Consider a numpy ndarray M of shape (say) (a,b,c) and a list of coordinates [(b0,c0),(b1,c1),...,(bn,cn)].
I want the following collection of vectors: [M[:,b0,c0],M[:,b1,c1],...,M[:,bn,cn]].
How can I achieve this with list comprehension?
EDIT: I need a solution that works for any number of dimensions, i.e. something that returns a list like the one above also in the case (say) M.shape = (a,b,c,d) and coordinates = [(b0,c0,d0),...,(bn,cn,dn)], and higher-dimensional cases as well.
You don't want to do this with a list comprehension. "Fancy indexing" can do the whole thing in one go. I'd recommend:
inds = [(b0,c0),(b1,c1),...,(bn,cn)]
#inds_array[0] = [b0, b1, b2, ...]
inds_array = np.moveaxis(np.array(inds), -1, 0)
M[np.index_exp[:] + tuple(inds_array)]
Demo:
>>> x, y, z = np.ogrid[:2,:4,:5]
>>> M = 100*x + 10*y + z
>>> M.shape
(2, 4, 5)
>>> inds = [(0, 0), (2, 1), (3, 4), (1, 2)]
>>> inds_array = np.moveaxis(np.array(inds), -1, 0); inds_array
array([[0, 2, 3, 1],
[0, 1, 4, 2]])
>>> M[np.index_exp[:] + tuple(inds_array)] # same as M[:, [0, 2, 3, 1], [0, 1, 4, 2]]
array([[ 0, 21, 34, 12],
[100, 121, 134, 112]])
slice(None), ..... I had to look up index_exp to see how it's different from s_, and without running an example I can't picture inds_array.If you want a list of these vectors you can simply use:
[M[:,bi,ci] for bi,ci in coordinates]where coordinates is of course your list:
coordinates = [(b0,c0),(b1,c1),...,(bn,cn)]
EDIT: in case you want multivariate slicing, you can call the __getitem__ method with a slice(None) and the remaining indices:
[M.[(slice(None),*coord)] for coord in coordinates]for python-3.5; or:
[M[(slice(None),)+coord] for coord in coordinates]for other python versions.
.__getitem__( directly. There only reason to do that is super, and this is not that