Practical Gradient Boosting: An deep dive into Gradient Boosting in Python

Well grounded and excellent coverage. Perhaps the only missing piece is ranking problems. Books should be focused like this one.

This book is an excellent resource for anyone who wants to learn more about gradient boosting in Python. It is clear that the author has a lot of experience in the area and all the explanations are clear and concise. I particularly like that most theory explanations are accompanied by code. This makes it much easier to understand the concepts, as you can see exactly how they work in practice. I particularly enjoyed the chapter "Understanding and explaining a model" which looks at using methods like feature importance and SHAP. In light of recent developments (i.e. EU AI Act) these skills are going to become even more critical. Understanding how your models work and being able to explain them to others is going to be an essential part of working with AI going forward. Overall, I would highly recommend this book to anyone who wants to learn more about gradient boosting or machine learning in general. It is well-written, easy to follow, and packed with practical advice and code examples.

This book is a great resource to really understand the gradient boosting. With very plain Python code, clear examples, and straightforward descriptions, Dr. Saupin walks us right through decision trees, ensembles, then gradient boosting. Even if you aren’t expert in the math, you can understand this book, and become comfortable with Gradients and Hessians as a bonus. Advanced topics like hyperparameter tuning, objective functions, and applications to time series are covered in later chapters. Especially nice is a section comparing the most popular Python implementations of the algorithm, xgboost, lighting, and catboost. As a machine learning practitioner, I highly recommend this book.

I had high hopes for this book. Boosting models are widely used and deep material about them is lacking. This book is written by someone with great knowledge of boosting. But its as if you closed the person in a room and ask him to write, from memory alone, a not too complicated assay on boosting. The math is mentioned. No attempt at explaining. What's the point of implementing something whose underpinnings you do not understand? A modern decision tree is not even graphed or seriously analyzed. As I skimmed through I thought well, maybe there will be some words of wisdom on the comparison between xgboost, lightgbm and catboost. True for the time the book was written but still interesting. Nothing. Just more of this assay format of well you can use this and you can use that. No proper discussion even of how to format your X matrix between the three approaches. Too bad. The author is clearly knowledgeable on the topic. He just needs an editor from the community. Instead you get what seems to be a self published assay on boosting at a rather expensive price for its content.