Deep Graph Convolutional Neural Network (DGCNN)

[oxfordblues]

Hi Emmanouil,

This is an exciting repository, which seems to be well written, and lightweight, without heavy dependencies!

I have taken a look at the code, and am wondering if it is possible to use this repository to predict the correct label for a previously unseen graph? By a previously unseen graph we mean one that was not used for training and validating of the model. An example for such a model can i.a. be found here: https://stellargraph.readthedocs.io/en/stable/demos/graph-classification/dgcnn-graph-classification.html

[maniospas]

Hi Eirik,

thanks a lot for the interest and suggestion!

I have not yet gotten around to implementing the sorting layer mentioned in the approach you link to, but plan to do so in the upcoming months. Before that point, I will push a version that supports earlier DGCNN approaches without the sorting layer but can still classify unseen graphs.

[oxfordblues]

Thank you for the answer, this is really interesting. There are very few JVM libraries that support GNN, and it will be exciting to see if this can work on unseen graphs!

[maniospas]

Forgot to timely update this issue.
Graph classification is now supported, after the discussion in #2 .
You can find the respective tutorial here: https://github.com/MKLab-ITI/JGNN/blob/main/tutorials/GraphClassification.md

(Feel free to reopen this issue if things are not working well in specific applications.)

[oxfordblues]

Thank you, great work!