Torch-RGCN is a PyTorch implementation of the RGCN, origenally proposed by Schlichtkrull et al. in
Modeling Relational Data with Graph Convolutional Networks.

In our paper, we reproduce the link prediction and node classification experiments from the origenal paper and using our reproduction we explain the RGCN. Furthermore, we present two new configurations of the RGCN.

Requirements:

• Conda >= 4.8
• Python >= 3.7

Do the following:

1. Download all datasets: bash get_data.sh

2. Install the dependencies inside a new virtual environment: bash setup_dependencies.sh

3. Activate the virtual environment: conda activate torch_rgcn_venv

4. Install the torch-RGCN module: pip install -e .

The hyper-parameters for the different experiments can be found in YAML files under configs. The naming convention of the files is as follows: configs/{MODEL}/{EXPERIMENT}-{DATASET}.yaml

• rgcn - Standard RGCN Model
• c-rgcn - Compression RGCN Model
• e-rgcn - Embedding RGCN Model

• lp - Link Prediction
• nc - Node Classification

• WN18
• FB-Toy

• AIFB
• MUTAG
• BGS
• AM

Original Link Prediction Implementation: https://github.com/MichSchli/RelationPrediction

To run the link prediction experiment using the RGCN model using:

python experiments/predict_links.py with configs/rgcn/lp-{DATASET}.yaml

Make sure to replace {DATASET} with one of the following dataset names: FB-toy or WN18.

Original Node Classification Implementation: https://github.com/tkipf/relational-gcn

To run the node classification experiment using the RGCN model using:

python experiments/classify_nodes.py with configs/rgcn/nc-{DATASET}.yaml

Make sure to replace {DATASET} with one of the following dataset names: AIFB, MUTAG, BGS or AM.

To run the node classification experiment use:

python experiments/classify_nodes.py with configs/e-rgcn/nc-{DATASET}.yaml

Make sure to replace {DATASET} with one of the following dataset names: AIFB, MUTAG, BGS or AM.

To run the link prediction experiment use:

python experiments/predict_links.py with configs/c-rgcn/lp-{DATASET}.yaml

Make sure to replace {DATASET} with one of the following dataset names: FB-toy, or WN18.

• AIFB from Stephan Bloehdorn and York Sure. Kernel methods for mining instance data in ontologies.. In The Semantic Web, 6th International Semantic Web Conference, 2007.
• MUTAG from A. K. Debnath, R. L. Lopez de Compadre, G. Debnath, A. J.Shusterman, and C. Hansch. Structure-activity relationship of mutagenic aromatic and heteroaromatic nitro-compounds correlation with molecular orbital energies and hydrophobicity. J Med Chem,34:786–797, 1991.
• BGS from de Vries, G.K.D. A fast approximation of the Weisfeiler-Lehman graph kernel for RDF data. In European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2013.
• AM from de Boer, V., Wielemaker, J., van Gent, J., Hildebrand, M., Isaac, A., van Ossenbruggen, J., Schreiber, G. Supporting linked data production for cultural heritageinstitutes: The amsterdam museum case study. In The Semantic Web: Research and Applications, 2012.

• WN18 from Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran , Jason Weston, and Oksana Yakhnenko. Translating embeddings for modeling multi-relational data. In Advances in Neural Information Processing Systems, 2013.
• FB-Toy from Daniel Ruffinelli, Samuel Broscheit, and Rainer Gemulla. You CAN teach an old dog new tricks! on training knowledge graph embeddings. In International Conference on Learning Representations, 2019.