Deep Learning with PyTorch

by Eli Stevens, Luca Antiga

Every other day we hear about new ways to put deep learning to good use: improved medical imaging, accurate credit card fraud detection, long range weather forecasting, and more. PyTorch puts these superpowers in your hands, providing a comfortable Python experience that gets you started quickly and then grows with you as you—and your deep learning skills—become more sophisticated. Deep Learning with PyTorch will make that journey engaging and fun.

Genres: Programming, Artificial Intelligence, Computer Science, Technology, Nonfiction, Mathematics, Technical

450 pages, Paperback

First published January 1, 2020

Reviews

[JJ Khodadadi · Rating 4 out of 5]

عالی بود
بنظرم منبع خیلی خوبی هست برای کار

[Sebastian Gebski · Rating 3 out of 5]

This book is full of contrasts. Some things it does tremendously well, some things it ... well, let's say it's far from succeeding in those. What precisely?

1. It's not a book about Deep Learning but Convolutional Neural Networks - so just a subset of DL. It's still a capacious topic, but the title is misleading. HOWEVER, I quite liked how it tackled the intro to CNNs.
2. If you expect to learn PyTorch here, I need to disappoint you, my friend ... Yes, you'll get familiar with some basics, but then you'll get thrown into very specific scenarios - which makes an overall impression that you learn just a fraction of the library, w/o getting familiar with its overall structure, philosophy, etc. I hate learning that way (call me old-fashioned if you want).
3. The practical part of the book (cancer detection scenario) also does some things very well: you can take a look at the overall composition of the solution that actually does resemble a real product. But ... the way you're getting navigated here makes me wonder - am I really taught here? Or demoed? Will I be able to create a similar solution for a different problem? Hmmm ...

So, to cut a long story short - I'm not really sure who this book is for. Who would benefit most from it? I don't know - in my case, I'm certain: I didn't find here what I was looking for.
If you don't mind this kind of learning (scenario-based) - feel free to add 1 star to the final rating.

[bimri · Rating 4 out of 5]

My best word on this, is this, part II of the book will build your intuitions on deep learning like no other book - I have encountered(I've been around the block).

The lung cancer case study really does a number on your brain... you trying to dissect ways on how to build a model(s) for detecting nodules on relatively huge dimension spaces. Which also led me to respect the work of radiologists. It's no small feat to recognize "these needles in haystacks". Chances of a deep learning model to accomplish this with orders of magnitude search spaces: is what lead me into this field in the first place.

This book does a very decent job to make sure you get the lay of the land. And, how to go around making meaningful explorations on your own.

If not to educate you; it does also show the complexity of real world problems. It's such a hustle - from the context of this challenge put forth on the book.

On the other hand: it only focuses on CNNs - so if you came to this book as a novice - I'd shun from the urge to begin reading it. It requires some intermediate levels of machine learning or deep learning; in order to appreciate its power.

With that said, I would be looking forward to the second edition; in order to get more insights to what is to come from its part III revamped (that is the section dealing with deployment). It was a little bit underwhelming knowing that there's more to be done in this space!


Things that didn't get lost on me:
1. The jokes(they were spot on!),

2. The illustrations at the beginning of every chapter(these were priceless - they really strengthen your intuitions and help with your grasp of the knowledge data points!),

3. The emphasis side note on discrimination("racism is learnt!" the authors didn't have to point that one out - based on the context of that subject matter in that section - but all the same went a long way to hold a stand with the Black Lives Matter movement.) That was quite something!

[Vicki · Rating 5 out of 5]

This is a really excellent and comprehensive view into deep learning with a PyTorch specific focus. The first part covers all the fundamentals you need to understand for ML with deep learning layers as well as PyTorch idioms and the second part goes into a project.

The project itself was not so relevant to me because I’m working mostly with text for my project but the first part is required reading and I got so much out of it. The third part on serving is fairly small but also gives a good overview of models in prod.

[Jeff · Rating 2 out of 5]

If you are competent with Python and understand deep learning, this is not the book for you. Waaaaay too much fluff. Honestly, you're probably 10x better off just opening an open source PyTorch project in one tab and the PyTorch documentation in another.

[Roy · Rating 2 out of 5]

The authors clearly know their business. They are very knowledgeable of the field and provide many tips along the way on how you'll be able to implement what you've read to your own projects.

However, not everyone with knowledge is also a good teacher. The book is extremely nondidactic. Yes, it teaches you the logic and math behind PyTorch. However, it has three major faults:
1. It always tries to find the longest path from A to B - this includes an enormous amount of unnecessary text, which could definitely be shortened and more to the point.
2. The best way to learn a programming language is by doing. Sadly, very little emphasis is given to the actual coding. The authors provide notebooks and mention modules that you should be using, but the actual practice is in the form of a few very short questions/exercises at the end of chapters.
3. The book does very little to build knowledge gradually. There is minimal mention of previous steps in previous chapters. Given the amount of extraneous text, it's very difficult to keep a tab on what was taught, and the book doesn't do much to help with that.

Part 2 of the book, the "practical" part, is a good example of points 1 and 2 above. It goes into extreme detail about a very specific project which might not be relevant to all readers, and seems to try very hard to teach new concepts in the most cumbersome way possible. The authors should stick to the title of the book rather than employ the most complex Python techniques to deliver their points. Otherwise, it's akin to using French to teach Chinese.

Admittedly, I haven't read the whole book. I gave up somewhere along part 2. In my defence - I really tried to like it.

In summary, my idea of a good book is one that makes complex ideas easy to understand. For me, this book made the ideas more complex, unnecessarily.

[Tolga Karahan · Rating 4 out of 5]

It's a good book if you're a beginner, so I just skipped basic stuff and only read parts that most interested me. If you a beginner to Deep Learning, and PyTorch it's a useful book.

[Bemmu · Rating 4 out of 5]

This undertakes a rather ambitious project: detecting lung cancer from CT scans by training a segmentation model to find possible cancer nodules, and then a classifier to determine if they are malignant or not.

Very practical, showing and discussing in detail in Python all the data mangling necessary to get in the data. But can get quite involved if you are like me just following along by reading instead of actually doing all the exercises, as the second part of the book is essentially one long exercise.

The final chapter discusses deployment, which was the most educational for me. How to serve a model on a website so that it doesn't block web requests, and how to deploy it in C++ or to a mobile device.

[Abdulkader Abdulrazzak · Rating 4 out of 5]

Although the book is great, the title should be convolutional neural networks with Pytorch and not deep learning. I can't say I'm disappointed. Brick by brick, the book walks you through building your NNs. Introduces you the historical problems by exposing you to them, then goes over the solutions and their implementations in Pytorch. After finishing the building blocks, you move to solve (an unsolved) complex real world problem (which requires breastly specs and a lot of storage).

If you're looking to go fast over the framework, this is not the book for you as it goes into the small details and very slowly builds up the facts.

Overall a great introduction for CNNs. I loved the jokes.

[Xianshun Chen · Rating 4 out of 5]

The part 1 and part 3 of the book meets my needs as I am familiar with concepts of deep learning as well as frameworks such as Tensorflow and MXNet but not so familiar with PyTorch ands basic operations in PyTorch. The part 2 is a bit redundant for me as most of the techniques covered in this part overlaps with what i already know

[eugene_ · Rating 3 out of 5]

This book provides a solid introduction to the fundamentals of Deep Learning, with a particular focus on Computer Vision and PyTorch. It is highly beneficial for beginners who are new to the field of Deep Learning or PyTorch. However, if you already have some experience, this book might not offer much new information for you.

[Simon · Rating 4 out of 5]

All time classic, the first half of the book is simply brilliant with great figures and explanations. But the second half it horrible in my opinion. It is hard to follow, and way too complicated for beginners.

[Hannah · Rating 5 out of 5]

A book on basic instructions of PyTorch.

[Silentroar · Rating 4 out of 5]

Four stars cause the complete version has not been published.

[Paolo Quadri · Rating 5 out of 5]

End to end guide to Pytorch, from tensor basics to production

[Youi · Rating 5 out of 5]

Loved it. Cannot wait for the authors' next book, especially one with recipes and techniques alluded to in the end.

[Kamal · Rating 4 out of 5]

First part is very well written and was exactly what I was looking for but the second part has alot of irrelevant stuff that makes reading and following a pian, I understand the authors intended to give an example of a real world project but they could choose a more intuitive problem and dataset, anyway the first part was good enough for me to give it four stars