Machine learning in a database

Aurora

Aurora runs on async write.
The storage is replicated and shared. The replica in next AZ just needs cache sync.
Writes to 6 storage instances across the 3 AZ's in the region
When 4 out of 6 writes done, it's quorate.

MySQL compatible
Supports serverless with an HTTP endpoint, or AWS SDK
Serverless scales easily.
Can create global database

Aurora integrates with ML

• Amazon comprehend
• Amazon Sagemaker
• Aurora MySQL

Use a select and the services run on the data selected.

direct SQL ML

One endpoint for read/write optimised, one for read only.
Attach AWS Lambda or Sagemaker or Comprehend Call directly in SQL, e.g.

select comment_text, aws_comprehend_detect_sentiment(comment_text, 'english') from comment

Sagemaker

Sagemaker autopilot: Pass a well prepared dataset and it does the ML Set up model and publish as an endpoint Write a function in the database to invoke the end point, passing the features, returning prediction. Then select, and call the function.

Workshop on amazon aurora looks useful: awsauroralabsmy.com

Easy to add ML to existing applications, just call an endpoint.

aws labs on github is a good resource.