AWS updates SageMaker, Secrets Manager, and Step Functions

What's going on with AWS this week? In this post, we'll bring you the finest in AWS news, including:

• SageMaker gets a smorgasbord of updates
• Secrets Manager now supports Rotation Windows
• Step Functions Local introduces mock responses
• Amazon Fraud Detector now includes Prediction Explanations
• Amazon Lex supports alternate transcriptions

Want to get the full details? Read on for more!

Get started with ACG and transform your career with courses and real hands-on labs in AWS, Microsoft Azure, Google Cloud, and beyond.

SageMaker continues to be a major area of focus for AWS, with several updates being released this week.

• SageMaker Autopilot, which helps build, train, and tune your models, has had a couple of updates. Autopilot now supports Apache Parquet files, and the maximum size for datasets is being raised from 10GB up to 100GB. You can also lodge requests with AWS Support to consume even larger datasets through Autopilot.

• SageMaker Data Wrangler, which helps streamline the processing of data for machine learning, has also added support for the popular JSON and ORC data formats, along with some new transformation capabilities.

• SageMaker JumpStart, which connects you to popular collections of machine learning models, can now deploy their endpoints into a custom VPC, and encrypt their data using custom KMS Settings. These are really important considerations for large enterprise customers with complex architectures.

AWS Secrets Manager now supports rotation windows for secrets that it manages.

Rotating secrets is critical to ensuring that our applications remain secure. Even when managed by a service like Secrets Manager, there’s still opportunities for compromise. The challenge is that when we rotate our secrets, everything doesn’t happen at exactly the same time. There may be short periods where applications using these secrets don’t match, and cause calls to be denied during the rotation.

Rotation Windows allow us to set specific times when we’d like our secrets to rotate during an expected maintenance window. This means if we do have issues for a short period of time, we can ensure it won’t impact as many users.

If your solution supports it, using alternating secrets in your rotation strategy can ensure higher reliability. But for solutions where only a single secret is used, having a Rotation Window can help make your solution more reliable.

Step Functions have been getting a lot of love over the last few months, and now we have the latest update with the release of service integration response mocking for Step Functions Local.

Last year, AWS upgraded Step Functions to support more than 200 AWS services and thousands of API actions. The ability to mock the responses from these service integrations during local testing makes means you can more throughly test your Step Functions in a local development environment.

Mock responses allow us to simulate the interactions we'd have with the service integrations. Now, these will only get us so far, and there are limitations (such as Step Functions Local not validating the format of the mock responses). But this does solve an existing gap when it comes to development with Step Functions, and overall, I expect Step Functions are only going to get more powerful and popular over time.

Fraud Detector has received a really interesting update, with the addition of new Prediction Explanations.

Like many machine learning products, Fraud Detector has given you a list of results which indicate the risk for potentially fraudulent activity. That’s cool, but there’s no real way to actually understand . . . why it was determined to be fraudulent. It’s the machine equivalent of asking someone a question, and them responding with, "I just know."

The inclusion of the new Prediction Explanations helps provide details into which particular variables impacted the prediction. Knowing the underlying factors behind a prediction both helps analysts assess the claim, and architects develop more proactive countermeasures. Just more proof that Amazon is continuing to invest in their managed machine learning services.

This feature is already available for models generated after June 30, 2021. If your model was generated before then, you’ll need to retrain your model to access this feature. If you’re interested in exploring Amazon Fraud Detector, check out the demo from AWS re:Invent 2020.

Amazon Lex has released a new feature which allows for multiple transcripts and confidence scoring of a users speech input.

We’ve all had the irritating experiences with voice interfaces where you say one thing, and the system interprets it as another.

This is nearly as much of a frustration for the people who build these systems as the people who use them. The beauty of this new feature with Amazon Lex is that it will offer multiple suggestions of how to transcribe it, along with confidence scoring.

With the confidence scoring and alternative interpretations, you can make your conversational interface a lot more robust — either by asking for clarification when it’s not very sure or by selecting a more likely alternate interpretation.

Any time you’re working with voice interfaces, it always pays to take the time to refine it for a smoother user experience. If you’re using Lex with speech input, this could be worth exploring as a refinement.

Want to keep up with all things AWS? Follow ACG on Twitter and Facebook, subscribe to A Cloud Guru on YouTube for weekly AWS updates, and join the conversation on Discord.

Now, go forth and learn all the things. (To help with that, check out our free courses — no credit card required!) And as always, keep being awesome, cloud gurus!