Scalecast

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'

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

Added
Changed
- Changed where logs are called back to when importing library due to warnings that don't get logged when unpickling object
Fixed

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

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

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