`numpy.sum` vs. `ndarray.sum`

Question

For a 1-D numpy array a, I thought that np.sum(a) and a.sum() are equivalent functions, but I just did a simple experiment, and it seems that the latter is always a bit faster:

In [1]: import numpy as np

In [2]: a = np.arange(10000)

In [3]: %timeit np.sum(a)
The slowest run took 16.85 times longer than the fastest. This could mean that an intermediate result is being cached.
100000 loops, best of 3: 6.46 µs per loop

In [4]: %timeit a.sum()
The slowest run took 19.80 times longer than the fastest. This could mean that an intermediate result is being cached.
100000 loops, best of 3: 5.25 µs per loop

Why is there a difference? Does this mean that we should always use the numpy.ndarray version of functions like sum, mean, std, etc.?

Mostly you are seeing a difference in one level of function redirection. In most of these cases the function version redirects the task to the method (look at the code). Don't worry about speed here - use the form that makes your code clearest (to you and your readers). You must use the function version if your input might be a list instead of an array. — hpaulj
– hpaulj, Commented Feb 23, 2018 at 6:48
Last year I answered something similar np.sum and np.add.reduce - in production, what do you use? — hpaulj
– hpaulj, Commented Feb 23, 2018 at 6:59

Daniel F · Accepted Answer · 2018-02-23 07:06:28Z

4

I'd imagine it is becasue np.sum() and the like ~~needs to explicitly convert the inputs to ndarray first (using np.asanyarray)~~ checks a few other .sum functions before settling on the ndarray.sum method in order to allow operation on lists, tuples, etc.

On the other hand, ndarray.sum() is a method of the ndarray class and thus doesn't need to do any checking.

edited Feb 23, 2018 at 7:06

answered Feb 23, 2018 at 6:41

Daniel F

14.5k2 gold badges34 silver badges59 bronze badges

Sign up to request clarification or add additional context in comments.

6 Comments

p-value Over a year ago

Thanks, but I thought the conversion should just boil down to a simple check if a is already an array, and shouldn't involve explicit copying, right?

Daniel F Over a year ago

You'd think so, but it actually seems to deduce that a is an ndarray only if nothing else's .sum method works, including generators and the old numeric multiarray

Daniel F Over a year ago

Which makes sense, you don't want to convert a big linked list to ndarray if you don't have to

Eric Over a year ago

@Hilbert: No copy is made - the overhead is O(1), not O(N).

javidcf Over a year ago

@Eric My guess is that is to maintain the functionality of the standard sum in case you do from numpy import * or from pylab import *.

|

Collectives™ on Stack Overflow

`numpy.sum` vs. `ndarray.sum`

1 Answer 1

6 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

6 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related