A PyTorch implementation of "Stable Fair Graph Representation Learning with Lipschitz Constraint"
SFG is a Lipschitz constraint-based method to maintain the stability of fair GNNs while preserving fgairness and accuracy performance The core idea behind SFG is to control the size of the encoder weights space in the presence of a generator.
We have provided a rar formatted compressed file, which you can simply extract directly.
To reproduce our results, please run:
bash run.shIf you want to visualize the result comparison, you can do it as follows:
-
Copy the log to the visualize folder:
Navigate to the logs directory and locate the training log file named XXXXX_ResLog.txt. Then, Copy the corresponding log to the visualize folder. -
Run the Plot Script: run plot_curve_30.ipynb.
For simplicity, we provide some sample reference data that you can use directly.
Note: It is important that the new file names are consistent to ensure the script reads the correct data.