How to reshape square numpy matrix into "diamond" of diagonals structure most efficiently

Question

Say I have a square numpy.matrix like this

[['78' '17' '53' '28']
 ['22' '75' '31' '67']
 ['15' '94' '03' '80']
 ['04' '62' '16' '14']]

and for my purposes I need a "diamond structure" like this of all of the diagonals in an upwards direction (note that a list of lists is fine for my purpose)

 [       ['78'], 
      ['22' '17'],
    ['15' '75' '53'],
  ['04' '94' '31' '28'],
    ['62' '03'  '67'],
       ['16' '80'],
         ['14'],      ]

What is the best way to do so?

I can't decide if using some incrementing magic with position accessing is required or some built in way making use of numpy.diagonal is possible, which I haven't been able to employ successfully yet.

Answer 1

import numpy as np

a=np.array([[78, 17, 53, 28],
    [22, 75, 31, 67],
    [15, 94, 03, 80],
    [04, 62, 16, 14]])

[np.diag(a[-1:-a.shape[0]-1:-1,:], i).tolist() for i in range(-a.shape[0]+1,a.shape[0])]

outputs [[78], [22, 17], [15, 75, 53], [4, 94, 31, 28], [62, 3, 67], [16, 80], [14]] as requested

it works very fast for 1000x1000 matrix as well