https://github.com/hbaniecki/compress-then-explain
Efficient and accurate explanation estimation with distribution compression (ICLR 2025 Spotlight)
https://github.com/hbaniecki/compress-then-explain
dalex explainable-ai feature-attribution goodpoints interpretable-machine-learning kernel-thinning pdp sage shap
Last synced: about 1 month ago
JSON representation
Efficient and accurate explanation estimation with distribution compression (ICLR 2025 Spotlight)
- Host: GitHub
- URL: https://github.com/hbaniecki/compress-then-explain
- Owner: hbaniecki
- License: mit
- Created: 2024-06-26T12:15:15.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2025-02-11T17:52:09.000Z (3 months ago)
- Last Synced: 2025-03-25T14:25:26.072Z (about 2 months ago)
- Topics: dalex, explainable-ai, feature-attribution, goodpoints, interpretable-machine-learning, kernel-thinning, pdp, sage, shap
- Language: Python
- Homepage: https://openreview.net/forum?id=LiUfN9h0Lx
- Size: 1.25 MB
- Stars: 6
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Compress Then Explain
This repository is a supplement to [the following paper](https://openreview.net/forum?id=LiUfN9h0Lx):
> Hubert Baniecki, Giuseppe Casalicchio, Bernd Bischl, Przemyslaw Biecek. *Efficient and Accurate Explanation Estimation with Distribution Compression*. **ICLR 2025 (Spotlight)**
### Start: examples
In `examples`, we provide 4 Jupyter notebooks with simple code examples on how to use CTE to improve the estimation of SHAP, SAGE, Expected Gradients, and Feature Effects.
### Details: experiments
In `experiments`, we provide code to reproduce the results reported in Section 4 of the paper.
### Citation
```bibtex
@inproceedings{baniecki2025efficient,
title = {Efficient and Accurate Explanation Estimation with Distribution Compression},
author = {Hubert Baniecki and
Giuseppe Casalicchio and
Bernd Bischl and
Przemyslaw Biecek},
booktitle = {International Conference on Learning Representations},
year = {2025},
url = {https://openreview.net/forum?id=LiUfN9h0Lx}
}
```
### Acknowledgements
This work was financially supported by the Polish National Science Centre grant number 2021/43/O/ST6/00347. Hubert Baniecki gratefully acknowledges scholarship funding from the Polish National Agency for Academic Exchange under the Preludium Bis NAWA 3 programme.
