I have finally found some time to read this free ebook offered by PyTorch’s team. Therefore, I thought this was a good occasion to give you a quick review about the book. The book is free to download here
Who is this book for ?
This book was written for people who are already performing in the datascience industry. Also, this book does not cover the theorical principles of Deep Learning. Thus, if you are new to deep learning, or if you have no experience with the PyStack (numpy, pandas, etc.), the cover may a bit tough.
But if you are already familiar with neural networks and the PyStack, this book will be an excellent first start with PyTorch.
About the book content
Regarding neural networks, and specially deep learning, notions can be easily hard to understand. The authors have made the effort to add many graphics, and hand-made drawings, which give a friendly style to the book, and obviously makes the book much easier to read. The book is split into chapters that cover the following parts :
- Tensors: what it is, how to use it
- Tensors: how to work with specific data, such as images, text, or time series
- Neural Network training: from fast-forward to back-propagation
- Fitting the model to data: hyper-parameters, activation functions, and loss calculation
You can consider that this book covers the main topics that one should necessary know in order to train a model with CNNs for instance. The authors could have gone deeper, but regarding the fact that the book is free, this is still a great deal!
Should you read it ?
I think this book is clearly worth reading, provided that you have the minimum background. Topics are well documented, illustrated and a lot of simple code is also added to make it all more explainable.
After reading the book, I could train a basic neural network in a Jupyter Notebook, and evaluate the accuracy of my model with calculation of the training loss. The was not straight-forward, but I made it. And this is precisely what I was expecting from this book.
Keep also in mind that for a free book, you will get 100 pages to get you hands on PyTorch, and this is a not so expensive way to discover PyTorch 🙂
You can download it here