Releases: interpretml/interpret-community
Releases · interpretml/interpret-community
release v0.5.0
- Removed tabindex from GridChart
- Flask service for visualization dashboard
- Accessibility updates for visualization dashboard
- minor change of info message to debug for regression models explained by blackbox explainers
release v0.4.1
The 0.4.1 patch release includes several minor updates:
- separate base chart lib from interpret dashboard
- several build infrastructure updates
- fix sphinx doc warnings
- fix backcompat for explain local and global with data
release v0.4.0
- update shap to 0.33.0
- string localization in ExplanationDashboard
- add sections to does quack methods
- remove scikit-learn pin to older version and fix issues with latest scikit-learn release
patch release v0.3.1
- fixed deserialization logic for mimic explainer with older versions of lightgbm (hotfix)
release v0.3.0
- Add aria-label to top k slider
- Pass theme arg to chart components, accepting 'light' | 'dark' | 'white' | 'black' now
- Add comments to aid localization team
- Pin scikit-learn to 0.21.3 due to breaking changes in 0.22
- Make mimic explainer load function backwards compatible with older models
release v0.2.3
Patch release with minor bugfix to use length of per class rank instead of values for calculating num classes
release v0.2.2
Patch release with several minor bugfixes:
- fixing breaking change by making features public again on mimic explainer
- handle single class global importance values in ExplanationDashboard widget
- fixed warning on linear explainer due to unspecified default value
patch release v0.1.0.3.3
hotfix for logging mechanism failing when directory in specified path empty
release v0.2.1
Patch release with several minor bugfixes:
- add checks for fields in raw features explanation derived from engineered explanation
- add dataset size, num features and examples, to explanations
- fix lightgbm import warning appearing multiple times
- fix local file logger if directory to log to is missing
release v0.2.0
- Only get the eval_y_predicted and eval_data from explanation in _local_data method used by data contract if _DatasetsMixin exists on the explanation. Required for explanation dashboard widget to work on downloaded explanations without data.
- Add engineered explanation flag
- fix indexes with pandas:
add pandas indexes support, specifically for explaining timeseries models
add datetime featurization
fix bug in string indexing with one-hot-encoding for models that can't handle categoricals natively - update SHAP dependency to 0.32.1 to get several new fixes
Note: Versions will be in format ^.^.^ moving forward in future releases, as major.minor.patch