This is the source code for paper: A Data-centric Framework to Endow Graph Neural Networks with Out-Of-Distribution Detection Ability. If you have any questions about the paper or code, please contact [email protected] .

Requirements:

• torch-geometric==2.0.4
• torch-scatter==2.0.93
• torch-sparse==0.6.15
• numpy==1.21.2
• pandas==1.3.0
• python==3.9.15
• scikit-learn=1.0.2
• scipy==1.9.3
• torch==1.11.0
• torchvision==0.12.0

Training:

Run our model with SSD scoring fuction in TU datasets:

python clmd.py -DS_pair xxxx 

We also provide the code to search hyper-parameters, you can use the following command (for ENZYMES dataset) to run it:

bash eme.sh