What do you think?
Rate this book
PyTorch Pocket Reference: Building and Deploying Deep Learning Models
This concise, easy-to-use reference puts one of the most popular frameworks for deep learning research and development at your fingertips. Author Joe Papa provides instant access to syntax, design patterns, and code examples to accelerate your development and reduce the time you spend searching for answers. Research scientists, machine learning engineers, and software developers will find clear, structured PyTorch code that covers every step of neural network development-from loading data to customizing training loops to model optimization and GPU/TPU acceleration. Quickly learn how to deploy your code to production using AWS, Google Cloud, or Azure and deploy your ML models to mobile and edge devices.
- GenresTechnology
307 pages, Paperback
Published June 15, 2021
4 people are currently reading
19 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 2 of 2 reviews
July 28, 2023
A succint guide beyond a reference
As an AI Engineering Manager, I found this pocket reference to be highly informative. It goes beyond being just an API reference, to provide solid usage guidance in a summarized and concise to-the-point manner.
Historically I learned TensorFlow in depth, and have made efforts to aclimatize to the paradigm shift of JAX, but it cannot be ignored that PyTorch dominates research (just look at PapersWithCode, HuggingFace). So more often than not, I find myself reading PyTorch transformer code and mentally mapping the diff. This book is a great boon for my understanding (and for productization migrations): there is only so much time you can invest in learning multiple frameworks that do similar things - this book afforded me the ability to rapidly grasp the delta.
Obviously with the velocity of ML research, any such book is already outdated before it hits the printing press - most of the links are dead, and we are now at PyTorch v2 (which in fairness is mainly internal optimization changes).
I really like the way that different API subsets were categorized. What could be added to such a reference would be comparisons of alternative algorithms (eg of loss functions, norm layers, activation functions) and of competing ecosystem frameworks.
As a side note, its unfortunate that O'Reilly no longer publishes many pocket refs for nieche tech stacks.
I am happy to have purchased this and I also appreciated the clear, non-woffling writing style and holistic perspective.
As an AI Engineering Manager, I found this pocket reference to be highly informative. It goes beyond being just an API reference, to provide solid usage guidance in a summarized and concise to-the-point manner.
Historically I learned TensorFlow in depth, and have made efforts to aclimatize to the paradigm shift of JAX, but it cannot be ignored that PyTorch dominates research (just look at PapersWithCode, HuggingFace). So more often than not, I find myself reading PyTorch transformer code and mentally mapping the diff. This book is a great boon for my understanding (and for productization migrations): there is only so much time you can invest in learning multiple frameworks that do similar things - this book afforded me the ability to rapidly grasp the delta.
Obviously with the velocity of ML research, any such book is already outdated before it hits the printing press - most of the links are dead, and we are now at PyTorch v2 (which in fairness is mainly internal optimization changes).
I really like the way that different API subsets were categorized. What could be added to such a reference would be comparisons of alternative algorithms (eg of loss functions, norm layers, activation functions) and of competing ecosystem frameworks.
As a side note, its unfortunate that O'Reilly no longer publishes many pocket refs for nieche tech stacks.
I am happy to have purchased this and I also appreciated the clear, non-woffling writing style and holistic perspective.
November 8, 2023
A good starting point for anyone who already has a grasp of deep learning, is proficient in object-oriented programming in Python, and wants to get started with Pytorch for Computer Vision.
Good:
- clear, actionable examples
- lots of resources and pointers for further learning
- clear writing style
- good learning curve: starts very simple and gets increasingly complex
Not so good:
- I found the NLP part very lacking and limited; this book has a very strong focus on computer vision. I get that NLP is a particular field, but I'd have liked some more examples and explanations about data loaders for text data, loading/fine-tuning of Huggingface models, etcetera.
- Sometimes I'd like more information about what kinds of data types certain functions and methods expect
- Python code gets hard to read when it is spread over multiple pages, because it is hard to see how much certain blocks are indented; in some examples, this made it hard to see which parts of the code were within a loop or function.
Good:
- clear, actionable examples
- lots of resources and pointers for further learning
- clear writing style
- good learning curve: starts very simple and gets increasingly complex
Not so good:
- I found the NLP part very lacking and limited; this book has a very strong focus on computer vision. I get that NLP is a particular field, but I'd have liked some more examples and explanations about data loaders for text data, loading/fine-tuning of Huggingface models, etcetera.
- Sometimes I'd like more information about what kinds of data types certain functions and methods expect
- Python code gets hard to read when it is spread over multiple pages, because it is hard to see how much certain blocks are indented; in some examples, this made it hard to see which parts of the code were within a loop or function.
Displaying 1 - 2 of 2 reviews