Productionizing AI: How to Deliver AI B2B Solutions with Cloud and Python

✍ Scribed by Barry Walsh

• Publisher: Apress
• Year: 2022
• Tongue: English
• Leaves: 390
• Edition: 1
• Category: Library

✦ Synopsis

This book is a guide to productionizing AI solutions using best-of-breed cloud services with workarounds to lower costs. Supplemented with step-by-step instructions covering data import through wrangling to partitioning and modeling through to inference and deployment, and augmented with plenty of Python code samples, the book has been written to accelerate the process of moving from script or notebook to app.

From an initial look at the context and ecosystem of AI solutions today, the book drills down from high-level business needs into best practices, working with stakeholders, and agile team collaboration. From there you’ll explore data pipeline orchestration, machine and deep learning, including working with and finding shortcuts using artificial neural networks such as AutoML and AutoAI. You’ll also learn about the increasing use of NoLo UIs through AI application development, industry case studies, and finally a practical guide to deploying containerized AI solutions.

The book is intended for those whose role demands overcoming budgetary barriers or constraints in accessing cloud credits to undertake the often difficult process of developing and deploying an AI solution.

What You Will Learn

• Develop and deliver production-grade AI in one month
• Deploy AI solutions at a low cost
• Work around Big Tech dominance and develop MVPs on the cheap
• Create demo-ready solutions without overly complex python scripts/notebooks

 Who this book is for:

Data scientists and AI consultants with programming skills in Python and driven to succeed in AI.

✦ Table of Contents

Table of Contents
About the Author
About the Technical Reviewer
Preface
Prologue
Chapter 1: Introduction to AI and the AI Ecosystem
The AI Ecosystem
The Hype Cycle
Historical Context
AI – Some Definitions
AI Today
Machine Learning
Deep Learning
What Is Artificial Intelligence
Cloud Computing
CSPs – What Do They Offer ?
The Wider AI Ecosystem
Full-Stack AI
AI Ethics and Risk: Issues and Concerns
The AI ecosystem: Hands-on Practise
Applications of AI
Machine Learning
Deep Learning
Portfolio, Risk Management, and Forecasting
Natural Language Processing (NLP)
Chatbots
Cognitive Robotic Process Automation (CRPA)
Other AI Applications
AI Applications: Hands-on Practice
Data Ingestion and AI Pipelines
AI Engineering
What Is a Data Pipeline?
Extract, Transform, and Load (ETL)
Extract
Transform
Load
Data Wrangling
Performance Benchmarking
AI Pipeline Automation – AutoAI
Build Your Own AI Pipeline: Hands-on Practice
Neural Networks and Deep Learning
Machine Learning
Supervised Machine Learning
Unsupervised Machine Learning
Reinforcement Learning
What Is a Neural Network?
The Simple Perceptron
Deep Learning
Convolutional Neural Networks
Recurrent Neural Networks
Autoencoders and Variational Autoencoders (VAEs)
Generative Adversarial Networks (GANs)
Neural Networks – terminology
Tools for Deep Learning
Introduction to Neural Networks and DL: Hands-on Practice
Productionizing AI
Compute and Storage
The CSPs – Why No-one Can Be Successful in AI Without Investing in Amazon, Microsoft, or Google
Compute Services
Storage Services
Containerization
Docker and Kubernetes
Productionizing AI: Hands-on Practice
Wrap-up
Chapter 2: AI Best Practice and DataOps
Introduction to DataOps and MLOps
DataOps
The Data “Factory”
The Problem with AI: From DataOps to MLOps
Enterprise AI
GCP/BigQuery: Hands-on Practice
Event Streaming with Kafka: Hands-on Practice
Agile
Agile Teams and Collaboration
Development/Product Sprints
Benefits of Agile
Adaptability
react.js: Hands-on Practice
VueJS: Hands-on Practise
Code Repositories
Git and GitHub
Version Control
Branching and Merging
Git Workflows
GitHub and Git: Hands-on Practice
Deploying an App to GitHub Pages: Hands-on Practice
Continuous Integration and Continuous Delivery (CI/CD)
CI/CD in DataOps
Introduction to Jenkins
Maven
Containerization
Docker and Kubernetes
Play With Docker: Hands-on Practice
Testing, Performance Evaluation, and Monitoring
Selenium
TestNG
Issue Management
Jira
ServiceNow
Monitoring and Alerts
Nagios
Jenkins CI/CD and Selenium Test Scripts: Hands-on Practice
Wrap-up
Chapter 3: Data Ingestion for AI
Introduction to Data Ingestion
Data Ingestion – The Challenge Today
The AI Ladder
Cloud Architectures/Cloud “Stack”
Scheduled (OLAP) vs. Streaming (OLTP) Data
APIs
Data Types (Structured vs. Unstructured)
File Types
Automated Data Ingestion: Hands-on Practice
Working with Parquet: Hands-on Practise
Data Stores for AI
Data Stores: Data Lakes and Data Warehouses
Lakehouses
Scoping Project Data Requirements
OLTP/OLAP – Determining the Best Approach
ETL vs. ELT
SQL vs. NoSQL Databases
Elasticity vs. Scalability
Data Stores for AI: Hands-on Practice
Cloud Services for Data Ingestion
Cloud (SQL) Data Warehouses
Data Lake Storage
Hadoop
Stream Processing and Stream Analytics
Simple Data Streaming: Hands-on Practise
Cloud services for Data Ingestion: Hands-on Practise
Data Pipeline Orchestration – Best Practice
Storage Considerations
Data Ingestion Schedules
Serverless Computing
End-of-day Processes
Data Import for Machine and Deep Learning
Building a Delivery Pipeline
Example: XenonStack
Example: Red Hat/IBM
Example: AWS Serverless Architecture
Example: Databricks with Apache Spark
Example: Snowflake Workload Management
Data Pipeline Orchestration: Hands-on Practice
Wrap-up
Chapter 4: Machine Learning on Cloud
ML Fundamentals
Supervised Machine Learning
Classification and Regression
Time Series Forecasting
Introduction to fbprophet: Hands-on Practice
Unsupervised Machine Learning
Clustering
Dimensionality Reduction
Unsupervised Machine Learning (Clustering): Hands-on Practice
Semisupervised Machine Learning
Machine Learning Implementation
Exploratory Data Analysis (EDA)
Data Wrangling
Feature Engineering
Shuffling and Data Partitioning/Splitting
Sampling
End-to-End Wrangling: Hands-on Practice
Algorithmic Modelling
Performance Benchmarking
Continual Improvement
Machine Learning Classifiers: Hands-on Practice
Model Selection, Deployment, and Inference
Inference: Hands-on Practice
Reinforcement Learning
Wrap-up
Chapter 5: Neural Networks and Deep Learning
Introduction to Deep Learning
What Is Deep Learning
Deep Learning – Why Now?
AI and Deep Learning Hype Cycle
High-Level Architectures
TensorFlow Playground: Hands-on Practice
Stochastic Processes
Generative vs. Discriminative
Random Walks
Markov Chains and Markov Processes
Other Stochastic Processes: Martingales
Implementing a Random Walk in Python: Hands-on Practice
Introduction to Neural Networks
Artificial Neural Networks (ANNs)
The Simple Perceptron
Multilayer Perceptron (MLP)
Convolutional Neural Networks (CNN)
Recurrent Neural Networks (RNN)
Long Short-Term Memory (LSTM) networks
Other Types of Neural Networks
Restricted Boltzmann Machines (RBMs)
Deep Belief Networks (DBNs)
Deep Boltzmann Machines (DBMs)
Autoencoders
Generative Adversarial Networks
A Simple Deep Learning Solution – MNIST: Hands-on Practice
Autoencoders in Keras: Hands-on Practice
Deep Learning Tools
Tools for Deep Learning
TensorFlow
Keras
PyTorch
Other Important Deep Learning Tools
Apache Spark
Frameworks for Deep Learning and Implementation
Tensors
Key TensorFlow Concepts
The Deep Learning Modeling Lifecycle
Sequential and Functional Model APIs
Sequential Model API
Functional Model API
Implementing a CNN
Implementing an RNN
LSTM Implementation for Time Series
Neural Networks – Terminology
Computing the Output of a Multilayer Neural Network
Convolutional Neural Networks with Keras and TensorFlow: Hands-on Practice
Recurrent Neural Networks – Time Series Forecasting: Hands-on Practice
Tuning a DL Model
Activation Functions
(Logistic) sigmoid function
Hyperbolic Tangent Function (tanh)
Rectified Linear Unit (ReLU)
Softmax
Gradient Descent and Backpropagation
Backpropagation
Other Optimization Algorithms
SGD with Momentum
AdaGrad, Adadelta, and RMSProp
Adam
Loss Functions
Improving DL Performance
Deep Learning Best Practice – Hyperparameters
Network Tuning
Deeper Network/More Layers/More Neurons
Activation Function
Neural Network Ensembles
Batch Normalization
Pooling
Image Augmentation
Process Tuning
Number of Epochs and Batch Size
Learning Rate
Regularization
Dropout
Early Stopping
Transfer Learning
Wrap-up
Softmax: Hands-on Practice
Early Stopping: Hands-on Practice
Chapter 6: AutoML, AutoAI, and the Rise of NoLo UIs
Machine Learning: Process Recap
Global Search Algorithms
Bayesian Optimization and Inference
Bayesian Inference: Hands-on Practice
Python-Based Libraries for Automation
PyCaret
auto-sklearn
Auto-WEKA
TPOT
Python Automation with TPOT: Hands-on Practice
AutoAI Tools and Platforms
IBM Cloud Pak for Data
Azure Machine Learning
Google Cloud Vertex AI
Google Cloud Composer
AWS SageMaker Autopilot
TensorFlow Extended (TFX)
Wrap-up
AutoAI with IBM Cloud Pak for Data: Hands-on Practice
Healthcare diagnostics with Google Teachable Machines: Hands-on Practice
TFX and Vertex AI Pipelines: Hands-on Practice
Azure Video Analyzer: Hands-on Practice
Chapter 7: AI Full Stack: Application Development
Introduction to AI Application Development
Developing an AI Solution
AI Apps – Up and Running
APIs and Endpoints
Distributed Processing and Clusters
Clusters
Graphical Processing Units (GPUs)
TensorFlow Processing Units (TPUs)
Sharding
Virtual Environments
Running Python from Terminal: Hands-on Practice
API Web Services and Endpoints: Hands-on Practice
AI Accelerators - GPUs: Hands-on Practise
Software and Tools for AI Development
AI Needs Data and Cloud
Cloud Platforms
AWS
Azure
GCP
IBM Cloud
Heroku
Python-Based UIs
Flask
Dash
Django
Other AI Software Vendors
ONNX (Open Neural Network Exchange)
C3
DataRobot
Introduction to Dash: Hands-on Practice
Flask: Hands-on Practice
Introduction to Django: Hands-on Practice
ML Apps
Developing Machine Learning Applications
Customer Experience
Fraud Detection and Cybersecurity
Operations Management, Decision, and Business Support
Risk Management, and Portfolio and Asset Optimization
Developing a Recommendation Engine: Hands-on Practice
Portfolio Optimization Accelerator: Hands-on Practise
DL Apps
Developing Deep Learning Applications
Key Deep Learning Apps
Computer Vision
Forecasting
IoT
Full-Stack Deep Learning: Hands-on Practice
Wrap-up
Chapter 8: AI Case Studies
Industry Case Studies
Business/Organizational Demand for AI
AI Enablers
AI Solutions by Vertical Industry
AI Use Cases – Solution Frameworks
Solution Architectures
Example: Azure
Telco Solutions
Specific Challenges
Solution Categories
Real-time Dashboards
Sentiment Analysis
Predictive Analytics
Connecting to the Twitter API from Python
Twitter API and Basic Sentiment Analysis: Hands-on Practice
Retail Solutions
Challenges in the Retail Industry
Churn and Retention Modelling
A Best Practice Approach to Modelling Churn
Model Design and Outcomes
Online Retail Predictive Analytics with GCP BigQuery: Hands-on Practice
Predicting Customer Churn: Hands-on Practise
Social Network Analysis: Hands-on Practise
Banking and Financial Services/FinTech Solutions
Industry Challenges
Fraud Detection
Case Study: AWS Fraud Detection
AWS Fraud Detection with AWS SageMaker: Hands-on Practice
Supply Chain Solutions
Challenges Across Supply Chains
Predictive Analytics Solutions
Supply Chain Optimization and Prescriptive Analytics
Supply Chain Optimization with IBM CloudPak/Watson Studio: Hands-on Practice
Oil and Gas/Energy and Utilities Solutions
Challenges in Energy, Oil, and Gas Sectors
AI Solutions in Energy – An Opportunity or a Threat?
Healthcare and Pharma Solutions
Healthcare – The AI Gap
Healthcare and Pharma Solutions
HR Solutions
HR in 2002
Sample HR Solutions
HR Employee Attrition: Hands-on Practice
Other Case Studies
Public Sector and Government
Manufacturing
Cybersecurity
Insurance/Telematics
Legal
DALL-E for the Creative Arts: Hands-on Practise
Wrap-up
Chapter 9: Deploying an AI Solution (Productionizing and Containerization)
Productionizing an AI Application
Typical Barriers to Production
Cloud/CSP Roulette
Simplifying the AI Challenge – Start Small, Stay Niche
Database Management in Python: Hands-on Practice
App Building on GCP: Hands-on Practice
PowerBI – Python Handshake: Hands-on Practice
AI Project Lifecycle
Design Thinking Through to Agile Development
Driving Development Through Hypothesis
Collaborate, Test, Measure, Repeat
Continual Process Improvement
Data drift
Automated Retraining
Hosting on Heroku – End-to-End: Hands-on Practice
Enabling Engineering and Infrastructure
The AI Ecosystem – The AI Cloud Stack
Data Lake Deployment – Best Practice
Data Pipeline Operationalization and Orchestration
Big Data Engines and Parallelization
Dask
Leveraging S3 File Storage: Hands-on Practise
Apache Spark Quick Start on Databricks: Hands-on Practice
Dask Parallelization: Hands-on Practice
Full Stack and Containerization…the final frontier
Full Stack AI – React and Flask Case Study
Deploying on Cloud with a Docker Container
Implementing a Continuous Delivery Pipeline
Wrap-up
DL App deployment with Streamlit and Heroku: Hands-on Practice
Deploying on Azure with a Docker Container: Hands-on Practice
Chapter 10: Natural Language Processing
Introduction to NLP
NLP Fundamentals
Historical Context and Development of NLP
NLP Goals and Sector-specific Use Cases
Key Industrial Applications
The NLP Lifecycle
From Parsing to Linguistic Analysis
Word Embeddings to Deep Learning
Creating a Word Cloud: Hands-on Practice
Preprocessing and Linguistics
Preprocessing/Initial Cleaning
Regular Expressions
Text Stripping (e.g., HTML tags)
Linguistics and Data Transformation
Lexical Analysis
Removing Stop Words
Tokenization
Syntactic Analysis
Switch to Lowercase
Part of Speech (POS) Tagging
Named Entity Recognition (NER)
Handling Contractions
Stemming
Semantic Analysis
Lemmatization
Disambiguation
N-grams
Text Parsing with NLTK: Hands-on Practice
Text Vectorization, Word Embeddings, and Modelling in NLP
Rule-Based/Frequency-Based Embedding
One Hot Encoding and Count Vectorization
Bag of Words (BoW)
Latent Semantic Analysis (LSA)
TF-IDF
A Word on Cosine Similarity
Word Embeddings/Prediction-Based Embedding
Word2Vec (Google)
Other Models
NLP Modeling
Text Summarization
Topic Modeling
Sequence Models
Transformers and Attention Models
Word Embeddings: Hands-on Practice
Seq2Seq: Hands-on Practice
PyTorch NLP: Hands-on Practice
Tools and Applications of NLP
Python Libraries
NLP Applications
Text Analytics
Text-to-Speech-to-Text
Social Media Sentiment Analysis/Opinion Mining
Chatbots, Conversational Assistants, and IVAs
NLP 2.0
Natural Language Generation
Debating
Auto-NLP
Wrap-up
WATSON Assistant Chatbot/IVA: Hands-on Practice
Transformers for Chatbots: Hands-on Practice
Postscript
Wrap-up
Epilogue
Index