/robust-ce-inn

Primary LanguageJupyter NotebookMIT LicenseMIT

robust-ce-inn

Codes for the AAAI 2023 paper - Formalising the Robustness of Counterfactual Explanations for Neural Networks.

  • /datasets: datasets used for experiments
  • /expnns: contains the scripts, utility classes, and results for experiments
    • dataset-name.ipynb: experiments for Section 5.2, 5.3, and Appendix C
    • dicedemo.ipynb: experiments for Appendix D
  • /inn.py, dataset.py: utility classes for MILP encoding
  • /optsolver.py: MILP encodings of problem definitions: CFX-MILP, INN
  • /expces: scripts for experiments finding CEs
  • /roar: codes for the baseline method ROAR, Upadhyay et al., "Towards Robust and Reliable Algorithmic Recourse", NeurIPS 2021. Adapted from https://github.com/AI4LIFE-GROUP/ROAR
  • /requirements.txt: required dependencies to run all experiments. Python version used: 3.7.13.