Book Image

Amazon SageMaker Best Practices

By : Sireesha Muppala, Randy DeFauw, Shelbee Eigenbrode
Book Image

Amazon SageMaker Best Practices

By: Sireesha Muppala, Randy DeFauw, Shelbee Eigenbrode

Overview of this book

Amazon SageMaker is a fully managed AWS service that provides the ability to build, train, deploy, and monitor machine learning models. The book begins with a high-level overview of Amazon SageMaker capabilities that map to the various phases of the machine learning process to help set the right foundation. You'll learn efficient tactics to address data science challenges such as processing data at scale, data preparation, connecting to big data pipelines, identifying data bias, running A/B tests, and model explainability using Amazon SageMaker. As you advance, you'll understand how you can tackle the challenge of training at scale, including how to use large data sets while saving costs, monitoring training resources to identify bottlenecks, speeding up long training jobs, and tracking multiple models trained for a common goal. Moving ahead, you'll find out how you can integrate Amazon SageMaker with other AWS to build reliable, cost-optimized, and automated machine learning applications. In addition to this, you'll build ML pipelines integrated with MLOps principles and apply best practices to build secure and performant solutions. By the end of the book, you'll confidently be able to apply Amazon SageMaker's wide range of capabilities to the full spectrum of machine learning workflows.
Table of Contents (20 chapters)
Section 1: Processing Data at Scale
Section 2: Model Training Challenges
Section 3: Manage and Monitor Models
Section 4: Automate and Operationalize Machine Learning

ML training at scale with SageMaker distributed libraries

Two common scale challenges with ML projects are scaling training data and scaling model size. While increased training data volume, model size, and complexity can potentially result in a more accurate model, there is a limit to the data volume and the model size that you can use with a single compute node, CPU, or GPU. Increased training data volumes and model sizes typically result in more computations, and therefore training jobs take longer to finish, even when using powerful compute instances such as Amazon Elastic Compute Cloud (EC2p3 and p4 instances.

Distributed training is a commonly used technique to speed up training when dealing with scale challenges. Training load can be distributed either across multiple compute instances (nodes), or across multiple CPUs and GPUs (devices) on a single compute instance. There are two strategies for distributed training – data parallelism and...