Data Science Book Reviews #002
I hope this helps the readers to choose the right book for their learning needs.
Book Review on the title: “Deep Learning using Python”
Authors: Dr. S. Lovely Rose, Dr. L .Ashok Kumar & Dr. D. Karthika Renuka
Edition: 1st edition@2019 — Reprint@2020
Publisher: Wiley India Pvt.Limited, New Delhi
List Price: Rs. 449/-
This book covers the basic concepts of neural networks, convolution neural networks, recurrent neural networks, autoencoder, restricted boltzmann machine, frameworks of deep learning. Few case studies of DL applications are also included. This book is good read for beginners to start with deep learning.
This book contains of 7 chapters. The first chapter introduces working of neural networks, perceptron and multilayer perceptron. This chapter is starting point to get to understand the important concepts such as activation functions, optimization, loss functions, learning rate, gradient descent algorithms, back propagation algorithm, weight regulation, etc.
Chapter 2 is about the building blocks of convolutional neural networks, different architectures and applications of CNN. Chapter 3 is devoted to recurrent neural networks (RNNs). Chapters 4, 5, 6 are introductory of autoencoder, RBM and frameworks of deep learning respectively. Last chapter: Applications of deep learning, comprises of 5 case studies of CNN and RNNs. Complete code of each of these case studies is printed in this chapter. Every important concept of case study is explained with separate section of codes.
Authors provided instructor resources on the home page of publisher: https://www.wileyindia.com
You need to register with official email on the site to access the solutions and power point presentations of this book.
Book Review Rating — 5/10
At the outset, I liked the way authors introduced the basics of the neural networks in the first chapter. Brief history of evolution of neural networks is good choice of authors to start with the artificial neural networks (ANN). The fundamentals of ANNs such as activation, learning algorithm, optimization, loss, gradient descent and back propagation are explained in layman’s language.
Convolving with filters (kernels), pooling, padding and feature maps are beautifully demonstrated with relevant examples and figures. The popular CNN architectures like LeNet, AlexNet, GoogleNet, VGG and ResNet are shown with diagrams. All diagrams are sourced either from research papers or third party sites.
Back Propagation through Time (BPTT) is the most important learning algorithm of RNNs. BPTT is elaborated with a good numerical example. Architectures of RNN, LSTM, GRU, Bi-LSTM are explained with proper diagrams.
Autoencoder is the one of best feature extraction /dimensionality reduction architectures. The theoretical explanation of how vanilla autocoder works is good. Various types of autoencoders mentioned. Restricted Boltamann Machines are of only introductory. The book mentioned about popular deep learning frameworks and hardware requirements for deep learning. Simple archetecture of model training in Keras API and Pytorch is elaborated with relevant codes. It gives you fair idea of using frameworks for training models with different frameworks.
Lastly, final chapter consists of 5 case studies — Image classification, Visual speech recognition, Stock market prediction, Next word prediction, and Tamil handwritten OCR. First four case studies are used Keras-TensorFlow and the last one is used Pytorch. Hence, you need to have some basic skills in these frameworks to understand and implement the codes. These case studies are good learning experience for the beginners, although. I suggest you to experiment with these models to learn further.
As figures are sourced not graphically reproduced, most of the CNN architecture diagrams are very messy. The print of CNN diagrams is also mediocre. It is difficult to understand the CNN architecture diagrams as they are poorly depicted. It is very difficult for the beginners to understand mathematical terms like Jacobian matrix. Majority mathematical expressions mentioned throughout this book are not explained well, hence it is not understandable unless you have much stronger skills in mathematics.
Print quality and binding of the book is good. Font and language is comfortable to the read. Photographs are in black and white. However, the book did not included the importing, transformation and processing of the data that is suitable for popular frameworks such as Pytorch and/or Keras-TensorFlow. Data processing is the first step in the machine learning life cycle. Due to the absence of this topics, the book lost essence of working with deep learning frameworks.
As documentation of Google TensorFlow 2.0 API on its site is very brief, it is cumbersome for beginners how to use utilities of the framework. If the book guided the readers how to navigate the TensorFlow API to choose the specific classes/functions, it would have definitely becomes one of the best books of deep learning to start with.
Overall, the book can be regarded as ready reckoner of basic theory. You would definitely learn lot of new concepts if you are beginner in deep learning. Most importantly, you can refer the codes whenever necessary written in this book, in case you are dealing with similar problems.
It’s very difficult to rate the technical books. However, in my opinion, the book can be rated at 5 on 1–10 scale.
You can explore other similar books: “Deep Learning with Python — Francois Chollet”(Next Book Review), “Deep Learning from Scratch: Building with Python from First Principles — Seth Weidman”, “Deep Learning for Coders with fastai and PyTorch — Jeremy Howard and Sylvine Gugger”, etc
If you find this book review useful, do care to give me a CLAP on the site — My goal is to reach 100 book reviews with your motivation.
Follow this series to keep posted more Data Science book reviews which terrifically useful for you without doubt, if you are a book lover!
