spd_learn.functional.gbw_chart#

spd_learn.functional.gbw_chart(x: Tensor, metric: Tensor, power=1.0) → Tensor[source]#

Map SPD matrices into power-deformed BW coordinates.

Parameters:

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

  • metric (torch.Tensor) – SPD metric of shape (…, n, n), broadcastable with x.

  • power (float or torch.Tensor, default=1.0) – Positive scalar deformation exponent.

Returns:

Real SPD matrices with broadcast leading axes.

Return type:

torch.Tensor

Notes

Computes M^(-1/2) X^power M^(-1/2). The common metric must broadcast against the input; host spectral floors apply.