1.3.3. Mapping AWS Services to the ML Pipeline (SageMaker, Data Wrangler, Model Monitor)

First Principle: AWS provides a portfolio of services that map directly to the stages of the ML lifecycle, enabling teams to build, train, and operate ML solutions efficiently on a unified platform.

Knowing which tool to use at which stage is a key skill for a practitioner.

• Data Prep (EDA, Pre-processing, Feature Engineering):
  • Amazon SageMaker Data Wrangler: A visual tool to explore, clean, and prepare data with minimal code.
• Model Training & Tuning:
  • Amazon SageMaker: The core service for training custom models, either with built-in algorithms or your own code. It also handles hyperparameter tuning automatically.
• Model Management & Governance:
  • Amazon SageMaker Feature Store: A central repository to store, share, and manage curated features for ML models.
  • Amazon SageMaker Model Registry: A central place to catalog, version, and manage your trained models for deployment.
• Deployment:
  • Amazon SageMaker: Deploys models to real-time endpoints or for batch processing.
• Monitoring:
  • Amazon SageMaker Model Monitor: Automatically detects data drift and model quality degradation in production.
• Orchestration:
  • Amazon SageMaker Pipelines: A service to build, automate, and manage end-to-end ML workflows.

Scenario: A team needs to build a custom classification model. They have raw data in Amazon S3 and want to use a managed, integrated workflow on AWS.

Reflection Question: How would you guide them to use Data Wrangler for data prep, SageMaker for training, and Model Monitor for post-deployment, explaining how these services work together in a cohesive pipeline?

💡 Tip: Think of Amazon SageMaker as the "workbench" for the ML practitioner. It contains specialized tools (like Data Wrangler, Feature Store, Model Monitor) for every phase of the custom model-building process. Use Amazon SageMaker Pipelines to automate and orchestrate these steps into a repeatable workflow.