Auto Graph Learning

An autoML framework & toolkit for machine learning on graphs.

Actively under development by @THUMNLab

Feel free to open issues or contact us at autogl@tsinghua.edu.cn if you have any comments or suggestions!

News!

2023.10.29 New version! v0.4.1 is here!
- We have simplified the dataset module so that users can build their own datasets more easily!
- We have developed an extended library: AutoGL-light, which is a lightweight version of AutoGL. Users can customize their own AutoML pipeline for graphs more easily!
- Minor bugs fixed.
2022.12.30 New version! v0.4.0-pre is here!
- We have proposed NAS-Bench-Graph (paper, code, tutorial), the first NAS-benchmark for graphs published in NeurIPS'22. By using AutoGL together with NAS-Bench-Graph, the performance estimation process of GraphNAS algorithms can be greatly speeded up.
- We have supported the graph robustness algorithms in AutoGL, including graph structure engineering, robust GNNs and robust GraphNAS. See robustness tutorial for more details.
- We have supported graph self-supervised learning! See self-supervised learning tutorial for more details.
2021.12.31 Version v0.3.0-pre is released
- Support Deep Graph Library (DGL) backend including homogeneous node classification, link prediction, and graph classification tasks. AutoGL is also compatible with PyG 2.0 now.
- Support heterogeneous node classification! See hetero tutorial .
- The module model now supports decoupled to two additional sub-modules named encoder and decoder. Under the decoupled design, one encoder can be used to solve all kinds of tasks.
- Enrich NAS algorithms such as AutoAttend, GASSO, hardware-aware algorithm, etc.
2021.07.11 Version 0.2.0-pre is released, which supports neural architecture search (NAS) to customize architectures, sampling to perform tasks on large datasets, and link prediction.
2021.04.16 Our survey paper about automated machine learning on graphs is accepted by IJCAI! See more here.
2021.04.10 Our paper AutoGL: A Library for Automated Graph Learning is accepted by ICLR 2021 Workshop on Geometrical and Topological Representation Learning! You can cite our paper following methods here.

Introduction

AutoGL is developed for researchers and developers to conduct autoML on graph datasets and tasks easily and quickly. See our documentation for detailed information!

The workflow below shows the overall framework of AutoGL.

AutoGL uses datasets to maintain datasets for graph-based machine learning, which is based on Dataset in PyTorch Geometric or Deep Graph Library with some functions added to support the auto solver framework.

Different graph-based machine learning tasks are handled by different AutoGL solvers, which make use of five main modules to automatically solve given tasks, namely auto feature engineer, neural architecture search, auto model, hyperparameter optimization, and auto ensemble.

Currently, the following algorithms are supported in AutoGL:

Feature Engineer

Model

NAS

HPO

Ensemble

Generators
Graphlets
EigenGNN
more ...

Selectors
SeFilterConstant
gbdt

Graph
netlsd
NxAverageClustering
more ...

Homo Encoders
GCNEncoder
GATEncoder
SAGEEncoder
GINEncoder

Decoders
LogSoftmaxDecoder
DotProductDecoder
SumPoolMLPDecoder
JKSumPoolDecoder

    Algorithms  

    Random  

    RL  

    Evolution  

    GASSO  

    [more ...](http://mn.cs.tsinghua.edu.cn/autogl/documentation/docfile/documentation/nas.html)  


    Spaces  

    SinglePath  

    GraphNas  

    AutoAttend  

    [more ...](http://mn.cs.tsinghua.edu.cn/autogl/documentation/docfile/documentation/nas.html)  


    Estimators  

    Oneshot  

    Scratch

Grid
Random
Anneal
Bayes
CAMES
MOCAMES
Quasi random
TPE
AutoNE

Voting
Stacking

This toolkit also serves as a framework for users to implement and test their own autoML or graph-based machine learning models.

Installation

Requirements

Please make sure you meet the following requirements before installing AutoGL.

Python >= 3.6.0
PyTorch (>=1.6.0)

see for installation.
Graph Library Backend

You will need either PyTorch Geometric (PyG) or Deep Graph Library (DGL) as the backend. You can select a backend following here if you install both.

3.1 PyTorch Geometric (>=1.7.0)

See for installation.

3.2 Deep Graph Library (>=0.7.0)

See for installation.

Installation

Install from pip

Run the following command to install this package through pip.

pip install autogl

Install from source

Run the following command to install this package from the source.

git clone https://github.com/THUMNLab/AutoGL.git
cd AutoGL
python setup.py install

Install for development

If you are a developer of the AutoGL project, please use the following command to create a soft link, then you can modify the local package without install them again.

pip install -e .

Documentation

Please refer to our documentation to see the detailed documentation.

You can also make the documentation locally. First, please install sphinx and sphinx-rtd-theme:

pip install -U Sphinx
pip install sphinx-rtd-theme

Then, make an html documentation by:

cd docs
make clean && make html

The documentation will be automatically generated under docs/_build/html

Cite

Please cite our paper as follows if you find our code useful:

@inproceedings{guan2021autogl,
  title={Auto{GL}: A Library for Automated Graph Learning},
  author={Chaoyu Guan and Ziwei Zhang and Haoyang Li and Heng Chang and Zeyang Zhang and Yijian Qin and Jiyan Jiang and Xin Wang and Wenwu Zhu},
  booktitle={ICLR 2021 Workshop on Geometrical and Topological Representation Learning},
  year={2021},
  url={https://openreview.net/forum?id=0yHwpLeInDn}
}

You may also find our survey paper helpful:

@article{zhang2021automated,
  title={Automated Machine Learning on Graphs: A Survey},
  author={Zhang, Ziwei and Wang, Xin and Zhu, Wenwu},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, {IJCAI-21}},
  year={2021},
  note={Survey track}
}

License

We follow Apache license across the entire codebase from v0.2.