python numpy argsort issue

Question

First I generated a random 1d array x, then I generated array y by swapping elements of x.

Now y is a permutation of x, in principle, if I apply numpy argsort, I should get different results, as it turned out, it was not the case.

Here are my python codes,

import numpy as np
x = np.random.random(10)
print(x)
y = x
y[0], y[1] = x[1], x[0]
y[3], y[4] = x[4], x[3]
print(y)

Then I got

[ 0.818  0.99   0.291  0.608  0.773  0.232  0.041  0.136  0.645  0.94 ]
[ 0.99   0.818  0.291  0.773  0.608  0.232  0.041  0.136  0.645  0.94 ]

Now

print(np.argsort(x))
print(np.argsort(y))

I got

[6 7 5 2 4 8 3 1 9 0]
[6 7 5 2 4 8 3 1 9 0]

@Arman I'm using python3, it might work on other versions of python. BTW x is a numpy array not a list. — Machine, Jan 12 '16 at 22:15
@Arman -- I'm not sure about that. First, `x[:]` will only have a chance at working for 1D arrays. Second, `x[:]` creates a view in numpy. I think that this _can_ create a copy under some circumstances, but it frequently doesn't. . . — mgilson, Jan 12 '16 at 22:32
I nominated this for re-opening because the dupe question specifically dealt with lists when OP is working with numpy arrays. The behavior in this regard is certainly different. — mgilson, Jan 12 '16 at 22:33

score 4 · Accepted Answer · answered Jan 12 '16 at 22:03

4

When you do:

y = x

You alias y and x. They're the same object which you can see by issuing the following statement

y is x  # True

Instead, you probably want:

y = x.copy()

answered Jan 12 '16 at 22:03

mgilson

Thanks! `y = x` and `y = x[:]` do NOT work; `y is x` and `y = x.copy()` both work. – Machine Jan 12 '16 at 22:13

1 Answers1