SHAP Explainer Docker Tool

Description

The SHAP Explainer Docker Tool is a containerized solution for performing post hoc explainability on handcrafted radiomics machine learning models using SHAP (SHapley Additive exPlanations). SHAP is a powerful tool for interpreting machine learning models and understanding the impact of individual features on model predictions.

This tool allows you to:

• Compute SHAP values for your handcrafted radiomics machine learning model.
• Generate and save global summary plots, local explanation plots, and dependence plots to gain insights into model behavior.

Usage

1. Clone this repository to your local machine.

2. Prepare your model and data:

   • Ensure you have a trained handcrafted radiomics machine learning model saved in a file (e.g., filename_svm.sav). Save the model using the following method: pickle.dump(svm, open(filename_svm, 'wb')).
   • Prepare your preprocessed handcrafted radiomics features in a format compatible with the model (e.g., a Numpy array) and save it to a file (e.g., test_file.npy). Save the preprocessed features using Numpy's np.save method.

3. Create a configuration file (config.ini) with the required parameters:

   [DEFAULT]
   model_filename = filename_svm.sav
   test_data_preprocess = test_file.npy
   save_dir_for_plots = test_run
   local_plt_indices_list = 0,1,2
   top_n_dependence_plots = 6
   top_n_dependence_interactions = 4
   feature_names = L_GLCM_clusTend, L_GLCM_correl1, L_GLDZM_DZN, ...
   

4. Build the Docker image using the provided Dockerfile:

   docker build -t shap-exp-tool-img .
   

5. Run the Docker container using the built image and the mounted volumes for data and plots:

   docker run -it --rm -v /path/to/local/data:/home/chaimeleon/datasets shap-exp-tool-img
   

6. The tool will compute SHAP values, generate and save global summary plots, local explanation plots, and dependence plots in the specified save_dir_for_plots directory.

Dummy Examples

Global Explanations:

Global explanations summarize the overall behavior of the machine learning model. Below is an example of Global explanation plot:

• Summary plot for Class

Local Explanations:

Local explanations provide insights into individual predictions. Below is an example of local explanation plot:

• Local explanation plot

Dependence Plots:

Dependence plots show the relationship between feature values and SHAP values. Below is an example of dependence plot:

• Dependence plot

For any questions or issues with the tool, feel free to reach out.