This archive includes code and data for the paper:
"Bounding the Fairness and Accuracy of Classifiers from Population Statistics"
by Sivan Sabato and Elad Yom-Tov, ICML 2020.

The code is released under the MIT Open Source license. 

The following files are provided:

* README - this file
* find_lb.m - the implementation of Alg. 1 from the paper. Type "help find_lb" for input and output arguments.



Experiment files:

The following scripts use the data file USCensus1990raw.data.mat which can be downloaded from the following link: https://archive.ics.uci.edu/ml/datasets/US+Census+Data+(1990)

* census_commands.m - the script for running the experiments with beta = 1 on the UC Census data set. 
* census_commands_beta.m - the script for running the experiments with variable beta on the UC Census data set. 


* cancer.m - a script for running the experiments on the classifiers generated from search-engine data. 

* poll_run.m - a script for running the poll experiments. This script requires downloading the relevant data; 
See explanataions on how to do this in the comment at the top of the script. 

* cancer_mortality.m - a script for running the cancer mortality experiments. This script requires downloading the relevant data; 
See explanataions on how to do this in the comment at the top of the script. 



Data files:
* cancer_data.m - the statistics of the classifiers generated by the search-engine data in the first reported experiment. Used by the cancer.m script

auxiliary files used by the files above:

* calcunfairness.m
* calculate_classifier.m
* find_best_alphas.m



Result files:
The following files contain the experiment results, which were used to generate the plots in the paper.

* census_commands.csv: results of census_commands.m 
* discdata.csv: results of census_commands_beta.m 
* cancer.csv: results of cancer.m
* polls.csv: results of poll_run.m
* cancer_mortality.csv: results of cancer_mortality.m