Certifair is a framework for verifiying individual fairness properties globally for Neural Networks.
We tested Certifair on Ubuntu 20.04.4 LTS. We recommend having a GPU for training Neural Networks. It's also recommended to have at least 8 CPU cores for fairness verification.
We recommend using conda for installing Certifair and we'll provide a step by step guide for the setup of conda environment on a Ubuntu.
Install the latest version of Anaconda.
Create a conda environment with all the dependencies by running
conda env create --name envname --file=environment.yml
make sure to replace envname with the environment name you'd like and that the environment was created without errors.
PeregriNN -the underlying verifier- relies on Gurobi commercial solver which isn't open source. However, they provide a free academic license. Please request an academic license from here
After installing Certifair and acquiring Gurboi license, we can test the installation by running
python certifair.py german property1 german_p1_base --lr 0.007 --fr 0.0 --no-bound_loss --layers "30,30,1"
You can reproduce the main results in table 2 by running
./scripts/tab2.sh
Similarly, table 3 results can be reproduced by running
./scripts/tab3.sh
The results are dumped in experiments/logs/${experiment_name}.log and the models are saved in experiments/models/model_ex_${experiment_name}.pth
For description of all args, run
python certifair.py -h