4

Given 3D arr, I would like to fill all dioganal element equal to 1.

np.random.seed(0)
arr=np.random.rand(3,4,4)

expected outpud

1,0.71519,0.60276,0.54488
0.42365,1,0.43759,0.89177
0.96366,0.38344,1,0.52889
0.56804,0.92560,0.07104,1



1,0.83262,0.77816,0.87001
0.97862,1,0.46148,0.78053
0.11827,0.63992,1,0.94467
0.52185,0.41466,0.26456,1



1,0.56843,0.01879,0.61764
0.61210,1,0.94375,0.68182
0.35951,0.43703,1,0.06023
0.66677,0.67064,0.21038,1

Assign the fill_diagonal as below

arr=np.fill_diagonal(arr, 1)

return an error

ValueError: All dimensions of input must be of equal length

May I know how to properly fill diagonal equal to 1 for a 3d array

What being tried so far

arr[:,:,0] = np.diag((1,1))

ValueError: could not broadcast input array from shape (2,2) into shape (3,4)

What to avoid

Using for-loop with the fill_diagonal

Improve this question
1
  • With np.einsum: np.einsum('ijj->ij', arr)[:] = 1.0 Commented Mar 11, 2022 at 9:05

3 Answers 3

Reset to default
3

try this:

r = np.arange(4)
arr[:, r, r] = 1

Example:

arr = np.arange(3*4*4).reshape(3,4,4)
r = np.arange(4)
arr[:, r, r] = 1

output:

array([[[ 1,  1,  2,  3],
        [ 4,  1,  6,  7],
        [ 8,  9,  1, 11],
        [12, 13, 14,  1]],

       [[ 1, 17, 18, 19],
        [20,  1, 22, 23],
        [24, 25,  1, 27],
        [28, 29, 30,  1]],

       [[ 1, 33, 34, 35],
        [36,  1, 38, 39],
        [40, 41,  1, 43],
        [44, 45, 46,  1]]])
Improve this answer
Sign up to request clarification or add additional context in comments.

1 Comment

very neat trick. :)
0

This is the best way

import numpy as np

np.random.seed(0)
arr=np.random.rand(3,4,4)

d1, d2, d3 = arr.shape

for i in range(d1):
    np.fill_diagonal(arr[i,:,:], 1)
Improve this answer

2 Comments

thanks for the effort. while it is obvious to use for loop, but I am looking for approach that does not use for-loop
you have to due to the limitation of fill_diagonal
0

You can use np.diag_indices to generate indices for a diagonal of a 2D subarray and then use indexing and a view in which you assign values:

import numpy

rng = np.random.default_rng(0)
arr = rng.random((3,4,4))

diag = np.diag_indices(4, ndim=2)
for d1 in range(arr.shape[0]):
    arr_view = arr[d1, :]
    arr_view[diag] = 1

If all your dimensions are equally-sized (shape as (n, n, n)), you can directly use the np.fill_diagonal function that you tried before.

EDIT:

Without using a for-loop, you can use

rng = np.random.default_rng(0)
arr = rng.random((3,4,4))

# Build a custom indexing with proper broadcasting 
diag = np.arange(arr.shape[0])[:, None], *np.diag_indices(arr.shape[1], ndim=2)

arr[diag] = 1

This works because the np.diag_indices(n, ndim=m) function only gives you a m-tuple of (n)-shaped ndarrays. Consequently, you can just augment the return value with a (k)-shaped ndarray if you have a (k, n, n) array. For Numpy to be able to broadcast the three arrays to the correct shape, you then only need to add a new axis (via the None) to the first ndarray.

Improve this answer

1 Comment

thanks for the effort. while it is obvious to use for loop, but I am looking for approach that does not use for-loop

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.