I'd like to know if there is a more efficient/pythonic way to add multiple numpy arrays (2D) rather than:
def sum_multiple_arrays(list_of_arrays):
a = np.zeros(shape=list_of_arrays[0].shape) #initialize array of 0s
for array in list_of_arrays:
a += array
return a
Ps: I am aware of np.add() but it works only with 2 arrays.
np.sum(list_of_arrays, axis=0)
should work. Or
np.add.reduce(list_of_arrays)
The simplest and most Pythonic solution is simply to use sum(), like so:
sum(list_of_arrays)
This may even be faster than other options, at least under some conditions (thanks to Loc Quan for pointing this out!). For example, on my laptop, with Python 3.10.2 and NumPy 2.1.3, I ran the following code:
n = 100000; m = 5; k = 5;
list_of_arrays = [np.random.random_sample((n,m)) for _ in range(k)]
%timeit np.sum(list_of_arrays, axis=0)
7.29 ms ± 459 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)
%timeit np.add.reduce(list_of_arrays)
7.04 ms ± 134 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)
%timeit reduce(np.add, list_of_arrays)
2.04 ms ± 86.2 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)
%timeit sum(list_of_arrays)
1.59 ms ± 44.9 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
Showing a huge advantage to sum() over np.sum() and np.add.reduce(), and a more modest but still significant speedup vs. reduce(np.add, ...).
sum([np.array([0,1]), np.array([1,0]), np.array([1.1, 1.1])])
sum() runs faster than np.sum(..., axis=0) and np.add.reduce() by ~25% (1.4ms compared to 1.8ms) for my list of 5 arrays of ~100000 rows and 5 columns. I have checked again with a list of np.random.sample((100_000, 5)) arrays and same results.i know the question says numpy, but here's a pure python answer in case someone needs it.
map(sum, zip(a,b))
which returns a generator.
you can also do list(map(sum, zip(a,b))) to get a list
np.sum(list_of_arrays, axis=0)should work. Ornp.add.reduce(list_of_arrays).