NumPy k-th diagonal indices

The indices of the k'th diagonal of a can be computed with

def kth_diag_indices(a, k):
    rowidx, colidx = np.diag_indices_from(a)
    colidx = colidx.copy()  # rowidx and colidx share the same buffer

    if k > 0:
        colidx += k
    else:
        rowidx -= k
    k = np.abs(k)

    return rowidx[:-k], colidx[:-k]

Demo:

>>> 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, 24]])
>>> a[kth_diag_indices(a, 1)]
array([ 1,  7, 13, 19])
>>> a[kth_diag_indices(a, 2)]
array([ 2,  8, 14])
>>> a[kth_diag_indices(a, -1)]
array([ 5, 11, 17, 23])

Here's a way:

  1. Create index value arrays.
  2. Get the daigonal index values you want.
  3. Thats it! :)

Like this:

>>> import numpy as np
>>> rows, cols = np.indices((3,3))
>>> row_vals = np.diag(rows, k=-1)
>>> col_vals = np.diag(cols, k=-1)
>>> z = np.zeros((3,3))
>>> z[row_vals, col_vals]=1
>>> z
array([[ 0.,  0.,  0.],
       [ 1.,  0.,  0.],
       [ 0.,  1.,  0.]])

A bit late, but this version also works for k = 0 (and does not alter the arrays, so does not need to make a copy).

def kth_diag_indices(a, k):
    rows, cols = np.diag_indices_from(a)
    if k < 0:
        return rows[-k:], cols[:k]
    elif k > 0:
        return rows[:-k], cols[k:]
    else:
        return rows, cols