This repository contains the code to reproduce the numerical results in Not all distributional shifts are equal: Fine-grained robust conformal inference.

• datasets/: includes all the datasets for real data analysis.
  • nslm_semi_synthetic.csv is the semi-synthetic data generated based on the NSLM dataset of Carvalho et al.
  • covid_original.csv and covid_replication.csv correspond to the results from the original and replication study on the sharing COVID-related information.
• scripts/: contains the python code for reproducing the simulation and real application results of the paper.
  • simulation.py reproduces the simulation results in Section 5.
  • nslm.py reproduces the results in Section 6.1.
  • acs_income.py reproduces the results in Sectin 6.2.
  • covid.py reproduces the results in Section 6.3.
• results/: stores all the output files.

To set up the virtual environment, run the following command in terminal:

source ./scripts/venv3115/bin/activate

To reproduce one run of the simulation, execute the following command in terminal:

cd scripts
python3 simulation.py 1

To reproduce one run of the experiment on the NSLM dataset, execute the following command in terminal:

python3 nslm.py 1

To reproduce one run of the experiment on the ACS income dataset, execute the following command in terminal:

python3 acs_income.py 1

To reproduce one run of the experiment on the COVID dataset, execute the following command in terminal:

python3 covid.py 1

• The code for estimating the conditional cumulative distribution function is from the qosa-index package.
• The ACS income dataset and corresponding models are obtained from WhyShift.
• The COVID information datasets are from awesome-replicability-data.