Feature request: one-and-done functions for operational nowcasting and backtesting

[brookslogan]

Context: existing nowcasting packages often just have one function that goes from input data to nowcasts. They often require certain column names or default to expecting certain column names. This is very convenient. (E.g., see NobBS package.)

Currently, using DelphiRF to generate nowcasts involves at least two functions and several parameters, plus manual post-processing:

• maybe some dplyr munging to get into an appropriate format, ref date x lag x values, with no report date, otherwise there will be complaints later
• running data_preprocessing with
  • several name-related parameters without defaults
  • ref_lag (this often is required in other packages, but we might be able to guess using epiprocess::revision_summary)
• ensuring that receiving directory exists
• calling DelphiRF with
  • testing_start_date
  • some (cache?) file naming parameters
• exponentiating prediction outputs to get back to original scale
• potentially some more data munging to get to Hubverse format, though this format vs. the existing wide prediction format may be a matter of what the user is trying to do, and there may be no clear winner. Having a function around to convert to Hubverse format may be helpful (potentially borrowing from related functions in epipredict.)

And this is ignoring the possibility of using CV to select parameters.

Feature request:

• a one-and-done function to produce operational nowcasts for today, with a minimal number of parameters needed, and which defaults to a particular column naming scheme without the need for the user to specify those column names
• a similar function to produce pseudoprospective nowcasts across many test dates, performing any smart computation that is required

I'm planning to prototype something like this for a simpler nowcasting system, and may be able to work on a DelphiRF-backed variant after.