Amazon Machine Learning

Introduction

What is Machine Learning

1. Common Issues
2. Why It Matters

What is Amazon Machine Learning

1. Key Elements
2. Interface
3. Regional Availability And Data Sources
4. Support

AML concepts

Types of Data Sources

1. S3
2. Redshift
3. RDS
4. External data
5. Conclusion

Loading Your Data

1. Real data
2. Prepping Your Data
3. IAM roles
4. Uploading Your Data To S3
5. Creating the Data Source
6. Statistics
7. Failure

Data Nuts and Bolts

1. Excluded Attributes
2. Known Data Types
3. Missing values
4. Data type differences
5. Row Identifiers
6. Statistics

Data considerations

1. How much data makes sense
2. Protecting your data
3. Server Side Encryption
4. Transformations Before You Create a Data Source
5. Anonymize your data

Initial Model Build

1. Building from the console
2. Custom Settings
3. Experiment

More About Your Model

1. Shuffling
2. Using recipes to modify the data your model sees
3. Transformations available
4. Example Recipe
5. Which recipe should you use

Batch vs real time predictions

1. How to decide
2. Batch
3. Real-time
4. Pick based on use case

More About Batch Predictions

1. Location of observations
2. Kicking off a batch prediction
3. How do you know when you’re done
4. Results

More About Real Time Predictions

Why Manage Your Models

1. Creating and Comparing Models With the Same Data Schema
2. Creating and Comparing Models With Different Schemas
3. Combining Model Results

Managing Via the API

Creating models via the API

Other Use Cases

Limits of AML

1. Read only models
2. Only structured text data
3. No model insight
4. Number of features
5. Category cardinality
6. Limited configurability
7. No model export
8. Server side encrypted data
9. Ill Fitting Use Cases

Alternatives

Conclusion