Why do the rows and columns seem to be reversed in to_sparse_tensor? Thanks a lot.

[storyandwine]

In to_sparse_tensor row = edge_index[1], col = edge_index[0]

pytorch_geometric/torch_geometric/transforms/to_sparse_tensor.py

Lines 62 to 66 in e8b6def

	store.adj_t = SparseTensor(
	row=store.edge_index[1], col=store.edge_index[0],
	value=None if self.attr is None or self.attr not in store else
	store[self.attr], sparse_sizes=store.size()[::-1],
	is_sorted=True, trust_data=True)

While in other functions like dropout_adj

pytorch_geometric/torch_geometric/utils/dropout.py

Line 47 in 2db5eff

row, col = edge_index

row = edge_index[0] ,col = edge_index[1]

[rusty1s]

Yes, this is due to ToSparseTensor saving the transposed adjacency matrix adj_t (which accelerates message passing). The transposed adjacency matrix is given by simply swapping row and col.