This repository is the DGL-implemention of GIPA (General Information Propagation Algorithm). The following figure illustrates the process of GIPA:

The protein dataset ogbn-proteins from Open Graph Benchmark.

To train the model:

source main.sh

Or:

python src/main.py --n-layers 6 --n-epochs 8000 --lr 0.01 --batch-size 8000 --use-label --if-save --preprocess --n-hop 1 --gpu 0 --eval-every 5 --seed 0

To use the pre-trained model to do inference: see the Jupyter notebook in /example/gipa_test.ipynb.

ogb==1.3.0
numpy==1.19.2
torch==1.7.1
dgl_cu110==0.5.3