drumpt/Graph-Transformer-Networks

Graph Transformer Networks

• This repository is a modification of original repository for Graph Transformer Networks(NeurIPS'19).
• This repository is used for lab freshman and undergraduate students seminar in MLILAB.

Download required libraries

pip install -r requirements.txt

Download DBLP, ACM, IMDB dataset

• Download data.zip dataset from this link and unzip it to the uppermost directory of this repository.

• Make sure that the data are located like below.

Graph-Transformer-Network
├── main.py
├── model.py
├── gcn.py
├── utils.py
├── data_preprocessing.ipynb
├── ACM.mat
├── data
    ├── ACM
        ├── edges.pkl
        ├── labels.pkl
        └── node_features.pkl
    ├── DBLP
        ├── edges.pkl
        ├── labels.pkl
        └── node_features.pkl
    └── IMDB
        ├── edges.pkl
        ├── labels.pkl
        └── node_features.pkl    
├── imgs
    └── GTN.png
├── requirements.txt
└── README.md

TODOs

TODO 1 : Fill in models.py.

TODO 2 : Run main.py with DBLP and ACM dataset using the commands and report the training curves and macro F1 scores.

• DBLP

$ python main.py --dataset DBLP --num_layers 3

• ACM

 $ python main.py --dataset ACM --num_layers 2 --adaptive_lr true

TODO 3 : Modify main.py to check which meta-paths are important in ACM dataset.

• You might refer to data_preprocessing.ipynb to see which edge is which index.