Changelog

All notable changes to this project are documented in this file since v0.1.8. The source code for most releases is available on GitHub.

[0.19.10] - 2024-10-14

Added

Changed

Fixed

Fixed the name of a paramter in auxmodels.mlp_stack (#234)

[0.19.9] - 2024-07-06

Added

Added Forecaster.synthesize_models()

Changed

Fixed

Made compatible with more recent versions of the shap and xgboost packages

[0.19.8] - 2024-01-22

Added

Changed

Fixed

Fixed notebook.results_vis() function when the dictionary keys are not strings.
Transfer learning no longer fails when fitted values cannot be generated – still investigating the root cause of this issue.

[0.19.7] - 2023-12-12

Added

Added multivariate transfer learning

Changed

must_keep in Forecaster.auto_Xvar_select can now be str type and will maintain the expected behavior.

Fixed

[0.19.6] - 2023-12-04

Added

Changed

Fixed

Fixed how ARIMA model was reading future xreg values.

[0.19.5] - 2023-10-30

Added

Added fi_try_order as an argument in tune_test_forecast().

Changed

Fixed

Got rid of the AdditiveExplainer and GPUTreeExplainer in Forecaster.save_feature_importance() which do not work with feature importance as it is currently set up.

[0.19.4] - 2023-10-27

Added

Added more feature importance options, all sourced through the shap library.

Changed

shap is now a requirement and eli5 is not.
Changed Forecaster.reduce_Xvars() to use only shap feature importance to rank features.
Removed fi_method argument from tune_test_forecast().

Fixed

Fixed how a pandas function was called that was raising a warning.
Fixed feature importance to use shap only with TreeExplainer, PermutationExplainer, and other explainers (#85). See the docs The eli5 package appears to be deprecated.

[0.19.3] - 2023-10-02

Added

Added 'verbose':[-1] to lightgbm grids.

Changed

Refactored code and changed library version requirements to eliminate warnings that were originating from the seaborn (#49) and pandas libraries.

Fixed

[0.19.2] - 2023-09-21

Added

Added set_aside_test_set argument to the util.find_optimal_transformation() function to prevent leakage.
Added the util.gen_rnn_grid() function.
Tensorflow models now save the ymin and ymax attributes in history to facilitate transfer learning.

Changed

Changed where boxcox_tr() and bocxcox_re() functions are located in src so that Transformer and Reverter objects can more easily be pickled and the output when printed is more descriptive.
Changed the lightgbm default grid to suppress output.

Fixed

Added **kwargs in util.backtest_for_resid_matrix() to make pipelines that require kwargs work.

[0.19.1] - 2023-09-16

Added

Added transfer learning for RNN and LSTM models.
Added more Xvars to infer in the util.infer_apply_Xvar_selection() function.
Added regr argument to Forecaster.transfer_predict().
Added Forecater.save_tf_model() and Forecaster.load_tf_model() functions.

Changed

Transferred models store Xvar info in history.

Fixed

Fixed an issue with RNN models where AR terms that were not sequential were not counted correctly.
Fixed an issue where Xvars were not being recorded correctly in history for RNN models.

[0.19.0] - 2023-09-15

Added

Added Forecaster.transfer_predict() method. Only univariate sklearn models supported for now (#77).
Added Forecaster.transfer_cis() method.
Added carry_fit_models attribute in Forecaster object that can be changed when object is initialized.
Added util.infer_apply_Xvar_selection() function.

Changed

Changed how many history attributes are stored for each evaluated model, making the Forecaster object more memory efficient.
Refactored forecasting code for sklearn models so that model evaluation is more efficient.
Changed the max_ar = 'auto' behavior in Forecaster.auto_Xvar_select().
Changed scikit-learn dependency to <1.3.0 due to it not working with the shap library.

Fixed

Fixed an issue with combo modeling where defaults were not working when a previous model had been run test only.

[0.18.16] - 2023-08-10

Added

Changed

Fixed

Fixed an issue where impute_lookback=None was not working with fill_strategy='moving_seasonal_average' (#70).

[0.18.15] - 2023-08-07

Added

Added models argument to util.backtest_metrics.

Changed

Seasonality is inferred with more functions.
Made find_optimal_transformation never evaluate confidence intervals to save resources.

Fixed

Made a more descriptive error when Forecaster/MVForecaster objects are initiated cis=True and an insufficient test size.

[0.18.14] - 2023-08-06

Added

Changed

Forecaster.add_ar_terms() now accepts collections as arguments and will add individual lags to the object according to what is passed there.
Prophet and Silverkite models now accept direct autoregressive forecasting by passing lagged variables (‘AR…’) to the Xvars arguments in the models. As long as the lag order is the same or greater than the forecast horizon, lags are accepted.
Cleaned up silverkite model code.

Fixed

[0.18.13] - 2023-08-03

Added

Added util.Forecaster_with_missing_vals() function (#66).
Added Forecaster.round() method.

Changed

Fixed

Fixed banking history for failed models. Empty dictionaries no longer saved to Forecaster.history nor MVForecaster.history. There were instances where plotting was failing as a result.

[0.18.12] - 2023-08-01

Added

Added callbacks to the prophet model (#65).
Added fourier_order as a parameter in the Forecaster.add_seasonal_regressors() and Forecaster.add_cycle() methods.

Changed

Fixed

[0.18.11] - 2023-07-31

Added

Changed

Fixed

Fixed an issue where Prophet was not adding additional regressors correctly (#63).

[0.18.10] - 2023-07-30

Added

Added the restore_series_length() function (#62).

Changed

Changed how in-sample metrics are evaluated. If the series length currently in the object and the predictions are different lengths, the prediction-length is truncated so that an in-sample metric can still be evaluated.

Fixed

Fixed documentation typos.

[0.18.9] - 2023-07-27

Added

Added the RobustScaler transformer and added it to the default optimal transformation search.
Added 'robust' as a valid normalizer argument when forecasting with scikit-learn esitmators.

Changed

Fixed

[0.18.8] - 2023-07-03

Added

Changed

Made an explicit error message when the RNN model does not have enough observations to evaluate (#58).
Changed the title of the loss plot from the RNN model.

Fixed

Fixed how the prophet model creates its externals regressors dataframe to avoid a ValueError (#55).
Fixed an error with the RNN model when forecast horizon is 1.

[0.18.7] - 2023-06-24

Added

Changed

Changed requirements to avoid a dask/lightgbm error when importing the Forecaster module (#46).

Fixed

Calling auto_forecast() without tuning a model first no longer raises an error when banking the model’s history (#52).
RNN and LSTM models no longer raise errors with default CV parameters.

[0.18.6] - 2023-06-01

Added

Added exclude argument to Forecaster.plot(), Forecaster.plot_test_set(), and Forecaster.plot_fitted().

Changed

Fixed

[0.18.5] - 2023-05-17

Added

Changed

The combo model can now accept one-model arguments to facilitate auto-selecting “best” models.

Fixed

Updated requirement version of lightgbm (#46)

[0.18.4] - 2023-04-28

Added

Changed

Took out infer_datetime_format args from pandas functions since they are deprecated.
Modified example grids.

Fixed

[0.18.3] - 2023-04-22

Added

Changed

Fixed

Fixed the test-set length check when determining min length required for confidence intervals, which was not processing floating points precisely enough.

[0.18.2] - 2023-04-19

Added

Added MVForecaster.add_signals() for multivariate stacking.
Added Forecaster.add_series() to make multivariate RNNs possible.
Added return_train_only in the util.find_optimal_transformation() function to add additional leaking-prevention measures to this function.
Added padding options in Forecaster.ingest_Xvars_df().
Added exclude_models argument to Reverter.fit_transform().

Changed

SeriesTransformer no longer deletes attributes after reverting, making the object more reusable.
Reverter objects always copy the passed Forecaster objects to make them more flexible and able to be used in more multivariate contexts.
Added __copy__() and __deepcopy__() methods to several objects.
Removed Forecaster.add_diffed_terms().
Changed how Forecaster.add_lagged_terms() deals with the introduction of N/A values – just warn instead of chopping observations.
Changed how train_only arguments are read in SeriesTransformer when there is no test_length in the Forecaster object. If test_length is 0, then train_only behaves as if it were False.

Fixed

Fixed how exogenous variables are read in util.find_optimal_lag_order().
train_only was being ignored in SeriesTransformer.DeseasonTransform().

[0.18.1] - 2023-04-14

Added

Added functions to create expanding dynamic intervals using the conformal framework with backtesting: util.backtest_for_resid_matrix(), util.get_backtest_resid_matrix(), and util.overwrite_forecast_intervals().
Added a check for the correct estimator type with Forecaster.auto_Xvar_select().

Changed

Fixed

Fixed Forecaster.export_Xvars_df(), which wasn’t working correctly when exogenous regressors were added with Forecaster.ingest_Xvars_df().

[0.18.0] - 2023-04-13

Added

Added tbats model.
MVForecaster can now have Xvars added to it.
Added multiseries.line_up_dates() function.
New argument test_again added to Forecaster.manual_forecast() and MVForecaster.manual_forecast(). auto_forecast() in the same objects has the same new argument.
model argument added to Forecaster.save_feature_importance() and doesn’t have to be called right after a model is evaluated.
Added verbose args to cross_validate() and util.find_optimal_transformation() functions to offer more transparency around these processes.
Added min_grid_size argument to Forecaster.limit_grid_size() and tune_test_forecast().

Changed

Distinction between level/non-level is no longer tracked within Forecaster and MVForecaster objects and all methods that facilitated with that have gone away (diff(), integrate(), etc). Use SeriesTransformer to gain the same plus more functionality.
Cross validation and tuning faster due to using more efficient objects.
Cross validation has more customization available.
RNNs and naive models can be optimized using grid search.
All forecasting functions rewritten to be more efficient by using more efficient objects and fewer loops.
Added Forecaster.chop_from_front() and Forecaster.chop_from_back() methods.
Less information stored in history.
Can only refer to series with default or user-selected names within MVForecaster.
The structure of MVForecaster has changed. mvf.y is a dictionary with series in it instead of being located in mvf.series1['y']. The old way was to keep better track of level/nonlevel series but that no longer exists.
test_only arg is gone in Forecaster.manual_forecast() in favor of Forecaster.test(), MVForecaster.test(). All out-of-sample methods facilited with chop_from_front() to split data and ensure no data leaks.
MVForecaster.add_optimizer_func() now accepts functions.
MVForecaster.manual_forecast() now accepts Xvars as an argument.
normalizer argument considered a hyperparameter where applicable and not given its own entry in history.
Got rid of several Forecaster methods that are never demonstrated in examples.
Got rid of Forecaster.backtest() and MVForecaster.backtest(). Pipeline.backtest() and MVPipeline.backtest() are better alternatives.

Fixed

The first element in m is taken if multiple are passed to util.find_optimal_transformation(). The code was taking the second.
Fixed the diffy arg in Forecaaster.adf_test().
cvkwargs were not being passed to the cross_validate() function in notebook.tune_test_forecast().
util.find_optimal_transformation() was only using first value of m for seasonal adjustments when multiple were passed.

[0.17.20] - 2023-04-02

Added

Changed

Fixed

Fixed the MVPipeline.backtest() method to return consistent results when a transformation is not taken on the Forecaster objects.

[0.17.19] - 2023-04-01

Added

Changed

util.find_optimal_transformation() now uses pipeline backtesting, making it harder to leak data into the decision and also to make cross validation possible.

Fixed

[0.17.17] - 2023-03-31

Added

Forecaster/MVForecaster objects pickled in now re-initiate warning logging for estimated models.

Changed

Fixed

[0.17.16] - 2023-03-29

Added

Added util.metrics.abias() function.

Changed

Revamped the RNN model. Now accepts exogenous regressors and other small changes that I believe will make it faster and more accurate, as well as allow for more customization.

Fixed

Fixed the Pipeline.backtest() function, which was adding too much space between consecutive training sets.
Fixed the train_only argument in SeriesTransformer.DetrendTransform() when loess = True.

[0.17.15] - 2023-03-27

Added

Added a LOESS detrender (from statsmodels) to the SeriesTransformer object. Called using SeriesTransformer.DetrendTransform().

Changed

SeriesTransformer can seasonally adjust the same series multiple times (to capture multiple seasonalities).
The util.find_optimal_transformation() function now tries a LOESS detrender by default.

Fixed

[0.17.14] - 2023-03-17

Added

Changed

“Differenced” no longer displayed when calling Forecaster.__repr__().

Fixed

Fixed plotting confidence intervals on separate axes in the Forecaster and MVForecaster objects.

[0.17.13] - 2023-03-12

Added

Added Forecaster.STL() method.
Added cilevel argument to Pipeline.backtest() method.
Added SeriesTransformer.DeseasonTransform() and corresponding revert function.
Added util.metrics.bias() function.
Added seasonal_adj in default try_order argument from util.find_optimal_transformation().

Changed

Fixed

Fixed how metrics are calculated after reverting a detrend transformation.

[0.17.12] - 2022-03-08

Added

Added a Forecaster.add_metric() and MVForecaster.add_metric() argument that can be used for custom metrics.
Added a default called argument in MVForecaster.add_optimizer_func() to be the function’s name.

Changed

Changed how metrics are set when initializing the objects and how they can be set later.
Changed how metrics can be accepted in util.backtest_metrics().

Fixed

Fixed the error message that is displayed when setting metrics goes wrong.

[0.17.11] - 2022-03-03

Added

Changed

Removed pandas-datareader as a dependency.

Fixed

[0.17.10] - 2023-03-02

Added

Changed

Small changes in documentation.

Fixed

Fixed so that Forecaster and MVForecaster objects can be pickled again (broke after 0.17.7 update). Added better testing of that feature.

[0.17.9] - 2023-02-26

Added

Added Forecaster.add_signals() method for custom model stacking.

Changed

Took out the LevelTestSetActuals column from MVForecaster.export('model_summaries').

Fixed

Fixed the default determine_best_by arg in Forecaster.export().

[0.17.8] - 2023-02-25

Added

Changed

Took out an error when setting validation metric to R2 and less than 2 obs in validation set. Because of cross validation, this error isn’t applicable.

Fixed

Fixed Forecaster.set_metrics() method.

[0.17.7] - 2023-02-24

Added

Changed

Code refactoring to gain efficiency and make coding base more maintainable. Better practices being followed.
Users can now choose which metrics to evaluate for any forecasting procedure. The defaults remain what they always were.
Changed the documentation in a few key spots to (hopefully) be clearer. For example, Forecaster._estimators_ global is now Forecaster().estimator attribute. This also makes pickling more convenient.

Fixed

[0.17.6] - 2023-02-23

Added

Changed

Updated requirements in test directory.

Fixed

Fixed auxmodels.mlp_stack() for multivariate forecasting.

[0.17.5] - 2023-02-22

Added

Added the catboost model to the out-of-the-box models. Made catboost a required dependency.
Added catboost grids to example grids files.
Added backtest() method for Pipeline and MVPipeline.
Added the util.backtest_metrics() function.
Added FutureWarnings to Forecaster.backtest() and MVForecaster.backtest() related methods. All backtesting will be moved to pipelines in the future.

Changed

Fixed

[0.17.4] - 2023-02-14

Added

Added an explicit error for when auxmodels.mlp_stack() is specified incorrectly.
Added ax argument to more plotting functions.
Added test scripts to the GitHub repository.
Added an explicit error in Forecaster.auto_Xvar_select() for situations when monitoring a test metric but no test set is specified.

Changed

Explicitly specify dtype = float for parts of code that raise a warning when this is not the case.
Changed to relative library imports wherever possible.
Changed the Forecaster.__deepcopy__() to be more explicit.
Changed how GridGenerator gets grids and added more grids files to choose from.
Got rid of some confidence-interval information when exporting model summary info from MVForecaster.
MVForecaster.export('lvl_test_set_predictions') no longer fails when there is no test set.

Fixed

Fixed the AnomalyDetector.EstimatorDetect() function.
Fixed util.break_mv_forecaster() for situations when MVForecaster does not have a test set.
Fixed how exporting works in MVForecaster so that errors are not returned when test set is 0 and default args are maintained.

[0.17.2] - 2023-02-09

Added

Changed

Explicitly specify dtype = float for parts of code that raise a warning when this is not the case.
Logging a warning from SeriesTransformer.DetrendTranform() instead of raising it.

Fixed

[0.17.1] - 2023-02-08

Added

Added FutureWarnings to functions that touch the Forecaster.diff() method. This method will be removed soon as the SeriesTransformer object is a better alternative that does the same thing.
Added diffy argument to Forecaster.adf_test().
Added a correct_residuals() function that corrects residuals for autocorrelation before building confidence intervals.

Changed

The train_only default argument in auxmodels.auto_arima() changed to False.
Changed how default arguments are parsed in Forecaster.export().
Refactored code to avoid a SettingWithCopyWarning from pandas in Forecaster.ingest_Xvars_df().

Fixed

Removed the print_attr argument from notebook.results_vis() that is no longer accepted in Forecaster.diff().
Removed a warning from Forecaster.ingest_Xvars_df() that shouldn’t have been raised.

[0.17.0] - 2023-02-06

Added

Added the cis argument to Forecaster__init__() and MVForecaster.__init__().

Changed

Only conformal intervals now supported in Forecaster and MVForecaster. By default, these will not be generated and can only be generated if there is a test set and it is specified to a sufficient length.
Removed Forecaster.proba_forecast() and MVForecaster.proba_forecast() and all probabilistic arguments in functions.
Default test_length argument in Forecaster.__init__() and MVForecaster.__init__() is now 0.
Took out the exog_coint argument from vecm.__init__() since it’s not actually usable.
Only model-specific and hyperparameter optimization warnings will be logged in warnings.log and other warnings will be printed.
Changed many categories of warnings from UserWarning to Warning.

Fixed

Added staticmethod decorators to util.metrics methods. This doesn’t change functionality, just the documentation.

[0.16.6] - 2023-02-03

Added

Changed

Fixed

Fixed the warning in Forecaster.ingest_Xvars_df().

[0.16.5] - 2023-01-31

Added

Changed

Fixed

Fixed an error that can arise from the util.break_mv_forecaster() function.
Fixed the AnomalyDetector.EstimatorDetect() function.

[0.16.4] - 2023-01-31

Added

Added mapie as a requirement.

Changed

Forecaster.proba_forecast() now uses conformal prediction from the mapie package to create confidence intervals and is much more efficient. This will be also implemented in MVForecaster soon.
Took out ‘CIPlusMinus’ as a history key in Forecaster and MVForecaster.

Fixed

[0.16.3] - 2023-01-29

Added

Added a naive/seasonal naive estimator to Forecaster.

Changed

Forecaster.add_diffed_terms no longer supports second differencing.

Fixed

Fixed the warning that gets passed for the RNN and Silverkite models when dynamic_testing is passed as False.

[0.16.0] - 2023-01-28

Added

Changed

It is now possible to skip model testing by setting test_length = 0 in Forecaster and MVForecaster.
Got rid of LastTestSetPrediction and LastTestSetActuals columns from model summary df.
Got rid of best_fcst DataFrame.
Gave documentation a once-over.
Refactored code to be slightly more efficient.

Fixed

The scaler in MVForecaster was not just being applied to the test-set inputs and was being called twice per model train.

[0.15.16] - 2023-01-25

Added

added ax argument to all forecast plotting functions.

Changed

Fixed

removed tensorflow from the list of requirements and added it and tensorflow-macos to the optional add-on list (#31)

[0.15.14] - 2022-01-23

Added

added a str representation for the SeriesTransformer object
added exclude_models argument to SeriesTransformer.Revert() and similar functions

Changed

Fixed

cleaned up some documentation links

[0.15.12] - 2022-12-12

Added

Changed

fill N/A for all forecasts with a forward fill to prevent some max-value errors, both when forecasting and reverting transformations

Fixed

[0.15.11] - 2022-12-06

Added

Changed

Fixed

fixed exception catching in Forecaster.auto_Xvar_select() and util.find_optimal_transformation()

[0.15.10] - 2022-12-05

Added

added dayofyear to search for in Forecater.auto_Xvar_select() when frequency is daily or lower.
added cycle_lens argument to Forecaster.add_seasonal_regressors().

Changed

Fixed

fixed how seasonalities are selected when they are not Fourier transformed in Forecaster.auto_Xvar_select().

[0.15.9] - 2022-11-03

Added

Changed

Fixed

fixed getting cis with backtesting. only the first backtest iteration was being used.

[0.15.8] - 2022-11-02

Added

added method argument to Forecaster.reeval_cis() and MVForecaster.reeval_cis() and included an option to get confidence intervals through backtesting.
added Forcaster.plot_backtest_values().

Changed

changed some error messages to be more descriptive and to encourage raising issues on github.
changed how level confidence intervals are obtained when calling SeriesTransformer.DiffRevert() to be more efficient.
the backtest_values DataFrame now includes dates in Forecaster and MVForecaster. the order of the dataframe is Date –> Actuals –> Preds for all iterations.

Fixed

[0.15.7] - 2022-10-28

Added

Changed

changed dynamic_testing = <int> to same way it was in 0.12.8 when it was introduced because it gives better and more efficient results in both Forecaster and MVForecaster.

Fixed

[0.15.6] - 2022-10-25

Added

Changed

the m argument in util.find_optimal_transformation can be a list and multiple seasonal differences can be tried in that function
changed how level plotting is peformed in Forecaster.plot() and Forecaster.plot_test_set()

Fixed

fixed how history[‘LevelY’] attribute is undifferenced in SeriesTransformer, which was causing plots to be incorrect.

[0.15.5] - 2022-10-22

Added

added the Pipeline.MVPipeline object.

Changed

Fixed

[0.15.4] - 2022-10-21

Added

Changed

Fixed

changed where import calls occur in the SeriesTransformer module to avoid a circular import that can happen
added AttributeError to the list of exceptions to catch in util.find_optimal_transformations() function.
fixed initiating MVForecaster() when merge_Xvars=='i'.

[0.15.3] - 2022-10-20

Added

increased documentation around forecasting different model types.

Changed

‘LevelY’ passed to history in util.break_mv_forecaster()
changed the optional dependency pip intall fbprophet to pip install prophet (#18)

Fixed

added IndexError to the list of exceptions to catch in util.find_optimal_transformations() function.
convert values in **kwargs from numpy.bool_ to bool type when forecasting with HWES (#19).

[0.15.2] - 2022-10-19

Added

added the util.find_optimal_transformation() function.

Changed

changed setup.py so that version needs to be added manually to make an update (#16).
Forecaster.integrate() now has **kwargs that are passed to Forecaster.adf_test().
changed some data processing in SeriesTransform.DiffRevert() to accomodate a future change from pandas.
changed the name of util.find_series_transformation() to util.find_statistical_transformation(). this function is new as of 0.15.0 so hopefully it has not been adopted widely yet and this change doesn’t cause issues for users.
added a try/except wrapper around the self.f.typ_set() code in SeriesTransformer.DetrendTransform()

Fixed

drops null values before running though Forecaster.normality_test().
changed the import in util to from scalecast import Pipeline as the util.find_series_transformation() was not working because of it.

[0.15.1] - 2022-10-18

Added

added the SeriesTransformer.DetrendTransform() and SeriesTransformer.DetrendRevert() functions.

Changed

Fixed

all reverter functions in SeriesTransformer revert level confidence intervals.

[0.15.0] - 2022-10-17

Added

added the Pipeline module.
added util.find_optimal_series_transformation()
added Forecaster.normality_test() function

Changed

changed the default full_res arg in Forecaster.adf_test() to True
changed links to some model documentation in readme

Fixed

[0.14.8] - 2022-10-14

Added

added must_keep arg to Forecater.auto_Xvar_select().
added SeriesTransformer.SqrtTransform() and SeriesTransformer.SqrtRevert() functions.

Changed

two-level differencing no longer natively supported in Forecaster and MVForecaster. it’s too much work to maintain two differencing and it is also supported more efficiently and dynamically through the SeriesTransformer object and that is available for who needs it.
- this changed the arguments in the following functions:
changed accepted values that can be passed to the probabilistic arg in the Forecaster.tune_test_forecast(), MVForecaster.tune_test_forecast(), and notebook.tune_test_forecast() functions so that some models can be forecasted probabilistically and others don’t have to be (speeds up processing time generally).

Fixed

issues with second-differencing have all been resolved since second differencing is no longer supported natively in the Forecaster object.
took out revert_fvs from SeriesTransformer.DiffRevert() function because it hasn’t been working.

[0.14.7] - 2022-10-05

Added

Changed

Fixed

added trend_estimator_kwargs to the Forecaster.auto_Xvar_select() function to avoid an error that occurs when mixing estimators and not using default hyperparameters

[0.14.6] - 2022-10-03

Added

Changed

changed the default figsize for all plotting functions to (12,6).

Fixed

[0.14.5] - 2022-09-30

Added

Changed

added figsize argument in notebook.results_vis() and notebook.results_vis_mv()

Fixed

[0.14.4] - 2022-09-23

Added

added the tf_model attribute to the Forecaster object for users to access and save rnn and lstm models (#12)
added figsize arg to all forecast plotting methods in Forecaster and MVForecaster
added a link to the M4 example in the readme

Changed

Fixed

took out the freq argument from auxmodels.vecm (which was being set automatically) since it is optional when dates are passed and was causing frequencies that statsmodels does not recognize to fail (#13)

[0.14.3] - 2022-09-16

Added

util.pdr_load() now accepts multiple series and returns an MVForecaster object of everything loaded together
added more arguments to the util.pdr_load() function
added auxmodels.vecm model, which is a model class that can be imported using the MVForecaster.add_sklearn_estimator() function (#11)
modified the source code in the MVForecaster object to accomodate new model classes (e.g. vecm)
added a vecm grid to the example grids
added the util.find_optimal_coint_rank() and util.find_optimal_lag_order() functions

Changed

changed scaling syntax in Forecaster and MVForecaster to circumvent a warning having to do with feature names–only numpy arrays are scaled now (not dataframes)

Fixed

added a call of Forecaster.Forecaster.typ_set() right after MVForecater.__init__(), before chopping dates to fix weird loading errors that occured once in a while
fixed the util function that wasn’t working in 0.14.2 and yanked that release. everything scheduled for 0.14.3 will now be part of 0.14.4.

[0.14.1] - 2022-09-09

Added

if there are not enough observations to use in cross validation (usually because too many AR terms were added), an error is raised when calling the Forecater.cross_validate() and MVForecaster.cross_validation() functions

Changed

no Forecaster.auto_Xvar_select() no longer raises errors if more AR terms passed to max_ar argument than the model is able to estimate

Fixed

Forecaster.determine_best_series_length() will no longer fail if the min_obs arg value is greater than the amount of observations in the series
found more instances where TypeErrors should not be raised (such as passing an int64 type when int is required)
fixed an issue that occurs after selecting Xvars with Forecaster.auto_Xvar_select() on an integrated series then loading to MVForecaster
fixed the error raised when 0 or less is passed to the Forecaster.set_validation_length() and MVForecaster.set_validation_length() functions

[0.14.0] - 2022-08-31

Added

added the Forecaster.auto_Xvar_select() method
added a check for NAs in Forecaster and MVForecaster when evaluating grids and validation metric is mape. a descriptive error is raised if NAs are found (#10)
added Forecaster.drop_all_Xvars()
added Forecaster.determine_best_series_length()
added Forecaster.restore_series_length()

Changed

removed deprecated functions identified and labeled in 0.13.1
Forecaster.keep_smaller_history() can now accept numpy int types as an argument.

Fixed

[0.13.11] - 2022-08-19

Added

added grids_file attribute to Forecaster and MVForecaster objects, as well as set_grids_file() method to both objects.

Changed

Fixed

[0.13.10] - 2022-08-15

Added

added probabilistic argument option to auxmodels.mlp_stack() function.
added Xvars argument to auxmodels.auto_arima() function.

Changed

made it so that an error is raised earlier when using Forecaster.ingest_Xvars_df() incorrectly.

Fixed

fixed an issue with auxmodels.mlp_stack() where **kwargs were not being passed correctly.

[0.13.9] - 2022-08-11

Added

added error arg to Forecaster.tune_test_forecast() and MVForecater.tune_test_forecast() methods.

Changed

took out an error check that was redundant and not even working

Fixed

fixed some documentation syntax for new objects added last dist

[0.13.8] - 2022-08-08

Added

added auxmodels module with auto_arima() and mlp_stack() functions
added pmdarima as an optional requirement to make auto_arima work
added util.metrics class
added Forecaster.reeval_cis() and MVForecaster.reeval_cis() methods

Changed

Fixed

[0.13.6] - 2022-08-04

Added

added error argument to the Forecaster.diff() method

Changed

took out the error that’s raised when trying to add AR terms after data has already been differenced using Forecaster.add_AR_terms()

Fixed

fixed an issue with util.break_mv_forecaster that was caused from adding future_dates arg to Forecaster.__init__() method

[0.13.5] - 2022-08-03

Added

added optional future_dates arg to Forecaster.__init__() method
added error argument to the Forecaster.drop_Xvars() and Forecaster.drop_regressors() methods

Changed

changed SeriesTransformer scaling functions to use only training data if train_only=True.
took out the error that’s raised when trying to add AR terms after data has already been differenced

Fixed

fixed an issue with the notebook.tune_test_forecast() function
MVForecaster no longer takes AR terms when merge_Xvars = 'u'
fixed an issue where util.break_mv_forecaster was not converting xreg dict correctly

[0.13.4] - 2022-07-29

Added

added util.pdr_load() function.
added limit_grid_size as an argument to Forecaster.tune_test_forecast(), MVForecaster.tune_test_forecast(), and notebook.tune_test_forecast() to support randomized grid search through this process.

Changed

changed dynamic window forecasting loop
if trying to cross validate with less data than it is possible to create the correct-sized folds for, the program will no longer raise an error but instead pass default parameters to the best_params attribute and log a warning.
made cross validation slightly more efficient

Fixed

changed some source code to reduce the amount of TypeErrors a user is likely to get (such as passing int.64 type when int type is required)

[0.13.3] - 2022-07-27

Added

Changed

SeriesTransformer.diffrevert() now supports an argument revert_fvs, which is True by default. since adding level cis, this is now possible.

Fixed

fixed an issue that caused model evaluation to fail if models were not tuned from the grid successfully. this was an issue since 0.12.3 due to how cross validation changed tuning.

[0.13.2] - 2022-07-24

Added

Changed

Fixed

Fixed an issue from 0.13.1 that was caused by running models test_only = True on integrated series.

[0.13.1] - 2022-07-24

Added

added level fitted values and default level confidence intervals for all models called through Forecaster and MVForecaster.

Changed

deprecated several export functions and rewrote Forecaster.export() and MVForecaster.export() to allow confidence intervals when cis=True. all deprecated functions should log a FutureWarning and will be removed in 0.14.0. all of these functionalities are now dupliated in Forecaster.export() and MVForecaster.export()
- Forecaster.export_test_set_preds_with_cis()
- Forecaster.export_test_set_preds_with_cis()
- MVForecaster.export_model_summaries()
- MVForecaster.export_forecasts()
- MVForecaster.export_test_set_preds()
- MVForecaster.export_level_forecasts()
- MVForecaster.export_level_test_set_preds()
made shap an optional add-on due to some installation issues by some users

Fixed

notebook.tune_test_forecast() was missing an argument in the function
fixed an issue with MVForecaster.backtest() causing some models to return a key error when backtested
fixed an issue where 'ValidationMetricValue' could not be passed to MVForecaster.set_best_model(determine_best_by)

[0.13.0] - 2022-07-19

Added

added probabilistic forecasting through Forecaster.proba_forecast() and MVForecaster.proba_forecast() methods
added level confidence intervals always for models run at level (so it won’t fail to generate cis anymore when passing Forecaster.plot(ci=True,level=True))
probabilistic forecasting also makes it possible to derive level confidence intervals even when model was run at difference
added probabilistic as a bool argument to Forecaster.tune_test_forecast(), MVForecaster.tune_test_forecast(), and notebook.tune_test_forecast() functions

Changed

changed how it was determined that a model was tuned for efficiency gains
changed so that “Dynamically” does not appear in the history[‘tuned’] attribute
changed the error that is raised when reduce_Xvars() doesn’t work due to feature importance not being supported by a given model so that it is more explicit
changed the default mlp grid so that random_state is no longer a value. this makes that model more ammenable to probabilistic forecasting

Fixed

[0.12.9] - 2022-07-15

Added

added suffix argument to Forecaster.tune_test_forecast(), MVForecaster.tune_test_forecast(), and notebook.tune_test_forecast() functions (#5)
added fi_method argument to notebook.tune_test_forecast() function

Changed

Fixed

[0.12.8] - 2022-07-11

Added

added AnomalyDetector.MonteCarloDetect_sliding() method

Changed

changed the dynamic_testing and dynamic_tuning arguments in Forecaster and MVForecaster so that window forecast evaluation is now supported. now, instead of having the choice between 1-step and arbitrary multi-step forecasting, any integer value is accepted as arguments in those parameters but True and False are still supported and do the same thing as always.

Fixed

[0.12.7] - 2022-07-08

Added

Changed

got rid of printing when calling ChangepointDetector.WriteCPtoXvars()

Fixed

fixed cross validation when Xvars = 'all' is passed as an argument

[0.12.6] - 2022-07-06

Added

added the ChangepointDetector object
added the AnomalyDetector.adjust_anom() method

Changed

Fixed

fixed last index span from AnomalyDetector.MonteCarloDetect()

[0.12.5] - 2022-07-01

Added

added jump_back parameter in Forecaster.backtest() and MVForecaster.backtest() methods
added the theta model from darts

Changed

changed how dataframes are grouped in the cross_validate() method. turned sorting off to prevent some failures, specifically in the theta model

Fixed

[0.12.4] - 2022-06-28

Added

Changed

changed how shap feature scores are sorted in reduce Xvars, no adjustment needed like with PFI

Fixed

[0.12.3] - 2022-06-27

Added

added shap feature importances in addition to pfi by allowing user to select method = ‘shap’ when calling Forecaster.save_feature_importance()
added shap library to dependencies list
added SeriesTransformer class
added AnomalyDetector class
added function to util that breaks an MVForecaster class into several objects of Forecaster class
can now init Forecaster object with require_future_dates = False. when using False, the object will not forecast into future dates and will not make you know values into the future for regressors passed through Forecast.ingest_Xvars_df().

Changed

took the ‘per’ key out of the history attribute
changed the order of some of the source code to be more efficient (very small gains)
changed the size of the dataset that pfi feature importance is called on to make it include all values previously seen by any given model passed to it. before, it sliced off the last couple observations only – this was more or less a mistake but I don’t expect results will be affected significantly for anyone using reduce_Xvars() due to how features are sorted in that function.
changed the Forecaster.reset() function so that it returns a true copy of the initiated object.
in Forecaster.save_feature_importance(), added the on_error arg to raise errors if the user prefers. The default is still to log errors as warnings so as not to break loops.

Fixed

test_only was not working with the lstm estimator, so fixed that
fixed an issue where the function didn’t ignore the argument passed to estimator with reduce_Xvars(method='l1')

[0.11.2] - 2022-06-20

Added

Changed

Fixed

fixed an issue where None wasn’t being accepted in grid with 'Xvars' as the key and using cross_validate() (#2)

[0.11.1] - 2022-06-15

Added

Changed

changed how the validation set length is calculated in history attributed – given na value if cross validation used to tune models

Fixed

fixed an issue caused by None values in hyperparam grids being changed to np.nan and therefore not accepted in some functions after cross validation has been called

[0.11.0] - 2022-06-14

Added

added cross_validate() methods to Forecaster and MVForecaster objects, which can now be used for the same purposes as tune() but with cross validation
added cross_validate as a (bool) argument to the Forecaster.tune_test_forecast(), Forecaster.reduce_Xvars(), MVForecaster.tune_test_forecast(), and notebook.tune_test_forecast() functions
added “CrossValidated” key to history dict in Forecaster and MVForecaster objects

Changed

if np.nan is passed as a normalizer value, it will convert to None so that it can be used

Fixed

[0.10.5] - 2022-06-07

Added

Changed

Fixed

fixed the set_best_model() method in MVForecaster, which was broken due to not being able to parse new updates from 0.10.1

[0.10.4] - 2022-06-05

Added

Changed

Fixed

sorting from pfi variable reduction method was incorrect after initial variable drop, causing the wrong sequence of variables to be dropped when method = ‘pfi’

[0.10.3] - 2022-06-05

Added

added lasso and ridge as default estimators and gave them default grids for both Forecaster and MVForecaster
added reduce_Xvars() method to Forecaster object
added util module with one function (for now) to plot error changes from calling the Forecaster.reduce_Xvars() method

Changed

Changed default hyperparameter grid values for knn, xgboost, and lightgbm estimators

Fixed

[0.10.2] - 2022-05-17

Added

Changed

Changed where logs are called back to when importing library due to warnings that don’t get logged when unpickling object

Fixed

[0.10.1] - 2022-05-16

Added

MVForecaster.set_optimize_on() now accepts user functions (like weighted averages) by leveraging new MVForecaster.add_optimizer_func() function

Changed

Suppressed future warnings on import because of a pandas warning caused from importing LightGBM. If LightGBM does not fix this issue with an update, we might need to pull it from the default list of sklearn estimators; users would still be able to use LightGBM by importing it manually.

Fixed

[0.10.0] - 2022-05-13

Added

added permutated feature dataset to history for sklearn models with feature importance

Changed

changed some pandas syntax to avoid some warnings
permutation feature importance applied on test set only and works better on models run test_only = True

Fixed

fixed the order in which Xvars are parsed, which was causing issues when Xvar names were added in a function in a different order than they appeared in the object

[0.9.9] - 2022-05-12

Added

Changed

warning logs now called after Forecaster and MVForecaster objects are initiated
No more get_funcs() method, a bad idea from the start

Fixed

[0.9.8] - 2022-05-11

Added

added the corr() and corr_lags() methods to the MVForecaster object

Changed

Fixed

[0.9.7] - 2022-05-10

Added

Changed

Fixed

fixed an issue where MVForecaster was not generating the correct forecast horizon worth of values for models run with lags only (no seasonality or trends)

[0.9.6] - 2022-05-06

Added

Changed

got rid of printing when calling silverkite

Fixed

fixed an issue with ARIMA where it didn’t work if no Xvars had been added first

[0.9.4] - 2022-04-29

Added

Changed

Fixed

fixed an issue where data types get changed from int to float when grid searching

[0.9.2] - 2022-04-29

Added

Added SGDRegressor from sklearn as a default estimator (sgd)

Changed

Changed “backcast” nomenclature to “backtest”, which is what it really is (oops!)

Fixed

[0.9.1] - 2022-04-21

Added

Changed

Fixed

fixed an issue in ARIMA model where future x regressors weren’t being read correctly
fixed an issue where save_feature_importance() was throwing an error sometimes

[0.9.0] - 2022-04-15

Added

added backcast() method to Forecaster and MVForecaster objects
added copy() and deepcopy() methods to the same objects
added test_only options to all models in Forecaster. ignored in combo models. a future dist will have this for MVForecaster
added plot_loss_test as arg in lstm and rnn models
added notebook.results_vis_mv() function

Changed

Fixed

[0.8.4] - 2022-04-11

Added

Changed

changed how names are looked for in the MVForecaster object

Fixed

[0.8.3] - 2022-04-05

Added

Changed

Fixed

fixed an issue when plotting in MVForecaster that occurs when you didn’t initially pass names to __init__()
fixed an instance where labels was spelled wrong

[0.8.1] - 2022-04-04

Added

Changed

Fixed

Fixed inconsistent indexing in MVForecaster that was causing some models to fail when different Forecaster objects’ dates didn’t align
Fixed an issue where level=True wasn’t working in plot_test_set() sometimes

[0.8.0] - 2022-04-03

Added

Added more description to the MVForecaster.__repr__() method

Changed

Changed arguments in the rnn estimator to make it more flexible and dynamic

Fixed

[0.7.6] - 2022-03-30

Added

Changed

Fixed

Noticed an error when forecasting with Silverkite on differenced data and fixed it.

[0.7.5] - 2022-03-28

Added

Changed

Fixed

MVForecaster.plot() wasn’t working after a recent update, fixed now

[0.7.4] - 2022-03-28

Added

Changed

Fixed

MVForecaster.export() wasn’t working after last update, fixed now

[0.7.3] - 2022-03-28

Added

Changed

changed “BestModel” key in MVForecaster.export() method to “best_model” to be consistent with Forecaster notation

Fixed

was calling the wrong function when the series weren’t the same length in the MVForecaster.__init__() method
fixed MVForecaster.set_optimize_on() so that it accepts “mean” as an argument without error

[0.7.2] - 2022-03-25

Added

Changed

Fixed

MVForecaster wasn’t combining xvars with AR terms correctly
MVForecaster wasn’t plotting some series correctly
MVForecaster wasn’t making copies of arrays in dictionaries when copying xvars

[0.7.1] - 2022-03-24

Added

added the MVForecaster object for multivariate vector forecasting
added mv_grids to GridGenerator, which is written out to working directory as MVGrids (due to the added “lags” arg)
added multiseries.keep_smallest_first_date() function to make the series of multiple Forecaster objects the same lengths

Changed

Fixed

[0.7.0] - 2022-03-18

Added

added the add_cycle() method for identification of irregular cycles

Changed

Fixed

[0.6.9] - 2022-03-15

Added

Changed

Fixed

fixed the default hidden_layers_type arg in _forecast_rnn()

[0.6.8] - 2022-03-15

Added

Changed

Fixed

fixed the rnn estimator which was using a dict to set hidden layers without unique keys, causing all layers except the last to be ignored, so all rnns were returning one layer results. Now the args in that function are different, but tested and it works to add more layers.

[0.6.7] - 2022-03-07

Added

added N_actuals to __repr__() function

Changed

Took pt normalizer out of all example grids since it still errors out sometimes without being obvious why
notebook.results_vis() no longer prints anything and automatically sets the plot title instead
Forecaster.set_test_length() now accepts fractional splits

Fixed

[0.6.6] - 2022-02-28

Added

added the add_sklearn_estimator() function so a user can add any sklearn regression model to forecast with.

Changed

Fixed

some of the examples weren’t plotting correctly after recent updates.

[0.6.4] - 2022-02-25

Added

added residuals to the function export_fitted_vals(). Now gives dates, actuals, fitted vals, and residuals
added multiseries.export_model_summaries() function

Changed

Fixed

notebook.results_vis() broke after last update and has now been fixed

[0.6.2] - 2022-02-25

Added

Changed

the plot(), plot_test_set(), and plot_fitted() functions now return figures instead of automatically plotting for you. This means more customization is now possible. Took out to_png and related args from these functions since that can be done now with matplotlib

Fixed

changed the insample metric evaluations for RNN and LSTM to be on the full training set instead of just the last few observations.

[0.6.1] - 2022-02-23

Added

Changed

Fixed

fixed an issue when calling RNN or LSTM models after a combo model that saved attributes of the combo model illogically in history
fixed an issue that caused test-set evaluation of ARIMA models to be inaccurate (causing severe underperformance)

[0.6.0] - 2022-02-09

Added

Added drop argument to several adder functions, giving the user the option to drop the original regressors after making certain transformations on them

Changed

add_ar_terms() now accepts 0 as an argument but it doesn’t do anything

Fixed

fixed issue with pop_using_criterion() function that wasn’t dropping models correctly in some instances
fixed the default parameters for the rnn model which weren’t working due to mislabeling of one of the parameters in TensorFlow. updated the docstring accordingly

[0.5.9] - 2022-02-01

Added

Added CurrentEstimator to __repr__() function.

Changed

Fixed

Made it impossible to forecast without adding future dates first

[0.5.8] - 2022-01-27

Added

Changed

Made it impossible to pass tune argument to manual_forecast()

Fixed

Fixed results_vis in notebook (wasn’t displaying one of the widgets)

[0.5.7] - 2022-01-26

Added

Changed

Cleaned up a lot of documentation

Fixed

Fixed links in readme

[0.5.6] - 2022-01-25

Added

Added documentation on Read the Docs
Added cilvel information to export functions

Changed

Fixed

[0.5.5] - 2022-01-20

Added

Added CILevel info to export model_summaries function

Changed

Fixed

Fixed an issue where plots were diplaying incorrect confidence levels if cilevel had been changed since training it

[0.5.4] - 2022-01-20

Added

Added the rnn estimator

Changed

plot_loss argument no longer considered a hyperparameter value for LSTM and RNN models

Fixed

Fixed an issue where “==” wasn’t being accepted in the evaluated_as argument in the pop_using_criterion() function
Scaler in history saved as ‘minmax’ instead of None for LSTM and RNN models

[0.5.3] - 2022-01-18

Added

EvaluatedModels to __repr__()

Changed

Does not call infer_freq() as often, making code more efficient

Fixed

sometimes the attribute ci_bootstrap_samples was being called bootstrap_samples, changed everything to bootstrap_samples only

[0.5.2] - 2022-01-12

Added

Changed

Fixed

Fixed an error that occured when calling the __repr__() method if no models had been evaluated first

[0.5.1] - 2022-01-12

Added

Added export_fitted_vals() function
Added ci option to the results_vis() function in notebook
Added the get_funcs() function

Changed

No Xvars in LSTM model, changed to lags (now model will only look at its own history)
No normalizer in LSTM model (always uses a minmax scaler now)
LSTM model can no longer be tuned
Got rid of all lstm model grids
changed __str__() and __repr__() so that they now offer better info

Fixed

Fixed the LSTM model by scaling the dependent variable and unscaling it (minmax) when it comes out and getting rid of other Xvars

[0.5.0] - 2022-01-10

Added

Added confidence intervals using bootstrapping
- set_cilvel() function (default .95)
- set_bootstrap_samples() function (default 100)
Added ci parameter to plot() and plot_test_set() function
Added UpperCI, LowerCI, TestSetUpperCI, TestSetLowerCI keys to history dict
Added export_forecasts_with_cis() and export_test_set_preds_with_cis() functions
Added source code (commented out) to get level confidence intervals – when I tested, the intervals were too large to implement but maybe in the future this will be revisited

Changed

Fixed

[0.4.4] - 2022-01-07

Added

added lstm grid in example grids
added EarlyStopping callback functionality for the LSTM model
added get_expanded_lstm_grid() to GridGenerator module which gives an example of a grid with early stopping

Changed

changed default paramaters for the lstm model
added **kwargs to the lstm model forecast function that are passed to the fit() function in TensorFlow, got rid of epochs and batch_size args consequently

Fixed

[0.4.25] - (quick fix) 2022-01-06

Added

Changed

Fixed

source code was using f instead of self when when calling pop_using_criterion()

[0.4.2] - 2022-01-06

Added

lstm estimator
added pop_using_criterion() function

Changed

Fixed an issue where sklearn models were being fit on the same data twice – does not change outcomes but the models run faster now
Output from _scale() function is now always a numpy matrix (could have been either that or pandas dataframe before)
sorted _estimators_ list
changed the error message for when importing a grid fails to account for one other possible reason the failure occured

Fixed

[0.4.1] - 2021-12-30

Added

Can now sort by metric value in export_all_validation_grids_to_excel()

Changed

Fixed

Fixed an issue where sometimes the incorrect AR terms were being propogated for test-set evaluation only

[0.4.0] - 2021-12-30

Added

Changed

deleted the ‘scale’ normalizer from the mlp grid

Fixed

Fixed an issue with the PowerTransformer normalizer that failed because of a DivideByZero error, now defaults to a StandardScaler when this issue is encountered and logs a warning

[0.3.9] - 2021-12-30

Added

Added init_dates and levely attributes
Added 'Observations' info to history and export()
Added 'lvl_test_set_predictions' to export dataframes

Changed

Got rid of first_obs and first_dates attributes and wrote more efficient code to do what they were there for
More information available when __str__() is called

Fixed

Fixed what became an issue with the last update in which when calling add_diffed_terms() or add_lagged_terms(), the level series wasn’t accurate due to how undifferencing was being executed. After examining this issue, it became evident that the previous way to undifference forecasts was less efficient than it should have been, this update fixed the issues from the last update and made the code more efficient
Fixed an issue where AR terms were manipulating the underlying xreg structures so that each forecast were using its own test-set propogated AR values instead of the correct AR values
Fixed the export_Xvars_df() method which wasn’t working correctly if at least one forecast hadn’t been called first

[0.3.8] - 2021-12-29

Added

added the following functions that can each add additional Xvars to forecast with:
- add_exp_terms() - for non polynomial exponential transformations
- add_logged_terms() - for log of any base transformations
- add_pt_terms() - for individual variable power transformations (box cox and yeo johnson available)
- add_diffed_terms() - to difference non-y terms
- add_lagged_terms() - to lag non-y terms
added the ‘pt’ normalizer for yeo-johnson normalization (in addition to ‘minmax’, ‘normalize’, and ‘scale’)
added the drop_Xvars() function that is identical to the drop_regressors() function

Changed

imports all sklearn models as soon as scalecast is imported
src code cleanup with better coding practices when it comes to forecasting sklearn models (no more copying and pasting new functions)
changed several set data types to lists in src code
changed the names of some hidden functions
other src code cleanup for readability and minor efficiency gains
better in-line comments and docstring documentation
got rid of quiet paramater in save_summary_stats() and save_feature_importance() and now these simply log any problems as warnings
time trends now start at 1 instead of 0 (makes log transformations possible)
observation dropping for AR terms in sklearn models now based on the number of N/A values in each AR term instead of just the AR number
changed some example grids to include the pt normalizer

Fixed

now logs all warnings

[0.3.7] - 2021-12-27

Added

dynamic_testing argument to manual_forecast() and auto_forecast() functions – this is True by default (makes all testing comparable between sklearn/non-sklearn models)
dynamic_tuning argument to tune() function – this is False by default to majorly improve speed in some applications

Changed

native Forecaster warnings will be logged

Fixed

[0.3.6] - 2021-12-14

Added

added tune_test_forecast() function to notebook module to create a progress bar when using a notebook

Changed

Fixed

fixed an issue with Forecaster.ingest_Xvars_df() when use_future_dates=False causing an error to be raised

[0.3.5] - 2021-12-07

Added

added include_traing parameter to notebook.results_vis() function

Changed

Fixed

fixed print_attr parameter default in notebook.results_vis()

[0.3.4] - 2021-12-07

Added

added results_vis() notebook function (requires ipywidgets)
added Forecaster.export_Xvars_df() function
added max_integration argument to the Forecaster.integrate() function

Changed

Fixed

[0.3.3] - 2021-11-26

Added

Changed

Now reloads Grids file each time ingest_grid() is called so that notebooks do not have to be rerun when a grid cannot be found

Fixed

Fixed an issue with some sklearn estimators that occurs when passing a subset of regressors in a list to the forecast function

[0.3.2] - 2021-11-01

Added

Changed

Fixed

Found an issue when using floor in Prophet

[0.3.1] - 2021-10-29

Added

Added the eCommerce example
In limit_grid_size(), users can now set random_seed parameter for consistent results

Changed

Fixed

Scikit-learn models were not accepting Xvars='all' as an arguments
Fixed an issue causing models run at different levels to error out sometimes when plotted
Fixed a plotting error that occured sometimes when setting models parameter to None

[0.3.0] - 2021-10-15

Added

Added an option to save to png in plot(), plot_test_set(), and plot_fitted() methods using plt.savefig() from matplotlib and calling with to_png = True

Changed

Made errors more descriptive, stripping out AssertionError types

Fixed

fixed typos in doc strings

[0.2.9] - 2021-09-27

Added

Changed

In plot() method, models=None is now accepted and will plot only actual values
Example grids are modified to prevent overfitting in some models

Fixed

Fixed the add_time_trend() method to not skip a time step in the first observation

[0.2.8] - 2021-08-27

Added

Added a descriptive error when all_feature_info_to_excel() or all_validation_grids_to_excel() fails

Changed

Using pd.shift() instead of np.roll() to create AR terms to avoid further issues with AR terms
Prophet, silverkite, and ARIMA have better Xvar validation mechanisms to ensure that autoregressive terms aren’t fed to them, which could cause errors and doesn’t add anything to the models that isn’t already built into them. Now, even if a user tries to feed AR terms only, it will pass no Xvars to these models

Fixed

AR terms were not dropping the correct first observations before being estimated with SKLEARN models, so we fixed that but it didn’t seem to make a noticeable difference in any of the examples

[0.2.7] - 2021-08-20

Added

added reset() function that deletes all regressors and resets the object to how it was initiated
added documentation and hints in the source code

Changed

changed readme documentation to be more concise

Fixed

in the documentation, it was stated that the ‘scale’ value passed to the ‘normalize’ parameter when calling manual_forecast() or auto_forecast() would use a StandardScaler from sklearn, but a Normalizer was actually being applied. Now, you can pass ‘scale’ to get the StandardScaler, ‘normalize’ to get the Normalizer, or ‘minmax’ to get the MinMaxScaler (unchanged from previous distributions). ‘minmax’ is still the default for all estimators that accept this argument

[0.2.6] - 2021-08-12

Added

added train_only argument to following functions to reduce data leakage in eda/preprocessing steps: integrate, adf_test, plot_acf, plot_pacf, plot_periodogram, seasonal_decompose – default argument is still False for these. Now it is suggested to set a test length before running any one of these methods and only examine the training set correlations to prevent data leakage

Changed

Fixed

[0.2.5] - 2021-08-09

Added

added integrate() method that can be used to automatically find the series appropriate level to achieve stationarity, according to augmented dickey fuller test

Changed

Fixed

[0.2.4] - 2021-08-03

Added

added tune_test_forecast() function that allows what used to take four lines to be aggregated into one, also allows more easy saving of feature information by setting feature_importance or summary_stats parameters to True
added all_feature_info_to_excel() function
added all_validation_grids_to_excel() function

Changed

Fixed

removed a duplicate column from the dataframe created when calling the export() method

[0.2.3] - 2021-07-19

Added

Changed

changed removed pandas-datareader from imports in setup.py since it is not a package dependency (change having to do with installation only and should not affect anything when applying the library)

Fixed

[0.2.2] - 2021-07-16

Added

added GridGenerator module so user can more easily create grids in working directory

Changed

changed all functions with diffy parameter (plot_acf, plot_pacf, seasonal_decompose) now accept True, False, 0, 1, or 2 as possible values

Fixed

fixed issues with two-level undifferencing where it was adding values exponentially because the level was being added to the first and second-level differences
fixed issues with two-level undifferencing where dates were being mixed up
fixed issues with one-level test-set evaluation where the incorrect initial value was set to undifference values in the test-set only, causing miscalculation of metrics, although the bias was in both directions so when rerunning avocados.ipynb, for example, the results were virtually the same with different models now outperforming others but metrics remaining more or less the same on average; forecasted values did not change

[0.1.9] - 2021-07-09

Added

added lightgbm and silverkite as estimators

Changed

changed ‘which’ parameter in set_valiation_metric() to ‘metric’ for clarity
changed ‘which’ parameter in set_estimator() to ‘estimator’ for clarity

Fixed

[0.1.8] - 2021-07-05

Added

Changed

Fixed

fixed an error in combo modeling that was causing incorrect applications of weights in weighted averaging – the weights were generating correctly but not being applied to the best-performing models in the correct order