Modular Counterfactuals

This repository contains the preliminary code of the modular counterfactual explanation framework MCE, supervised by Nadia el Bekri at Fraunhofer IOSB, Karlsruhe.

To Try yourself

Clone the repository into some directory, create a virtual environment with

virtualenv pyvenv

activate the environment with

source pyvenv/bin/activate

and install the required packages with pip:

pip install -r requirements.txt

from inside the source code directory you can then try a test run using the experiments.py

python experiments.py --ex ids

using the --ex ids/kdd flag, you can choose between running the experimental evaluation on KDD or IDS datasets, if provided correctly.

Experiment Environment - Sacred

To keep track of our experiments, we use sacred. How to use sacred for other experiments as well can be seen in the experiments.py. We use sacred with the MongoObserver and display results using Omniboard.

Code Refactoring

Along with adding consistent docs we ought to improve code formatting and the general architecture to be more API-like, modular and accessible. A great example for a good Python DataScience project in this regard is LIME, which is largely object-oriented.

Adhere to the PEP8 Style Guide as much as possible.

Generally:

• Look out for consistent variable naming and expressive names
• remove legacy code and commented-out code
• Make stuff simpler
• Write explanatory comments inside the code to explain magic steps.