I have a numpy.array, called grid, with shape:

grid.shape = [N, M_1, M_2, ..., M_N]

The values of N, M_1, M_2, ..., M_N are known only after initialization.

For this example, let's say N=3 and M_1 = 20, M_2 = 17, M_3 = 9:

grid = np.arange(3*20*17*9).reshape(3, 20, 17, 9)

I am trying to loop over this array, as follows:

for indices, val in np.ndenumerate(grid[0]):print indices_some_func_with_N_arguments(*grid[:, indices])

At the first iteration, indices = (0, 0, 0) and:

grid[:, indices] # array with shape 3,3,17,9

whereas I want it to be an array of three elements only, much like:

grid[:, indices[0], indices[1], indices[2]] # array([   0, 3060, 6120])

which however I cannot implement like in the above line, because I don't know a-priori what is the length of indices.

I am using python 2.7, but a version-agnostic implementation is welcome :-)

I think you want something like this:

In [134]: x=np.arange(24).reshape(4,3,2)In [135]: x
Out[135]: 
array([[[ 0,  1],[ 2,  3],[ 4,  5]],[[ 6,  7],[ 8,  9],[10, 11]],[[12, 13],[14, 15],[16, 17]],[[18, 19],[20, 21],[22, 23]]])In [136]: for i,j in np.ndindex(x[0].shape):...:     print(i,j,x[:,i,j])...:     
(0, 0, array([ 0,  6, 12, 18]))
(0, 1, array([ 1,  7, 13, 19]))
(1, 0, array([ 2,  8, 14, 20]))
(1, 1, array([ 3,  9, 15, 21]))
(2, 0, array([ 4, 10, 16, 22]))
(2, 1, array([ 5, 11, 17, 23]))

where the 1st line is:

In [142]: x[:,0,0]
Out[142]: array([ 0,  6, 12, 18])

Unpacking the index tuple as i,j and using that in x[:,i,j] is the simplest way of doing this index. But to generalize it to other numbers of dimensions I'll have to play with tuples a bit. x[i,j] is the same as x[(i,j)].

In [147]: for ind in np.ndindex(x.shape[1:]):...:     print(ind,x[(slice(None),)+ind])...:     
((0, 0), array([ 0,  6, 12, 18]))
((0, 1), array([ 1,  7, 13, 19]))
...

with enumerate:

for ind,val in np.ndenumerate(x[0]):print(ind,x[(slice(None),)+ind])