Bootstrap: mechanism to only include data that entered the original problem

[barrettk]

We currently use nm_data() to get the original data needed for resampling (ref).

By default we should only be bootstrapping data that made it into the final problem (i.e. not ignored subjects). However, the advantage of nm_data() is that it doesnt require the previous/based_on model to have been submitted.

Some ideas:

• Default to filtering out subjects automatically, which could be done via the following ways:
  • Use nm_join() instead of nm_data(). This is the easiest approach, but it would require that the user included $TABLE records, which likely isn't a good idea.
    • Potentially allow the use of nm_data() for un-submitted based_on models (with a warning), though default to nm_join() for executed models. This has the same concern as above though.
  • Attempt to filter the output of nm_data() ourselves internally via an expected ignore column, that would be specified by the user.
  • Use nmrec to parse the $DATA record and return ignore content in a useable form.
    • E.g., parse IGNORE=@ content:

      $DATA ../../../../extdata/acop.csv IGNORE=@
      IGNORE(ID.EQ.2)

    • Unclear how difficult this would be given that im not sure how variable these blocks can be.
• Maybe allow users to pass in their own dataset. This would require more function arguments, and may also increase the possibility of human error & need for additional checks. We are leaning away from this option, but it's still a possibility.

[barrettk]

This has been addressed in https://github.com/metrumresearchgroup/bbr/pull/671

• we currently use nm_join to get around this, and warn if that cant be done (we dont check for .join_col though). This could be improved at a later date, but likely via passing a dataset (which can be a separate issue)