SageMaker Training

The mental model for where training and tuning sit in SageMaker's lifecycle - between data prep and deployment - and which SDK calls mark each stage.

Launch a training job against a built-in algorithm via the SDK - point at S3 data, set resources and hyperparameters, wait for completion, and read back the model artifact.

Choosing between AWS's built-in algorithm images and your own training container for a SageMaker training job - and how the same CreateTrainingJob call routes to each.

Organize training runs into experiments and trials with CreateExperiment/CreateTrial, associate a training job automatically via ExperimentConfig, then compare trial components to reproduce and rank runs.

A checklist for launching SageMaker training jobs from the SDK - job setup, cost controls, distributed and spot training, checkpointing, and reproducibility - for boto3 and AWS SDK v3.