ACFX is Actionable Counterfactual eXplainer. It is a model-agnostic XAI algorithm that provides actionable counterfactual explanations for any type of machine-learning model.

To be able to use the acfx package, install it using pip from PyPi

conda create --name acfx_env python=3.11
conda activate acfx_env
conda install pip
pip install acfx

...or directly from source code

git clone https://github.com/sbobek/acfx
cd acfx/src
pip install .

To be able to run benchmark, install dependencies by running

conda create --name acfx_env python=3.11
conda activate acfx_env
conda install pip
pip install acfx[benchmark]

...or directly from source code

git clone https://github.com/sbobek/acfx
cd acfx/src
pip install .[benchmark]

To run benchmark, type:

git clone https://github.com/sbobek/acfx
cd acfx/src/
python -m acfx.benchmark.main

Make sure to run the main.py file as a module using python -m and execute it from the acfx/src/acfx directory. This ensures that Python treats the acfx folder as a package, allowing relative imports inside the script to work correctly. Running the script directly (e.g., python benchmark/main.py) or from a different location may result in import errors due to incorrect module resolution (and) or invalid filepaths given in some of the benchmark methods.