spd_learn.functional.off_diag

spd_learn.functional.off_diag(X: Tensor) → Tensor

Extract the hollow off-diagonal part of a matrix (zeroing the diagonal).

Given a square matrix \(X\), this returns:

\[\text{off\_diag}(X) = X - \text{diag}(\text{diag}(X))\]

preserving the shape and off-diagonal entries while setting all diagonal elements to zero.

Parameters:

X (torch.Tensor) – Input tensor of shape (…, n, n).

Returns:

Hollow matrix with zero diagonal, shape (…, n, n).

Return type:

torch.Tensor

Examples

>>> import torch
>>> from spd_learn.functional import off_diag
>>> X = torch.ones(3, 3)
>>> off_diag(X)
tensor([[0., 1., 1.],
        [1., 0., 1.],
        [1., 1., 0.]])