Implementation of "Causal Generative Explainers using Counterfactual Inference".
The Morpho-MNIST data used in the paper can be downloaded here.
Models can be downloaded here. Once downloaded, this is what you should pass as the --model-dir argument to many scripts.
Precomputed shap values can be downloaded here for the --shap-value-dir argument to certain scripts.
Precomputed metrics for IM1/IM2/oracle scores can be downloaded here. These are passed as the --metrics-csv argument to appropriate scripts.
If any of the above links do not work, please reach out to me at [email protected]
Sometimes it expects the directory of precomputed CFs rather than the directory with the whole dataset. The scripts where this is the case are image_shap_evolution.py, mnist_cf_comparisons.py, and contrastive_evolution.py.