should remove_variables be exported?

[n-kall]

I was recently looking for a way to remove unwanted variables from a draws object, and I found that there is a remove_variables function (mentioned by in #331) but it does not seem to be exported. Is this intentional?

Here, the generic is not exported, but the individual methods appear to be:
https://github.com/stan-dev/posterior/blob/1210f4a42eeb6b942572097d60a8cc2fb8540adf/R/remove_variables.R#L1C1-L14C2

Perhaps another option would be to add an exclude parameter in subset_draws (e.g. subset_draws(example_draws(), variable = "theta", exclude = TRUE) would remove all theta variables)

[paul-buerkner]

I think we can export the generic and thus the methods actually. However, then we would have to add all kinds of validity checks, which we currently don't have in place there. I like the idea of adding an exclude option to subset_draws.
This could internally call remove_variables then.

BTW: the fact that the remove_variables methods are exported just ensures that they are registered as S3 methods to the generic. They won't actually be exported this way.

[n-kall]

I think an exclude option for subset_draws could be implemented with setdiff like this. If this seems ok, I can make a PR

subset_draws.draws_array <- function(x, variable = NULL, iteration = NULL,
                                     chain = NULL, draw = NULL, regex = FALSE,
                                     unique = TRUE, exclude = FALSE, ...) {
  if (all_null(variable, iteration, chain, draw)) {
    return(x)
  }

  x <- repair_draws(x)
  variable <- check_existing_variables(variable, x, regex = regex)
  iteration <- check_iteration_ids(iteration, x, unique = unique)
  chain <- check_chain_ids(chain, x, unique = unique)
  draw <- check_draw_ids(draw, x, unique = unique)

 ########## code for exclude option
  if (exclude) {
    if (!is.null(variable)) {
      variable <- setdiff(variables(x), variable)
    }
    if (!is.null(iteration)) {
      iteration <- setdiff(iteration_ids(x), iteration)
    }
    if (!is.null(chain)) {
      chain <- setdiff(chain_ids(x), chain)
    }
    if (!is.null(draw)) {
      draw <- setdiff(draw_ids(x), draw)
    }
  }
 #############

  x <- prepare_subsetting(x, iteration, chain, draw)
  if (!is.null(draw)) {
    iteration <- draw
  }
  x <- .subset_draws(x, iteration, chain, variable, reserved = TRUE)
  if (!is.null(chain) || !is.null(iteration)) {
    x <- repair_draws(x, order = FALSE)
  }
  x
}

[paul-buerkner]

Yes, that looks sensible to me. Thank you!

[n-kall]

Closed by #336