Domain 1 of 5 · Chapter 3 of 3

ML Development Lifecycle

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing AWS Certified AI Practitioner AIF-C01 premium course — no separate purchase.

Included in this chapter:

  • The lifecycle stage by stage: what actually happens
  • AWS mechanics: a SageMaker AI capability for each stage
  • Exam-pattern recognition: stem → stage, tool, and metric

ML lifecycle stage → Amazon SageMaker AI capability → purpose

Lifecycle stageSageMaker AI capabilityWhat it does
EDA & data pre-processingSageMaker Data WranglerVisually explore, clean, and transform tabular data with low/no code before training
Feature engineeringSageMaker Feature StoreCentrally store, version, and share engineered features across teams for training and inference
Model sourcing / quick startSageMaker JumpStartDeploy or fine-tune open-source and pre-trained models instead of training from scratch
Model training & tuningSageMaker training jobs (with automatic model tuning)Run managed, scalable training and hyperparameter optimization jobs
Evaluation (bias & explainability)SageMaker ClarifyDetect bias and explain feature importance during evaluation
Model governanceSageMaker Model RegistryVersion models and gate deployment with approval status for MLOps
Orchestration / CI-CDSageMaker PipelinesAutomate the end-to-end workflow as a repeatable, auditable pipeline
Production monitoringSageMaker Model MonitorDetect data and model drift on live endpoints to trigger retraining

Decision tree

Build a custom model,or task already solved?Managed API fitsManaged AI serviceComprehend / Rekognition:skip lifecycleOwn the modelPre-deployment build,or run in production?Building / evaluatingWhich build stagedo you need?DeployedDrift watch, or make itrepeatable / governed?EDA / prepData Wranglerclean & transformReuse featuresFeature Storeshare engineered featuresHead startJumpStartpre-trained / fine-tuneBias / explainClarifybias & explainabilityLive driftModel Monitordetect drift,trigger retrainMLOps CI/CDPipelines +Model RegistryAlways: the lifecycle loops — monitoring feeds retraining
Cheat sheet

Cheat sheet

  • The ML lifecycle is an iterative loop, not a one-way pipeline
  • Deployment is the start of monitoring, not the finish line
  • EDA understands the data before any model is built
  • Hold the test set out of all training and tuning
  • Use Data Wrangler to prep and explore data with little to no code
  • Feature Store reuses features once to kill training-serving skew
  • Feature Store offers an online store, an offline store, or both
  • JumpStart starts you from pre-trained models instead of from scratch
  • A model comes from scratch or from a pre-trained starting point
  • Parameters are learned during training; hyperparameters are set before it
  • Automatic Model Tuning searches hyperparameters for you
  • Clarify is the answer for bias detection and explainability
  • Model Registry versions models and gates promotion with approval
  • Pipelines orchestrate the ML workflow into repeatable CI/CD
  • Model Monitor detects drift on models already in production
  • Managed API vs self-hosted API: who runs the endpoint
  • MLOps applies DevOps discipline to the model lifecycle
  • Precision vs recall: trade off false positives against false negatives
  • Accuracy misleads on imbalanced data; use F1 or AUC
  • Business metrics, not just model metrics, decide success

Unlock with Premium — includes all practice exams and the complete study guide.

References

  1. Prepare data with Amazon SageMaker Data Wrangler
  2. Amazon SageMaker Feature Store
  3. Amazon SageMaker JumpStart
  4. Deploy models for inference with Amazon SageMaker AI
  5. Automatic model tuning with Amazon SageMaker AI
  6. Fairness, model explainability and bias detection with SageMaker Clarify
  7. Register and deploy models with Model Registry
  8. Amazon SageMaker Pipelines
  9. Monitor models for data and model quality with Amazon SageMaker Model Monitor
  10. Real-time inference with SageMaker AI endpoints
  11. Batch transform for inference with Amazon SageMaker AI
  12. MLOps with Amazon SageMaker AI