Recently, a lot of friends have written to me, mainly about two questions. One question is what books should I read if I want to start machine learning? The second question is, are there any benefits to share? Recently I happened to be doing some sorting work, so I took some time to sort out all the e-books I had read before, and selected a few books of high quality which I personally think to recommend to you.

These books are ranked in order of douban rating from highest to lowest.

The first Python deep learning book


This book is less famous than other basic books, but it is very popular abroad. It was written by the famous Chollet, the author of the Keras framework, known as the father of Keras, and a top bull in the industry. The quality of a book written by such a big bull can be imagined.

I read this book last year, and it took me more than a month to read and practice it. I remember clearly that I did not touch the PS4 for more than a month, so you can imagine its charm. The book is strictly in the realm of deep learning, with little to no traditional machine learning content. As a result, it is more focused, introducing almost all the commonly used algorithms and models in the field of deep learning, and also attaching detailed code demonstrations. The code demos and case studies are excellent and enjoyable to follow. And the book uses Keras framework implementation, the overall model is very straightforward, no redundant code, no anti-human design. It’s a two-for-one deal to learn deep learning and to get started with Keras.

The downside, if any, is Keras, which is relatively less popular than TensorFlow and Pytorch. So AFTER I finished, I hardly ever used Keras again, so I discounted it a little bit. But now that TensorFlow2.0 has Keras embedded in it, learning Keras also helps with learning TensorFlow2.0, so it’s not too much of a problem.

Douban score: 9.5

Recommended index: Takes in the following information

Scikit-learn and TensorFlow machine Learning Practical guide


This is a book by the great OREILLY Publishing house, which is very well known in the industry and produces almost all the best books. Different from other books that cover everything, there are not many models involved in this book. I checked the catalogue and found only LR, SVM, decision tree, random forest, DNN and CNN. In fact, the core models of machine learning are the ones we use most often in our daily lives. Each model in the book is explained in great depth, with formulas, principles, and code demonstrations. In fact, I used the demo data from the book in my previous articles.

Another good thing about this book is that it’s very well explained around the concept of machine learning, which is not just building or implementing a model and then using it. In the real industrial scenario, the machine learning model is only a small part, and algorithm engineers are faced with the whole link from data to application to reflux and iteration. There are very few books that go beyond the model, but it’s actually very important. The book does this very well, and if you’re a beginner, you’ll come away with a much better understanding of machine learning itself. Although the model may not be very many, the use of tool library is not proficient, but at least in the field of machine learning cognition is qualified.

There are advantages and disadvantages. One of the disadvantages is that the Chinese version is relatively few. I only found the open source version translated by netizens. Relatively speaking, the quality of typesetting will be poor, and some words are not as clear as English. Another problem is that the deep learning section is relatively superficial, and many models are not deep enough.

Douban score: 9.2

Recommended index: Takes in the following information

The third statistical learning method

img

I have been in touch with this book for a long time. It was recommended to me by a big bull when I was an intern in my junior year. I put it in kindle before I finished the first chapter. Some time ago, I took it out and reviewed it again, completely overturning my previous cognition. There is nothing wrong with this book, just that I was too lazy before. In fact, this book is very high status in the industry, and the quality of the content is very good.

The title of this book doesn’t seem to have anything to do with machine learning. Actually, it’s not. It’s statistics, but it’s all about machine learning methods. Probably because it was so old, before machine learning was a buzzword, these algorithms were relegated to statistics. This book has a total of more than 200 pages, covering more than a dozen models, with an average of more than 20 pages per model. Each model is clearly explained and basically all the knowledge points are covered with high quality. As you can see, this book is very dry and refined.

But its disadvantages are also obvious. The whole book is unreadable and very easy to discourage. It’s hard for beginners to do it all in one sitting because it’s so much like notes or lecture notes. And there is no demo related to the code, pure theory, after reading it to practice will feel at a loss to start. I personally feel that it is more suitable for checking and filling gaps as a reference book, convenient to refer to at any time.

Douban score: 9.0

Recommendable index: Takes place outside China

The fourth watermelon book


This is a star book, in the industry is also famous, nTU famous professor Zhou Zhihua wrote, because the cover is full of watermelons called watermelon book. His artificial intelligence laboratory is very famous, there are postgraduate entrance examination or postgraduate students can consider it. Before, my friends always joked with me that because of Zhou Zhihua, the School of Computer science in NTU has almost become the school of artificial intelligence.

Professor Zhou is well known and the quality of his books is certainly good. This book has 440+ pages, including more than ten algorithms of machine learning, not only explains algorithms, data processing, model evaluation methods and some common indicators are also covered. There are footnotes in the footer for some terms or technical terms, detailed illustrations and derivation of formulas, and, like Statistical Learning Methods, it is full of dry stuff.

As with “Statistics,” too much dry stuff can lead to indigestion, especially for beginners. The no. 1 comment on Douban pointed out that the book is inappropriate as a primer for self-study, and I didn’t actually rely on the book to get started. Some netizens said that this book is a collection of professor Zhou’s usual teaching content, so it may be better to coordinate with the teaching effect, and it will be difficult to understand just by reading. In addition, the translation of some symbols and terms in this book is not rigorous enough. I can’t think of any examples at the moment. If you have read foreign English materials and then read this book, you must have such a feeling.

To sum up, this book is a good one, but not very suitable for beginners to teach themselves how to get started. Need to look up a lot of information in the process of reading.

Douban score: 8.6

Recommendable index: Takes place outside China

The fifth machine learning actual combat

img

This book is also well known, and I have recommended it on Zhihu. There’s actually some debate online about the quality of this book, and I personally have read it twice, the first time a few years ago when I was getting started. It was amazing to read at the time, because every model in the book had a detailed code implementation that could be clicked and run locally. In addition, some models are explained in detail, such as SVM. Besides rigorous formula derivation, SMO algorithm is also implemented. It’s very beginner friendly at this point.

But when I read the book for the second time this year, something changed, and the biggest feeling was that the code in the book was really bad. It’s not half bad. It’s very bad. There are always some confusing acronyms, some of the code doesn’t look too lean, there is obviously a more elegant way to write it, and there is almost no object oriented section, the author wrote it as if it were process oriented. Another big problem is that some of the book is out of date, and some models, such as decision trees, are not covered in depth enough. There is also no mention of the use of the popular SkLearn library, leaving readers to find their own apis.

HHHHH said so much, feel black much more than boast. I didn’t want to put this book up, but then I saw the first comment on this book in Douban: A qualified reader should know what he needs. I thought about it and it made a lot of sense, and it’s actually a pretty good book for a clear-headed reader who knows what they want. If you just want a quick fix, a shortcut, and a couple of days to become a “good user” of machine learning models, then it’s obviously not for you, and if you want to dig a little deeper and know what the code implementation behind those models looks like, then you’ll probably have a good time (regardless of the code quality).

Douban score: 8.1

Recommendable index: Takes place outside the country

conclusion

To summarize, I recommend reading sciKit-Learn and TensorFlow in the following order: Machine Learning in Action > Python Deep Learning > Statistical Learning Methods > Watermelon. Of course, you can also read the other way around, from difficult to easy, and get a feeling of going from hell to heaven [fog].

I personally think that if you can finish reading these books and try out the code for yourself, I don’t think it will be a problem for you to get started with machine learning. If you have better books or materials, you can also share them with me in the background or in the comments.

Finally, I will share the open source materials of these books with you in the background of the official account reply machine Learning Introduction.

If you like this article, if you can, please click the following, give me a little encouragement, but also convenient access to more articles.

This article is formatted using MDNICE