An option for when to apply the postprocessor

[topepo]

This is related to discussions regarding back-transforming an outcome column.

Recipe steps have the skip option, which, when TRUE, does not apply the step to new data (i.e., only to the training set).

While not the same thing, we might think about having a mandatory flag called something like devel_only that specifies that we only apply it during the model development cycle (i.e., in the tuning/resampling/last_fit/best_fit cycles).

We could default to devel_only = FALSE for almost everything, but it would allow users to only execute it during functions like predict() or augment() when used outside of the aforementioned functions.

(I'm not dead set about the proposed argument name)

[topepo]

I like new_data_only more than devel_only.

I think that the steps would be

• Add this argument to every adjustment
• Update the constructor to require it
• Update predict() to conditionally apply the fitted adjustments.

Right now I think that we would only have this affect predict() (and not fit()).