What's changed: Deepened MLS-C01 Chapter 6 (component roles/spot training/Experiments/Debugger/Pipelines, AI service Custom features & decision/Textract vs Rekognition/Forecast vs DeepAR, Ground Truth workforces/auto-labeling/build-buy spectrum + tables, scenarios, FAQ, traps; ja figures)
6.3Data Labeling and Build Strategy
The final call—understand Ground Truth (labeling), AI services vs custom, the cost/effort trade-off, and choosing wisely. Decide build vs buy smartly.
Supervised learning needs labeled data. Use Ground Truth for labeling, and decide whether to build at all by checking if an AI service suffices.
6.3.1Labeling and build-vs-buy
- SageMaker Ground Truth: streamline labeling training data with human + automated labeling.
- Prefer AI services: if a managed API meets the need, it is fastest and least effort, no training.
- Custom only when needed: for unique problems/data, build a custom model on SageMaker (full control, more effort).
- Decision axis: weigh cost, effort, and required accuracy/control to choose buy (AI services) vs build (SageMaker).
Common on MLS-C01: labeling training data = SageMaker Ground Truth, common tasks → AI services first (no training), and unique requirements → custom model on SageMaker. "Fast/low-effort" → AI services; "custom with your own data" → SageMaker.
Continue reading — free sign-up
You're reading the free preview. Sign up free to read this section in full, plus every chapter (including 4+) and all questions.