MDGNN for MAG240M-LSC 2022

The code is about MDGNN solution for MAG240M-LSC 2022

Getting Started

Requirements

• Python==3.7
• PyTorch==1.9.0+cu111
• dgl-cuda111==0.7.0
• ogb==1.3.4
• sklearn==1.0.2

Running this code

Preprocess graph and generate features

bash run_preprocess.sh

After this step, you will get .npy feature files in MAG_FEAT_PATH.

Get embeddings of other methods:

We extract the metapath2vec embeddings of arXiv papers, which are provided by R-UNIMP. We also try to reproduce the code of Deepmind and gain the arXiv paper embedding. The arXiv paper embeddings of the above two methods can be obtained from the following link (password 3oot):

https://pan.baidu.com/s/1EVlhs2jClCJHwbH-ht2vqw

Please download the x_m2v_64.npy and x_jax_153.npy. Then, put them into MAG_FEAT_PATH.

The pre-processing and generating features are time-consuming, and we also provide all features as downloadable options so you can choose the ones you need.

Train our model

We selected different features for different models, the corresponding relationships are listed in features/config_feats.py.

bash run_folds.sh

Run ensemble

bash run_ensemble.sh

This step contains 15 models (14 trained models + Deepmind embedding), for more details please refer to our technical report.