5. AML in a Nutshell
• Supervised
• Structured CSV text
• Little to no coding needed
• Hosted
• Limited configurability
• Data insights
• Pay as you go
6. AML Limitations
• “medium data”
• No model export
• Only AWS
• 4 input data types
• 3 prediction data types
• 2 regions
• 1 algorithm
7.
8. Training data
31, Private, 60229, HS-grad, 9,
Married-civ-spouse, Machine-op-
inspct, Husband, White, Male, 0, 0,
40, United-States, TRUE
<many more records>
20. Where does AML make sense?
• Easy intro to ML
• No coding needed
• Exploration of data
• Structured text data (esp if in AWS)
• Don’t need unlimited config knobs
• Don’t want to run infrastructure
• Don’t want to pay a lot of money
21. Alternatives
AWS:
• Domain specific (Rekognition, Lex)
• Deep learning AMI
• Sagemaker
• SparkML
Non AWS:
• Google, Azure, startups
• Open source (scikit-learn, tensorflow)
22. Thanks
• UCI for the data
Where to find out more
• https://aws.amazon.com/aml/
• AML O’Reilly video course:
https://bitly.com/introtoaml
• Code:
https://github.com/mooreds/amazonmachi
nelearning-anintroduction/