Muhan Zhang

Hi @aaron951117 , could you provide more details so that I can reproduce the bug? Such as your dataset, command, and torch version? Thanks.

The latent_dim does not matter here. pytorch_DGCNN supports list of latent_dim as input. See [here](https://github.com/muhanzhang/pytorch_DGCNN/blob/50f504131ca66382f3f078ae98e59aa3ae35b795/DGCNN_embedding.py#L22). Also, the importing of main.py is before setting custom cmd_args in Main.py. So setting custom...

Hi, I am not familiar with Colab, but it seems to support reading files from Google drive. You may save the c++ files in the drive, and compile them on...

In your case, both subgraph and embedding features are not applicable to new nodes (the enclosing subgraph is empty with only the target nodes; while embeddings also do not generalize...

No. SEAL is an inductive method. After training SEAL on a given network, you can apply it to other networks, since the learned graph structure features that work on one...

Could you try to turn off the parallelization by appending "--no-parallel" to your command?

Thanks for your interest in this work. Unfortunately I don't have saved model files now. If Facebook is taking very long time, you may try to reduce the number training...

Hi! Yes. SEAL can be applied to directed graphs. When working on directed graphs, you should include both incoming and outgoing neighbors when extracting enclosing subgraphs. I haven't compared with...

After training, in this [line](https://github.com/muhanzhang/pytorch_DGCNN/blob/50f504131ca66382f3f078ae98e59aa3ae35b795/main.py#L234), you can add a `torch.save(classifier.state_dict(), your_model_save_path)` to save the trained model. Then, next time, you can skip the training and load the saved model by...

Hi! See #36.