Skip to main content

AWS SageMaker: training, deploying and monitoring models on AWS

How to run a real machine learning project on AWS SageMaker, from data on S3 to a monitored endpoint, with the invoice in view at every step.

Course Duration: 7h​

What You'll Learn​

  • Position SageMaker's components against the ML lifecycle you already know
  • Configure Studio, notebooks and IAM with just enough privilege
  • Store data on S3 in a shape SageMaker training jobs can consume
  • Launch training jobs with a built-in container, then with your own script
  • Tune hyperparameters and deploy real-time, serverless or batch inference
  • Automate the whole flow with SageMaker Pipelines and the Model Registry
  • Watch data drift with Model Monitor and cap the invoice with budgets

Prerequisites​

  • Course 20 (MLOps): experiment tracking, model registry, monitoring
  • Working knowledge of AWS S3 and IAM
  • Python, pandas, scikit-learn and XGBoost basics

Course Modules​

  1. Overview of SageMaker and its components
  2. Studio, notebooks and working environments
  3. Data on S3 and the Feature Store
  4. Training jobs and built-in containers
  5. Custom containers and training scripts
  6. Automatic hyperparameter tuning
  7. Real-time and serverless endpoints
  8. Batch transform for large volumes
  9. SageMaker Pipelines and the Model Registry
  10. Monitoring, alerts and cost control

The running example​

The churn model from course 20 is ported to SageMaker, one module at a time. Data lands on S3 in module 3, XGBoost trains on it in module 4, a custom scikit-learn script replaces it in module 5, tuning refines it in module 6, endpoints serve it in modules 7 and 8, a pipeline orchestrates everything in module 9, and Model Monitor plus a budget close the loop in module 10. Every module prints the estimated cost of what you launched.

Assessment and certificate​

The course ends with a 40-question exam covering every module. On success, a certificate of completion is issued; its number is verifiable on the platform.

Free courses, by contrast, end with a 5-question quiz and a preview of the certificate, without certification.