Scalecast

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)`

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

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

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

Added
Changed
- got rid of printing when calling `ChangepointDetector.WriteCPtoXvars()`
Fixed
- fixed cross validation when `Xvars = 'all'` is passed as an argument

Added
- added the `ChangepointDetector` object
- added the `AnomalyDetector.adjust_anom()` method
Changed
Fixed
- fixed last index span from `AnomalyDetector.MonteCarloDetect()`