numpy.empty giving nonempty array

Question

When I create an empty numpy array using foo = np.empty(1) the resulting array contains a float64:

>>> foo = np.empty(1)
>>> foo
array([ 0.])
>>> type(foo[0])
<type 'numpy.float64'>

Why doesn't it just return array([])?

• Ghilas BELHADJ: you specified size 1
• hpaulj: np.empty() is not the same as a list []. It requires a parameter which you gave it. It sounds like you want a 0 size array. But why? That's rarely useful. See stackoverflow.com/questions/44872454/…
• G Warner: @GhilasBELHADJ The first input of np.empty() is shape. Shouldn't the code I used just generate an empty array of dimensions 1? By my understanding, if I said np.empty(2) I'd expect to get array([[]]. How would I get that?
• BrenBarn: @GWarner. That would be an empty array of dimension 2, not 1. Also, shape is not the same as dimension. Arrays [1], [1, 2], and [1, 2, 3] are all one-dimensional, but they have different shapes.
• hpaulj: empty is probably not the best name for this function, given that an empty list has a very different meaning. np.zeros or np.ones might be a better starting point.

Answer 1

You didn't read the documentation, which says:

Return a new array of given shape and type, without initializing entries.

empty has nothing to do with creating an array that is "empty" in the sense of having no elements. It just means the array doesn't have its values initialized (i.e., they are unpredictable and depend on whatever happens to be in the memory allocated for the array).

• G Warner: OK thanks, I'm trying to make an empty matrix that I can add dimensions to. Are traditional python arrays more suited to this than numpy?
• BrenBarn: @GWarner: Probably, although the best way depends on exactly what you're doing. Adding dimensions (or elements) to numpy arrays is an inefficient operation. Numpy arrays achieve their speed by using a fixed memory layout, so the equivalent of something like list.append requires creating a whole new array. Because of this, numpy arrays are not well suited for tasks that require arrays of dynamically changing size.

Answer 2

foo = np.empty(0) 

is what you need instead

you specified its size as 1, you need to set it as 0 instead like they told you in the comments

Answer 3

If you want an size=0 array of dimension N, you can use np.empty((0,)*N):

>>> np.empty((0,)*1)
array([], dtype=float64)
>>> np.empty((0,)*2)
array([], shape=(0, 0), dtype=float64)

Of course, if the array has .size == 0, then empty is the same as zeros