This MR fixes the problems introduced by the torch sparse transformation during the GNN filtering stage
Fixing #119