OpenHGNN

启智社区（中文版） | OpenHGNN [CIKM2022] | Space4HGNN [SIGIR2022] | Benchmark&Leaderboard | Slack Channel

This is an open-source toolkit for Heterogeneous Graph Neural Network based on DGL [Deep Graph Library] and PyTorch. We integrate SOTA models of heterogeneous graph.

News

Key Features

• Easy-to-Use: OpenHGNN provides easy-to-use interfaces for running experiments with the given models and dataset. Besides, we also integrate optuna to get hyperparameter optimization.
• Extensibility: User can define customized task/model/dataset to apply new models to new scenarios.
• Efficiency: The backend dgl provides efficient APIs.

Get Started

Requirements and Installation

• Python >= 3.6

• PyTorch >= 2.3.0

• DGL >= 2.2.1

• CPU or NVIDIA GPU, Linux, Python3

1. Python environment (Optional): We recommend using Conda package manager

conda create -n openhgnn python=3.6
source activate openhgnn

2. Install Pytorch: Follow their tutorial to run the proper command according to your OS and CUDA version. For example:

pip install torch torchvision torchaudio

3. Install DGL: Follow their tutorial to run the proper command according to your OS and CUDA version. For example:

pip install dgl -f https://data.dgl.ai/wheels/repo.html

4. Install openhgnn:

• install from pypi

  pip install openhgnn

• install from source

  git clone https://github.com/BUPT-GAMMA/OpenHGNN
  # If you encounter a network error, try git clone from openi as following.
  # git clone https://git.openi.org.cn/GAMMALab/OpenHGNN.git
  cd OpenHGNN
  pip install .

5. Install gdbi(Optional):

• install gdbi from git

  pip install git+https://github.com/xy-Ji/gdbi.git

• install graph database from pypi

  pip install neo4j==5.16.0
  pip install nebula3-python==3.4.0

Running an existing baseline model on an existing benchmark dataset

python main.py -m model_name -d dataset_name -t task_name -g 0 --use_best_config --load_from_pretrained

usage: main.py [-h] [--model MODEL] [--task TASK] [--dataset DATASET] [--gpu GPU] [--use_best_config][--use_database]

optional arguments:

-h, --help show this help message and exit

--model -m name of models

--task -t name of task

--dataset -d name of datasets

--gpu -g controls which gpu you will use. If you do not have gpu, set -g -1.

--use_best_config use_best_config means you can use the best config in the dataset with the model. If you want to set the different hyper-parameter, modify the openhgnn.config.ini manually. The best_config will override the parameter in config.ini.

--load_from_pretrained will load the model from a default checkpoint.

--use_database get dataset from database

---mini_batch_flag train model with mini-batchs

---graphbolt mini-batch training with dgl.graphbolt

---use_distributed train model with distributed way

e.g.:

python main.py -m GTN -d imdb4GTN -t node_classification -g 0 --use_best_config

python main.py -m RGCN -d imdb4GTN -t node_classification -g 0 --mini_batch_flag --graphbolt

Note: If you are interested in some model, you can refer to the below models list.

Refer to the docs to get more basic and depth usage.

Use TensorBoard to visualize your train result

tensorboard --logdir=./openhgnn/output/{model_name}/

e.g.：

tensorboard --logdir=./openhgnn/output/RGCN/

Note: To visualize results, you need to train the model first.

Use gdbi to get grpah dataset

take neo4j and imdb dataset for example

• construct csv file for dataset(node-level:A.csv,edge-level:A_P.csv)

• import csv file to database

  LOAD CSV WITH HEADERS FROM "file:///data.csv" AS row
  CREATE (:graphname_labelname {ID: row.ID, ... });

• add user information to access database in config.py file

  self.graph_address = [graph_address]
  self.user_name = [user_name]
  self.password = [password]

• e.g.:

python main.py -m MAGNN -d imdb4MAGNN -t node_classification -g 0 --use_best_config --use_database

Models

Supported Models with specific task

The link will give some basic usage.

Candidate models

• Heterogeneous Graph Attention Networks for Semi-supervised Short Text Classification[EMNLP 2019]
• Heterogeneous Information Network Embedding with Adversarial Disentangler[TKDE 2021]

Contributors

OpenHGNN Team[GAMMA LAB], DGL Team and Peng Cheng Laboratory.

See more in CONTRIBUTING.

Cite OpenHGNN

If you use OpenHGNN in a scientific publication, we would appreciate citations to the following paper:

@inproceedings{han2022openhgnn,
  title={OpenHGNN: An Open Source Toolkit for Heterogeneous Graph Neural Network},
  author={Hui Han, Tianyu Zhao, Cheng Yang, Hongyi Zhang, Yaoqi Liu, Xiao Wang, Chuan Shi},
  booktitle={CIKM},
  year={2022}
}