Introducing Amazon SageMaker Feature Store - a fully managed repository to store, discover, share and serve machine learning features
We’re excited to announce Amazon SageMaker Feature Store, a new capability of Amazon SageMaker to ingest, store, share, reuse, and serve features for real time and batch machine learning (ML) applications.
In machine learning, features are data signals that ML models rely on to make accurate predictions. During training, features are stored in batches to train multiple variations. The same features need to be available in real-time during inference for accurate predictions. Maintaining consistency between training and inference is challenging and may lead to inaccurate predictions or require additional coding.
Amazon SageMaker Feature Store is a fully managed repository that helps maintain consistency between features used at the time of inference and model training, so you can confidently deploy models in production with more predictable behavior allowing you to operate ML models at scale. Amazon SageMaker Feature Store enables metadata management and discovery of features with easy tagging and search, so data science teams can simply reuse an existing feature instead of having to rewrite and process features for each new model. For real time predictions, features can be served with low millisecond latency or extracted for model training or batch prediction use cases from the feature store. Amazon SageMaker Feature Store manages historical records of feature data so that features can easily be reproduced at a specific point in time. With Amazon SageMaker Feature Store, you can accelerate machine learning, increase productivity and scale across thousands of models.
Amazon SageMaker Feature Store is now generally available in all AWS regions in the Americas and Europe, and some regions in Asia Pacific with additional regions coming soon. You can find the details of the specific regions here. Read the documentation for more information and for sample notebooks. To learn how to use the feature visit the blog post.