MLOps Engineering at Scale

MLOps Engineering at Scale
brief contents
contents
preface
acknowledgments
about this book
	Who should read this book
	How this book is organized: A road map
	About the code
	liveBook discussion forum
about the author
about the cover illustration
Part 1—Mastering the data set
	1 Introduction to serverless machine learning
		1.1 What is a machine learning platform?
		1.2 Challenges when designing a machine learning platform
		1.3 Public clouds for machine learning platforms
		1.4 What is serverless machine learning?
		1.5 Why serverless machine learning?
			1.5.1 Serverless vs. IaaS and PaaS
			1.5.2 Serverless machine learning life cycle
		1.6 Who is this book for?
			1.6.1 What you can get out of this book
		1.7 How does this book teach?
		1.8 When is this book not for you?
		1.9 Conclusions
		Summary
	2 Getting started with the data set
		2.1 Introducing the Washington, DC, taxi rides data set
			2.1.1 What is the business use case?
			2.1.2 What are the business rules?
			2.1.3 What is the schema for the business service?
			2.1.4 What are the options for implementing the business service?
			2.1.5 What data assets are available for the business service?
			2.1.6 Downloading and unzipping the data set
		2.2 Starting with object storage for the data set
			2.2.1 Understanding object storage vs. filesystems
			2.2.2 Authenticating with Amazon Web Services
			2.2.3 Creating a serverless object storage bucket
		2.3 Discovering the schema for the data set
			2.3.1 Introducing AWS Glue
			2.3.2 Authorizing the crawler to access your objects
			2.3.3 Using a crawler to discover the data schema
		2.4 Migrating to columnar storage for more efficient analytics
			2.4.1 Introducing column-oriented data formats for analytics
			2.4.2 Migrating to a column-oriented data format
		Summary
	3 Exploring and preparing the data set
		3.1 Getting started with interactive querying
			3.1.1 Choosing the right use case for interactive querying
			3.1.2 Introducing AWS Athena
			3.1.3 Preparing a sample data set
			3.1.4 Interactive querying using Athena from a browser
			3.1.5 Interactive querying using a sample data set
			3.1.6 Querying the DC taxi data set
		3.2 Getting started with data quality
			3.2.1 From “garbage in, garbage out” to data quality
			3.2.2 Before starting with data quality
			3.2.3 Normative principles for data quality
		3.3 Applying VACUUM to the DC taxi data
			3.3.1 Enforcing the schema to ensure valid values
			3.3.2 Cleaning up invalid fare amounts
			3.3.3 Improving the accuracy
		3.4 Implementing VACUUM in a PySpark job
		Summary
	4 More exploratory data analysis and data preparation
		4.1 Getting started with data sampling
			4.1.1 Exploring the summary statistics of the cleaned-up data set
			4.1.2 Choosing the right sample size for the test data set
			4.1.3 Exploring the statistics of alternative sample sizes
			4.1.4 Using a PySpark job to sample the test set
		Summary
Part 2—PyTorch for serverless machine learning
	5 Introducing PyTorch: Tensor basics
		5.1 Getting started with tensors
		5.2 Getting started with PyTorch tensor creation operations
		5.3 Creating PyTorch tensors of pseudorandom and interval values
		5.4 PyTorch tensor operations and broadcasting
		5.5 PyTorch tensors vs. native Python lists
		Summary
	6 Core PyTorch: Autograd, optimizers, and utilities
		6.1 Understanding the basics of autodiff
		6.2 Linear regression using PyTorch automatic differentiation
		6.3 Transitioning to PyTorch optimizers for gradient descent
		6.4 Getting started with data set batches for gradient descent
		6.5 Data set batches with PyTorch Dataset and DataLoader
		6.6 Dataset and DataLoader classes for gradient descent with batches
		Summary
	7 Serverless machine learning at scale
		7.1 What if a single node is enough for my machine learning model?
		7.2 Using IterableDataset and ObjectStorageDataset
		7.3 Gradient descent with out-of-memory data sets
		7.4 Faster PyTorch tensor operations with GPUs
		7.5 Scaling up to use GPU cores
		Summary
	8 Scaling out with distributed training
		8.1 What if the training data set does not fit in memory?
			8.1.1 Illustrating gradient accumulation
			8.1.2 Preparing a sample model and data set
			8.1.3 Understanding gradient descent using out-of-memory data shards
		8.2 Parameter server approach to gradient accumulation
		8.3 Introducing logical ring-based gradient descent
		8.4 Understanding ring-based distributed gradient descent
		8.5 Phase 1: Reduce-scatter
		8.6 Phase 2: All-gather
		Summary
Part 3—Serverless machine learning pipeline
	9 Feature selection
		9.1 Guiding principles for feature selection
			9.1.1 Related to the label
			9.1.2 Recorded before inference time
			9.1.3 Supported by abundant examples
			9.1.4 Expressed as a number with a meaningful scale
			9.1.5 Based on expert insights about the project
		9.2 Feature selection case studies
		9.3 Feature selection using guiding principles
			9.3.1 Related to the label
			9.3.2 Recorded before inference time
			9.3.3 Supported by abundant examples
			9.3.4 Numeric with meaningful magnitude
			9.3.5 Bring expert insight to the problem
		9.4 Selecting features for the DC taxi data set
		Summary
	10 Adopting PyTorch Lightning
		10.1 Understanding PyTorch Lightning
			10.1.1 Converting PyTorch model training to PyTorch Lightning
			10.1.2 Enabling test and reporting for a trained model
			10.1.3 Enabling validation during model training
		Summary
	11 Hyperparameter optimization
		11.1 Hyperparameter optimization with Optuna
			11.1.1 Understanding loguniform hyperparameters
			11.1.2 Using categorical and log-uniform hyperparameters
		11.2 Neural network layers configuration as a hyperparameter
		11.3 Experimenting with the batch normalization hyperparameter
			11.3.1 Using Optuna study for hyperparameter optimization
			11.3.2 Visualizing an HPO study in Optuna
		Summary
	12 Machine learning pipeline
		12.1 Describing the machine learning pipeline
		12.2 Enabling PyTorch-distributed training support with Kaen
			12.2.1 Understanding PyTorch-distributed training settings
		12.3 Unit testing model training in a local Kaen container
		12.4 Hyperparameter optimization with Optuna
			12.4.1 Enabling MLFlow support
			12.4.2 Using HPO for DcTaxiModel in a local Kaen provider
			12.4.3 Training with the Kaen AWS provider
		Summary
Appendix A—Introduction to machine learning
	A.1 Why machine learning?
	A.2 Machine learning at first glance
	A.3 Machine learning with structured data sets
	A.4 Regression with structured data sets
	A.5 Classification with structured data sets
	A.6 Training a supervised machine learning model
Appendix B—Getting started with Docker
	B.1 Getting started with Docker
	B.2 Building a custom image
	B.3 Sharing your custom image with the world
index
	Symbols
	A
	B
	C
	D
	E
	F
	G
	H
	I
	J
	K
	L
	M
	N
	O
	P
	Q
	R
	S
	T
	U
	V
	W
	X
	Y
	Z