R/var_is_numeric_ish_stuff.R

In ejanalysis/analyze.stuff: Miscellaneous Tools for Analyzing Data in Rows and Columns

#' which columns of data.frame can be interpreted as/coerced to numeric?
#' @param x data.frame (or vector)
#' @param only.if.already.numeric optional, whether to leave as character even if could coerce to number
#'
#' @return logical vector of length NROW(x)
#' @export
#'
var_is_numeric_ish_stuff <- function(x, only.if.already.numeric=FALSE) {

  # treat a vector differently than a matrix/data.frame/data.table
  # even if those nonvectors are just 1 row (multiple indicators) like results_overall,
  # or just 1 column (single indicator) of a table (e.g., subset of df where drop=F)
  # For a vector we might want to round each element differently and maybe only some are even roundable.
  # For a table, each column is treated as an indicator where it is roundable and rounded just 1 way for all rows of the column.

  roundable_by_element_of_vector <- function(vect) {
    function(x) { !inherits(try(round(x), silent = TRUE), "try-error")}
  }

  if (is.vector(x)) {
    #  names were provided using var parameter

    if (only.if.already.numeric) {
      return(is.numeric(x))
    } else {

      # if (strip.characters.before.coerce) {
      #   warning('strip.characters.before.coerce=T is not implemented yet')
      #   ### try to remove spaces, percent signs, etc. like latlon_as.numeric() does?
      #   # x <- latlon_as.numeric(x) # not tested at all ***
      # }

      # note as.numeric() will convert integer vector to numeric vector, which are
      # not identical() but are all.equal()

      # if it is just NA, treat it as potentially a number.
      # and then if as.numeric() results in an NA when it was NOT initially NA, it is truly not a potential number
      x[is.na(x)] <- 0
      answer <-  suppressWarnings(x == as.numeric(x))
      answer[is.na(answer)] <- FALSE # will not report NA as roundable, possibly numeric-
      # it is NA when failed to do as.numeric, and also if it started as NA which arguably can be made numeric

      return(answer)
    }

  } else {
    # table not vector    *** could optimize / vectorize for speed here:

    # if (data.table::is.data.table(x)) {data.table::setDF(x); wasdt <- TRUE} else {wasdt <- FALSE}

    answer <- vector(length = NCOL(x))
    x[is.na(x)] <- 0 # this works cell by cell in a table, faster than in loop
    for (i in 1:NCOL(x)) {

      answer[i] <-  all(suppressWarnings(x[ , i] == as.numeric(x[ , i])))
      answer[i][is.na(answer[i])] <- FALSE  # will not report a vector of NAs as roundable, possibly numeric
    }

    # if (wasdt) {data.table::setDT(x)}
    return(answer)
  }

}