Scalecast

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

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

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

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 `TypeError`s a user is likely to get (such as passing `int.64` type when `int` type is required)

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.

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