`train` vs `fit`

[Simon P. Couch]

The method to fit a tailor is called fit() and a new helper function calls that operation fitting:

tailor/R/utils.R

Lines 76 to 81 in e98131d

 #' @export 
 #' @keywords internal 
 #' @rdname tailor-internals 
 tailor_requires_fit <- function(x) { 
   any(purrr::map_lgl(x$operations, tailor_operation_requires_fit)) 
 } 

At the same time, some other functionality (printing, the slot $trained, the following helper) uses train:

tailor/R/utils.R

Lines 64 to 70 in e98131d

 tailor_fully_trained <- function(x) { 
   if (length(x$operations) == 0L) { 
     return(FALSE) 
   } 
  
   all(purrr::map_lgl(x$operations, tailor_operation_trained)) 
 } 

train follows the precedent of recipes::fully_trained(), print methods elsewhere, etc. I'm not sure if this might be a point of confusion for users.

[Max Kuhn]

I'm wondering if we should differentiate between "trained", "untrained", and "n/a". The last case is things like probability thresholding that don't estimate anything.

[Simon P. Couch]

The requires_fit slot can help us tell which operations don't estimate anything!