torchcell.models.equivariant_cell_graph_transformer.compute_smoothness

compute_smoothness(X: Tensor) → float

Compute smoothness of node features (oversmoothing diagnostic).

Lower values indicate oversmoothing (features collapsing toward mean). Higher values indicate feature diversity is preserved.

Parameters:

X – Node feature matrix [N, d]

Returns:

Frobenius norm of deviation from mean features