See How Fiddler Works in Amazon SageMaker AI
Fiddler offers a native integration with AWS SageMaker AI, enabling users to observe ML models such as binary classification, multi classification, regression, ranking, and more from within their secure AWS environment. This integration offers a seamless ML lifecycle experience with full data privacy and rapid, secure scaling — all within your secure AWS environment. Watch the video to learn how to set up and run Fiddler from within SageMaker Studio using a free trial.
Fiddler and AWS SageMaker Integration
[00:00:00] Fiddler is an AI Observability and security platform that helps you monitor, analyze, and protect AI models. Fiddler offers a native integration with AWS SageMaker AI enabling users to observe ML models from within their secure AWS environment.
Key Benefits of Fiddler-SageMaker Integration
[00:00:17] This integration brings a few key benefits. One, a smooth workflow from model building to deployment to production monitoring. It all happens seamlessly inside your stage maker. Instance. Two, full data privacy. Your data stays inside your environment. No third party sharing
[00:00:33] And three faster deployments because of the integration security. There's less red tape for you around procurement and implementation.
[00:00:40]
Step-by-Step Integration Process
[00:00:41] To access the Fiddler integration, start by logging into your SageMaker portal and select the partner AI apps.
[00:00:47] Then choose view details under Fiddler.
[00:00:52] Here we can see the details of the Fiddler trial. And when you're ready, click the Go to Marketplace to subscribe button.
[00:00:59] From here, we can subscribe to the Fiddler Free Trial using the try for free button. Once completed, our last step is to return to the SageMaker portal and quickly configure our new Fiddler instance.
Using Fiddler Charts and Dashboards
[00:01:12] Now that we've deployed Fiddler within our SageMaker studio, let's take a look around. We can configure alerts and build custom charts and dashboards to help monitor model performance and gain insights. For example, using the insights tab, I can click into the custom dashboard for my churn prediction model.
[00:01:34] Looking at my initial charts and graphs, I can see that we have a major drop in revenue for our users in Hawaii.
[00:01:40] If I continue scrolling, I can see the accuracy of the Hawaii model dropped around May 16th.
[00:01:48] And I can also tie into a spike in model drift.
[00:01:51] To understand what's causing this drift. I can launch root cause analysis from my custom chart, which breaks down feature level impacts.
[00:02:03] Here I can see that Fiddler has found the likely cause of my model drift based on our prediction drift impact score.
[00:02:09] I can do the same for data integrity problems like missing values type, or range violations. Ready to try Fiddler for yourself? Request a demo or visit the Fiddler Listing in the AWS Marketplace for a free trial.
