Permit sparse matrices in Laplacian filter

[klane]

The line to enforce symmetry does not support sparse matrices. A potential fix is either to convert to dense and back to sparse or raise an exception if the user selects the Laplacian filter with a sparse adjacency matrix.

torchTS/torchts/utils/graph.py

Line 34 in 600b8c3

adj_mx = np.maximum.reduce([adj_mx, adj_mx.T])

[akashshah59]

@klane There is an unusual L.astype('float32') present at the end of the scaled laplacian.
I have often faced errors when my adjacency matrix type is float64 and I attribute it to this enforcement. Do you see a reason to keep this typecast?

[klane]

@klane There is an unusual L.astype('float32') present at the end of the scaled laplacian.
I have often faced errors when my adjacency matrix type is float64 and I attribute it to this enforcement. Do you see a reason to keep this typecast?

Good catch @akashshah59. This should probably be removed.