Fairness-in-Learning-of-LDS

This is the source code for the paper Fairness in Forecasting of Observations of Linear Dynamical Systems:

@misc{https://doi.org/10.48550/arxiv.2209.05274,
  doi = {10.48550/ARXIV.2209.05274},
  url = {https://arxiv.org/abs/2209.05274},
  author = {Zhou, Quan and Marecek, Jakub and Shorten, Robert N.},
  title = {Fairness in Forecasting of Observations of Linear Dynamical Systems},
  publisher = {arXiv},
  year = {2022},
}

Dependencies

1. Mosek/9.2 https://www.mosek.com/downloads/list/9/

2. Python scripts:

• Python/3.9.6

• inputlds https://raw.githubusercontent.com/jmarecek/OnlineLDS/master/inputlds.py

• ncpol2sdpa 1.12.2 https://ncpol2sdpa.readthedocs.io/en/stable/index.html

• AIF360 (for post-processing) https://github.com/Trusted-AI/AIF360

3. Julia scripts:

• Julia/1.8.5

• TSSOS https://github.com/wangjie212/TSSOS

4. The COMPAS dataset:

• It was downloaded from https://www.propublica.org/datastore/dataset/compas-recidivism-risk-score-data-and-analysis
• It is named as ``data/compas-scores-two-years.csv''.

The structure of scripts:

• F1.py: implementation and plotting of Figure 1.
• F2.py: implementation and plotting of Figure 2.
• F3_ncpol2sdpa.py F3_tssos.py F3_tssos_compas.py F3_sparsity.py: implementation of Figure 3.
• F3_plot.ipynb: plotting of Figure 3.
• F4.py: implementation and plotting of Figure 4.
• PostProcess_1.py PostProcess_2.py PostProcess_aif360.ipynb: implementation of Figure 5-7.
• PostProcess_plot.ipynb: plotting of Figure 5-7.
• functions.py: functions used for implementation.
• fairncpop.batch: not relevant.