Explain Machine Learning Models With Shapley Values

This course teaches you one of the most powerful methods to explain predictions of machine learning models. Shapley values have a solid theoretical foundation in game theory and are model-agnostic — they can be applied to any machine learning model.

What the Course Will Teach You

• Understand and interpret Shapley values correctly.
• Hands-on application of Shapley values with R and Python
• Combining Shapley explanations to get global model interpretations such as feature importance, interactions and dependence plots.
• Optional: Deep dive into the mathematical and game-theoretical foundations.

Who is this Course for?

The course has three types of goals: building intuition, hands-on code examples and theoretical foundations. It‘s an all-inclusive package which should satisfy your needs, at whatever complexity you work with Shapley values. This course is for:

• Data Scientists who want to explain their model and build it into a dashboard.
• Researcher who want to find out what the heck their machine learning model learned.
• Machine learning engineers who make explanations part of their modeling pipeline.
• Operator of machine learning models that actually have to make sense of explanations that are produced by Shapley values.

Some Course Contents

• What are Shapley values?
• How can a prediction be explained with game theory?
• What the heck are SHAP, TreeSHAP, KernelSHAP?
• How do I use Shapley values in R or Python?
• How to combine individual explanations to get an understanding of the model behavior: Shapley based feature importance, interactions and dependence plots.
• How to apply Shapley values for text and image classification tasks.
• Shapley values and feature correlation.
• Creating model explanation dashboards quickly with shapash.

Why should you trust me (Christoph Molnar)?

• I have written a book about about various way to interpret machine learning models in a free online book on Interpretable Machine Learning.
• The book already has two chapters about Shapley values and SHAP. So you can check out if you like my style of explanation and trust me expertise.
• I have studied statistics and also worked as statistician. I have also done research about interpretable machine learning and worked as data scientist. I am positioned between academia and application, which is a good place to be to translate research into practical advice.
• I am not selling Shapley values as a miracle tool. I will be very honest about the situations in which they fail and what the caveats are.