why do we need adj.t()?

[liangbright]

I am curious to know why we need to input "adj.t()" to a layer:
"https://pytorch-geometric.readthedocs.io/en/latest/notes/sparse_tensor.html
out2 = conv(x, adj.t())"

It seems to be easy for the user to input adj, and the conv layer can transpose it if necessary

[Padarn]

I had a similar confusion, but the transposed matrix is the more useful one to store. You can take a look at an earlier discussion where I discussed his with @rusty1s #4248 (comment).

As for he API taking adj.. I guess I'm not qualified to say, but it seems better to generally encourage the user to keep adj_t around rather than adj given the above reasoning.