This repository is the code supplied with our paper submission for the 21st Workshop on Mining and Learning with Graphs (MLG) at ECML-PKDD 2024.
The layer is intended to be submitted to Pytorch Geometric library (PyG) seperatly after acceptence.
Install all required packages using the requirements.yml as:
conda env create -f environment.yml
The repository does not include all the Datasets as they are to large. The user is expected to download them from https://chrsmrrs.github.io/datasets/. They must then be converted to the pytorch tensor format, as shown as an example in Datasets/PROTEINS/PROTEINS_full/convert-protein.py.
To run an experiment, make sure to run the code on a server with Slurm available. Then simply pick an experiment (data set) to rerun. For example, PROTEIN:
run-sbatch.sh batch-protein.sh
Although we aim to release the layer on PyG, in the mean time the layer can be used by simply importing the cluster_pool.py file into your code, which contains the ClusterPooling class. This class can be used in the same way as many other pooling layers.
The other dir contains some temporary code used to convert certain formats to create some needed information for plotting graphs etc. It is included for the advanced reader, as it is undocumented.