spd_learn.functional.ledoit_wolf#

spd_learn.functional.ledoit_wolf(covariances: Tensor, shrinkage: Tensor, shrink_mat: Tensor, size: int) → Tensor[source]#

Applies Ledoit-Wolf shrinkage to a batch of covariance matrices.

Parameters:

  • covariances (Tensor) – A batch of covariance matrices with shape (…, n, n).

  • shrinkage (Tensor) – Unconstrained shrinkage parameters, which will be passed through a sigmoid function.

  • shrink_mat (Tensor) – The target “shrink” matrices, which are often identity matrices.

  • size (int) – The dimensionality of the covariance matrices.

Returns:

The shrunken covariance matrices.

Return type:

Tensor