Mastering Azure Machine Learning: Execute Large-Scale End-to-end Machine Learning with Azure

✍ Scribed by Christoph Korner; Marcel Alsdorf

• Publisher: Packt Publishing, Limited
• Year: 2022
• Tongue: English
• Leaves: 624
• Edition: 2
• Category: Library

✦ Synopsis

Supercharge and automate your deployments to Azure Machine Learning clusters and Azure Kubernetes Service using Azure Machine Learning services

Key Features:

• Implement end-to-end machine learning pipelines on Azure
• Train deep learning models using Azure compute infrastructure
• Deploy machine learning models using MLOps

Book Description:

Azure Machine Learning is a cloud service for accelerating and managing the machine learning (ML) project life cycle that ML professionals, data scientists, and engineers can use in their day-to-day workflows. This book covers the end-to-end ML process using Microsoft Azure Machine Learning, including data preparation, performing and logging ML training runs, designing training and deployment pipelines, and managing these pipelines via MLOps.

The first section shows you how to set up an Azure Machine Learning workspace; ingest and version datasets; as well as preprocess, label, and enrich these datasets for training. In the next two sections, you'll discover how to enrich and train ML models for embedding, classification, and regression. You'll explore advanced NLP techniques, traditional ML models such as boosted trees, modern deep neural networks, recommendation systems, reinforcement learning, and complex distributed ML training techniques - all using Azure Machine Learning.

The last section will teach you how to deploy the trained models as a batch pipeline or real-time scoring service using Docker, Azure Machine Learning clusters, Azure Kubernetes Services, and alternative deployment targets.

By the end of this book, you'll be able to combine all the steps you've learned by building an MLOps pipeline.

What You Will Learn:

• Understand the end-to-end ML pipeline
• Get to grips with the Azure Machine Learning workspace
• Ingest, analyze, and preprocess datasets for ML using the Azure cloud
• Train traditional and modern ML techniques efficiently using Azure ML
• Deploy ML models for batch and real-time scoring
• Understand model interoperability with ONNX
• Deploy ML models to FPGAs and Azure IoT Edge
• Build an automated MLOps pipeline using Azure DevOps

Who this book is for:

This book is for machine learning engineers, data scientists, and machine learning developers who want to use the Microsoft Azure cloud to manage their datasets and machine learning experiments and build an enterprise-grade ML architecture using MLOps. This book will also help anyone interested in machine learning to explore important steps of the ML process and use Azure Machine Learning to support them, along with building powerful ML cloud applications. A basic understanding of Python and knowledge of machine learning are recommended.

✦ Table of Contents

Cover
Title Page
Copyright and Credits
Contributors
Table of Contents
Preface
Section 1: Introduction to Azure Machine Learning
Chapter 1: Understanding the End-to-End Machine Learning Process
Grasping the idea behind ML
Problems and scenarios requiring ML
The history of ML
Understanding the inner workings of ML through the example of ANNs
Understanding the mathematical basis for statistical analysis and ML modeling
The case for statistics in ML
Basics of statistics
Understanding bias
Classifying ML algorithms
Analyzing errors and the quality of results of model training
Discovering the end-to-end ML process
Excavating data and sources
Preparing and cleaning data
Defining labels and engineering features
Training models
Deploying models
Developing and operating enterprise-grade ML solutions
Summary
Chapter 2: Choosing the Right Machine Learning Service in Azure
Choosing an Azure service for ML
Navigating the Azure AI landscape
Consuming a managed AI service
Building a custom AI service
What is the Azure Machine Learning service?
Managed ML services
Azure Cognitive Services
Custom Cognitive Services
Azure Applied AI Services
Custom ML services
Azure Machine Learning Studio (classic)
Azure Machine Learning designer
Azure Automated Machine Learning
Azure Machine Learning workspace
Custom compute services for ML
Azure Databricks
Azure Batch
Data Science Virtual Machines
Summary
Chapter 3: Preparing the Azure Machine Learning Workspace
Technical requirements
Deploying an Azure Machine Learning workspace
Understanding the available tooling for Azure deployments
Deploying the workspace
Exploring the Azure Machine Learning service
Analyzing the deployed services
Understanding the workspace interior
Surveying Azure Machine Learning Studio
Running ML experiments with Azure Machine Learning
Setting up a local environment
Enhancing a simple experiment
Logging metrics and tracking results
Scheduling the script execution
Running experiments on a cloud compute
Summary
Section 2: Data Ingestion, Preparation, Feature Engineering, and Pipelining
Chapter 4: Ingesting Data and Managing Datasets
Technical requirements
Choosing data storage solutions for Azure Machine Learning
Organizing data in Azure Machine Learning
Understanding the default storage accounts of Azure Machine Learning
Exploring options for storing training data in Azure
Creating a datastore and ingesting data
Creating Blob Storage and connecting it with the Azure Machine Learning workspace
Ingesting data into Azure
Using datasets in Azure Machine Learning
Tracking datasets in Azure Machine Learning
Accessing data during training
Using external datasets with open datasets
Summary
Chapter 5: Performing Data Analysis and Visualization
Technical requirements
Understanding data exploration techniques
Exploring and analyzing tabular datasets
Exploring and analyzing file datasets
Performing data analysis on a tabular dataset
Initial exploration and cleansing of the Melbourne Housing dataset
Running statistical analysis on the dataset
Finding and handling missing values
Calculating correlations and feature importance
Tracking figures from exploration in Azure Machine Learning
Understanding dimensional reduction techniques
Unsupervised dimensional reduction using PCA
Supervised dimensional reduction using LDA
Non-linear dimensional reduction using t-SNE
Generalizing t-SNE with UMAP
Summary
Chapter 6: Feature Engineering and Labeling
Technical requirements
Understanding and applying feature engineering
Classifying feature engineering techniques
Discovering feature transformation and extraction methods
Testing feature engineering techniques on a tabular dataset
Handling data labeling
Analyzing scenarios that require labels
Performing data labeling for image classification using the Azure Machine Learning labeling service
Summary
Chapter 7: Advanced Feature Extraction with NLP
Technical requirements
Understanding categorical data
Comparing textual, categorical, and ordinal data
Transforming categories into numeric values
Orthogonal embedding using one-hot encoding
Semantics and textual values
Building a simple bag-of-words model
A naïve bag-of-words model using counting
Tokenization – turning a string into a list of words
Stemming – the rule-based removal of affixes
Lemmatization – dictionary-based word normalization
A bag-of-words model in scikit-learn
Leveraging term importance and semantics
Generalizing words using n-grams and skip-grams
Reducing word dictionary size using SVD
Measuring the importance of words using TF-IDF
Extracting semantics using word embeddings
Implementing end-to-end language models
The end-to-end learning of token sequences
State-of-the-art sequence-to-sequence models
Text analytics using Azure Cognitive Services
Summary
Chapter 8: Azure Machine Learning Pipelines
Technical requirements
Using pipelines in ML workflows
Why build pipelines?
What are Azure Machine Learning pipelines?
Building and publishing an ML pipeline
Creating a simple pipeline
Connecting data inputs and outputs between steps
Publishing, triggering, and scheduling a pipeline
Parallelizing steps to speed up large pipelines
Reusing pipeline steps through modularization
Integrating pipelines with other Azure services
Building pipelines with Azure Machine Learning designer
Azure Machine Learning pipelines in Azure Data Factory
Azure Pipelines for CI/CD
Summary
Section 3: The Training and Optimization of Machine Learning Models
Chapter 9: Building ML Models Using Azure Machine Learning
Technical requirements
Working with tree-based ensemble classifiers
Understanding a simple decision tree
Combining classifiers with bagging
Optimizing classifiers with boosting rounds
Training an ensemble classifier model using LightGBM
LightGBM in a nutshell
Preparing the data
Setting up the compute cluster and execution environment
Building a LightGBM classifier
Scheduling the training script on the Azure Machine Learning cluster
Summary
Chapter 10: Training Deep Neural Networks on Azure
Technical requirements
Introduction to Deep Learning
Why Deep Learning?
From neural networks to deep learning
DL versus traditional ML
Using traditional ML with DL-based feature extractors
Training a CNN for image classification
Training a CNN from scratch in your notebook
Generating more input data using augmentation
Training on a GPU cluster using Azure Machine Learning
Improving your performance through transfer learning
Summary
Chapter 11: Hyperparameter Tuning and Automated Machine Learning
Technical requirements
Finding the optimal model parameters with HyperDrive
Sampling all possible parameter combinations using grid search
Testing random combinations using random search
Converging faster using early termination
Optimizing parameter choices using Bayesian optimization
Finding the optimal model with Automated Machine Learning
The unfair advantage of Automated Machine Learning
A classification example with Automated Machine Learning
Summary
Chapter 12: Distributed Machine Learning on Azure
Technical requirements
Exploring methods for distributed ML
Training independent models on small data in parallel
Training a model ensemble on large datasets in parallel
Fundamental building blocks for distributed ML
Speeding up deep learning with data-parallel training
Training large models with model-parallel training
Using distributed ML in Azure
Horovod – a distributed deep learning training framework
Implementing the HorovodRunner API for a Spark job
Training models with Horovod on Azure Machine Learning
Summary
Chapter 13: Building a Recommendation Engine in Azure
Technical requirements
Introduction to recommendation engines
A content-based recommender system
Measuring the similarity between items
Feature engineering for content-based recommenders
Content-based recommendations using gradient boosted trees
Collaborative filtering – a rating-based recommender system
What is a rating? Explicit feedback versus implicit feedback
Predicting the missing ratings to make a recommendation
Scalable recommendations using ALS factorization
Combining content and ratings in hybrid recommendation engines
Automatic optimization through reinforcement learning
Summary
Section 4: Machine Learning Model Deployment and Operations
Chapter 14: Model Deployment, Endpoints, and Operations
Technical requirements
Preparations for model deployments
Understanding the components of an ML model
Registering your models in a model registry
Auto-deployments of registered models
Customizing your deployment environment
Choosing a deployment target in Azure
Deploying ML models in Azure
Building a real-time scoring service
Deploying to Azure Kubernetes Services
Defining a schema for scoring endpoints
Managing model endpoints
Controlled rollouts and A/B testing
Implementing a batch-scoring pipeline
ML operations in Azure
Profiling models for optimal resource configuration
Collecting logs and infrastructure metrics
Tracking telemetry and application metrics
Detecting data drift
Summary
Chapter 15: Model Interoperability, Hardware Optimization, and Integrations
Technical requirements
Model interoperability with ONNX
What is model interoperability and how can ONNX help?
Converting models to ONNX format with ONNX frontends
Native scoring of ONNX models with ONNX backends
Hardware optimization with FPGAs
Understanding FPGAs
Comparing GPUs and FPGAs for deep neural networks
Running DNN inferencing on Intel FPGAs with Azure
Integrating ML models and endpoints with Azure services
Integrating with Azure IoT Edge
Integrating with Power BI
Summary
Chapter 16: Bringing Models into Production with MLOps
Technical requirements
Ensuring reproducible builds and deployments
Version-controlling your code
Registering snapshots of your data
Tracking your model metadata and artifacts
Scripting your environments and deployments
Validating the code, data, and models
Testing data quality with unit tests
Integration testing for ML
End-to-end testing using Azure Machine Learning
Continuous profiling of your model
Building an end-to-end MLOps pipeline
Setting up Azure DevOps
Continuous integration – building code with pipelines
Continuous deployment – deploying models with release pipelines
Summary
Chapter 17: Preparing for a Successful ML Journey
Remembering the importance of data
Starting with a thoughtful infrastructure
Automating recurrent tasks
Expecting constant change
Thinking about your responsibility
Interpreting a model
Fairness in model training
Handling PII data and compliance requirements
Summary
Index
Other Books You May Enjoy