Scalecast

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.

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()`.

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.

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

Added
Changed
- Updated requirements in test directory.
Fixed
- Fixed `auxmodels.mlp_stack()` for multivariate forecasting.

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 `FutureWarning`s to `Forecaster.backtest()` and `MVForecaster.backtest()` related methods. All backtesting will be moved to pipelines in the future.
Changed
Fixed