NumPy, why equality check does not work for an array of objects?

Question

With the following arrays of objects:

a = np.array([[1], [1, 2], [1, 2, 3], [1], [1]], dtype=object)
b = np.array([(1,), (1, 2), (1, 2, 3), (1,), (1,)], dtype=object)

The following equality checks do not work:

a==[1]
#array([False, False, False, False, False], dtype=bool)

b==(1,)
#array([False, False, False, False, False], dtype=bool)

if I use strings instead:

c = np.array(['[1]', '[1, 2]', '[1, 2, 3]', '[1]', '[1]'])

the equality check works:

c == '[1]'
#array([ True, False, False,  True,  True], dtype=bool)

why the array check behaves like that?

If we iterate over a or b and perform the checks it also gives the expected result:

[i==[1] for i in a]
#[True, False, False, True, True]

[i==(1,) for i in b]
#[True, False, False, True, True]

Thank you!

user2357112 · Accepted Answer · 2013-08-03 07:27:23Z

6

NumPy is designed to automatically treat array-like objects as arrays in many situations. Here, NumPy sees that [1] and (1,) are array-like objects and applies the broadcasting rules. Length-1 axes on either side are expanded out to the length of the other object's corresponding axis, and if one object has less dimensions than the other, missing dimensions are filled in on the left with the other object's lengths in those dimensions. Thus,

a == [1]

gives the same result as

a == numpy.array([1, 1, 1, 1, 1])

which is an array of 5 Falses.

answered Aug 3, 2013 at 7:27

user2357112

286k32 gold badges490 silver badges569 bronze badges

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

4 Comments

Saullo G. P. Castro Over a year ago

thank you! Do you know a way to avoid the broadcasting rules such that a is compared to [1] element-wise?

user2357112 Over a year ago

Nope. My advice is to avoid using sequences (or arrays) as array elements. You might try wrapping them in some sort of definitely-not-a-list object, or see if masked arrays fit your use case.

Robert Kern Over a year ago

You can make an empty shape-() array, then fill it with your [1] list. c=np.empty([], dtype=object); c[()] = [1]. That avoids the automatic attempt to interpret the list as its own array. Comparing a with c will give you the results you are after.

user2357112 Over a year ago

@RobertKern: Clever idea, but it makes the code a lot bigger, and if you don't do it everywhere, the places you forgot fail silently. I wouldn't recommend it.

Collectives™ on Stack Overflow

NumPy, why equality check does not work for an array of objects?

1 Answer 1

4 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Your Answer

Sign up or log in

Post as a guest

Related