Machine Learning Engineering on AWS

Course Description

Machine Learning (ML) Engineering on AWS is a 3-day deep dive into the practicalities of moving ML models from experimental scripts to production environments. This course addresses the challenges of scaling ML tasks, teaching you how to use Amazon SageMaker AI for model building and deployment, alongside Amazon EMR for large-scale data processing. You will learn to orchestrate full-stack ML workflows, automate your release cycles with CI/CD, and ensure the long-term reliability of your applications through advanced monitoring and security practices.

Who is this course for

This course is intended for builders who want to master the "Engineering" in Machine Learning Engineering:

• Machine Learning Engineers who need to operationalize models on the AWS Cloud.

• DevOps Engineers tasked with automating ML workflows (MLOps).

• Developers and SysOps Engineers seeking to understand the infrastructure and deployment patterns specific to AI/ML workloads.

Course Objectives

• Foundations & Data: Explain ML fundamentals and use AWS services to process, transform, and engineer data for training.

• Model Selection: Select appropriate algorithms and modeling approaches based on specific business requirements and the need for model interpretability.

• Pipeline Orchestration: Design and implement scalable ML pipelines for training, deployment, and automated orchestration.

• MLOps & CI/CD: Create automated continuous integration and delivery pipelines tailored for ML workflows.

• Security & Governance: Discuss and implement appropriate security measures for protecting ML resources on AWS.

• Production Monitoring: Implement monitoring strategies to detect data drift and maintain model performance after deployment.

Prerequisites

• Technical Skills: Working knowledge of Python and common data science libraries (NumPy, Pandas, Scikit-learn).

• AI/ML Knowledge: Familiarity with basic machine learning concepts.

• Cloud Knowledge: Basic understanding of cloud computing and familiarity with the AWS platform.

• Optional: Experience with version control systems like Git is beneficial.

Course outline

Section 1: Introduction to Machine Learning (ML) on AWS

• Introduction to ML

• Amazon SageMaker AI

• Responsible ML

Section 2: Analyzing Machine Learning (ML) Challenges

• Evaluating ML business challenges

• ML training approaches

• ML training algorithms

Section 3: Data Processing for Machine Learning (ML)

• Data preparation and types

• Exploratory data analysis

• AWS storage options and choosing storage

Section 4: Data Transformation and Feature Engineering

• Lab 1: Analyze and Prepare Data with Amazon SageMaker Data Wrangler and Amazon EMR

• Lab 2: Data Processing Using SageMaker Processing and the SageMaker Python SDK

• Handling incorrect, duplicated, and missing data

• Feature engineering concepts

• Feature selection techniques

• AWS data transformation services

Section 5: Choosing a Modeling Approach

• Amazon SageMaker AI built-in algorithms

• Selecting built-in training algorithms

• Amazon SageMaker Autopilot

• Model selection considerations

• ML cost considerations

Section 6: Training Machine Learning (ML) Models

• Lab 3: Training a model with Amazon SageMaker AI

• Model training concepts

• Training models in Amazon SageMaker AI

Section 7: Evaluating and Tuning Machine Learning (ML) models

• Lab 4: Model Tuning and Hyperparameter Optimization with Amazon SageMaker AI

• Evaluating model performance

• Techniques to reduce training time

• Hyperparameter tuning techniques

Section 8: Model Deployment Strategies

• Lab 5: Shifting Traffic A/B

• Deployment considerations and target options

• Deployment strategies

• Choosing a model inference strategy

• Container and instance types for inference

Section 9: Securing AWS Machine Learning (ML) Resources

• Access control

• Network access controls for ML resources

• Security considerations for CI/CD pipelines

Section 10: Machine Learning Operations (MLOps) and Automated Deployment

• Lab 6: Using Amazon SageMaker Pipelines and the Amazon SageMaker Model Registry with Amazon SageMaker Studio

• Introduction to MLOps

• Automating testing in CI/CD pipelines

• Continuous delivery services

Section 11: Monitoring Model Performance and Data Quality

• Lab 7: Monitoring a Model for Data Drift

• Detecting drift in ML models

• SageMaker Model Monitor

• Monitoring for data quality and model quality

• Automated remediation and troubleshooting