Week 8 covers the latter half of Domain 3 (Modeling) — "how to train models, how to tune them, and how to generalize them." Today we dive into SageMaker Training Jobs, the runtime foundation of learning. After selecting an algorithm (Week 6), the next question is "on what compute, with what data transfer method, and how cheaply can we run this training?" The exam tests Estimator configuration, input modes (File/Pipe/FastFile), distributed learning strategies, and Spot training cost optimization as key indicators.
The flow of a SageMaker training job is always the same.
1. Define Estimator (image, instance, hyperparameters, output path)
2