On Wed, Dec 3, 2014 at 4:02 PM, Stefan van der Walt <stefan@sun.ac.za> wrote:
Hi Catherine
On 2014-12-04 01:12:30, Moroney, Catherine M (398E) < Catherine.M.Moroney@jpl.nasa.gov> wrote:
I have an array "A" of shape (NX, NY, NZ), and then I have a second array "B" of shape (NX, NY) that ranges from 0 to NZ in value.
I want to create a third array "C" of shape (NX, NY) that holds the "B"-th slice for each (NX, NY)
Those two arrays can broadcast if you expand the dimensions of B:
A: (NX, NY, NZ) B: (NX, NY, 1)
Your result would be
B = B[..., np.newaxis] # now shape (NX, NY, 1) C = A[B]
For more information on this type of broadcasting manipulation, see
http://nbviewer.ipython.org/github/stefanv/teaching/blob/master/2014_assp_sp...
and
I don't think this would quite work... Even though it now has 3 dimensions (Nx, Ny, 1), B is still a single array, so when fancy indexing A with it, it will only be applied to the first axis, so the return will be of shape B.shape + A.shape[1:], that is (Nx, Ny, 1, Ny, Nz). What you need to have is three indexing arrays, one per dimension of A, that together broadcast to the desired shape of the output C. For this particular case, you could do: nx = np.arange(A.shape[0])[:, np.newaxis] ny = np.arange(A.shape[1]) C = A[nx, ny, B] To show that this works:
A = np.arange(2*3*4).reshape(2, 3, 4) A 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]]])
B = np.random.randint(4, size=(2, 3)) B array([[2, 1, 2], [2, 0, 3]]) nx = np.arange(A.shape[0])[:, None] ny = np.arange(A.shape[1]) A[nx, ny, B] array([[ 2, 5, 10], [14, 16, 23]])
Jaime -- (\__/) ( O.o) ( > <) Este es Conejo. Copia a Conejo en tu firma y ayúdale en sus planes de dominación mundial.