the timings of creation/unpacking/indexing for short
tuples/lists and loops for longer tuples/lists. In Python 2.1, 2.2 and
2.3 creation and unpacking of short tuples is 2x/3x faster which is
what you may expact knowing they are fixed size, but indexing is
slightly slower for tuples which I didn't expect. Whether you loop
over tuple or list does matter only for 2.2 (tuples are 30% slower).
Where there is 2x difference or 30% difference it is reproducable in
subsequent runs of of the script given below.
lupan@psi:[~/dyplom/src]$ sh tuple-vs-list.sh
2.1
creation
tuple 1000000 loops, best of 3: 0.442 usec per loop
list 1000000 loops, best of 3: 1.1 usec per loop
unpacking
tuple 1000000 loops, best of 3: 0.58 usec per loop
list 1000000 loops, best of 3: 1.23 usec per loop
indexing
tuple 1000000 loops, best of 3: 0.369 usec per loop
list 1000000 loops, best of 3: 0.339 usec per loop
looping
tuple 10000 loops, best of 3: 170 usec per loop
list 10000 loops, best of 3: 173 usec per loop
2.2
creation
tuple 1000000 loops, best of 3: 0.341 usec per loop
list 1000000 loops, best of 3: 0.96 usec per loop
unpacking
tuple 1000000 loops, best of 3: 0.457 usec per loop
list 1000000 loops, best of 3: 1.09 usec per loop
indexing
tuple 1000000 loops, best of 3: 0.286 usec per loop
list 1000000 loops, best of 3: 0.264 usec per loop
looping
tuple 10000 loops, best of 3: 149 usec per loop
list 10000 loops, best of 3: 114 usec per loop
2.3
creation
tuple 1000000 loops, best of 3: 0.286 usec per loop
list 1000000 loops, best of 3: 0.672 usec per loop
unpacking
tuple 1000000 loops, best of 3: 0.387 usec per loop
list 1000000 loops, best of 3: 0.761 usec per loop
indexing
tuple 1000000 loops, best of 3: 0.204 usec per loop
list 1000000 loops, best of 3: 0.19 usec per loop
looping
tuple 10000 loops, best of 3: 74.6 usec per loop
list 10000 loops, best of 3: 76.3 usec per loop
# tuple-vs-list.py
TIMEIT=/usr/lib/python2.3/timeit.py
# they are big as I have fast machine
N=1000000
N2=10000
for ver in 2.1 2.2 2.3; do
echo $ver
echo -ne 'creation\ntuple '
python$ver -O $TIMEIT -n $N '(1,2,3)'
echo -n 'list '
python$ver -O $TIMEIT -n $N '[1,2,3]'
echo -ne 'unpacking\ntuple '
python$ver -O $TIMEIT -n $N 'a,b,c = (1,2,3)'
echo -n 'list '
python$ver -O $TIMEIT -n $N 'a,b,c = [1,2,3]'
echo -ne 'indexing\ntuple '
python$ver -O $TIMEIT -n $N -s 'x = (1,2,3)' 'x[1]'
echo -n 'list '
python$ver -O $TIMEIT -n $N -s 'x = [1,2,3]' 'x[1]'
echo -ne 'looping\ntuple '
python$ver -O $TIMEIT -n $N2 -s 'r = tuple(range(1000))' 'for x in r: pass'
echo -n 'list '
python$ver -O $TIMEIT -n $N2 -s 'r = range(1000)' 'for x in r: pass'
done
I
Ivan Voras
Ok, I gather touples should be smaller & faster as I thought 
I
Ivan Voras
Lukasz said:
Thank you, this is exactly what I was looking for.
P
Peter Hansen
Ivan said:Ok, I gather touples should be smaller & faster as I thought
Don't make a decision based on performance. Use tuples if you
require the immutability, or if you want to follow Guido's "advice" (?)
to treat them as simple unnamed structs.
Use lists any time you want lists of stuff, even if you think you
might benefit by the supposedly higher performance of a tuple.
Anyone reading your code will likely be much happier, and you won't
find yourself getting caught in a bind because you've used an
immutable tuples where you really should have used a list.
-Peter
I
Ivan Voras
Peter said:
I know that. I was just wandering if they also behave faster. Lists are only
interesting to me in case I need mutability.
P
Peter Hansen
Ivan said:
I'd suggest you have that backwards. Lists should _always_ be interesting
to you. Tuples should be interesting only in the case where you need
*immutability*.
-Peter
I
Ivan Voras
Peter said:
I disagree.
I would always use tuples except when I explicitely needmutability. That doesn't meen that I would intentionaly (badly) restructure
code just to use tuples, only that in many cases my structures don't need to
be modified.
P
Peter Hansen
Ivan said:I disagree.
mutability. That doesn't meen that I would intentionaly (badly) restructure
code just to use tuples, only that in many cases my structures don't need to
be modified.
That's your personal choice, I suppose, and I won't try any more to dissuade
you any more except to point out that most Python programmers, I believe, do
not see tuples as their first choice, but use them only in certain special cases.
I could be wrong. It happens. ;-)
-Peter
D
Donn Cave
Peter Hansen said:
These two positions may be more similar than different. The
notion of a sequence that never needs to be modified is certainly
a special case, and if we were looking at the actual applications
for these sequences we might all agree that they're more or less
appropriately cast as tuples. One argues about principles like
this at some risk of silliness.
The thing that eventually pushed me towards lists and away from
tuples, for sequences of like items, is the notation. It may
seem like a awfully prosaic consideration, but lists are usually
easier to mentally parse out of an expression, because the tuple's
parentheses are shared by other syntactical groupings. And then
there's the awkward notation for a single-element tuple. Lists
are easier to write and easier to read.
Donn Cave, (e-mail address removed)