Examples for Training and Inference on Link Prediction GNN models.

Homogeneous (CORA)

We use the CORA dataset as an example for sampling against a homogeneous dataset.

homogeneous_inference.py and homogeneous_training.py are example inference and training loops for the CORA dataset, the MNIST of graph models, and available via the PyG Planetoid dataset.

You can follow along with cora.ipynb to run an e2e GiGL pipeline on the CORA dataset. It will guide you through running each component: config_populator -> data_preprocessor -> trainer -> inferencer

Heterogeneous (DBLP)

We use use the DBLP dataset as an example for sampling against a heterogeneous dataset.

heterogeneous_inference.py and heterogeneous_training.py are example inference and training loops for the DBLP dataset. The DBLP dataset is avaialble at the PyG DBLP dataset.

You can follow along with dblp.ipynb to run an e2e GiGL pipeline on the DBLP dataset. It will guide you through running each component: config_populator -> data_preprocessor -> trainer -> inferencer