Last week we prepared data and organized it as features. Starting this week, we actually train models with that data. The basic unit of model training in SageMaker is a Training Job. Instead of directly running for epoch in ... in a notebook, you submit a job: "train this data in this container with this instance", SageMaker spins up instances, completes training, saves results to S3, and automatically terminates instances.
In the MLA-C01 exam, Training Job appears as keywords like "Estimator configuration", "input channels (S3/FSx)", "instance selection", "cost reduction via Spot training". Today we cover the structure of training jobs and four axes for managing costs.