Learn Keras for deep neural networks: a fast-track approach to modern deep learning with Python

✍ Scribed by Moolayil, Jojo John

• Publisher: Apress
• Year: 2018;2019
• Tongue: English
• Leaves: 192
• Category: Library

✦ Synopsis

Learn, understand, and implement deep neural networks in a math- and programming-friendly approach using Keras and Python. The book focuses on an end-to-end approach to developing supervised learning algorithms in regression and classification with practical business-centric use-cases implemented in Keras.

The overall book comprises three sections with two chapters in each section. The first section prepares you with all the necessary basics to get started in deep learning. Chapter 1 introduces you to the world of deep learning and its difference from machine learning, the choices of frameworks for deep learning, and the Keras ecosystem. You will cover a real-life business problem that can be solved by supervised learning algorithms with deep neural networks. You'll tackle one use case for regression and another for classification leveraging popular Kaggle datasets.

Later, you will see an interesting and challenging part of deep learning: hyperparameter tuning; helping you further improve your models when building robust deep learning applications. Finally, you'll further hone your skills in deep learning and cover areas of active development and research in deep learning.

At the end ofLearn Keras for Deep Neural Networks, you will have a thorough understanding of deep learning principles and have practical hands-on experience in developing enterprise-grade deep learning solutions in Keras.

What You'll Learn

Master fast-paced practical deep learning concepts with math- and programming-friendly abstractions. Design, develop, train, validate, and deploy deep neural networks using the Keras framework Use best practices for debugging and validating deep learning models Deploy and integrate deep learning as a service into a larger software service or product Extend deep learning principles into other popular frameworksWho This Book Is For

Software engineers and data engineers with basic programming skills in any language and who are keen on exploring deep learning for a career move or an enterprise project.

✦ Table of Contents

Table of Contents......Page 4
About the Author......Page 8
About the Technical Reviewer......Page 9
Acknowledgments......Page 10
Introduction......Page 11
Introduction to DL......Page 14
Demystifying the Buzzwords......Page 15
Decomposing a DL Model......Page 18
Exploring the Popular DL Frameworks......Page 21
Theano......Page 22
MxNet......Page 23
High-Level DL Frameworks......Page 24
A Sneak Peek into the Keras Framework......Page 26
Training the Model and Making Predictions......Page 28
Summary......Page 29
Selecting the Python Version......Page 30
Installing Python for Windows, Linux, or macOS......Page 31
Installing Keras and TensorFlow Back End......Page 32
Input Data......Page 34
Neuron......Page 36
Activation Function......Page 37
Sigmoid Activation Function......Page 38
ReLU Activation Function......Page 39
Layers......Page 41
Dense Layer......Page 42
Dropout Layer......Page 43
Other Important Layers......Page 44
The Loss Function......Page 45
Optimizers......Page 48
Adam......Page 50
Other Important Optimizers......Page 51
Model Configuration......Page 52
Model Training......Page 53
Model Evaluation......Page 56
Putting All the Building Blocks Together......Page 58
Summary......Page 65
Getting Started......Page 66
Problem Statement......Page 68
Why Is Representing a Problem Statement with a Design Principle Important?......Page 69
Designing an SCQ......Page 70
Designing the Solution......Page 72
Exploring the Data......Page 73
Looking at the Data Dictionary......Page 76
Finding Data Types......Page 79
Working with Time......Page 80
Predicting Sales......Page 82
Exploring Numeric Columns......Page 83
Understanding the Categorical Features......Page 87
Data Engineering......Page 91
Defining Model Baseline Performance......Page 97
Designing the DNN......Page 98
Improving the Model......Page 102
Increasing the Number of Neurons......Page 106
Plotting the Loss Metric Across Epochs......Page 110
Testing the Model Manually......Page 111
Summary......Page 112
Getting Started......Page 113
Problem Statement......Page 114
Designing the Solution......Page 115
Exploring the Data......Page 116
Data Engineering......Page 122
Defining Model Baseline Accuracy......Page 130
Designing the DNN for Classification......Page 131
Standardize, Normalize, or Scale the Data......Page 136
Transforming the Input Data......Page 138
DNNs for Classification with Improved Data......Page 139
Summary......Page 146
The Problem of Overfitting......Page 148
So, What Is Regularization?......Page 150
L2 Regularization......Page 151
Dropout Regularization......Page 152
Hyperparameter Tuning......Page 153
Number of Neurons in a Layer......Page 154
Number of Layers......Page 155
Weight Initialization......Page 156
Learning Rate......Page 157
Approaches for Hyperparameter Tuning......Page 158
Grid Search......Page 159
Further Reading......Page 162
Tailoring the Test Data......Page 163
Saving Models to Memory......Page 165
Retraining the Models with New Data......Page 166
Online Models......Page 167
Delivering Your Model As an API......Page 168
Putting All the Pieces of the Puzzle Together......Page 169
Summary......Page 170
What’s Next for DL Expertise?......Page 171
CNN......Page 172
RNN......Page 177
CNN + RNN......Page 180
Why Do We Need GPU for DL?......Page 181
Other Hot Areas in DL (GAN)......Page 184
Concluding Thoughts......Page 186
Index......Page 187