Hands-On Artificial Intelligence on Google Cloud Platform

Hands-On Artificial Intelligence

on Google Cloud Platform

Build intelligent applications powered by TensorFlow, Cloud

AutoML, BigQuery, and Dialogflow

Anand Deshpande

Manish Kumar

Vikram Chaudhari

BIRMINGHAM - MUMBAI

Packt.com

Subscribe to our online digital library for full access to over 7,000 books and videos, as well

as industry leading tools to help you plan your personal development and advance your

career. For more information, please visit our website.

Why subscribe?

Spend less time learning and more time coding with practical eBooks and Videos

from over 4,000 industry professionals

Improve your learning with Skill Plans built especially for you

Get a free eBook or video every month

Fully searchable for easy access to vital information

Copy and paste, print, and bookmark content

Did you know that Packt offers eBook versions of every book published, with PDF and

ePub files available? You can upgrade to the eBook version at www.packt.com and as a print

book customer, you are entitled to a discount on the eBook copy. Get in touch with us at

[email protected] for more details.

At www.packt.com, you can also read a collection of free technical articles, sign up for a

range of free newsletters, and receive exclusive discounts and offers on Packt books and

eBooks.

Hands-On Artificial Intelligence on Google

Cloud Platform

Copyright © 2020 Packt Publishing

All rights reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form

or by any means, without the prior written permission of the publisher, except in the case of brief quotations

embedded in critical articles or reviews.

Every effort has been made in the preparation of this book to ensure the accuracy of the information presented.

However, the information contained in this book is sold without warranty, either express or implied. Neither the

authors, nor Packt Publishing or its dealers and distributors, will be held liable for any damages caused or alleged to

have been caused directly or indirectly by this book.

Packt Publishing has endeavored to provide trademark information about all of the companies and products

mentioned in this book by the appropriate use of capitals. However, Packt Publishing cannot guarantee the accuracy

of this information.

Acquisition Editor: Devika Battike

Content Development Editor: Nazia Shaikh

Senior Editor: Ayaan Hoda

Technical Editor: Utkarsha S. Kadam

Copy Editor: Safis Editing

Language Support Editor: Storm Mann

Project Coordinator: Aishwarya Mohan

Proofreader: Safis Editing

Indexer: Rekha Nair

Production Designer: Joshua Misquitta

First published: March 2020

Production reference: 1050320

Published by Packt Publishing Ltd.

Livery Place

35 Livery Street

Birmingham

B3 2PB, UK.

ISBN 978-1-78953-846-5

www.packt.com

About the authors

Anand Deshpande has over 19 years' experience with IT services and product

development. He is currently working as Vice President of Advanced Analytics and

Product Development at VSquare Systems Pvt. Ltd. (VSquare). He has developed a special

interest in data science and an algorithmic approach to data management and analytics and

co-authored a book entitled Artificial Intelligence for Big Data in May 2018.

This book and anything worthwhile in my life is possible only with the blessings of my

spiritual guru, my parents, in-laws and the unconditional love and support from my wife

Mugdha and my daughters Devyani and Sharvari. Special thanks to my friends and co￾authors of this book, Manish Kumar and Vikram Chaudhari. And lastly, I would like to

thank Mr. Sunil Kakade and Mr. Ajay Deshmukh for mentoring and their support.

Manish Kumar works as Director of Technology and Architecture at VSquare. He has over

13 years' experience in providing technology solutions to complex business problems. He

has worked extensively on web application development, IoT, big data, cloud technologies,

and blockchain. Aside from this book, Manish has co-authored three books (Mastering

Hadoop 3, Artificial Intelligence for Big Data, and Building Streaming Applications with Apache

Kafka).

I would like to thank my parents, Dr. N.K. Singh and Dr. Rambha Singh, for their

blessings. The time spent on this book has taken some precious time from my wife, Mrs.

Swati Singh, and my adorable son, Lakshya Singh and I would like to thank them for their

support throughout this time. And lastly, I would like to thank my friends and co-authors,

Anand Deshpande and Vikram Chaudhari.

Vikram Chaudhari works as Director of Data and Advanced Analytics at VSquare. He has

over 10 years' IT experience. He is a certified AWS and Google Cloud Architect and has

completed multiple implementations of data pipelines with Amazon Web Services and

Google Cloud Platform. With implementation experience on multiple data pipelines across

platforms, Vikram is instrumental in creating reusable components and accelerators that

reduce costs and implementation time.

I would like to thank my mother, Mrs. Jyoti Chaudhari for her encouragement and

blessings, my wife Amruta Chaudhari for her continuous support and love, my precious

little gem Aahana, as she had to compromise on some of her weekend plans when I was

busy writing the book. Finally, I would like to thank my mentor, Anand Deshpande, who

always guided me and helped me to get to the next level in my career and, Mr. Manish

Kumar, who is the co-author and one of the best Architects I have worked with. Lastly, I

would want to thank the wonderful Packt team for their wonderful support throughout the

project.

About the reviewers

Arvind Ravulavaru is a full stack architect and consultant with over 11 years' experience in

software development and 2 years' experience in hardware and product development. For

the last 5 years, he has been working extensively on JavaScript, both on the server side and

the client side and, for the last few years, in IoT, AI, ML, and big data.

Alexey Bokov is an experienced Azure architect and has been a Microsoft technical

evangelist since 2011. He works closely with Microsoft's top-tier customers all around the

world to develop applications based on the Azure cloud platform. Building cloud-based

applications in challenging scenarios is his passion, along with helping the development

community to upskill and learn new things through hands-on exercises and hacking. He's a

long-time contributor to, and co-author and reviewer of, many Azure books, and, from time

to time, speaks at Kubernetes events.

Judy T Raj is a Google Certified Professional Cloud Architect with extensive experience

across the three leading cloud platforms—AWS, Azure, and the GCP. She has co-authored

a book on Google Cloud Platform with Packt Publishing. She has also worked with a wide

range of technologies, such as data science, deep learning, computer vision, blockchain, big

data, and IoT and has published many relevant academic papers. Currently employed as a

Data and AI Engineer, she is a passionate coder, a machine learning practitioner, and a

computer vision enthusiast.

Packt is searching for authors like you

If you're interested in becoming an author for Packt, please visit authors.packtpub.com

and apply today. We have worked with thousands of developers and tech professionals,

just like you, to help them share their insight with the global tech community. You can

make a general application, apply for a specific hot topic that we are recruiting an author

for, or submit your own idea.

Table of Contents

Preface 1

Section 1: Section 1: Basics of Google Cloud Platform

Chapter 1: Overview of AI and GCP 8

Understanding the Cloud First strategy for advanced data analytics 9

Advantages of a Cloud First strategy 10

Anti-patterns of the Cloud First strategy 12

Google data centers 13

Overview of GCP 15

AI building blocks 16

Data 17

Storage 17

Processing 18

Actions 18

Natural language processing 18

Speech recognition 19

Machine vision 19

Information processing and reasoning 19

Planning and exploring 20

Handling and control 20

Navigation and movement 20

Speech generation 21

Image generation 21

AI tools available on GCP 21

Sight 21

Language 23

Conversation 24

Summary 25

Chapter 2: Computing and Processing Using GCP Components 26

Understanding the compute options 27

Compute Engine 27

Compute Engine and AI applications 28

App Engine 29

App Engine and AI applications 29

Cloud Functions 30

Cloud Functions and AI applications 30

Kubernetes Engine 30

Kubernetes Engine and AI applications 31

Diving into the storage options 31

Table of Contents

[ ii ]

Cloud Storage 32

Cloud Storage and AI applications 33

Cloud Bigtable 33

Cloud Bigtable and AI applications 34

Cloud Datastore 34

Cloud Datastore and AI applications 35

Cloud Firestore 35

Cloud Firestore and AI applications 36

Cloud SQL 36

Cloud SQL and AI applications 37

Cloud Spanner 37

Cloud Spanner and AI applications 38

Cloud Memorystore 38

Cloud Memorystore and AI applications 39

Cloud Filestore 39

Cloud Filestore and AI applications 40

Understanding the processing options 40

BigQuery 40

BigQuery and AI applications 41

Cloud Dataproc 41

Cloud Dataproc and AI applications 42

Cloud Dataflow 42

Cloud Dataflow and AI applications 43

Building an ML pipeline 43

Understanding the flow design 44

Loading data into Cloud Storage 45

Loading data to BigQuery 46

Training the model 48

Evaluating the model 49

Testing the model 50

Summary 51

Section 2: Section 2: Artificial Intelligence with Google Cloud

Platform

Chapter 3: Machine Learning Applications with XGBoost 53

Overview of the XGBoost library 54

Ensemble learning 54

How does ensemble learning decide on the optimal predictive model? 55

Reducible errors – bias 55

Reducible errors – variance 55

Irreducible errors 56

Total error 56

Gradient boosting 57

eXtreme Gradient Boosting (XGBoost) 60

Training and storing XGBoost machine learning models 61

Using XGBoost trained models 64

Table of Contents

[ iii ]

Building a recommendation system using the XGBoost library 68

Creating and testing the XGBoost recommendation system model 70

Summary 71

Chapter 4: Using Cloud AutoML 72

Overview of Cloud AutoML 72

The workings of AutoML 74

AutoML API overview 75

REST source – pointing to model locations 75

REST source – for evaluating the model 77

REST source – the operations API 78

Document classification using AutoML Natural Language 78

The traditional machine learning approach for document classification 79

Document classification with AutoML 79

Navigating to the AutoML Natural Language interface 80

Creating the dataset 81

Labeling the training data 82

Training the model 83

Evaluating the model 84

The command line 85

Python 86

Java 86

Node.js 88

Using the model for predictions 88

The web interface 89

A REST API for model predictions 89

Python code for model predictions 90

Image classification using AutoML Vision APIs 90

Image classification steps with AutoML Vision 91

Collecting training images 91

Creating a dataset 91

Labeling and uploading training images 93

Training the model 96

Evaluating the model 98

The command-line interface 100

Python code 100

Testing the model 101

Python code 102

Performing speech-to-text conversion using the Speech-to-Text API 103

Synchronous requests 104

Asynchronous requests 111

Streaming requests 111

Sentiment analysis using AutoML Natural Language APIs 113

Summary 117

Chapter 5: Building a Big Data Cloud Machine Learning Engine 118

Understanding ML 119

Understanding how to use Cloud Machine Learning Engine 120

Google Cloud AI Platform Notebooks 121

Table of Contents

[ iv ]

Google AI Platform deep learning images 121

Creating Google Platform AI Notebooks 123

Using Google Platform AI Notebooks 127

Automating AI Notebooks execution 129

Overview of the Keras framework 134

Training your model using the Keras framework 137

Training your model using Google AI Platform 141

Asynchronous batch prediction using Cloud Machine Learning

Engine 145

Real-time prediction using Cloud Machine Learning Engine 150

Summary 151

Chapter 6: Smart Conversational Applications Using DialogFlow 152

Introduction to DialogFlow 153

Understanding the building blocks of DialogFlow 153

Building a DialogFlow agent 159

Use cases supported by DialogFlow 170

Performing audio sentiment analysis using DialogFlow 171

Summary 172

Section 3: Section 3: TensorFlow on Google Cloud Platform

Chapter 7: Understanding Cloud TPUs 175

Introducing Cloud TPUs and their organization 175

Advantages of using TPUs 178

Mapping of software and hardware architecture 179

Available TPU versions 180

Performance benefits of TPU v3 over TPU v2 181

Available TPU configurations 182

Software architecture 184

Best practices of model development using TPUs 185

Guiding principles for model development on a TPU 186

Training your model using TPUEstimator 186

Standard TensorFlow Estimator API 187

TPUEstimator programming model 187

TPUEstimator concepts 188

Converting from TensorFlow Estimator to TPUEstimator 188

Setting up TensorBoard for analyzing TPU performance 189

Performance guide 191

XLA compiler performance 191

Consequences of tiling 191

Fusion 192

Understanding preemptible TPUs 193

Steps for creating a preemptible TPU from the console 193

Preemptible TPU pricing 195

Table of Contents

[ v ]

Preemptible TPU detection 196

Summary 196

Chapter 8: Implementing TensorFlow Models Using Cloud ML Engine 197

Understanding the components of Cloud ML Engine 198

Training service 198

Using the built-in algorithms 198

Using a custom training application 208

Prediction service 210

Notebooks 215

Data Labeling Service 219

Deep learning containers 220

Steps involved in training and utilizing a TensorFlow model 221

Prerequisites 221

Creating a TensorFlow application and running it locally 223

Project structure recommendation 223

Training data 224

Packaging and deploying your training application in Cloud ML

Engine 226

Choosing the right compute options for your training job 229

Choosing the hyperparameters for the training job 231

Monitoring your TensorFlow training model jobs 233

Summary 234

Chapter 9: Building Prediction Applications 235

Overview of machine-based intelligent predictions 235

Understanding the prediction process 236

Maintaining models and their versions 239

Taking a deep dive into saved models 244

SignatureDef in the TensorFlow SavedModel 246

TensorFlow SavedModel APIs 250

Deploying the models on GCP 252

Uploading saved models to a Google Cloud Storage bucket 255

Testing machine learning models 256

Deploying models and their version 257

Model training example 264

Performing prediction with service endpoints 278

Summary 279

Section 4: Section 4: Building Applications and Upcoming

Features

Chapter 10: Building an AI application 281

A step-by-step approach to developing AI applications 281

Problem classification 282

Classification 283

Table of Contents

[ vi ]

Regression 283

Clustering 283

Optimization 283

Anomaly detection 284

Ranking 284

Data preparation 284

Data acquisition 284

Data processing 285

Problem modeling 285

Validation and execution 286

Holdout 286

Cross-validation 287

Model evaluation parameters (metrics) 287

Classification metrics 287

Model deployment 288

Overview of the use case – automated invoice processing (AIP) 290

Designing AIP with AI platform tools on GCP 292

Performing optical character recognition using the Vision API 296

Storing the invoice with Cloud SQL 300

Creating a Cloud SQL instance 300

Setting up the database and tables 302

Enabling the Cloud SQL API 304

Enabling the Cloud Functions API 304

Creating a Cloud Function 305

Providing the Cloud SQL Admin role 309

Validating the invoice with Cloud Functions 311

Scheduling the invoice for the payment queue (pub/sub) 313

Notifying the vendor and AP team about the payment completion 315

Creating conversational interface for AIP 316

Upcoming features 316

Summary 319

Other Books You May Enjoy 320

Index 323

Preface

We are at an interesting juncture in the journey of computation; the majority of enterprise

workloads (databases, computation, and analytics) are moving over to the cloud. Cloud

computing is making it feasible and easy for anyone to have access to high-end computing

power. As a result, we are seeing a fundamental shift in the way computational resources

are utilized by individuals and organizations of all sizes. The providers of cloud computing

infrastructure have also enabled the Software as a Service (SaaS) paradigm. With this,

various services related to storage, compute, machine learning, and visualization are

available without having to perform any server management or administration. The

fundamental shift we are seeing based on our experience is toward a serverless

architecture.

At this juncture, Google Cloud Platform (GCP) is a complete platform provided by Google

based on years of innovation and research that contributed toward building the most

powerful search engine available to date. The same technology stack is now made

commercially available by Google and as a result, everyone has access to massive

computational power. With the tools available on GCP, it is very easy to build complex

data management pipelines as well as machine learning/AI workflows. This

democratization brings enormous power to the development community in building

cutting-edge, innovative, intelligent applications that complement human intelligence and

increase human capabilities many times.

This book is an attempt to make it easy to get started with building intelligent AI systems

using GCP. We have taken a hands-on approach to explain various concepts and

components available on GCP that facilitate AI development. We hope that this book will

be a good starting point to begin exploring an exciting and ever-expanding world of

computation to build AI-enabled applications on GCP.

Preface

[ 2 ]

Who this book is for

This book is for software developers, technical leads, and architects who are planning to

build AI applications with GCP. In addition to that, students and anyone who has a great

idea for an AI application and wants to understand the available tools and techniques to

quickly build prototypes and eventually production-grade applications will benefit from

this book. This book is also useful for business analysts and anyone who understands the

data landscape from the business perspective. Without a great deal of prior hands-on

experience, it is possible to follow this book to build AI-enabled applications based on

domain knowledge. We have attempted to provide the instructions in a step-by-step

manner so the reader finds them easy to follow and implement.

What this book covers

We have divided this book into four sections based on the logical grouping of the content

covered. Section 1 provides the fundamental introduction to GCP and introduces the reader

to various tools available on GCP.

Chapter 1, Overview of AI on GCP, sets the context for serverless computation on the cloud

and introduces the reader to GCP.

Chapter 2, Computing and Processing Using GCP Components, introduces the reader to

various tools and applications available on GCP for end-to-end data management, which is

essential for building AI applications on GCP.

Chapter 3, Machine Learning Applications with XGBoost, shows how one of the most popular

machine learning algorithms, XGBoost, is utilized on GCP. The idea is to enable readers to

understand that machine learning algorithms can be used on GCP without having to worry

about the underlying infrastructure and computation resources.

Chapter 4, Using Cloud AutoML, will help us take our first step toward the democratization

of machine learning. AutoML intends to provide machine learning as a service and makes it

easy for anyone with a limited understanding of machine learning models and core

implementation details to build applications with machine learning models. We will be

introduced to natural language and vision interfaces using AutoML.

Preface

[ 3 ]

Chapter 5, Building a Big Data Cloud Machine Learning Engine, will explore some of the

fundamentals of machine learning in the cloud. There is a paradigm shift with regard to

machine learning models on the cloud. We need to understand various concepts of cloud

computing and how storage and compute are leveraged to build and deploy models. This is

an essential chapter to understand if we want to optimize the costs of training and running

machine learning models on the cloud. We will take a look at various frameworks, such as

Keras, and we'll see how to use them on GCP.

Chapter 6, Smart Conversational Applications Using DialogFlow, discusses conversational

interfaces to machine intelligence, which is an essential component of overall AI

capabilities. In this chapter, we will understand how to use DialogFlow to build

conversational applications. DialogFlow provides an easy web and API interface for

quickly building a conversational application. It is possible to provide human-like verbal

communication once the model is trained for a large number of conversation paths.

Chapter 7, Understanding Cloud TPUs, discusses Tensor Processing Units (TPUs), which are

the fundamental building blocks behind the machine learning models on GCP. In this

chapter, we will introduce readers to TPUs and discuss their organization and significance

for accelerating machine learning workflows on GCP. If we want to optimize performance

and increase speed, it is imperative to utilize the strength of TPUs.

Chapter 8, Implement TensorFlow Models Using Cloud ML Engine, will further explore ML

Engine and shows how to build TensorFlow models on ML Engine. We will take a step￾wise approach to training and deploying the machine learning models on GCP. We will

take a look at recommended best practices for building a machine learning pipeline on

GCP.

Chapter 9, Building Prediction Applications, explains the process of building prediction

applications on GCP. We begin by discussing the basics of the prediction process and take a

step-by-step, hands-on approach to building a prediction application with GCP. We will

train and deploy a model on the platform and utilize the API layer to interface with the

deployed model.

Chapter 10, Building an AI Application, will utilize various components of GCP to build an

end-to-end AI application. We will illustrate this with an example use case: automating an

invoice processing workflow using the tools on GCP.