I think we want 1. probably

1. lets us have something that converts from a common data format but also interesting enables converting from one model data format to another which could be useful.

If we did keep both its unclear to me what prepare would also do.

@seabbs on call:

• This adds the ability to convert between different data formats
• "i.e as_latent.delay_model would be a s3 method to map from a "delay_model" to the latent model"
• Do what is standard unless there is a reason not to
• Call validate when you put it into new functions at the start and also at the end of the conversion
• See example in scoringutils

Here is some info on potential problems with a manual validate_ function: epiforecasts/scoringutils#816

Do what is standard unless there is a reason not to

If I had a bigger forehead I might get this tattooed

If I had a bigger forehead I might get this tattooed

Male pattern baldness could make this feature more attainable within our current roadmap

Judging by my family we are in luck

@seabbs here are my preliminary designs for the new functions:

as_latent_individual <- function(...) {
  UseMethod("as_latent_individual")
}
 
as_latent_individual.default <- function(data) {
  NULL
  # Takes data as input
  # Outputs an object of class latent_individual
  # The default one would take in data
  # Other ones could take in other model data formats
  # Advanced R suggests a constructor and a helper but I think we might just
  # want this single as function
}

validate_latent_individual <- function() {
  # Checks that the object of class latent_individual has the correct properties
  # Should be put in the top of any function that uses such objects
  # Also put it at the end of as_latent_individual methods
}

Let me know what you think.

• Add is_latent_variable using inherits function (see scoringutils)
• Get rid of prepare_epidist default
• Could have no default and require data.frame or data.table
  • All data.table are data.frame (as with tibble)
• Either be from scratch, with checkmate, or with coercedt (probably with checkmate)
• This PR should also have unit tests and documentation (all from now on)