Book Sharing of "Hands-on Deep Learning (PyTorch Edition)" 1: Overview of the Book

kit7828

Book Sharing of "Hands-on Deep Learning (PyTorch Edition)" 1: Overview of the Book [Copy link]

Thanks again to EEWord for providing the opportunity to read and share books.

1. Publication Information

"Hands-On Deep Learning (PyTorch Edition)" was edited by Aston Zhang and others, translated by He Xiaoting and others, and published by People's Posts and Telecommunications Press in February 2004. The book has 572 pages and 972,000 words.

The homepage of the book is printed in black and white. In addition to the cover, there are 4 pages of coated paper with many pictures printed on them, including deep learning training, test pictures, and effect visualization pictures. The printing quality is very good.

II. Praise for this book

The book has received praise from many big names, and several academicians have written recommendations. The author should have a very strong background. The one I am more familiar with is Zhou Zhihua, the author of Xigua Book.

Recommendations from the industry include NVIDIA's Huang Renxun, Horizon Robotics' Yu Kai, and others

3. Table of Contents and Book Structure

The catalog and book structure are mainly composed of three parts with different color differences, but on the black and white printed pages, the color levels are not obvious enough.

Compared with the color picture provided on the website, it is very different

Part 1 (Chapter 1 to Chapter 4): Covers the basics and preparatory knowledge. Chapter 1 provides an introductory course in deep learning. Chapter 2 quickly introduces the prerequisites required to practice deep learning, such as how to store and process data, and how to apply various numerical operations based on basic concepts such as linear algebra, calculus, and probability. Chapters 3 and 4 cover the basic concepts and techniques of deep learning, such as linear regression, multilayer perceptrons, and regularization.

Part 2 (Chapters 5 to 10): Focuses on modern deep learning techniques. Chapter 5 describes the various key components of deep learning computations and lays the foundation for our subsequent implementation of more complex models. Chapters 6 and 7 introduce convolutional neural networks (CNNs), powerful tools that form the backbone of most modern computer vision systems. Chapters 8 and 9 introduce recurrent neural networks (RNNs), which are models that exploit temporal or sequential structure in data and are commonly used in natural language processing and time series forecasting. Chapter 10 introduces a new class of models that employ a technique called attention mechanisms, which have recently begun to replace recurrent neural networks in natural language processing. This section will help you quickly understand the basic tools behind most modern deep learning applications.

Part III (Chapters 11 to 15): Discusses scalability, efficiency, and applications. Chapter 11 discusses several commonly used optimization algorithms for training deep learning models. Chapter 12 explores several key factors that affect the computational performance of deep learning codes. Chapter 13 shows the main applications of deep learning in computer vision. Chapters 14 and 15 show how to pre-train language representation models and apply them to natural language processing tasks.

4. Learning environment configuration

The code related to this book needs to be run in a Python environment, so you need to configure the corresponding development environment, mainly including the installation of MiniConda, the installation of PyTorch, the installation of the D2L software package, and the acquisition of the book's supporting code, etc.

Since AnaConda is installed on my computer, I first checked the differences between MiniConda and

conda is a package and environment management tool that can not only manage packages, but also isolate and manage environments of different Python versions. It is similar to the nvm tool that manages the nodejs environment.

Anaconda and miniconda are both distributions of conda, but they contain different packages.

Anaconda includes more than 180 scientific packages such as conda and python and their dependencies, and is relatively large. But you may not use many of them, so there is a mini version.

miniconda is the smallest conda installation environment, which only contains conda+python+pip+zlib and some other commonly used packages, and is very small.

Pip is also called a package manager. The difference between it and conda is that pip only manages Python packages, while conda can install packages of all languages. Moreover, conda can manage Python environments, but pip cannot.

The website of MiniConda is https://docs.anaconda.com/miniconda/ . You can download it directly. Just pay attention to the operating system of your computer.

The website of PyTorch is https://pytorch.org /. When downloading, you need to confirm your operating system and whether your graphics card has a GPU.

Direct installation code is given for both with and without GPU:

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip3 install torch torchvision torchaudio

Since the book was published earlier, the versions used are CUDA 1.12 and CPU 0.13.

D2L software can be downloaded directly using PiP in the Python development environment

pip install d2l==0.17.6

The code for D2L is https://zh-v2.d2l.ai/d2l-zh-2.0.0.zip and can be downloaded directly

5. Electronic version of books

When searching for the D2L software package, I accidentally found the electronic version of this book.

The website is: https://zh.d2l.ai/index.html