Full of lengthy code examples, PyTorch recipes is a very good textbook for beginner and intermediary neural network developers using PyTorch.

The book chapters are structured identically: introduce a topic and a problem (usually a technical one); describe a solution to it; and then present the code that goes together with the solution. The topics covered include probability distribution, convolutional neural networks (CNNs), recurrent neural networks (RNNs), neural networks overall, supervised learning, deep learning models, natural language processing (NLP), distributed modeling, model optimization and deployment, data augmentation and feature engineering, and sklearn. All of these topics are extensively discussed and presented in light of how they are tackled in PyTorch.

The book’s introduction briefly compares PyTorch and TensorFlow. If fails to provide a clearcut analysis of which platform should be used for which tasks, instead giving a summary of the differences between the two platforms--these are minimal.

The sections in the chapters provide step-by-step guidance on how to use and adopt the printed code. However, they do not provide very deep insights into the problems to be addressed, nor explanations about the code examples themselves. That is why this book is qualified as a hands-on practical tutorial to using PyTorch, which corresponds to the word “recipes” in the title. The fact that the extensive code examples are printed and not provided in digital form may render the usage of the insights taught a bit difficult.

That being said, the book covers all important facets of neural network implementation and modeling, and could definitely be useful to students and developers keen for an in-depth look at how to build models using PyTorch, or how to engineer particular neural network features using this platform.