[BUG] HolidayFeatures crashes if dataframe doesn't contain specified date about sktime HOT 6 OPEN

I think this is still an issue, reopening. @fnhirwa, you are still welcome to pick this up if you´d like.

Yeah you can assign me I will start to work on it

from sktime.

made minor edits to include the fully reproducible example, hope that is ok.

Confirmed on current main, windows, python 3.11

from sktime.

@guimalo As per the current logic, "holidays" present in the "holiday_windows" argument is expected to be present in the index of test dataframe as well, otherwise ValueError is raised.

from sktime.

@VyomkeshVyas, thanks for clarifying!

I would still consider this a "bug" in the sense that it is not obvious that the estimator will break (e.g., from docstring), and when it breaks it does not give clear feedback.

It is also not expected, as the holiday_windows could in-principle intersect with the observed period, even if the event itself is not there.

I do think, hence, that the optimal state is:

• allow events outside the observed period - otherwise you will have to update the calendar continuously, and the estimator will fail for naive windowing or backtesting (!), because some events end up outside windows or folds
  • a naive solution would be just to subset the calendar
• ensure that holiday_windows intersecting with observed data are properly treated
  • that might be more difficult, especially for non-contiguous dates. A simple solution might be to "grow" the holiday dates by their windows, then check if intersections come up empty

The above is my "optimal" resolution, the "minimal" would be a clear error message that explains to the user what they have to do. I would still consider that quite sub-optimal because the calendar has to vary with the data.

from sktime.

I am going to start looking into this item if no one is currently working on it

from sktime.

I think this is still an issue, reopening. @fnhirwa, you are still welcome to pick this up if you´d like.

from sktime.