:seedling: :seedling: Currently we are building this project with more models and datasets. We welcome your questions and suggestions.:seedling::seedling:
:milky_way::milky_way: Our paper has been accepted by CIKM'23 as resource paper: Boshen Shi, Yongqing Wang, Fangda Guo, Jiangli Shao, Huawei Shen, and Xueqi Cheng. 2023. OpenGDA: Graph Domain Adaptation Benchmark for Cross-network Learning. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management (CIKM ’23), Octo- ber 21–25, 2023, Birmingham, United Kingdom. ACM, New York, NY, USA, 5 pages. https://doi.org/10. :milky_way::milky_way:
OpenGDA is a benchmark which integrates 1) datasets for evaluating diverse cross-network learning tasks and 2) state-of-the-art graph domain adaptation models.
cross-network learning task: To alleviate the lack of high-quality labels and the sparse graph structure, reseachers build cross-network learning task by introducing relevent source graphs to transfer labeling and structural knowledge to target graphs. The goal of cross-network learning task is improving task performance on target graphs by transferring knowledge from source graphs.
graph domain adaptation: Researchers improve domain adaptation techinique by taking the properties of structured graph data into account.
Currently, there mainly exist two limitations in evaluating graph domain adaptation models.
As comprehensively assessing models could enhance model practicality in real-world applications, we propose a benchmark, known as OpenGDA. It provides abundant pre-processed and unified datasets for different types of tasks (node, edge, graph). They originate from diverse scenarios, covering web information systems, urban systems and natural systems. Furthermore, it integrates state-of-the-art models with standardized and end-to-end pipelines. Overall, OpenGDA provides a user-friendly, scalable and reproducible benchmark for evaluating graph domain adaptation models.
We have tested under multiple environments and here we list some of them:
We mainly require pytorch
as neural network framework and torch-geometric
as GNN framework.
Other related packages you can find in the requirements.txt
OpenGDA framework is shown below:
OpenGDA workflow is shown below:
:fireworks:Currently we provide Airport dataset as it is relatively small, other datasets please refer to their original studies, and we will provide a copy on cloud drive asap.
:bangbang: For instructions to data resources, please refer to node-level
:sun_with_face:To run a model, like ASN, you need to change your path to \model\ASN
, and run it with:
python start_nc.py --dataset_name airport --src_name usa --tgt_name brazil --cuda 0
For command line args, please refer to start_nc.py
for more details.
:bangbang: For instructions to data resources, please refer to edge-level
:fireworks:Currently we provide LetterHigh-LetterLow dataset as it is relatively small, other datasets please refer to their original studies, and we will provide a copy on cloud drive asap.
:bangbang: For instructions to data resources, please refer to graph-level
:sun_with_face:To run a model, like GRADE, you need to change your path to \model\GRADE
, and run it with:
python start_gc.py --dataset_name TUDataset --src_name Letter-high --tgt_name Letter-low --cuda 0
For command line args, please refer to start_gc.py
for more details.