This week covered the latter half of Domain 3 (Modeling) — how to train models (Day1), tune them (Day2), generalize them (Day3), and make them converge well (Day4). Today we tie these four flows into a single decision chain and organize how to narrow answers from clues during the exam.
[Day1] Training Runtime Infrastructure
Estimator → input modes (File/Pipe/FastFile) → distributed (data/model parallel) → Spot+checkpoints
[Day2] Hyperparameter Tuning (AMT)
search space, target metrics → strategies (Bayesian/Random/Grid/Hyperband) → early stopping → warm start