This repository contains the code to replicate the experiments conducted for the paper "A SHAP Quotient Game for explaining Raman Spectroscopy classification models", submitted to AIME 2025.
The dataset folder contains pickle files with the datasets used for the experiments.
The covid and pd-ad folders contain the weights of the models trained using the leave-one-patient-out paradigm in Keras).
The utils folder contains utility scripts used in the experiments:
- Dataset.py: Functions for reading and preparing the datasets for experiments.
- GradCam.py: Core functions for applying Grad-CAM.
- quotient_game.py: Implementation of the Quotient Game method.
- kernel_shap.py: A re-implementation of the SHAP method using the SHAP library.
- cnn_models.py: Functions to create the CNN model used for the experiments using keras python library
Description of the remaining files in the repository:
- shap.py: Experiments using the SHAP method.
- lime.py: Experiments using the LIME method.
- qg_experiment.py: Experiments using the Quotient Game method.
- grad_cam.py: Experiments using the Grad-CAM method.
The file experiments.py provides an example experiment for each method applied to the PD-AD dataset. It leverages all the available files in the repository.
For any questions or further information, please contact Marco Piazza ([email protected]) or Enza Messina ([email protected]).