CLI version 11.3.0
Release Date: 22 February 2023
TruEra release notes are structured by product functional area and further organized as follows:
Product Area: DIAGNOSTICS
- Stability diagnostic functionality renamed to Drift to more accurately reflect its analytics
- Python SDK
- Baseline model creation now supported in local workspace
- Support for training baseline models as data splits are ingested
- Python SDK
- Prediction caches now supported for non-virtual models.
- Default project upload behavior updated to support non-virtual models.
- Interesting segments now operate on labels.
- Point-wise metrics support — mean squared error (MSE)/mean squared logarithmic error. (MSLE) — for interesting segments.
- Public SDK now includes local add-on package.
- Support added for new classification metrics using suggest_high_error_segments().
- New “interesting segment” type added based on under/oversampling.
- Support for changing column names of output for high error segments.
- Visualization packages plotly and nbformat are now pre-installed with the truera-local whl
- Improved validation of materialize calls during data ingestion to check for columns that don't exist.
- XGBoost is now used in load_model() to deserialize xgb models in quickstart.
- Model Test
- More robust handling of virtual models in Model Test; relative threshold behavior updated to treat the threshold as a percentage of the reference score.
- Code samples/examples added to SDK method and function definitions throughout the Python SDK Technical Reference.
- Additional definition/description added for get_model() functions in custom model wrappers in the Python SDK Technical Reference.
- In order to prevent indeterministic behavior during ingestion, users can no longer create data sources with duplicate names; the error is now trapped.
Please sign in to leave a comment.