Scalecast

Added
Changed
- the `plot()`, `plot_test_set()`, and `plot_fitted()` functions now return figures instead of automatically plotting for you. This means more customization is now possible. Took out `to_png` and related args from these functions since that can be done now with matplotlib
Fixed
- changed the insample metric evaluations for RNN and LSTM to be on the full training set instead of just the last few observations.

Added
Changed
Fixed
- fixed an issue when calling RNN or LSTM models after a combo model that saved attributes of the combo model illogically in history
- fixed an issue that caused test-set evaluation of ARIMA models to be inaccurate (causing severe underperformance)

Added
- Added drop argument to several adder functions, giving the user the option to drop the original regressors after making certain transformations on them
Changed
- `add_ar_terms()` now accepts 0 as an argument but it doesn't do anything
Fixed
- fixed issue with `pop_using_criterion()` function that wasn't dropping models correctly in some instances
- fixed the default parameters for the rnn model which weren't working due to mislabeling of one of the parameters in TensorFlow. updated the docstring accordingly

Added
- Added CurrentEstimator to `__repr__()` function.
Changed
Fixed
- Made it impossible to forecast without adding future dates first

Added
Changed
- Made it impossible to pass tune argument to `manual_forecast()`
Fixed
- Fixed results_vis in notebook (wasn't displaying one of the widgets)

Added
Changed
- Cleaned up a lot of documentation
Fixed
- Fixed links in readme