Intended behaviour of np.roll?

[johannes-graeter]

Hi Simon,
I came across a line of code that made me think:

UnFlow/src/e2eflow/core/input.py

Lines 178 to 179 in ddb4bd3

	shift = shift % len(filenames_extended)
	filenames_extended = list(np.roll(filenames_extended, shift))

If I understand correctly np.roll with axis=None will first flatten the array and then roll.
Is that the intention? In order to shift the training examples to the desired training position it seems to me that np.roll(..., axis=0) would be the desired behaviour (in order to roll complete rows of the array).

Thanks :)
Johannes