[IJCAI2025] Let's Group: A Plug-and-Play SubGraph Learning Method for Memory-Efficient Spatio-Temporal Graph Modeling

Module

The SubGraph Learning method is located in the SGL module folder.

Model

We have integrated the SubGraph Learning method into the code frameworks of models such as DDGCRN-main to ensure fair comparisons.

Running the Model

Please refer to the README file in each model framework for instructions on setting up the environment and running the models.

How to Use the SGL Method

We have embedded the parameters for the SGL method into the model's parameters:

• In DDGCRN-main、DGCNet-main and GMAN, the related parameter settings are in the config_file.
• In STAEformer, the related parameter settings are in model/STAEformer.yaml.
• In STWave, the related parameter settings are in baselines/STWave/PEMS0X.py.
• In DGCRN-main, the settings are in main.py.

To disable the SGL method, set use_subgraph to False.
The parameters memory_node and topk control the number of subgraphs and the number of nodes in each subgraph, respectively.

Dataset

The relevant datasets can be downloaded from Google Drive.
The files are already named according to the models. To use a dataset, simply place the corresponding files into the appropriate framework folder.

• For DDGCRN and DGCNet, move the data folder inside data_DDGCRN_DGCNet to the root directory of both models.
• For GMAN, move the data folder inside data_GMAN to the root directory of GMAN.
• For DGCRN, move the data folder inside data_DGCRN to the root directory of DGCRN-main.
• For STAEformer, move the data folder inside data_STAEformer to the root directory of STAEformer-main.
• For STWave, move the datasets folder inside data_STWave to the root directory of STWave.