spd_learn.functional.bwcm_mean#

spd_learn.functional.bwcm_mean(X: Tensor, weights: Tensor | None = None, *, theta: float = 0.5, dim: int = 0, keepdim: bool = False) → Tensor#

Compute closed-form Fréchet mean in Power-Cholesky coordinates.

Parameters:

  • X (torch.Tensor) – SPD matrices of shape (…, n, n) with at least 3 dims.

  • weights (torch.Tensor, optional) – Weights along dim, normalized to sum to 1.

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

  • dim (int, default=0) – Batch dimension along which mean is computed.

  • keepdim (bool, default=False) – Whether to retain dim.

Returns:

Weighted Fréchet mean SPD matrix.

Return type:

torch.Tensor