Skip to content

Release Notes

22.3.5

Enhancements

  • Upgraded scikit-learn and TensorFlow packages

22.3.2

Enhancements

  • Optimized authorization session handling

Bug Fixes

  • Fixed a bug with messages missing from event queues

22.3.0

Features

  • Added support for Tree SHAP
  • Added confidence intervals for PDPs

Bug Fixes

  • Fixed a bug with fallback values
  • Fixed a bug with feature attributions when using some column names
  • Fixed a bug with deleting projects
  • Fixed a bug with type violation alerts

22.1.0

Features

  • Added the ability to update ground truth labels multiple times

Enhancements

  • Refactored project homepage
  • Upgraded XGBoost from 0.90 to 1.5.1
  • Improved performance for monitoring metric API calls from the UI
  • Improved PDP endpoint performance

Bug Fixes

  • Fixed a bug with event ingestion for ranking models
  • Fixed a bug with field names containing * in SQL queries
  • Fixed a bug with SSO login sessions
  • Fixed a bug resulting in 400 errors when trying to change a password
  • Fixed a bug with sorting of alerts
  • Fixed a bug with confusion matrices for multiclass classification models

August Release (5136)


Features

  • New Homepage and Search Bar: This feature is available in preview mode. It provides users with quick-links to fiddler docs, provides details on recently viewed projects/models/datasets, starred projects and much more including the capability to search for projects/models/datasets using the search bar.
  • Fiddler ARchive(FAR) and MLFlow: These features are available in preview mode. These features allow users to upload containerized models to fiddler either using FAR or MLFLow method.
  • Model Monitoring Summary Dashboard: This feature is available in preview mode. It provides user summary statistics of model traffic, drift, data integrity violations and triggered alerts.
  • Ranking Model Monitoring Accuracy: This feature is available in preview mode. It introduces 2 performance metrics for ranking models: MAP (Mean Average Precision) and NDCG (Normalized Discounted Cumulative Gain).
  • Fiddler on Azure: Support for Fiddler Azure on-prem deployments.
  • Avro dataset upload: This feature allows users to upload Avro datasets using fiddler-client.

Enhancements/Bug Fixes

  • Product hardening fixes
  • Filtering on metadata columns from explain
  • Monitoring config support for lower bin size(5 min)
  • Fixed issue with feature sensitivity charts
  • Handling of uppercase column names for production data

Note

  • Fiddler-Client: New version(0.8) has been released, update to the latest version to access the new features.
  • Ranking Model Monitoring: As this feature is still in preview mode following points are to be noted,
    • feature importance and Evaluate tab are work in progress
    • to use Ranking Monitoring only publish_events_batch api is supported to publish events as of now, and it is very important that the events belonging to the same group ID are together in the file and not spread out. Otherwise, the NDCG and MAP (performance metrics for ranking models) computations will be wrong
    • Missing parts for Ranking Models will be added in upcoming releases.

May Release (3850)


Features

  • RBAC: Role Based Access Control, this feature allows admins to assign roles and permissions to users and teams at a project level
  • Ping Federate: Integration with Ping Federate for user authenication and SSO. Uses user info for RBAC authorization.
  • Time Range Selector: This feature allows the users to select custom time ranges.
  • Model Fairness: This feature is available in preview mode. Check with Fiddler team to enable this feature
  • TTL: Added configurable data retention for monitoring events and aggregate metrics

Enhancements/Bug Fixes

  • Product hardening fixes
  • Minor enhancements:

    • Long feature names
    • Improved error messages
    • RHEL-8 Support

April Release (3613)


Features/Enhancements

  • Onboarding enhancement: New onboarding workflow that allows users to attach a model artifact to an already registered model. See update_model API for more details.
  • Dataset refactor: With Dataset refactor, Project is the higher level entity, model and dataset are sub-entities under project.
  • ETL Batch mode - Introducing a new API publish_events_batch for ingesting production events into fiddler in batch mode. It supports different batch sources.
  • Product hardening fixes

March Release (3338)


Features/Enhancements

  • All-purpose Explanation AI (AXAI): Check this blog post for more information.
  • Onboarding enhancement: New onboarding workflow that allows users to upload a dataset and register a model in Fiddler using register_model API. See the register_model for details
  • Monitoring pre-flight check: Set the dry_run to True and run the samples events to make sure it matches with model inputs and output. This is only client side testing, no event ingested. See details in publish_event
Back to top