Created a preprocessing step that limits the size of the training window

[rachlobay]

Addressed issue 36 by creating a preprocessing step, step_training_window(), that limits the size of the training window to the n most recent observations per location. See training_window.R and test-step_training_window.R for some tests.

[dajmcdon]

It looks like this is failing because of cmu-delphi/epiprocess#96. Blocked for the moment.

[rachlobay]

@dajmcdon, when using training data in bake() (that is, bake(new_data = NULL) such as in the tests for this function in particular, we now lose the epi_df class of the resulting data frame. In that case, the baked data gets demoted to a plain old tibble.

I think the issue here is that when we put new_data = NULL in bake, bake.recipe sends us to juice() and since the composition is a tibble (by default), then new_data <- tibble::as_tibble(new_data) on line 845-ish of juice() here: https://github.com/tidymodels/recipes/blob/main/R/recipe.R. This is why I think that all was well with the class when I used bake(new_data = NULL) along w/that old as_tibble.epi_df function that was removed… What do you think?

[dajmcdon]

I suspect you're right. We're going to need a new bake.epi_recipe() method. Ugh.