Interpretable Machine Learning

A collection of code, notebooks, and resources for training interpretable machine learning (ML) models, explaining ML models, and debugging ML models for accuracy, discrimination, and security.

Want to contribute your own examples/code/resources? Just make a pull request.

Setup

cd interpretable-ml
virtualenv -p python3.6 env
source env/bin/activate
pip install -r python/jupyter-notebooks/requirements.txt

** Note: if using Ubuntu, you may have to manually install gcc. Try the following 
1. sudo apt-get update
2. sudo apt-get install gcc
3. sudo apt-get install --reinstall build-essential

Contents

• Presentations
  • Responsible Machine Learning
  • Overview of Interpretable Machine Learning Techniques (incomplete list)
  • Discrimination in Machine Learning
  • Secure Machine Learning
• Jupyter Notebooks
  • Binary Classification
    • Shapley, PDP, & ICE
    • Decision Tree Surrogate and Leave One Covariate Out
    • LIME
    • Residual Analysis
    • Sensitivity Analysis
    • Disparate Impact Analysis
    • Disparate Impact Analysis w/ Python datatable
  • Multinomial Classification
    • Shapley, PDP, & ICE
    • Disparate Impact Analysis

• Simulated Data for Testing Purposes
  • Binary Classfication
    • Shapley, PDP, & ICE
  • Multinomial Classification
    • Shapley, PDP, & ICE
    • Shapley, PDP, ICE, & Decision Tree Surrogate

Further reading:

• Books/Articles
  • Responsible Machine Learning: Actionable Strategies for Mitigating Risks & Driving Adoption
  • An Introduction to Machine Learning Interpretability, 2nd Edition
  • A Responsible Machine Learning Workflow with Focus on Interpretable Models, Post-hoc Explanation, and Discrimination Testing
  • On the Art and Science of Explainable Machine Learning
  • Proposals for model vulnerability and security
  • Proposed Guidelines for the Responsible Use of Explainable Machine Learning
  • Real-World Strategies for Model Debugging
  • Warning Signs: Security and Privacy in an Age of Machine Learning
  • Why you should care about debugging machine learning models

Resources

• Awesome Machine Learning Interpretability