Scalecast

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

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

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

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

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

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