Skip to content

Releases: THUMNLab/AutoGL

AutoGL Release v0.4.0

30 Dec 06:11
Compare
Choose a tag to compare
  • 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.

AutoGL Release v0.3.1

22 Apr 06:08
Compare
Choose a tag to compare

v0.3.1 is here!

- We have also released Chinese tutorial for the first time!
- AutoGL now support Deep Graph Library (DGL) backend to be interface-friendly for DGL users! All the homogeneous node classification task, link prediction task, and graph classification task are currently supported under DGL backend. AutoGL is also compatible with PyG 2.0 now.
- The heterogeneous node classification tasks are now supported! 
- To make the library more flexible, the module model now supports decoupled to two additional sub-modules named encoder` and decoder
- AutoGL supports neural architecture search (NAS) to customize architectures for the given datasets and tasks. 
- The link prediction task is now also supported! 

Feel free to open issues and PR!

AutoGL Release v0.1.1

23 Dec 08:15
Compare
Choose a tag to compare

v0.1.1 is here!

Fix the following bugs in v0.1.0:

  1. AutoNE support for node classification
  2. Change the interface of solver <-> trainer from Data to Dataset (to better support AutoNE)
  3. OGB node classification dataset support
  4. Some typos in the framework

Known issues:

  1. Currently, the autogl cannot directly used if you do not import second level package (when installed through pip):
>>> import autogl
>>> autogl.datasets
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'autogl' has no attribute 'dataset'

Please also import the second level package when using (import autogl.datasets). This bug will be fixed in future version.

  1. Some part of feature engineering in autogl 0.1.1 is not compactible with torch_geometric > 1.6.1 yet. Please use torch_geometric==1.6.1 or lower to use feature engineering. [See issue #10]

Feel free to open issues and PR!

AutoGL Release v0.1.0

07 Dec 12:26
Compare
Choose a tag to compare

Dear all:

We are happy to release the toolkit we are developing: AutoGL, an autoML framework & toolkit for machine learning on graphs.

AutoGL is developed for researchers and developers to quickly conduct autoML on the graph datasets & tasks. Refer to our documentation for more details. Please have fun playing with this first release and propose issues or pull request. Or contact us through [email protected].

Known issues

  1. OGB dataset is not fully supported, you may encounter bugs when conducting experiments on OGB datasets.
  2. AutoNE in HPO now have some bugs in implementation.

Future work

  1. Add NAS support
  2. Add data/model training sampling support
  3. Support more tasks, datasets and models