nvidia-china-sae/WholeGraph

Prerequisites

Download the source codes:

git clone https://gitlab-master.nvidia.com/xiaonans/wholegraph_github.git

Build docker from Dockerfile.pytorch and run the container:
```
cd wholegraph_github
sh docker.sh
```

Download the ogbn-papers100M dataset and convert it to binary format:

mkdir dataset
cd build
python3 ../examples/tools/ogb_data_convert.py -d ogbn-papers100M -r ../dataset

Make sure you are under the build dir.

For the score of "gnn_benchmark", run the following command.

sh run_benchmark.sh

For the score of "gnn_addfeature", run the following command.

sh run_addfeature.sh

Running summaries can be got at result/papers100m.

Method	Test Accuracy	Valid Accuracy	Parameters	Hardware
gnn_benchmark	0.6693 ± 0.0010	0.7111 ± 0.0002	713,754	7*A100(40GB)
gnn_addfeature	0.6736 ± 0.0010	0.7172 ± 0.0005	883,378	16*V100(32GB)