How to make np.loadtxt return multidimensional arrays even the file has only one dimensional?

Question

I need to get the last four column data of a ndarray, most of time code arr[:, -4:] is ok, but if the array just has one dimension, this will throw IndexError: too many indices.

My data is get with arr = np.loadtxt('test.txt'), so if test.txt has more than one line, like

0 1 2 3 4
0 10 20 30 40

everything is ok, but if test.txt has just one line, like

0 1 2 3 4

this will return array([ 0, 1, 2, 3, 4]), then arr[:, -4:] will throw exception, because it should be arr[-4:], so how to make loadtxt return array([[ 0, 1, 2, 3, 4]])?

Answer 1

Use the Ellipsis (...) instead of an empty slice (:) for your first index:

>>> a = np.arange(30).reshape(3, 10)
>>> a[:, -4:]
array([[ 6,  7,  8,  9],
       [16, 17, 18, 19],
       [26, 27, 28, 29]])
>>> a[..., -4:]  # works the same for the 2D case
array([[ 6,  7,  8,  9],
       [16, 17, 18, 19],
       [26, 27, 28, 29]])

>>> a = np.arange(10)
>>> a[..., -4:]  # works also in the 1D case
array([6, 7, 8, 9])
>>> a[:, -4:]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
IndexError: too many indices for array

EDIT If you want the return to be 2D also for the single row case, then this should do the trick:

>>> np.atleast_2d(a[..., -4:])
array([[6, 7, 8, 9]])

Answer 2

Just found it here.

You can ask it to have at least 2 dimensions with:

arr = np.loadtxt('test.txt', ndmin=2)