How to get the batch features of line graph

[StefanIsSmart]

As we all know, the pyg could convert a graph to a linegraph.
But finished converting, we cannot get the new batch features of the edges. The batch is still representative the original nodes rather than the nodes that converted from edges.

How to get the batch features of new graph?

[rusty1s]

What do you mean by batch features?

[rusty1s]

That is correct; most transforms in PyG are designed to operate on single graphs only. The current procedure would be to convert each graph separately via the LineGraph transform, and then batch the line-graphs together afterwards. That is what is happening if you use, e.g., Dataset.transform or Dataset.pre_transform. Would that be applicable in your use-case?

[StefanIsSmart]

Thank you very much and I will have a try and tell you the result!

So the transformation of a batch of node graph to line-graph will be correct right?
I ask this question because you said: most transforms in PyG are designed to operate on single graphs only.
I worry I converting a batch of node graph will have some Errors. :p

Thank you again for your reply!

[rusty1s]

Yes, in general it is not guaranteed that transforms work on a batch of graphs (i.e., batch vector is not maintained). This may not always be a problem, but for some transformations it is. For example, Center would center on the center of all nodes in the batch, not just the nodes inside each graph.

[StefanIsSmart]

As your answer in another discussion.
The batch info of ling_graph could get from :
batch[edge_index[0]]
Right?

[rusty1s]

Yes.