I think that this work cannot be applied to large-scale graphs for the reason that calculating the adj through your method needs eigen decomposition and to_dense() method needs large memory available.
eig_value, left_vector = scipy.linalg.eig(p_ppr.numpy(),left=True,right=False) p_dense = torch.sparse.FloatTensor(edge_index, p, torch.Size([num_nodes,num_nodes])).to_dense()
hosseinghorbanzadeh
commented
7 months ago
I think that this work cannot be applied to large-scale graphs for the reason that calculating the adj through your method needs eigen decomposition and to_dense() method needs large memory available.
eig_value, left_vector = scipy.linalg.eig(p_ppr.numpy(),left=True,right=False) p_dense = torch.sparse.FloatTensor(edge_index, p, torch.Size([num_nodes,num_nodes])).to_dense()