Day 1- Business problem: Churn prediction
Load and display the dataset
Assess features and determine which Amazon SageMaker algorithm to use
Use Amazon Sagemaker to train, evaluate, and automatically tune the model
Deploy the model
Assess relative cost of errors
, Module 1: Introduction to Machine Learning
- Benefits of machine learning (ML)
- Types of ML approaches
- Framing the business problem
- Prediction quality
- Processes, roles, and responsibilities for ML projects
Module 2: Preparing a Dataset
- Data analysis and preparation
- Data preparation tools
- Demonstration: Review Amazon SageMaker Studio and Notebooks
- Hands-On Lab: Data Preparation with SageMaker Data Wrangler
Module 3: Training a Model
- Steps to train a model
- Choose an algorithm
- Train the model in Amazon SageMaker
- Hands-On Lab: Training a Model with Amazon SageMaker
- Amazon CodeWhisperer
- Demonstration: Amazon CodeWhisperer in SageMaker Studio Notebooks
Module 4: Evaluating and Tuning a Model
- Model evaluation
- Model tuning and hyperparameter optimization
- Hands-On Lab: Model Tuning and Hyperparameter Optimization with Amazon SageMaker
Module 5: Deploying a Model
- Model deployment
- Hands-On Lab: Deploy a Model to a Real-Time Endpoint and Generate a Prediction
Module 6: Operational Challenges
- Responsible ML
- ML team and MLOps
- Automation
- Monitoring
- Updating models (model testing and deployment)
Module 7: Other Model-Building Tools
- Different tools for different skills and business needs
- No-code ML with Amazon SageMaker Canvas
- Demonstration: Overview of Amazon SageMaker Canvas
- Amazon SageMaker Studio Lab
- Demonstration: Overview of SageMaker Studio Lab
- (Optional) Hands-On Lab: Integrating a Web Application with an Amazon SageMaker Model Endpoint, Introduction to data prep and SageMaker, Introduction to machine learning, Training and evaluating a model, Problem formulation and dataset preparation, Data analysis and visualization, Deployment / production readiness, Automatically tune a model, Amazon SageMaker architecture and features, Relative cost of errors