The Kaggle Book: Data analysis and machine learning for competitive data science

A more accurate title would be "Tricks and tips for Kaggle competitions". In case you are thinking about participating in Kaggle competitions or already have, you should get this book. The book cannot guarantee that you'll win the competition. After all, winning requires a lot of creative thinking. But using common practices will definitely help you climb the leaderboard. The book is the first of its kind and I would definitely buy it for my home library. Often overly detailed, this is a great practical guide for Kaggle competitions, including Kaggle platform overview, many lines of Python code, strategies and best practices. It does not discuss machine learning algorithms or machine learning theory in general; for that, you should look for specialized machine learning theory books. However, the authors provide a list of most popular algorithms used in competitions, as well as their key features and most important parameters. The chapters about Bayesian Optimization and Blending/Stacking are probably the best I have seen so far. The book has a lot of blurbs with interviews from Kagglers, which is the most entertaining part for me. In my opinion, these blurbs can be converted to another great book. In spite of the fact that this book is fantastic now, I expect it to become outdated pretty quickly given the level of details provided by the authors and the pace at which machine learning is progressing. It would be wise for the authors to consider a new edition when it becomes outdated. I would recommend this book to people who are interested in machine learning competitions and are familiar with machine learning theory.

I received the Book on time and the Book Packaging was good. I'll also recommend Amazon to my Friends and Relatives.

I've been in ML industry for years, but still I learned lots of new things, thanks to this book.It is well explained what to do when competition metrics is not one of the standard ones in tensorflow or pytorch, in section "custom metrics, and custom objective funciton" and "post-processing".Adversarial validation to estimate difference between distributions of training and test datasetpseudo-labeling: to label some test datasetHyperparameter search with Halving (HalvingGridSearchCV, HalvingRandomSearchCV)Bayesian optimization using scikit-optimize, kerasTuner, TPE, optunaJoin the book’s Discord workspace for a monthly Ask me Anything session with the authors:Data augmentation for NLP (albumentations)This book also interviewed lots of Kaggle experts for in-depth insights and technical/professional tips.

At first, I received the wrong book for which I placed a replacement. Today I returned the wrong book and got the replacement of the right one.Coming on to the condition of the book:The pages are glossy and smooth as you can see in the pictures.The print is of great quality.The pictures inside the book might not be that clear but it is still readable. (notice the 2nd pic)

This is the first book that I've come across that is singularly focused on the rules, format, tips, and best practices for Kaggle ML/Data Science competitions. As such, this book is well-deserving of your dollars and attention.Before even delving into specific aspects of Machine Learning, the authors chose to spend a great deal of time (chapters 1-5) outlining the basics of Kaggle competitions from the history of the platform, to teams, datasets, notebooks, discussion forums, etiquette, and the different types of competitions available on the site. Complete beginners to Kaggle would get the most use of these chapters, it sure beats trying to figure all of this stuff out on your own.The remaining chapters start getting increasingly advanced in terms of subjects and techniques. I definitely appreciate the authors discussing the importance of the design of good model validation before delving deeper into hyperparameter tuning, walk before you run!The later chapters really drill into more advanced techniques such as using hyperparameter studies and Bayesian optimization to extract the best combination of values for your specific model. Ensembling and stacking are presented as clearly as I've seen anywhere, along with the most helpful snippets of code to date on a ML book. This alone might be worth the price for some. Intermediate and advanced users will get the most of these chapters.A nice extra is the Q&A sections in each chapter with "Kaggle Masters", people who have either won competitions in the past or who regularly place very high in many competitions. These are done informally and provide a lot of great tips.Now, who is this book really for? If you are new to Machine Learning, I'd say that perhaps this would not be the best place to start. While the book is great for what it sets out to do (teach you to become a better competitor) it is not perfect.Some information that could be helpful to beginners is grossly glossed over, such as the explanation of specific hyperparams. It is very odd how they chose to handle this. Case-and-point: when going over XGBoost hyperparams such as "n_estimators", they describe it as "usually an integer ranging from 10 to 5,000". Compare this with Corey Wade's explanation("Gradient Boosting with XGBoost and SciKit Learn", also from Packt ), "The number of trees in the ensemble/the number of trees trained on the residuals after each boosting round. Increasing might improve accuracy on larger datasets". Which is more useful you think? You either explain it clearly for the benefit of all or just leave it out. Giving the domain and range is not a proper substitution. Obviously, the author's expect the reader to have had some exposure to algorithms and modeling as the pace of several sections move a little too quickly for the complete beginner. As such, I would say this is a perfect book for semi-intermediate to advanced users looking to extract the most out of their models.All in all, this is an excellent resource that will be sure to help countless current and aspiring data scientists in their journeys to become masters of their crafts. I wish I had access to this text five years ago...Highly Recommended!