Description
Interpretable Machine Learning is a book dedicated to making machine learning models and their decisions interpretable. The book addresses the common issue of computers not explaining their predictions, which can lead to trust issues and undetected bugs. It explores the concepts of interpretability and introduces simple models like decision trees and linear regression. The focus is on model-agnostic methods such as LIME and Shapley values for explaining individual predictions, and permutation feature importance for understanding general feature-prediction relationships. The book also covers methods specific to deep neural networks.
Each interpretation method is explained in depth, discussing how they work, their strengths and weaknesses, and how to interpret them. This enables readers to select and apply the most suitable interpretation method for their machine learning applications. The book is recommended for machine learning practitioners, data scientists, statisticians, and anyone interested in making models interpretable.
The author, Christoph Molnar, began writing the book as a side project while working as a statistician. Surprised by the lack of comprehensive resources on the topic, he created this book to deepen his understanding and share insights. The book has become a go-to resource, cited by researchers and used by students and instructors. It has also been foundational in the author's career, inspiring his PhD and his work as a writer and consultant.
The 3rd edition includes updates such as new chapters on LOFO and Ceteris Paribus, reorganized introductions, deeper data examples, and practical tips. The book is licensed under the CC BY-NC-SA 4.0 license.
Book Details
Covers model-agnostic methods
Includes methods for deep neural networks
Explains LIME and Shapley values
Discusses permutation feature importance
Provides critical discussion of methods
Suitable for practitioners and researchers
3rd edition includes new chapters
Licensed under CC BY-NC-SA 4.0
Who This Book Is For
- Model interpretability
- Data science education
- Research citation
- Industry application
- Consulting
