Using a model registry

A model registry allows you to centrally track key metadata for each model version. The granularity of metadata tracked is often dependent on the chosen implementation (Amazon SageMaker's model registry, a custom solution, or a third-party solution).  

Regardless of the implementation, the key metadata to consider includes model version identifiers, and the following information about each model version registered:

• Model inputs: These include metadata related to the inputs and versions of those inputs used to train the model. This can include inputs such as the name of the Amazon S3 bucket storing the training data, training hyperparameters, and the Amazon Elastic Container Registry (ECR) repository or container image used for training.
• Model performance: This includes model evaluation data such as training and validation metrics.
• Model artifact: This includes metadata about the training model artifact. At a minimum, this includes...