Replace diagonals of a 2D array with python [duplicate]

You can use np.fill_diagonal(..) for that. Like the documentation says:

numpy.fill_diagonal(a, val, wrap=False)

Fill the main diagonal of the given array of any dimensionality.

For example:

np.fill_diagonal(A, 20)

We here thus broadcast 20 over the entire diagonal.

You can also fill the diagonal with different values, like:

np.fill_diagonal(A, [0,2,15,20])

For example:

>>> a = np.zeros((4,4))>>> np.fill_diagonal(a, [0,2,15,20])>>> aarray([[ 0.,  0.,  0.,  0.],       [ 0.,  2.,  0.,  0.],       [ 0.,  0., 15.,  0.],       [ 0.,  0.,  0., 20.]])

In case you want to change other diagonals, then it is a matter of mirroring the array. For example for the antidiagonal, we can use:

np.fill_diagonal(A[::-1], -20)

Then we thus get:

>>> A = np.zeros((4,4))>>> np.fill_diagonal(A[::-1], -20)>>> Aarray([[  0.,   0.,   0., -20.],       [  0.,   0., -20.,   0.],       [  0., -20.,   0.,   0.],       [-20.,   0.,   0.,   0.]])

If we do not take superdiagonals and subdiagonals into account, a n dimensional matrix, has n×(n-1) diagonals. We can assign this by mirroring one or more dimensions.

Checkout the numpy docs on indexing into multidimensional arrays.

A[np.arange(A.shape[0]), np.arange(A.shape[1])] = [0,2,15,20]

Note: @WillemVanOnsem's answer is the best answer for filling in main diagonal but this is the best general way for getting/setting any subset of elements in a multidimensional array!

For example to change the other diagonal:

A[-np.arange(1, A.shape[0] + 1), np.arange(A.shape[1])] = [0,2,15,20]