Diving deep on SageMaker Autopilot

In this section, we're going to learn in detail how SageMaker Autopilot processes data and trains models. If this feels too advanced for now, you're welcome to skip this material. You can always revisit it later once you've gained more experience with the service.

First, let's look at the artifacts that SageMaker Autopilot produces.

The job artifacts

Listing our S3 bucket confirms the existence of many different artifacts:

$ aws s3 ls s3://sagemaker-us-east-2-123456789012/sagemaker/DEMO-autopilot/output/my-first-autopilot-job/

We can see many new prefixes. Let's figure out what's what:

PRE data-processor-models/
PRE documentation/
PRE preprocessed-data/
PRE sagemaker-automl-candidates/
PRE transformed-data/
PRE tuning/
PRE validations/

• The preprocessed-data/tuning_data prefix contains the training and validation splits generated from the input dataset. Each split is broken down further into small...