vijaydwivedi75/gnn-lspe

Source code for GNN-LSPE (Graph Neural Networks with Learnable Structural and Positional Representations), ICLR 2022

Graph Neural Networks with
Learnable Structural and Positional Representations

Source code for the paper "Graph Neural Networks with Learnable Structural and Positional Representations" by Vijay Prakash Dwivedi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio and Xavier Bresson, at the Tenth International Conference on Learning Representations (ICLR) 2022.

We propose a novel GNN architecture in which the structural and positional representations are decoupled, and are learnt separately to learn these two essential properties. The architecture, named MPGNNs-LSPE (MPGNNs with Learnable Structural and Positional Encodings), is generic that it can be applied to any GNN model of interest which fits into the popular 'message-passing framework', including Transformers.

1. Repo installation

Follow these instructions to install the repo and setup the environment.

2. Download datasets

Proceed as follows to download the benchmark datasets.

3. Reproducibility

Use this page to run the codes and reproduce the published results.

4. Reference

📃 Paper on arXiv
🎥 Video by @vijaydwivedi75 on YouTube
🎥 Video by @xbresson on YouTube

@inproceedings{dwivedi2022graph,
  title={Graph Neural Networks with Learnable Structural and Positional Representations},
  author={Vijay Prakash Dwivedi and Anh Tuan Luu and Thomas Laurent and Yoshua Bengio and Xavier Bresson},
  booktitle={International Conference on Learning Representations},
  year={2022},
  url={https://openreview.net/forum?id=wTTjnvGphYj}
}

gnn-lspe's Issues

Faced Issue while reproducing the results

I got the following error after following the step 1 and step 2 for the installation.
ImportError: dlopen(/Users/abhishekgadekar/opt/anaconda3/envs/gnn_lspe/lib/python3.7/lib-dynload/_ctypes.cpython-37m-darwin.so, 0x0002): Library not loaded: @rpath/libffi.6.dylib
This error occured when I ran the following command:

Run the main file (at the root of the project)

python main_ZINC_graph_regression.py --config 'configs/GatedGCN_ZINC_LSPE.json' # for CPU
python main_ZINC_graph_regression.py --gpu_id 0 --config 'configs/GatedGCN_ZINC_LSPE.json' # for GPU

It would be really helpful if you could provide a fix for this issue. I also checked by installing libffi by using pip, but that also couldn't solve the issue.

Migrate to partial node classification tasks

Dear Author,

Thanks so much for the code!

I don't know if LSPE can be used for sub-graph node classification (prediction on part of the nodes). I checked another repo you wrote (https://github.com/graphdeeplearning/benchmarking-gnns) and it seems I have to generate the features of all the nodes/edges first.