Part of this repository contains source code necessary to reproduce some of the main results in the paper.
Moreover, an extension have been developed to complete my master's degree thesis.
Some running examples can be found in Run folder. As an example, the file run.py show the running command:
python nifty_sota_gnn-fairattr.py --drop_edge_rate_1 0.1 --drop_edge_rate_2 0.1 --dropout 0.5 --hidden 16 --lr 1e-3 --epochs 2500 --model ssf --encoder gcn --dataset german --sim_coeff 0.6' + ' --drop_feature_rate_1 ' + str(p) + ' --drop_feature_rate_2 ' + str(p) + ' --seed ' + str(r) + ' --fairdrop 1' + " --delta " + str(d) + ' --fairattr 1
where:
- nifty_sota_gnn-fairattr.py is the script that can run the entire solution with Biased Edge Dropout AND Fair Related Features
- drop_edge_rate_1 and 2 is the drop edge rate used by NIFTY (values should be the same)
- drop_feature_rate_1 and 2 same as the previous point but w.r.t. nodes attributes
- delta is the delta hyperparameter as defined in the paper
- fairattr is a boolean indicating wheter using or not the fair related features extension
- fairdrop is a boolean indicating wheter using or not the biased edge dropout extension
Values for hyperparameters can be found both in the paper and in the thesis
Be sure to install the correct versions for the packages used as indicated in requirements.txt
For any questions feel free to mail me: Federico Caldart