v0.65.0

v0.65.0 Jan. 3, 2023

Enhancements

• Added the ability to retrieve prediction intervals for estimators that support time series regression #3876
• Added utils to handle the logic for threshold tuning objective and resplitting data #3888
• Integrated OrdinalEncoder into AutoMLSearch #3765
• Fixed DefaultAlgorithm adding an extra OneHotEncoder when a categorical column is not selected #3914

Fixes

• Fixed ARIMA not accounting for gap in prediction from end of training data #3884

Changes

• Added a threshold to DateTimeFormatDataCheck to account for too many duplicate or nan values #3883
• Changed treatment of Boolean columns for SimpleImputer and ClassImbalanceDataCheck to be compatible with new Woodwork inference #3892
• Split decomposer seasonal_period parameter into seasonal_smoother and period parameters #3896
• Excluded catboost from the broken link checking workflow due to 403 errors #3899
• Pinned scikit-learn version below 1.2.0 #3901
• Cast newly created one hot encoded columns as bool dtype #3913

Documentation Changes

• Hid non-essential warning messages in time series docs #3890

Testing Changes

v0.64.0

v0.64.0 Dec. 8, 2022

Fixes

• Allowed the DFS Transformer to calculate feature values for Features with a dataframe_name that is not "X" #3873
• Stopped passing full list of DFS Transformer features into cloned pipeline in partial dependence fast mode #3875

Changes

• Remove Int64Index after Pandas 1.5 Upgrade #3825
• Reduced the threshold for setting use_covariates to False for ARIMA models in AutoMLSearch #3868
• Pinned woodwork version at <=0.19.0 #3871
• Updated minimum Pandas version to 1.5.0 #3808

v0.63.0

v0.63.0 Nov. 23, 2022

Enhancements

• Added fast mode to partial dependence #3753
• Added the ability to serialize featuretools features into time series pipelines #3836

Fixes

• Fixed TimeSeriesFeaturizer potentially selecting lags outside of feature engineering window #3773
• Fixed bug where TimeSeriesFeaturizer could not encode Ordinal columns with non numeric categories #3812
• Updated demo dataset links to point to new endpoint #3826
• Updated STLDecomposer to infer the time index frequency if it's not present #3829
• Updated _drop_time_index to move the time index from X to both X.index and y.index #3829
• Added TimeSeriesPipeline.should_skip_featurization to fix bug where data would get featurized unnecessarily #3849
• Fixed bug where engineered features lost their origin attribute in partial dependence, causing it to fail #3830
• Fixed bug where partial dependence's fast mode handling for the DFS Transformer wouldn't work with multi output features #3830
• Allowed target to be present and ignored in partial dependence's DFS Transformer fast mode handling #3830

Changes

• Consolidated decomposition frequency validation logic to Decomposer class #3811
• Removed Featuretools version upper bound and prevent Woodwork 0.20.0 from being installed #3813
• Updated min Featuretools version to 0.16.0, min nlp-primitives version to 2.9.0 and min Dask version to 2022.2.0 #3823
• Rename issue templates config.yaml to config.yml #3844

Documentation Changes

• Added information about STL Decomposition to the time series docs #3835
• Removed RTD failure on warnings #3864

v0.62.0

v0.62.0 Nov. 1, 2022

Fixes

• Fixed bug with datetime conversion in get_time_index #3792
• Fixed bug where invalid anchored or offset frequencies were including the STLDecomposer in pipelines #3794
• Fixed bug where irregular datetime frequencies were causing errors in make_pipeline #3800

Changes

• Capped dask at < 2022.10.1 #3797
• Uncapped dask and excluded 2022.10.1 from viable versions #3803
• Removed all references to XGBoost's deprecated _use_label_encoder argument #3805
• Capped featuretools at < 1.17.0 #3805
• Capped woodwork at < 0.21.0 #3805

v0.61.1

v0.61.1 Oct. 27, 2022

Fixes

• Fixed bug where TimeSeriesBaselinePipeline wouldn't preserve index name of input features #3788
• Fixed bug in TimeSeriesBaselinePipeline referencing a static string instead of time index var #3788

Documentation Changes

• Updated Release Notes #3788

v0.61.0

v0.61.0 Oct. 26, 2022

Enhancements

• Added the STL Decomposer #3741
• Integrated STLDecomposer into AutoMLSearch for time series regression problems #3781
• Brought the PolynomialDecomposer up to parity with STLDecomposer #3768

Changes

• Cap Featuretools at < 1.15.0 #3775
• Remove Featuretools upper bound restriction and fix nlp-primitives import statements #3778

v0.60.0

v0.60.0 Oct. 19, 2022

Enhancements

• Add forecast functions to time series regression pipeline. #3742

Fixes

• Fix to allow IDColumnsDataCheck to work with IntegerNullable inputs. #3740
• Fixed datasets name for main performance tests. #3743

Changes

• Use Woodwork's dependence_dict method to calculate for TargetLeakageDataCheck #3728

Documentation Changes

Testing Changes

Warning

Breaking Changes

• TargetLeakageDataCheck now uses argument mutual_info rather than mutual #3728

v0.59.0

v0.59.0 Sep. 27, 2022

Enhancements

• Enhanced Decomposer with determine_periodicity function to automatically determine periodicity of seasonal target. #3729
• Enhanced Decomposer with set_seasonal_period function to set a Decomposer object's seasonal period automatically. #3729

Fixes

• Fixed holdout warning message showing when using default parameters #3727
• Fixed bug in Oversampler where categorical dtypes would fail #3732

Changes

• Automatic sorting of the time_index prior to running DataChecks has been disabled #3723

Documentation Changes

Testing Changes

• Update job to use new looking glass report command #3733

v0.58.0

v0.58.0 Sep. 20, 2022

Enhancements

• Defined get_trend_df() for PolynomialDecomposer to allow decomposition of target data into trend, seasonality and residual. #3720
• Updated to run with Woodwork >= 0.18.0 #3700
• Pass time index column to time series native estimators but drop otherwise #3691
• Added errors attribute to AutoMLSearch for useful debugging #3702

Fixes

• Removed multiple samplers occurring in pipelines generated by DefaultAlgorithm #3696
• Fix search order changing when using DefaultAlgorithm #3704

Changes

• Bumped up minimum version of sktime to 0.12.0. #3720
• Added abstract Decomposer class as a parent to PolynomialDecomposer to support additional decomposers. #3720
• Pinned pmdarima < 2.0.0 #3679
• Added support for using downcast_nullable_types with Series as well as DataFrames #3697

Documentation Changes

Testing Changes

• Updated pytest fixtures and brittle test files to explicitly set woodwork typing information #3697
• Added github workflow to run looking glass performance tests on merge to main #3690
• Fixed looking glass performance test script #3715
• Remove commit message from looking glass slack message #3719

v0.57.0 Sept. 6, 2022

v0.57.0 Sept. 6, 2022

• Enhancements
  • Added KNNImputer class and created new knn parameter for Imputer :pr:3662
• Fixes
  • IDColumnsDataCheck now only returns an action code to set the first column as the primary key if it contains unique values :pr:3639
  • IDColumnsDataCheck now can handle primary key columns containing "integer" values that are of the double type :pr:3683
  • Added support for BooleanNullable columns in EvalML pipelines and imputer :pr:3678
  • Updated StandardScaler to only apply to numeric columns :pr:3686
• Changes
  • Unpinned sktime to allow for version 0.13.2 :pr:3685
  • Pinned pmdarima < 2.0.0 :pr:3679