RobustFair Evaluation

Code for RobustFair: Adversarial Evaluation through Fairness Confusion Directed Gradient Search

Package Requirements:

Python 3.8,

tensorflow 2.4.1,

numpy 1.19.5,

keras 2.4.3,

scikit-learn 1.0.2,

pandas 1.4.3,

RobustFair for evaluating accurate fairness

This package provides code for evaluating accurate fairness using the RobustFair method. The following steps outline the process for conducting experiments on the Adult dataset:

1.Unzip the dataset.zip file.

2.Run adult_train_mode.py to train the baseline model.

3.Run adult_prepare_seeds.py to get the experimental seeds.

4.1.Run compare_experiment_RF.py to evaluate the accurate fairness.

4.2.Run check_loss_change_RobustFair.py to analyze the loss function trend during accurate fairness evaluation.

5.Run adult_retrain_model.py to retrain the model using the RobustFair evaluation from training dataset.

6.Run adult_check_BL_model.py and adult_check_retrain_model.py to check the models on the original testing data.

Exporting experiment results

7.Use the adult_get_result.py script to export the experiment results as worksheets.

If you have any questions or need further assistance, please reach out to us.