Product DocumentationAPI and Python SDK ReferenceRelease Notes
Schedule a Demo
Schedule a Demo
Back to All

Arthur Platform 3.7.0

8 months ago by ReadMe API

Release notes for Arthur Platform 3.7.0

Important Callouts:

  • Alert Notification Emails for SSO Users - We now require administrators of SSO-enabled environments to configure the set of allowed email domains to ensure that all emails from Arthur are sent to approved audiences, which follows information security best-practices. All users subscribed to Alert Rules MUST have a domain which matches this whitelist. Model Owners subscribing users to Alert Rules will not be able to subscribe an email to an Alert Rule unless the email address domain matches a domain on the whitelist.
    • This configuration can be modified as a comma-separated list of email domains in the Administration Console under “Advanced Settings → Email Domain Whitelist”

New Features:

  • Custom Performance Metrics: Users are now able to see custom vs. autogenerated metrics within the Performance chart section. Users are able to create Custom Performance Metrics using the Arthur Query Language and the Arthur SDK.
  • Ground Truth Labeling: Users can now label ground truths for individual inferences in the UI. A user can identify and select any of the predicted labels as the ground truth for a certain inference and make that update in the UI from the inference deep dive table.

UI Enhancements:

  • Added default metric for “Average Predicted Probability” which can be seen in the Performance Chart in the model overview page
  • API Login attempts that fail 10 times within 10 minutes will now lock the user out and show a lockout message in the UI.
  • Added messaging around Null values and calculating Anomaly scores to better indicate the loading and calculation processes
  • Users can now click and scroll the scrollbar in dropdown menus to view & select all options.

Bug Fixes:

  • Fixed the ability for users to filter by individual model tags on the Model Overview Page
  • The create model API now returns a nice error message when users do not mark the input attribute of an NLP model as categorical.

API Improvements:

  • Accuracy functions can now be used as inputs to other functions in the query API.
  • Added more sorting capabilities to the V4 metrics APIs such as name, created_at, and updated_at.
  • Added new events to the audit log for creating/updating/deleting custom metrics.
  • Custom metrics automatically track the user who last updated them.

SDK Improvements:

  • The SDK now uses the V4 metrics APIs for managing custom metrics.
  • Reference Dataset Validation: Added more data validation checks and made errors easier to understand for an all-around easier model onboarding experience.
  • In addition to being able to upload a directory of parquet files, users can now provide file objects to the upload function.