Is your feature request related to a problem? Please describe.
After PR #697 within the XGBoost and GBLinear forecasting models there is a filtering for nans in the fit method. If this is not done, the fit will crash for example on moments where DST changes and NANs are created, or other reasons why NANs in the label end up in the pipeline.
Describe the solution you'd like
This filtering logic should be moved from inside each model to a central place within the pipeline.