Learning on Graphs with Out-of-Distribution Nodes

GitHub Link

The GitHub link is https://github.com/songyyyy/kdd22-oodgat

Introduce

The repository "SongYYYY/KDD22-OODGAT" contains the implementation of "OODGAT," a method presented in the paper "Learning on Graphs with Out-of-Distribution Nodes." The implementation involves running scripts within a 'pyg_data' folder and requires dependencies like torch_geometric, networkx, sklearn, numpy, and scipy. The article provides instructions for downloading datasets, but notes changes in directory structure for recent versions of torch_geometric. The code supports testing on various datasets by modifying parameters in the 'train.py' script, including dataset names, splits, and hyperparameters. The hyperparameters for different datasets are listed in a table. Graph Neural Networks (GNNs) are state-of-the-art models for performing prediction tasks on graphs.

Content

This is the repository for paper 'Learning on Graphs with Out-of-Distribution Nodes'. To run the scripts, please create a folder named 'pyg_data' in the root directory. In the first run, datasets will be automatically downloaded to the folder from torch_geometric. When dataset is changed, the 'splits', 'ID_classes' and 'continuous' should also be changed accordingly. Hyperparameters for all datasets are listed in the following table:

---