v0.74.0

v0.74.0 Apr. 19, 2023

Enhancements

• Saved computed additional_objectives computed during search to AutoML object #4141
• Remove extra naive pipelines #4142

Fixes

• Fixed usage of codecov after uploader deprecation #4144
• Fixed issue where prediction intervals were becoming NaNs due to index errors #4154

Changes

• Capped size of seasonal period used for determining whether to include STLDecomposer in pipelines #4147

v0.73.0

v0.73.0 Apr. 11, 2023

Enhancements

• Allowed InvalidTargetDataCheck to return a DROP_ROWS DataCheckActionOption #4116
• Implemented prediction intervals for non-time series native pipelines using the naïve method #4127

Changes

• Removed unnecessary logic from imputer components prior to nullable type handling #4038, #4043
• Added calls to _handle_nullable_types in component fit, transform, and predict methods when needed #4046, #4043
• Removed existing nullable type handling across AutoMLSearch to just use new handling #4085, #4043
• Handled nullable type incompatibility in Decomposer #4105, :pr:`4043
• Removed nullable type incompatibility handling for ARIMA and ExponentialSmoothingRegressor #4129
• Changed the default value for null_strategy in InvalidTargetDataCheck to drop #4131
• Pinned sktime version to 0.17.0 for nullable types support #4137

Testing Changes

• Fixed installation of prophet for linux nightly tests #4114

v0.72.0

• Enhancements
  * Updated pipeline.get_prediction_intervals() to add trend prediction interval information from STL decomposer #4093
• Fixes
  • Fixed ensemble pipelines not working with generate_pipeline_example #4102
• Changes
  • Pinned ipywidgets version under 8.0.5 #4097
  • Calculated partial dependence grid values for integer data by rounding instead of truncating fractional values #4096
• Testing Changes
  • Updated graphviz installation in GitHub workflows to fix windows nightlies #4088

v0.71.0

v0.71.0 Mar. 18, 2023

Fixes

• Fixed error in PipelineBase._supports_fast_permutation_importance with stacked ensemble pipelines :pr:4083

v0.70.0

v0.70.0 Mar. 16, 2023

Changes

• Added Oversampler nullable type incompatibility in X #4068
• Removed nullable handling from objective functions, roc_curve, and correlation_matrix #4072
• Transitioned from prophet-prebuilt to prophet directly #4045

v0.69.0

v0.69.0 Mar. 15, 2023

Enhancements

• Move black to regular dependency and use it for generate_pipeline_code #4005
• Implement generate_pipeline_example #4023
• Add new downcast utils for component-specific nullable type handling and begin implementation on objective and component base classes #4024
• Add nullable type incompatibility properties to the components that need them #4031
• Add get_evalml_requirements_file #4034
• Pipelines with DFS Transformers will run fast permutation importance if DFS features pre-exist #4037
• Add get_prediction_intervals() at the pipeline level #4052

Fixes

• Fixed generate_pipeline_example erroring out for pipelines with a DFSTransformer #4059
• Remove nullable types handling for OverSampler #4064

Changes

• Uncapped pmdarima and updated minimum version #4027
• Increase min catboost to 1.1.1 and xgboost to 1.7.0 to add nullable type support for those estimators #3996
• Unpinned networkx and updated minimum version #4035
• Increased scikit-learn version to 1.2.2 #4064
• Capped max holidays version to 0.21 #4064
• Stop allowing knn as a boolean impute strategy #4058
• Capped nbsphinx at < 0.9.0 #4071

Testing Changes

• Use release.yaml for performance tests on merge to main #4007
• Pin github-action-check-linked-issues at v1.4.5 #4042
• Updated tests to support Woodwork's object dtype inference for numeric columns #4066
• Updated TargetLeakageDataCheck tests to handle boolean targets properly #4066

v0.68.0

v0.68.0 Feb. 15, 2023

Enhancements

• Integrated determine_periodicity into AutoMLSearch #3952
• Removed frequency limitations for decomposition using the STLDecomposer #3952

Changes

• Remove requirements-parser requirement #3978
• Updated the SKOptTuner to use a gradient boosting regressor for tuning instead of extra trees #3983
• Unpinned sktime from below 1.2, increased minimum to 1.2.1 #3983

Testing Changes

• Add pull request check for linked issues to CI workflow #3970, #3980
• Upgraded minimum IPython version to 8.10.0 #3987

v0.67.0

v0.67.0 Feb. 1, 2023

Fixes

• Re-added TimeSeriesPipeline.should_skip_featurization to fix bug where data would get featurized unnecessarily #3964
• Allow float categories to be passed into CatBoost estimators #3966

Changes

• Update pyproject.toml to correctly specify the data filepaths #3967

Documentation Changes

• Added demo for prediction intervals #3954

v0.66.1

v0.66.1 Jan. 26, 2023

Fixes

• Updated LabelEncoder to store the original typing information #3960
• Fixed bug where all-null BooleanNullable columns would break the imputer during transform #3959

v0.66.0

v0.66.0 Jan. 25, 2023

Enhancements

• Improved decomposer determine_periodicity functionality for better period guesses #3912
• Added dates_needed_for_prediction for time series pipelines #3906
• Added RFClassifierRFESelector and RFRegressorRFESelector components for feature selection using recursive feature elimination #3934

Fixes

• Fixed set_period() not updating decomposer parameters #3932
• Removed second identical batch for time series problems in DefaultAlgorithm #3936
• Fix install command for alteryx-open-src-update-checker #3940
• Fixed non-prophet case of test_components_can_be_used_for_partial_dependence_fast_mode #3949

Changes

• Updated PolynomialDecomposer to work with sktime v0.15.1 #3930
• Add ruff and use pyproject.toml (move away from setup.cfg) #3928
• Pinned `category-encoders`` to 2.5.1.post0 #3933
• Remove requirements-parser and tomli from core requirements #3948