I.  Overview:

In this certification & training course, learn how to solve a real-world use case with machine learning and produce actionable results using Amazon SageMaker. This course teaches you how to use Amazon SageMaker to cover the different stages of the typical data science process, from analyzing and visualizing a data set, to preparing the data and feature engineering, down to the practical aspects of model building, training, tuning and deployment.

II.  Duration:  01 day

III. Objectives:

In this course, you will learn to:

• Apply Amazon SageMaker to a specific use case and dataset
• Practice all the steps of the typical data science process
• Visualize and understand the dataset
• Explore how the attributes of the dataset relate to each other
• Prepare the dataset for training
• Use built-in algorithms
• Train models with Amazon SageMaker using built-in algorithms
• Explore results and performance of the model, and demonstrate how it can be tuned and executed outside of SageMaker
• Run predictions on a batch of data with Amazon SageMaker
• Deploy a model to an endpoint in Amazon SageMaker for real-time predictions
• Learn how to configure an endpoint for serving predictions at scale
• Understand Hyperparameter Optimization (HPO) with Amazon SageMaker to find optimal model parameters
• Understand how to perform A/B model testing using Amazon SageMaker
• Perform the domain-specific cost of errors analysis to further tune the model threshold in order to maximize model utility expressed in financial terms

IV. Intended Audience:

This course is intended for:

• Data science practitioners
• Machine learning practitioners
• Developers and engineers
• Systems architects

V.    Prerequisites

We recommend that attendees of this course have:

• Experience with Python programming language
• Familiarity with NumPy and Pandas Python libraries is a plus
• Familiarity with fundamental machine learning algorithms
• Familiarity with productionizing machine learning models

VI. Course outlines:

      Module 1: Introduction to machine learning

-        Types of ML

-        Job Roles in ML

-        Steps in the ML pipeline

      Module 2: Introduction to data prep and SageMaker

-        Training and test dataset defined

-        Introduction to SageMaker

-        Demonstration: SageMaker console

-        Demonstration: Launching a Jupyter notebook

      Module 3: Problem formulation and dataset preparation

-        Business challenge: Customer churn

-        Review customer churn dataset

      Module 4: Data analysis and visualization

-        Demonstration: Loading and visualizing your dataset

-        Exercise 1: Relating features to target variables

-        Exercise 2: Relationships between attributes

-        Demonstration: Cleaning the data

      Module 5: Training and evaluating a model

-        Types of algorithms

-        XGBoost and SageMaker

-        Demonstration: Training the data

-        Exercise 3: Finishing the estimator definition

-        Exercise 4: Setting hyper parameters

-        Exercise 5: Deploying the model

-        Demonstration: hyper parameter tuning with SageMaker

-        Demonstration: Evaluating model performance

      Module 6: Automatically tune a model

-        Automatic hyper parameter tuning with SageMaker

-        Exercises 6-9: Tuning jobs

      Module 7: Deployment / production readiness

-        Deploying a model to an endpoint

-        A/B deployment for testing

-        Auto Scaling

-        Demonstration: Configure and test auto scaling

-        Demonstration: Check hyper parameter tuning job

-        Demonstration: AWS Auto Scaling

-        Exercise 10-11: Set up AWS Auto Scaling

      Module 8: Relative cost of errors

-        Cost of various error types

-        Demo: Binary classification cutoff

      Module 9: Amazon SageMaker architecture and features

-        Accessing Amazon SageMaker notebooks in a VPC

-        Amazon SageMaker batch transforms

-        Amazon SageMaker Ground Truth

-        Amazon SageMaker Neo