TruEra v1.29 Release Notes

​​Release: 1.29

CLI version 9.8.1

Release Date: 28 September 2022

Product Area: DIAGNOSTICS

NEW

• Python SDK

• • Feature Importance model test added
  • When packaging models, you can now pull the pip version from the current Python environment
  • Get and Set support for the type of performance metric desired added for remote workspaces

ENHANCEMENTS

• Web App

• • Points page now shows id and timestamp metadata if id and timestamp were ingested; these fields can also be exported from the Points page
  • Sorting by rank now saved into the query on the All Features page, allowing users to share a preferred rank-sorted link/query with others

• Python SDK

• • Transforms now supported in local environments
  • Models with transforms can now be downloaded
  • Support for uploading models with transforms from local to remote
  • Model test harness now allows test type filtering
  • Support for using both reference model and split at the same time for relative performance threshold

FIXES

• Web App

• • Correct error displayed when token API fails
  • Error thrown when a segment is selected from a Data Collection different than the Data Collection originally specified for the model
  • Model test summary action button now sets model comparison when clicked
  • Model selection issue resolved for aggressively selected models
  • Performance of Project Settings page corrected
  • Backend empty-segment error handling corrected 
  • Ad hoc Reports, Calibration, Swap Analysis, Feature Associations and Automated Segments removed from standard TruEra deployments
  • Global split drop-down disabled to correct error-handling of no splits in compare mode 
  • When selected split doesn’t contain ground truths, disclaimer on the Fairess page now shows the split name instead of the model

• Python SDK
   
  • Project Settings no longer change when resetting context during download/upload
  • Duplicate check for model stability test eliminated
  • Split and Model deletion now also deletes associated model tests

• Ingestion

• • Users can no longer create data sources with duplicate names, this prevents indeterministic behavior during ingestion.