spd_learn.functional.pcm_coords#

spd_learn.functional.pcm_coords(X: Tensor, theta: float = 0.5) → Tensor[source]#

Compute global deformed coordinates under Power-Cholesky geometry.

\[\tilde{\phi}_\theta(X) = \lfloor L \rfloor + \text{Diag}\left(\frac{1}{\theta} d^\theta\right)\]
Parameters:

  • X (torch.Tensor) – SPD matrices of shape (…, n, n).

  • theta (float, default=0.5) – Power deformation parameter (\(\theta \neq 0\)).

Returns:

Deformed coordinates of shape (…, n, n).

Return type:

torch.Tensor