spd_learn.functional.trace_normalization

spd_learn.functional.trace_normalization(covariances: Tensor, epsilon: float | None = None) → Tensor

Performs trace normalization on a batch of covariance matrices.

This function normalizes a batch of covariance matrices by their trace. It also adds a small identity matrix for numerical stability.

Parameters:

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

• epsilon (float, optional) – A small value to add to the diagonal of the covariance matrices for numerical stability. If None, uses the value from the unified numerical configuration. Default: None.

Returns:

The trace-normalized covariance matrices.

Return type:

Tensor