Arthur Platform 3.9.0

Release notes for Arthur Platform 3.9.0

• New Features:

  • Users on SSO-enabled environments can now copy their session token from the user-profile popup (bottom left corner), which can be used for creating sessions using our SDK/API (eg: in a notebook).
  • Ingestion Support for Time Series Inputs: Ability to ingest time series inputs and view the time series values at the inference level.
  • Added Support for Ranked List Outputs: Users are now able to visualize drift metrics and create new alerts on performance and prediction drift values for models with ranked list outputs.

• Product Enhancements:

  • A better error message is displayed upon log-in when users do not have access to any organizations.
  • Improved the way we calculate drift metrics on low inference counts to make them more interpretable.
  • Time interval selection is now hidden when in snapshot mode to avoid confusion between snapshot & time series modes on the overview page.
  • Users are now able to share and send alert specific instances to other SSO users.
  • Improved UX when selecting and viewing date and time in the global filters window.
  • Users now have the option to visualize the performance chart metrics in bar chart format in the UI.

• Bug Fixes:

  • Restored the ability to grab the scroll bar in drop down menus.
  • Segmenting the performance chart now works when the selected attribute category is not present in any inferences. For this case, the chart will now display that no data was selected, indicating to the user that no inferences match that specific attribute value.
  • Fixed formatting when filtering by batch from the global filters window.
  • Fairness plots now appear as expected for combo attributes in batch models.
  • Clicking on “view in overview” for a triggered alert now directs the user to the model overview page, filtered on the specific inferences captured in the alert.
  • Graph titles in the Alerts UI now correctly display the metric names.
  • Clicking on an alert in the “recent alerts” section now directs the user to the alerts page as expected.
  • Users are now able to add raw anomaly score as a column to the inference deep dive table for NLP models to sort by.
  • The “Inference Count by Class” performance metric now accurately accounts for configured classifier threshold values when shown on the Performance Chart.
  • Resolved how predicted values are showing up for predictions less-than 0.
  • The alert_rules API endpoint now returns a correctly formatted API response when there are no alert rules defined for a model.
  • Users are now able to sort by insights on the insights page.