Flattening without having to worry about current depth

[Superharz]

Allow ak.flatten(array, axis= 0). As of awkward 0.4.5 I get:

ValueError: axis=0 not allowed for flatten

This would make it possible to pass already flat arrays (it would just return the array as is was) and you wouldn't have to worry about the depth of array.

[jpivarski]

I couldn't exactly reproduce this, even going back to 0.4.5. If you pass axis=0 to ak.flatten, the case is handled in Python code and you don't get this error message that comes from C++. Tinkering with it, I did find an unhandled corner-case: if you pass the negative axis that is equivalent to axis=0 (such as axis=-2 for a singly jagged array or axis=-1 for a flat array), it missed the check for axis == 0 and went into C++. As far as I can see, that's the only way to encounter this error message.

(The C++ code was written under the assumption that axis == 0 would never be allowed, but it made sense to add that case, though its behavior is entirely different from any other axis value. Since axis == 1 flattens a possibly jagged two-dimensional array into a one-dimensional array, axis == 0 does not change the dimension of any array. However, all flattening operations remove None values, so axis == 0 only removes None from the top level. This is the only behavior that's consistent with the pattern set by all other axis values.)

The only way to express "flatten if not already flat" without exceptions is axis=None:

>>> ak.flatten(ak.Array([[1, 2, 3], [], [4, 5]]), axis=None)
<Array [1, 2, 3, 4, 5] type='5 * int64'>
>>> ak.flatten(ak.Array([1, 2, 3, 4, 5]), axis=None)

Be sure it's what you want, though. Unlike a single-axis flatten, it flattens all levels of depth:

>>> ak.flatten(ak.Array([[[1, 2], [], [3]], [], [[4, 5]]]), axis=None)
<Array [1, 2, 3, 4, 5] type='5 * int64'>

and even flattens records:

>>> ak.flatten(ak.Array([{"x": 1.1, "y": [1]}, {"x": 2.2, "y": [1, 2]}]), axis=None)
<Array [1.1, 2.2, 1, 1, 2] type='5 * float64'>

[jpivarski]

Oh, I also meant to say that you can also just

>>> ak.Array([[1, 2, 3], [], [4, 5]]).ndim
2
>>> ak.Array([1, 2, 3, 4, 5]).ndim
1

if you don't want something quite as drastic as ak.flatten with axis=None.

[Superharz]

Thank you for the explanation.
You are right, the error arrived with axis=-1 and an already flat array.

I have to admit that I never actually tried axis=0 and only assumed that it wouldn't work with it because it didn't work with axis=-1 on a flat array.