Function that guarantees a minimum number of dimensions (ndim) for a numpy.ndarray

Question

There are many situations where slicing operations in 2D arrays produce a 1D array as output, example:

a = np.random.random((3,3))
# array([[ 0.4986962 ,  0.65777899,  0.16798398],
#        [ 0.02767355,  0.49157946,  0.03178513],
#        [ 0.60765513,  0.65030948,  0.14786596]])
a[0,:]
# array([ 0.4986962 ,  0.65777899,  0.16798398])

There are workarounds like:

a[0:1,:]
# or
a[0,:][np.newaxis,:]
# array([[ 0.4986962 ,  0.65777899,  0.16798398]])

Is there any numpy built in function that transforms an input array to a given number of dimensions? Like:

np.minndim(a, ndim=2)

Answer 1

There is np.array(array, copy=False, subok=True, ndmin=N). np.atleast_1d, etc. actually use the reshape method, probably to better support some weird subclasses such as matrix.

For most slicing operations in 2-D you could actually use the matrix class, though I would strongly suggest limiting the usage to those few points in code where its features are heavly used.

Answer 2

You can use np.atleast_1d, np.atleast_2d and np.atleast_3d. Unfortunately I don't think there's currently an N-dimensional version.