Skip to content

scripts

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

scripts

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
__pycache__
__pycache__
 
 
README.md
README.md
 
 
adversarial_fairness.ipynb
adversarial_fairness.ipynb
 
 
attack_model.py
attack_model.py
 
 
bayesian_model.py
bayesian_model.py
 
 
gradient_reversal.py
gradient_reversal.py
 
 
initial-data-discovery.ipynb
initial-data-discovery.ipynb
 
 
main.py
main.py
 
 
model.json
model.json
 
 
naive_model.h5
naive_model.h5
 
 
notebook_pytorch_attack.ipynb
notebook_pytorch_attack.ipynb
 
 
notebook_tsne.ipynb
notebook_tsne.ipynb
 
 
plot_confusion_matrix.py
plot_confusion_matrix.py
 
 
script_attack_secML.py
script_attack_secML.py
 
 
script_eval_perf_classif_age_category_decil_score.py
script_eval_perf_classif_age_category_decil_score.py
 
 
script_evaluate_parameter_secML.py
script_evaluate_parameter_secML.py
 
 
script_factory_classifier.py
script_factory_classifier.py
 
 
script_factory_classifier_secML.py
script_factory_classifier_secML.py
 
 
script_incorporate_adversarial_points.py
script_incorporate_adversarial_points.py
 
 
script_sanitize_data.py
script_sanitize_data.py
 
 
script_sanitize_data_age_value.py
script_sanitize_data_age_value.py
 
 
script_secML_eval_perf_classif_age_category_decil_score.py
script_secML_eval_perf_classif_age_category_decil_score.py
 
 
script_test_example_secML.py
script_test_example_secML.py
 
 
test_secML_attack_and_evaluate_impact_on_Keras_model.ipynb
test_secML_attack_and_evaluate_impact_on_Keras_model.ipynb
 
 
unbiased_model.h5
unbiased_model.h5
 
 

README.md

Gradient reversal

The relevant code of the gradient reversal method presented by Adel et al. (2019), Raff et al. (2018) and Ganin et al. (2017).

Relevant files

gradient_reversal.py contains the main model. Pretrained models are available for COMPAS as naive_model.h5 and unbiased_model.h5. These can be used by:

gr_naive = GradientReversalModel()
gr_naive.load_trained_model(path=naive_model.h5, hp_lambda=0)
Y_pred_n = gr_naive.predict(X_test)

A demo notebook illustrating the method is provided in adversarial_fairness.ipynb.

Utility scripts are provided in:

  • bayesian_model.py: For the evaluation of fairness metrics in a compacter form.
  • plot_confusion_matrix.py: To plot confusion matrices...