Arthur Platform 3.8.0
7 months ago by ReadMe API
Release notes for Arthur Platform 3.8.0
New Features:
- Added support for ranked list output type models for recommender systems and search use cases. This release includes support for 6 performance metrics for ranked list type models: Precision @k, Recall @k, Mean Average Precision @k, nDCG @K, Mean Reciprocal Rank, and Ranked List AUC.
- Added statuses for model enrichments in the API. Users who want to ensure that enrichments are ready before sending data can use the wait_for_enrichments=True flag in the send_inferences and send_bulk_inferences SDK functions.
- Cluster administrators can now adjust the memory resources used for serving anomaly detection enrichments.
Improvements:
- Reduced the training time of the anomaly detection enrichment by up to 30x.
- Improved user experience for selecting model tags to filter on in the UI
- Improved user experience managing notifications for SSO Users
Bug Fixes:
- Alerts that are set up as “Less Than” will be categorized correctly as “Less Than” on the alerts page
- Users can now successfully create an alert rules from the overview page
- The UI will no longer show users the option to delete a model unless they have permissions to do so
- Alerts created using custom metrics in the notebook can now be edited in the UI
- Improvements to sorting Inference Deep Dive Table
- Alert plot dates now appear in the correct order on the alerts overview page