Skip to content

Latest commit

 

History

History
 
 

group_agnostic_fairness

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
data
data
 
 
data_utils
data_utils
 
 
README.md
README.md
 
 
__init__.py
__init__.py
 
 
adversarial_subgroup_reweighting_model.py
adversarial_subgroup_reweighting_model.py
 
 
adversarial_subgroup_reweighting_model_test.py
adversarial_subgroup_reweighting_model_test.py
 
 
baseline_model.py
baseline_model.py
 
 
baseline_model_test.py
baseline_model_test.py
 
 
fairness_metrics.py
fairness_metrics.py
 
 
fairness_metrics_test.py
fairness_metrics_test.py
 
 
ips_reweighting_model.py
ips_reweighting_model.py
 
 
ips_reweighting_model_test.py
ips_reweighting_model_test.py
 
 
main_trainer.py
main_trainer.py
 
 
main_trainer_test.py
main_trainer_test.py
 
 
requirements.txt
requirements.txt
 
 
robust_learning_model.py
robust_learning_model.py
 
 
robust_learning_model_test.py
robust_learning_model_test.py
 
 
run.sh
run.sh
 
 

README.md

Installation

Run the following:

virtualenv -p python3 .
source ./bin/activate

pip3 install -r group_agnostic_fairness/requirements.txt

Data Preparation

The data provided in the './group_agnostic_fairness/data/toy_data directory is dummy, and is only for testing the code. For meaningful results, please follow the steps below.

Pre-process COMPAS dataset and create train and test files:

Download the COMPAS dataset from: https://github.com/propublica/compas-analysis/blob/master/compas-scores-two-years.csv and save it in the './group_agnostic_fairness/data/compas' folder.

Run './group_agnostic_fairness/data_utils/CreateCompasDatasetFiles.ipynb' notebook to process the dataset, and create files required for training.

Pre-process UCI Adult (Census Income) dataset and create train and test files:

Download the Adult train and test data files can be downloaded from: https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.data https://archive.ics.uci.edu/ml/machine-learning-databases/adult/adult.test and save them in the './group_agnostic_fairness/data/uci_adult' folder.

Run './group_agnostic_fairness/data_utils/CreateLawSchoolDatasetFiles.ipynb' notebook to process the dataset, and create files required for training.

Pre-process Law School Admissions Council (LSAC) Dataset and create train and test files:

Download the Law School dataset from: (http://www.seaphe.org/databases.php), convert SAS file to CSV, and save it in the ./group_agnostic_fairness/data/law_school folder.

Run CreateLawSchoolDatasetFiles.ipynb to process the dataset, and create files required for training.

Generate synthetic datasets used in the paper:

To generate various synthetic datasets used in the paper run './group_agnostic_fairness/data_utils/CreateUCISyntheticDataset.ipynb' notebook.

Training and Inference

Training and evaluation for the adversarial robust learning model can be run as:

python -m group_agnostic_fairness/main_trainer

Refer to the test cases in <model_name>_model_test.py files to understand the workflow.

Disclaimer: This is not an official Google product.