QuestionQ84

VerifiedDeployment and Orchestration of ML Workflows

A company has created a new ML model. Before fully releasing the model to production, the company requires online model validation on 10% of traffic. The company serves the model through an Amazon SageMaker endpoint behind an Application Load Balancer (ALB).

Which solution will configure the required online validation with the LEAST operational overhead?

Explanation

Amazon SageMaker production variants provide built-in A/B testing by distributing endpoint invocation traffic among model variants according to their assigned weights. Adding the new model as a variant on the existing endpoint and assigning it a 10% traffic weight avoids a separate endpoint and custom ALB traffic-routing configuration. Amazon CloudWatch provides invocation metrics for monitoring the variants.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!