FDM: Fair Diversity Maximization in Streaming and Sliding-Window Models

In this repository, we implement FDM in Streaming and Sliding-Window Models. Algorithms(Fair-Swap, Fair-Flow and Fair-GMM) in Diverse Data Selection under Fairness Constraints and Fair-Greedy-Flow in Improved Approximation and Scalability for Fair Max-Min Diversification are also implemented, which are treated as baseline.

Datasets

Four real-world datasets ,i.e. Adult, CelebA, Census, Lyrics and one sythetic dataset are used in our experiments. Download links are included in the paper. To generate sythetic dataset, run the command like the following :

python generate_synthetic_dataset.py -n 1000 -m 2 -d 2

Dataset	Group	n	m	#feats	Distance Metric
Adult	Sex	48, 842	2	6	Euclidean
	Race		5
	S+R		10
CelebA	Sex	202, 599	2	41	Manhattan
	Age		2
	S+A		4
Census	Sex	2, 426, 116	2	25	Manhattan
	Age		7
	S+A		14
Lyrics	Genre	122, 448	15	15	Angular
SYN	-	1000-10,000,000	2-20	2	Euclidean

Requirements

• Ubuntu 20.04 (or higher version)
• Python 3.8 (or higher version)

Experiments

Once you have downloaded the datasets and put them in the directory datasets, you can reproduce the experiments by simply running the following commands:

python run_exp_varying_k_m_n_stream.py
python run_exp_varying_k_window.py 
python run_exp_varying_m_w_window.py