Practical Data Science with Amazon SageMaker

Agenda

1 - Introduction to Machine Learning

• Types of ML
• Job Roles in ML
• Steps in the ML pipeline

2 - Introduction to Data Prep and SageMaker

• Training and Test dataset defined
• Introduction to SageMaker
• Demonstration: SageMaker console
• Demonstration: Launching a Jupyter notebook

3 - Problem formulation and Dataset Preparation

• Business Challenge: Customer churn
• Review Customer churn dataset

4 - Data Analysis and Visualization

• Demo: Loading and Visualizing your dataset
• Exercise 1: Relating features to target variables
• Exercise 2: Relationships between attributes
• Demo: Cleaning the data

5 - Training and Evaluating a Model

• Types of Algorithms
• XGBoost and SageMaker
• Demo 5: Training the data
• Exercise 3: Finishing the Estimator definition
• Exercise 4: Setting hyperparameters
• Exercise 5: Deploying the model
• Demo: Hyperparameter tuning with SageMaker
• Demo: Evaluating Model Performance

6 - Automatically Tune a Model

• Automatic hyperparameter tuning with SageMaker
• Exercises 6-9: Tuning Jobs

7 - Deployment / Production Readiness

• Deploying a model to an endpoint
• A/B deployment for testing
• Auto Scaling Scaling
• Demo: Configure and Test Autoscaling
• Demo: Check Hyperparameter tuning job
• Demo: AWS Autoscaling
• Exercise 10-11: Set up AWS Autoscaling

8 - Relative Cost of Errors

• Cost of various error types
• Demo: Binary classification cutoff

9 - Amazon SageMaker architecture and features

• Accessing Amazon SageMaker notebooks in a VPC
• Amazon SageMaker batch transforms
• Amazon SageMaker Ground Truth
• Amazon SageMaker Neo