• Paper link: arXiv OpenReview
• Author's code repo: https://github.com/weihua916/powerful-gnns.

• MXNet 1.5+
• sklearn
• tqdm

bash pip install torch sklearn tqdm

An experiment on the GIN in default settings can be run with

DGLBACKEND=mxnet python main.py

An experiment on the GIN in customized settings can be run with

DGLBACKEND=mxnet python main.py [--device 0 | --disable-cuda] --dataset COLLAB \
               --graph_pooling_type max --neighbor_pooling_type sum

Run with following with the double SUM pooling way: (tested dataset: "MUTAG"(default), "COLLAB", "IMDBBINARY", "IMDBMULTI")

DGLBACKEND=mxnet python main.py --dataset MUTAG --device 0  \
                --graph_pooling_type sum --neighbor_pooling_type sum