Arthur Platform 3.11.0

Release notes for Arthur Platform 3.11.0

Release Notes 3.11.0

API Update:

• We have removed the async_model_provisioning query parameter on the POST /api/v3/models endpoint. Consequentially, all model creation request are now handled asynchronously, which is an improved workflow for the reliability of model onboarding. Asynchronous model onboarding has been the default for SDK-onboarded models, and our API now reflects this behavior as well. Customers using the API to onboard models should take note in this change of behavior and update any automation/scripts that use the API to either adhere to the asynchronous pattern (see our documentation for more details).

New Features:

• There is a new permission available, update on the ground_truth resource, which allows users to label ground truth in the UI without also granting the ability to write on inference resources.
• Added a new deciles function to the Arthur query API, allowing users to retrieve decile values for numerical attributes. Refer to our documentation for more information on how to use the API to query these values.
• Administrators will have the ability to save and edit column presets by pinning or unpinning columns on the inference deep dive page at the model level. For those with non-admin permissions, users will see the columns pinned by the admin by default and are able to pin columns to their view only. Refer to our documentation for more information.
• Users now have the ability to see enrichment statuses in the UI under the “Model Details” and “Model Status” sections, providing insight on if newly uploaded inferences will be enriched.

Product Enhancements:

• Infrastructure Upgrades: Arthur now supports Kubernetes 1.25 to 1.27.
• Deprecation: The older classic UI has been removed from the platform. Any related DNS records can be safely deleted.
• Users will now see the name of the alert on triggered alert notification emails.
• The Fairness tab is now hidden for any non-classification models.
• The ingestion APIs now ignores all null ground truth records contained in bulk inference uploads.
• Users can now view the “model description” in the UI in the model details window
• Users are now able to add and view the objects detected & ground truth columns for bounding box models in the inference deep dive page.
• Users can now view class labels on the objects detected column in the inference deep dive page for bounding box models.

Bug Fixes:

• Fixed a formatting issue in the “view inferences” pop up on the insights page.
• Prevented the page from breaking when attempting to edit a performance alert.
• The BYTES and COUNT columns now appear in the correct order on the usage page.
• Fixed an issue where alerts were sometimes triggered after the alert rule had been disabled.
• Added the option to scroll when selecting an organization.
• Saving new tags now works as expected.
• Made improvements to the x-axis labels on the performance and drift plots.
• The Time Series Icon now appears when expected, under the “Model Activity” section.
• Fixed the formatting on the “Create alert rule” drop down on the overview page.
• The correct error message will now be shown when there is no data available to display on the performance chart.
• Uploading ground truth via parquet files now works correctly with ground truth class labels.
• Replaced broken resource links.
• Fixed an issue in the insights section where the view inference model did not move past the first page of inferences.