Awesome

python implementation for the GIME method in IJCAI 2021 "Learning Groupwise Explanations for Black-Box Models"

Packages

• python 3.6.9
• torch 1.2.0
• numpy 1.16.2
• scikit-learn 0.20.2
• scipy 1.2.1

Structure

• dataset
• gime, the main implementation of our method
• utils.py, helper functions
• run.py, the pipeline to run our method

Run

python run.py, which loads the dataset (e.g., Wine), trains the target model (e.g., SVR), generates groupwise explanations, and evaluates the generalized fidelity.