README.md

Recourse-using-Incomplete-Causal-Graph

The following has been tested with python 3.10

Installation Instructions

git clone https://github.com/cmaspi/Recourse-using-Incomplete-Causal-Graph.git
pip install virtualenv
cd Recourse-using-Incomplete-Causal-Graph/
source _venv/bin/activate # or with equivalent shell extension
# Making Changes in site-packages
rm -r _venv/lib/python3.10/site-packages/numpy/testing/
svn checkout https://github.com/numpy/numpy/branches/maintenance/1.14.x/numpy/testing _venv/lib/python3.10/site-packages/numpy/testing/
sed -i '4s/.*/from collections.abc import Iterable/' _venv/lib/python3.10/site-packages/causalgraphicalmodels/cgm.py

Performing Causal Discovery

python causal_discovery/main.py --scm_class diamond

Finding the recourses and generating the tables

python recourse/main.py --scm_class diamond --classifier_class lr --lambda_lcb 2. --optimization_approach grad_descent --grad_descent_epochs 1000 --batch_number 0 --sample_count 100 --experimental_setups m0_true m1_alin m1_akrr m1_gaus

Here, the argument scm_class can take the arguments diamond, lin4v, or german-credit. Change this argument to get the corresponding table.