After reading Kaggle ni Idomu Deep Learning Programming no Gokui

I finished reading the book in the title, so here are my impressions. In one sentence, it is a good, easy-to-understand book that compresses the essence of machine learning into a short "index", whether or not you care about Kaggle. I recommend it both to machine learning beginners and Kaggle beginners. I am technically a Kaggle Competitions Master, and even so it helped me understand several things that I had only understood shallowly, and it introduced techniques I did not know. Machine learning engineers and people familiar with Kaggle should also get a lot from it.

The book, Kaggle ni Idomu Deep Learning Programming no Gokui, was given to me by one of its authors, Shotaro Ishihara. Thank you.

---

One of the good things about this book is that it is short. Excluding the index, it is about 200 pages. Many machine learning books are thick, so this one is nicely compact. Books that explain theory properly tend to become long because they need careful explanation, and bad books often become long because they explain difficult material in a confusing and redundant way.

This book explains things accurately and simply, so you can get a broad overview. Looking at the book's table of contents, it covers many techniques needed not only for Kaggle but also for practical machine learning in general, and it explains why each technique is needed and when to use it. From chapter 3 onward, the book applies techniques and ways of thinking to actual Kaggle competition tasks and improves the score, so it is easy to see that the methods really have an effect. If you want more detail, you can follow the URLs and papers scattered throughout the text. It also gives you important keywords, so you can deepen your knowledge through search or other books.

The main models it covers are also models that are strong enough to consider early in practical work: gradient boosted decision trees, neural networks such as CNNs, RNNs, and Transformers, and linear models for ensembles. This focused selection is probably one reason the book does not become redundant.

If I had encountered this book when I first started learning machine learning, I could have learned about models and algorithms that are good in terms of performance without wandering too far off course, and I could have learned simply how to look at data and how to validate models. Beginners will not understand everything just by reading it lightly, but at the beginning you often do not even know which keywords matter. This book exposes you to many keywords that feel important and worth looking up when needed. That alone is valuable.

If I had encountered it when I started Kaggle, it would also have answered many of the questions that come up when beginning Kaggle competitions. Fortunately, my first Kaggle competition was a team effort, and much stronger teammates taught me the basics, so I was able to understand the overall flow. With this book, I think I could have grasped the flow of Kaggle and started my first competition with more understanding.

As I wrote at the beginning, this is a good, easy-to-understand book that compresses the essence of machine learning into a short index, beyond Kaggle itself. I hope many people read it.