Link Level Implicit Graph Neural Networks

This is the official implementation of Link Level Implicit Graph Neural Networks (LIGCN).

Requirements

• pytorch
• torch-geometric
• tqdm
• numpy
• sklearn

Synthetic experiments

To run the synthetic experiments

python train_synthetic.py

PCQM-Contact

The dataset is too big to be provided in this folder. Please first follow the instructions in Long Range Graph Benchmark and download the PCQM-Contact files. Then, put the raw folder (containing train.pt, valid.pt, test.pt) into ./data/pcqm-contact/raw.

To run the molecular property prediction experiments

python train_pcqm.py

Knowledge graph completion

To run the KGC experiments

python train_kg.py --dataset <dataset name>

For example,

python train_kg.py --dataset GraIL-BM_WN18RR_v1 --out_bias --cuda
python train_kg.py --dataset GraIL-BM_fb237_v1 --cuda
python train_kg.py --dataset GraIL-BM_nell_v1 --cuda