book

MLOps Engineering at Scale

Name: MLOps Engineering at Scale
Author: Carl Osipov
ISBN: 9781617297762

by Carl Osipov

February 2022

Intermediate to advanced

344 pages

9h 41m

English

Manning Publications

Audiobook available

Read now

Unlock full access

MLOps Engineering at Scale
Copyright
contents
front matter
prefaceacknowledgmentsabout this bookWho should read this bookHow this book is organized: A road mapAbout the codeliveBook discussion forumabout the authorabout 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 platform1.3 Public clouds for machine learning platforms1.4 What is serverless machine learning?1.5 Why serverless machine learning?1.5.1 Serverless vs. IaaS and PaaS1.5.2 Serverless machine learning life cycle1.6 Who is this book for?1.6.1 What you can get out of this book1.7 How does this book teach?1.8 When is this book not for you?1.9 ConclusionsSummary
2 Getting started with the data set
2.1 Introducing the Washington, DC taxi rides data set2.1.1 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 set2.2 Starting with object storage for the data set2.2.1 Understanding object storage vs. filesystems2.2.2 Authenticating with Amazon Web Services2.2.3 Creating a serverless object storage bucket2.3 Discovering the schema for the data set2.3.1 Introducing AWS Glue2.3.2 Authorizing the crawler to access your objects2.3.3 Using a crawler to discover the data schema2.4 Migrating to columnar storage for more efficient analytics2.4.1 Introducing column-oriented data formats for analytics2.4.2 Migrating to a column-oriented data formatSummary
3 Exploring and preparing the data set
3.1 Getting started with interactive querying3.1.1 Choosing the right use case for interactive querying3.1.2 Introducing AWS Athena3.1.3 Preparing a sample data set3.1.4 Interactive querying using Athena from a browser3.1.5 Interactive querying using a sample data set3.1.6 Querying the DC taxi data set3.2 Getting started with data quality3.2.1 From “garbage in, garbage out” to data quality3.2.2 Before starting with data quality3.2.3 Normative principles for data quality3.3 Applying VACUUM to the DC taxi data3.3.1 Enforcing the schema to ensure valid values3.3.2 Cleaning up invalid fare amounts3.3.3 Improving the accuracy3.4 Implementing VACUUM in a PySpark jobSummary
4 More exploratory data analysis and data preparation
4.1 Getting started with data sampling4.1.1 Exploring the summary statistics of the cleaned-up data set4.1.2 Choosing the right sample size for the test data set4.1.3 Exploring the statistics of alternative sample sizes4.1.4 Using a PySpark job to sample the test setSummary
Part 2 PyTorch for serverless machine learning

5 Introducing PyTorch: Tensor basics
5.1 Getting started with tensors5.2 Getting started with PyTorch tensor creation operations5.3 Creating PyTorch tensors of pseudorandom and interval values5.4 PyTorch tensor operations and broadcasting5.5 PyTorch tensors vs. native Python listsSummary
6 Core PyTorch: Autograd, optimizers, and utilities
6.1 Understanding the basics of autodiff6.2 Linear regression using PyTorch automatic differentiation6.3 Transitioning to PyTorch optimizers for gradient descent6.4 Getting started with data set batches for gradient descent6.5 Data set batches with PyTorch Dataset and DataLoader6.6 Dataset and DataLoader classes for gradient descent with batchesSummary
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 ObjectStorageDataset7.3 Gradient descent with out-of-memory data sets7.4 Faster PyTorch tensor operations with GPUs7.5 Scaling up to use GPU coresSummary
8 Scaling out with distributed training
8.1 What if the training data set does not fit in memory?8.1.1 Illustrating gradient accumulation8.1.2 Preparing a sample model and data set8.1.3 Understanding gradient descent using out-of-memory data shards8.2 Parameter server approach to gradient accumulation8.3 Introducing logical ring-based gradient descent8.4 Understanding ring-based distributed gradient descent8.5 Phase 1: Reduce-scatter8.6 Phase 2: All-gatherSummary
Part 3 Serverless machine learning pipeline
9 Feature selection
9.1 Guiding principles for feature selection9.1.1 Related to the label9.1.2 Recorded before inference time9.1.3 Supported by abundant examples9.1.4 Expressed as a number with a meaningful scale9.1.5 Based on expert insights about the project9.2 Feature selection case studies9.3 Feature selection using guiding principles9.3.1 Related to the label9.3.2 Recorded before inference time9.3.3 Supported by abundant examples9.3.4 Numeric with meaningful magnitude9.3.5 Bring expert insight to the problem9.4 Selecting features for the DC taxi data setSummary
10 Adopting PyTorch Lightning
10.1 Understanding PyTorch Lightning10.1.1 Converting PyTorch model training to PyTorch Lightning10.1.2 Enabling test and reporting for a trained model10.1.3 Enabling validation during model trainingSummary
11 Hyperparameter optimization
11.1 Hyperparameter optimization with Optuna11.1.1 Understanding loguniform hyperparameters11.1.2 Using categorical and log-uniform hyperparameters11.2 Neural network layers configuration as a hyperparameter11.3 Experimenting with the batch normalization hyperparameter11.3.1 Using Optuna study for hyperparameter optimization11.3.2 Visualizing an HPO study in OptunaSummary
12 Machine learning pipeline
12.1 Describing the machine learning pipeline12.2 Enabling PyTorch-distributed training support with Kaen12.2.1 Understanding PyTorch-distributed training settings12.3 Unit testing model training in a local Kaen container12.4 Hyperparameter optimization with Optuna12.4.1 Enabling MLFlow support12.4.2 Using HPO for DcTaxiModel in a local Kaen provider12.4.3 Training with the Kaen AWS providerSummary
Appendix A. Introduction to machine learning
A.1 Why machine learning?A.2 Machine learning at first glanceA.3 Machine learning with structured data setsA.4 Regression with structured data setsA.5 Classification with structured data setsA.6 Training a supervised machine learning model
Appendix B. Getting started with Docker
B.1 Getting started with DockerB.2 Building a custom imageB.3 Sharing your custom image with the world
index

Content preview from MLOps Engineering at Scale

6 Core PyTorch: Autograd, optimizers, and utilities

This chapter covers

Understanding automatic differentiation
Using automatic differentiation with PyTorch tensors
Getting started with PyTorch SGD and Adam optimizers
Applying PyTorch to linear regression with gradient descent
Using data set batches for gradient descent
PyTorch Dataset and DataLoader utility classes for batches

In chapter 5, you learned about the tensor, a core PyTorch data structure for n-dimensional arrays. The chapter illustrated the significant performance advantages of PyTorch tensors over native Python data structures for arrays and introduced PyTorch APIs for creating tensors as well as performing common operations on one or more tensors.

This chapter teaches another ...

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.

Read now

Unlock full access

More than 5,000 organizations count on O’Reilly

O’Reilly covers everything we've got, with content to help us build a world-class technology community, upgrade the capabilities and competencies of our teams, and improve overall team performance as well as their engagement.

Julian F.

Head of Cybersecurity

I wanted to learn C and C++, but it didn't click for me until I picked up an O'Reilly book. When I went on the O’Reilly platform, I was astonished to find all the books there, plus live events and sandboxes so you could play around with the technology.

Addison B.

Field Engineer

I’ve been on the O’Reilly platform for more than eight years. I use a couple of learning platforms, but I'm on O'Reilly more than anybody else. When you're there, you start learning. I'm never disappointed.

Amir M.

Data Platform Tech Lead

I'm always learning. So when I got on to O'Reilly, I was like a kid in a candy store. There are playlists. There are answers. There's on-demand training. It's worth its weight in gold, in terms of what it allows me to do.

Mark W.

Embedded Software Engineer

Publisher Resources

ISBN: 9781617297762Publisher Support Publisher Website

Cloud Computing

Data Engineering

Data Science

AI & ML

Programming Languages

Software Architecture

IT/Ops

Security

Design

Business

Soft Skills

MLOps Engineering at Scale

by Carl Osipov

6 Core PyTorch: Autograd, optimizers, and utilities

Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.