Amazon SageMaker Best Practices by Sireesha Muppala PhD Randy DeFauw and Shelbee Eigenbrode

Amazon SageMaker Best Practices by Sireesha Muppala PhD Randy DeFauw and Shelbee Eigenbrode

Author:Sireesha Muppala, PhD, Randy DeFauw and Shelbee Eigenbrode
Language: eng
Format: epub
Publisher: Packt Publishing Pvt Ltd
Published: 2021-08-20T00:00:00+00:00


Amazon SageMaker's model registry is fully managed, meaning there are no servers to manage. It also natively integrates into SageMaker Pipelines, providing the ability to integrate directly with the model registry as a native step in your model build pipeline. It does this using the RegisterModel step.

For example, if you build a model build pipeline that contains the automated steps for data processing, training, and model evaluation, you can add a conditional step to validate the evaluation metric. If the evaluation metric is above a specified threshold (for example, accuracy > 90%), the pipeline can then be configured to automatically register your model.

SageMaker's model registry also integrates natively with SageMaker Pipelines projects. Projects allow you to automatically provision MLOps pipelines and provision patterns that take advantage of the model registry. SageMaker projects can be used to automatically set up the model package group, as well as the approval workflows that can be used to trigger the pre-configured downstream deployment pipeline.

Important note

Amazon SageMaker Pipelines is covered in more detail in Chapter 12, Machine Learning Automated Workflows. The model registry is a component within SageMaker Pipelines but can be used independently of SageMaker Pipelines.

Many of the parameters passed as input to the CreateModelPackage API are tailored for Amazon SageMaker use and integrations with other Amazon SageMaker features. For example, data that can be associated with model metrics has a direct correlation with metrics produced with features such as Amazon SageMaker Clarify, model statistical bias metrics, Amazon SageMaker Model Monitor, and data quality constraint metrics. In another example, the validation specification relates specifically to a SageMaker batch transform job run to evaluate the SageMaker model package.

In this section, we reviewed the high-level architecture and usage of the Amazon SageMaker model registry to provide a basis for comparison against other options that will be covered in the next sections. Multiple options are being covered in this chapter. This is in order to support a variety of use cases and to help you choose the right option for your specific use case.



Download



Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.