Should slide() for epi_archive be given access to less than the most up-to-date snapshots?

[ryantibs]

Currently the slide() method for the epi_archive class only uses the most up-to-date values of each signal as of the reference time for the sliding computation. To do this, it calls the as_of() method for the epi_archive class.

Some computations, e.g., backcasting or nowcasting, will want to access more than just the most up-to-date signal values. Should we extend slide() and as_of() to accommodate this? We could do it with an extra argument all_versions = TRUE. The default should be FALSE so the methods continue to behave as-is.

Or should these be separate functions entirely and keep slide() and as_of() so that they just consider/deliver the most recent snapshots?

@lcbrooks @elray1 @dajmcdon @jacobbien Do you guys have opinions?

[brookslogan]

I'd lean toward separate functions, but not too strongly.

Pro for using parameter: doesn’t crowd package namespace

Pro for using separate functions: keeps the types involved a little more consistent:

• as_of : fn (epi_archive, version_type) -> epi_df
• archive_as_of : fn (epi_archive, version_type) -> epi_archive
• slide: fn (epi_archive, formula/tidymutation of epi_df) -> epi_df
• slide_by_archive: fn (epi_archive, formula/tidymutation of epi_archive) -> epi_df
• slide_by_archive_different_semantics: fn (epi_archive, fn (epi_archive) -> epi_df) -> epi_archive
  Rather than expecting/producing different types based on the value of a string argument.

The interface is the most important, but code-wise we might also think that the parameter approach might provide more immediate input validation or avoid repeated validation of the same inputs, while the separate-functions approach would have simpler control flow (no branching on the string argument; instead, we’d probably have as_of call archive_as_of internally, and slide call slide_by_archive internally).

[ryantibs]

Thanks! Reading what you wrote, and thinking about it myself, I'm actually favoring implementing it as a parameter all_versions. I think that would be very simple to do.

@dajmcdon @jacobbien @elray1 Opinions?

[brookslogan]

The all_versions approach already seems to have been implemented. Closing.

[ryantibs]

Just curious, which branch/PR?

[brookslogan]

Sorry, good catch. Got mixed up with all_rows.

[nmdefries]

Closed by #259