A data scientist needs to package code into a container that produces both a new scikit-learn model forecast and a benchmark forecast, and wants AWS to handle operational maintenance of the container. Which packaging approach satisfies this requirement?

Extending the Amazon SageMaker scikit-learn container allows the data scientist to leverage AWS's operational maintenance for the base container while adding custom code for both the new scikit-learn model forecast and the benchmark forecast. This approach provides the flexibility to include additional logic beyond a simple training script, such as generating multiple forecasts, without building a container entirely from scratch. Providing code as a training script inside the existing container is too restrictive for generating two distinct forecasts. Building a custom container image and pushing it to Amazon ECR or deploying it to AWS Fargate would require the data scientist to manage the entire container's operational maintenance, which contradicts the requirement for AWS to handle it.