TruEra v1.22 Release Notes

Release: 1.22

CLI version 9.3.0

Release Date: June 14, 2022

TruEra release notes are structured by product functional area and further organized as follows:

NEW – functionality introduced for the first time in this release of the software
ENHANCEMENTS – improvements/extensions to existing functionality
FIXES – resolution of reported/discovered issues and bugs

NEW

Tests added to model test harness for Stability and Fairness
Users can define tests with thresholds on metric relative to different splits. Ex: Accuracy of test split should be in a range of 10% of accuracy on train split

ENHANCEMENTS

During ingestion, users can trigger computations for predictions and feature influence
Identification of high error segments now more tuneable according to segment size
For error segment identification, you can now choose between AUC or classification accuracy as your optimization metric.
add_feature_metadata now uses add_data_collection to set feature maps
Models packaged locally are automatically verified
Python SDK available in distinct public and licensed modules

Optimized for size, segment cards are now pinned to the top of Segments (persisted) for improved UX

FIXES

Missing “key contributing feature” now included in Fairness Overview
Switched metric of interest from RMSE to MAE for high error segments in regression models
Better formatting of numbers across the product
Feature dropdowns enabled to support any number of features
Graceful failures when running models to get predictions
Bar graph in Points display now renders correctly

FIXES

Dashboard links from Monitoring panels to root cause analysis in Diagnostics fixed (navigated via right click to Explore in Diagnostics)
Data availability edge case bugs fixed in Stability when no valid data is available for computation

ENHANCEMENTS

Expanded and improved documentation content covering bug fixes in quick start data files, SHAP-to-TruEra feature influence comparison notebooks, and R2 discussions
REST API documentation now published
Neural Network model ingestion automatically detects the ML framework used (PyTorch or TensorFlow) in the model object, making model ingestion easier
Number of time steps validation for ground truth labels added in Neural Network model wrapper verification
Support for DataRobot API v2 models (Important: This constitutes a breaking change for ingestion of DataRobot API v1 models; a new --api_version argument must be specified via the CLI)