python - Numpy roll in several dimensions -

i need shift 3d array 3d vector of displacement algorithm. of i'm using (admitedly ugly) method :

shiftedarray = np.roll(np.roll(np.roll(arraytoshift, shift[0], axis=0)                                      , shift[1], axis=1),                              shift[2], axis=2)   

which works, means i'm calling 3 rolls ! (58% of algorithm time spent in these, according profiling)

from docs of numpy.roll:

parameters:
shift : int

axis : int, optional

no mention of array-like in parameter ... can't have multidimensional rolling ?

i thought call kind of function (sounds numpy thing do) :

np.roll(arraytoshift,3dshiftvector,axis=(0,1,2)) 

maybe flattened version of array reshaped ? how compute shift vector ? , shift same ?

i'm surprised find no easy solution this, thought pretty common thing (okay, not that common, ...)

so how --relatively-- efficiently shift ndarray n-dimensional vector ?

i think scipy.ndimage.interpolation.shift want, docs

shift : float or sequence, optional

the shift along axes. if float, shift same each axis. if sequence, shift should contain 1 value each axis.

which means can following,

from scipy.ndimage.interpolation import shift import numpy np  arraytoshift = np.reshape([i in range(27)],(3,3,3))  print('before shift') print(arraytoshift)  shiftvector = (1,2,3) shiftedarray = shift(arraytoshift,shift=shiftvector,mode='wrap')  print('after shift') print(shiftedarray) 

which yields,

before shift [[[ 0  1  2]   [ 3  4  5]   [ 6  7  8]]   [[ 9 10 11]   [12 13 14]   [15 16 17]]   [[18 19 20]   [21 22 23]   [24 25 26]]] after shift [[[16 17 16]   [13 14 13]   [10 11 10]]   [[ 7  8  7]   [ 4  5  4]   [ 1  2  1]]   [[16 17 16]   [13 14 13]   [10 11 10]]]