Numpy argsort behavior for equal numbers

Question

array of odd length ( 15 ):

a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
np.argsort(a)

output = [ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14]

array of odd length ( 17 ):

a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
np.argsort(a)

output = [ 0 14 13 12 11 10 9 15 8 6 5 4 3 2 1 7 16]

array of odd length ( 35 ):

a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
np.argsort(a)

output = [ 0 19 20 21 22 23 24 18 25 27 28 29 30 31 32 26 33 17 15 1 2 3 4 5 6 16 7 9 10 11 12 13 14 8 34]

array of even length (16):

a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
    np.argsort(a)

output = [ 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15]

array of even length ( 18 ):

a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
print len(a)

output = [ 0 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 16 17]

array of even length ( 36 ):

a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
np.argsort(a)

output = [0 20 21 22 23 24 25 19 26 28 29 30 31 32 33 27 18 17 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 34 35]

So would somebody explain me how this algorithm works in case of equal numbers ?

its random for identical numbers, use `kind='mergesort'` with it. — Divakar, Nov 22 '17 at 17:24
I did that many times and same numbers appears , so how its random ? — Ahmed Ramzy, Nov 22 '17 at 17:26
I guess we need to dig deep into `quicksort` implementation in it to find out. — Divakar, Nov 22 '17 at 17:27
For the default quicksort, unspecified behavior. The people who wrote the algorithm don't care, and if you care, you shouldn't pick quicksort. — user2357112, Nov 22 '17 at 17:28
@Divakar 'arbitrary' or 'undefined' would be better wording than 'random' — Chris_Rands, Nov 22 '17 at 17:32

score 4 · Accepted Answer · answered Nov 22 '17 at 17:29

argsort() uses a configurable sorting algorithm; you can pick between quicksort, mergesort or heapsort. From the argsort documentation:

numpy.argsort(a, axis=-1, kind='quicksort', order=None)

[...]

kind : {‘quicksort’, ‘mergesort’, ‘heapsort’}, optional
Sorting algorithm.

The default is quicksort, and as Wikipedia states:

In efficient implementations it is not a stable sort, meaning that the relative order of equal sort items is not preserved.

Numpy's implementation is efficient and it does not preserve the relative order of equal values.

If you need a stable sort, use mergesort instead:

>>> a = np.array([5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5])
>>> np.argsort(a, kind='mergesort')
array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34])

Heapsort is not stable either.

Numpy argsort behavior for equal numbers

1 Answers1

Linked