Feather-Graph with attributed graphs

[sarmaier]

Hi,

First off, thank you maintaining this library, its awesome! Also, I apologize if this question is redundant. I found a few similar closed questions, but was still a bit confused on how to go about solving my question

Background: I have a few thousand small, attributed graphs. For each node, I have a set of 6 continuous feature variables. I would like to embed these graphs, considering the node features for use in downstream ML projects. From my understanding, the Feather-Graph class as implemented in karateclub only considers topological features (it would ignore the 6 continuous feature variables I have assigned to each node).

Question: I would like to embed my graphs using either the FeatherGraph class in karateclub or the FeatherG class from FEATHER repo (https://github.com/benedekrozemberczki/FEATHER) . Additionally I would like the graph-level embedding to account for node attributes. Would it be reasonable for me to pass the output of the FEATHER class (node embeddings) to a pooling function to get a graph-level representation that accounts for the 6 continuous node features.

Again, thank you so much for all your work! I've loved using the library so far.