Posted On: Sep 25, 2023
You can now use Amazon SageMaker Model Monitor to run a single execution of a monitoring job, enabling you to get results for your machine learning and data performance on demand. Running a monitoring job on-demand provides you the flexibility to monitor specific aspects of your ML as required and allows adaptability for situations with irregular monitoring needs, such as troubleshooting.
Using SageMaker Model Monitor, customers can monitor their machine learning models and data in four key areas: data quality, model quality, model bias drift, and explainability (i.e., feature attribution drift). They can establish scheduled model monitoring jobs to track their ML system at regular intervals (e.g., every 1-24 hrs), and with this new capability they can also run on-demand monitoring jobs as needed. For example, customers can obtain monitoring results for their batch inference models and data within minutes by initiating a single monitoring job right after the batch inference job concludes, or quickly troubleshoot and iterate by running on-demand monitoring jobs.
One-time monitoring jobs with Amazon SageMaker Model Monitor is available in all Amazon Web Services regions where Amazon SageMaker Model Monitor is currently available.