Overview

This repository contains the code and instructions for reproducing the experiments in the paper Interventional Fairness on Partially Known Causal Graphs: A Constrained Optimization Approach (ICLR 2024). The experiments are performed on synthetic dataset and real dataset (Student and Credit).

Installation and Setup

Prerequisites

• Python 3.9
• R 4.2.1

Before running the scripts, ensure that you have the required environments set up.

Dependencies

Install the necessary Python packages using:

pip install numpy==1.24.2 pandas==1.5.3 networkx==2.8.4 torch==2.0.0 python-igraph==0.10.4 scipy==1.10.1 scikit-learn==0.24.2

For R, install the required packages by running the following in your R console:

install.packages(c("reshape", "ggplot2", "ggpubr", "e1071", "condMVNorm", "latex2exp"))

Data and Scripts

For the real-world data, download the Student and Credit datasets from the UCI Machine Learning Repository and place them in the ./data folder. The Student dataset can be found here and the Credit dataset can be found here.

For the convenience for the reader, we provide the processed data in the ./data folder.

Usage

Generating Data and Results

Run the provided shell script to generate data and results. Modify parameters within run.sh as necessary:

bash run.sh

Parameters explanation (as seen in run.sh):

• adm: Number of admissible variables
• s: Number of edges
• j: Number of nodes, iterates from 10 to 40 in steps of 10.
• i: Graph number, iterates from 0 to 9.

Generating Plots

Tradeoff Plots

To reproduce the tradeoff plots, execute plot_tradeoff.R:

1. Change the working directory in the script to where your data is located.
2. Set the variable i to one of the four graph settings (different combinations of nodes and edges) you wish to plot.

setwd('PATH_TO_DATA')
...
i = 3  # one of the four settings that you want to get

The output will be saved in the ./Repository_adm={adm} folder as Tradeoff_MnodesNedges_truth.pdf and Tradeoff_MnodesNedges.pdf, where M and N denote the number of nodes and edges respectively.

Density Plots

For density plots, run plot_density.R after setting the working directory and parameters:

setwd('PATH_TO_DATA')
...
dd = 5  # Number of nodes
ss = 8  # Number of edges
adm = 0 # Number of admissible variables

Output will be stored in the ./Repository_adm={adm} folder as Density_MnodesNedgesKgraph.pdf, where K is the number of graphs, and M and N denote the number of nodes and edges respectively.

Real Data Analysis (Student and Credit)

Generate interventional data and results as follows: (Taking Student dataset as an example)

cd student  # or 'cd credit'
python gen_data_contingency.py    # generate data
python separate_data.py     # align input format
python soft_constraint.py   # train the model

Tradeoff plots can be reproduced using Simu_results_filter_and_processor_real.R within the respective folders. Remember to set the working directory appropriately:

setwd('PATH_TO_DATA')
...