spd_learn.functional.normalize_spd_adjacency

spd_learn.functional.normalize_spd_adjacency(adjacency: Tensor) → Tensor

Normalize nonnegative adjacency weights with unit self-loops.

Parameters:

adjacency (torch.Tensor) – Finite nonnegative adjacency matrices of shape (…, nodes, nodes).

Returns:

Row-stochastic propagation matrices with the input shape, dtype and device.

Return type:

torch.Tensor

Notes

Existing diagonal weights are retained before adding the self-loops.