Arthur Platform 3.10.0

Release notes for Arthur Platform 3.10.0

Release Notes 3.10.0

New Features:

• Added support for calculating and viewing data drift on Time Series features.
• The query API now supports filtering Time Series attributes by the range of timestamps contained in the series.
• The ingestion API and SDK functions now support uploading reference data for Time Series models.
• Users are now able to save preferred column selections in the inference deep dive table to a user's local session.

Product Enhancements:

• The “All Time” global filter now filters from the UTC unix epoch (Jan 1, 1970).
• Users will now see an error message when they attempt to do something that's not permitted by their role.
• Upgraded the Confluent Platform from 6.x to 7.x for improved stability and security.
• The get organizations API can now show roles and limits for each org with an optional “expand” query parameter.
• Binary classifiers are now required to have a positive predicted attribute, and requests to onboard these models will fail unless a predicted attribute is specified as a positive predicted attribute.
• Users are now able to see predicted objects detected for computer vision models inline on the Inference Deep Dive table.
• Instituted a function that returns the max amount of K values for filtering and metrics selection for Ranked List Outputs.
• Users are now able to view the ranked item name in the inference Deep Dive Table.
• Introduced a maximum limit for attribute filters. Users can now select up to 100 attribute filters at one time.

Bug Fixes:

• Logging into SSO environments now works correctly when users are only in one organization.
• The Inference Deep Dive Table now filters to zero for features that do not exist.
• Replaced broken resource links.
• Global Filter state is now stored in local session storage instead of in the cookie, resolving issues seen under certain configurations where the ingress proxy would reject requests when the cookie was too large.
• Filtering for a model type that does not exist now works as expected.
• The Attribute Distribution chart now correctly shows the rate of categories (previously it was 2x the correct value).
• Improved filtering by data types in the schema drop down.
• The “interpretation” value when expanding an individual inference on the inference deep dive page now renders correctly for classifier models for true/false positive/negative values.
• Resolved an issue where in some cases the Global Filters timestamp was defaulting to an incorrect year.
• When an alert rule is created for a model that had no alert rules, it immediately shows on the list of alert rules as expected and users do not need to refresh their page.