jayeshchoudhari / fair-k-means-clustering

Code of the paper Fair k-Means Clustering

Fair-Lloyd algorithm

Matlab implementation of the Fair-Lloyd algorithm. Description of the files:

Folder: Fair-Lloyd

a. load and pre-rpocess the data

• loadData: load the working dataset • preProcessEductaionVector: only for the credit dataset. Outputs the sensitive feature vector indicating higher-educated and lower-educated • normalizeData : make the mean 0 and variance one for each feature • projectData: project the data using PCA or Fair-PCA • fairPCA: project the data using the Fair-PCA algorithm of Samadi et al. 2018 (look below for subroutines) • mw (subroutine for fairPCA): multiplicative weight update • optApprox (subroutine for mw): finds the optimal rank-d approximation of a given matrix • re (subroutine for mw): calculates reconstruction error of a matrix

b. Fair-Lloyd

• lloyd: implementation of lloyd’s and Fair-Lloyd algorithm (look below for subroutines) • giveRandCenters: generate initial random centres • findCenters: given a clustering, find the centers • findClustering: given a set of centers, find the clustering • compCost: given a clustering of the data, computing (fair) cost • kmeansCost_S_C (subroutine for compCost): given a clustering and a set of centers, calculate the k-means cost. • bSearch: binary search procedure to find the fair centers

c. evaluate the results using lloyd’s and Fair-Lloyd

• maink: main file for running the experiments Folder: plotsCode

• plotting the results using the output data by Fair-Lloyd and Lloyd’s You can download the resultsData folder from https://drive.google.com/file/d/1d2ChCuUHRpTFjPeAvHnsiw0mkP359WI3/view?usp=sharing