Inconsistent Numpy array aliasing behavior

Question

The following behavior is expected and is what I get. This is consistent with how aliasing works for native Python objects like lists.

>>> x = np.array([1, 2, 3])
>>> y = x
>>> x
array([1, 2, 3])
>>> y
array([1, 2, 3])
>>> x = x + np.array([2, 3, 4])
>>> x
array([3, 5, 7])
>>> y
array([1, 2, 3])

But the following behavior is unexpected by changing x = x + np.array([2, 3, 4]) to x += np.array([2, 3, 4])

>>> x += np.array([2, 3, 4])
>>> x
array([3, 5, 7])
>>> y
array([3, 5, 7])

The Numpy version is 1.16.4 on my machine. Is this a bug or feature? If it is a feature how x = x + np.array([2, 3, 4]) differs from x += np.array([2, 3, 4])

Answer 1

Your line y = x doesn't create a copy of the array; it simply tells y to point to the same data as x , which you can see if you look at their id s:

x = np.array([1,2,3])
y = x
print(id(x), id(y))

(140644627505280, 140644627505280)

x = x + np.array([2, 3, 4]) will do a reassignment of x to a new id , while x += np.array([2, 3, 4]) will modify it in place. Thus, the += will also modify y, while x = x +... won't.

x += np.array([2, 3, 4])
print(id(x))
print(x, y)

x = x + np.array([2, 3, 4])
print(id(x))
print(x, y)

140644627505280
[3 5 7] [3 5 7]
140644627175744
[ 5  8 11] [3 5 7]