Stochastic Amortization for feature attributions

This repository implements an amortized approach to feature attribution from this paper published in NeurIPS 2024. The code for data valuation is available in this repo.

Installation

conda create -n xai-amortization python=3.9
conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia
pip install transformers==4.32.0 evaluate==0.4.0 datasets==2.14.3 accelerate==0.20.3 scikit-learn==1.3.0
pip install git+https://github.com/huggingface/transformers@e9ad51306fdcc3fb79d837d667e21c6d075a2451
pip install fsspec==2023.6.0
pip install jupyterlab jupyter_contrib_nbextensions wandb ipdb gpustat seaborn
ipython kernel install --name "xai-amortization" --user

# Set up git filter if you want to strip the output of Jupyter notebooks before committing
git config filter.strip-notebook-output.clean 'jupyter nbconvert --ClearOutputPreprocessor.enabled=True --ClearMetadataPreprocessor.enabled=True --to=notebook --stdin --stdout --log-level=ERROR'

Usage

Train amortization models

• scripts/imagenette/classifier.sh: train classifiers.
• scripts/imagenette/shapley.sh: train amortization models for Shapley value feature attribution.
• sripts/imagenette/lime.sh: train amortization models for LIME feature attribution.
• scripts/imagenette/banzhaf.sh: train amortization models for Banzhaf feature attribution.

Evaluate amortization models

• notebooks/visualize_results.ipynb: evaluate and visualize the results of the amortization models.

Note

This repository was inspired by the following repositories:

• FastSHAP
• ViT-Shapley
• Huggingface example - Image classification